1 //===- AsmParser.cpp - Parser for Assembly Files --------------------------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 // 9 // This class implements a parser for assembly files similar to gas syntax. 10 // 11 //===----------------------------------------------------------------------===// 12 13 #include "llvm/ADT/APFloat.h" 14 #include "llvm/ADT/APInt.h" 15 #include "llvm/ADT/ArrayRef.h" 16 #include "llvm/ADT/None.h" 17 #include "llvm/ADT/STLExtras.h" 18 #include "llvm/ADT/SmallSet.h" 19 #include "llvm/ADT/SmallString.h" 20 #include "llvm/ADT/SmallVector.h" 21 #include "llvm/ADT/StringExtras.h" 22 #include "llvm/ADT/StringMap.h" 23 #include "llvm/ADT/StringRef.h" 24 #include "llvm/ADT/Twine.h" 25 #include "llvm/BinaryFormat/Dwarf.h" 26 #include "llvm/DebugInfo/CodeView/SymbolRecord.h" 27 #include "llvm/MC/MCAsmInfo.h" 28 #include "llvm/MC/MCCodeView.h" 29 #include "llvm/MC/MCContext.h" 30 #include "llvm/MC/MCDirectives.h" 31 #include "llvm/MC/MCDwarf.h" 32 #include "llvm/MC/MCExpr.h" 33 #include "llvm/MC/MCInstPrinter.h" 34 #include "llvm/MC/MCInstrDesc.h" 35 #include "llvm/MC/MCInstrInfo.h" 36 #include "llvm/MC/MCObjectFileInfo.h" 37 #include "llvm/MC/MCParser/AsmCond.h" 38 #include "llvm/MC/MCParser/AsmLexer.h" 39 #include "llvm/MC/MCParser/MCAsmLexer.h" 40 #include "llvm/MC/MCParser/MCAsmParser.h" 41 #include "llvm/MC/MCParser/MCAsmParserExtension.h" 42 #include "llvm/MC/MCParser/MCAsmParserUtils.h" 43 #include "llvm/MC/MCParser/MCParsedAsmOperand.h" 44 #include "llvm/MC/MCParser/MCTargetAsmParser.h" 45 #include "llvm/MC/MCRegisterInfo.h" 46 #include "llvm/MC/MCSection.h" 47 #include "llvm/MC/MCStreamer.h" 48 #include "llvm/MC/MCSymbol.h" 49 #include "llvm/MC/MCTargetOptions.h" 50 #include "llvm/MC/MCValue.h" 51 #include "llvm/Support/Casting.h" 52 #include "llvm/Support/CommandLine.h" 53 #include "llvm/Support/ErrorHandling.h" 54 #include "llvm/Support/MD5.h" 55 #include "llvm/Support/MathExtras.h" 56 #include "llvm/Support/MemoryBuffer.h" 57 #include "llvm/Support/SMLoc.h" 58 #include "llvm/Support/SourceMgr.h" 59 #include "llvm/Support/raw_ostream.h" 60 #include <algorithm> 61 #include <cassert> 62 #include <cctype> 63 #include <climits> 64 #include <cstddef> 65 #include <cstdint> 66 #include <deque> 67 #include <memory> 68 #include <sstream> 69 #include <string> 70 #include <tuple> 71 #include <utility> 72 #include <vector> 73 74 using namespace llvm; 75 76 MCAsmParserSemaCallback::~MCAsmParserSemaCallback() = default; 77 78 extern cl::opt<unsigned> AsmMacroMaxNestingDepth; 79 80 namespace { 81 82 /// Helper types for tracking macro definitions. 83 typedef std::vector<AsmToken> MCAsmMacroArgument; 84 typedef std::vector<MCAsmMacroArgument> MCAsmMacroArguments; 85 86 /// Helper class for storing information about an active macro 87 /// instantiation. 88 struct MacroInstantiation { 89 /// The location of the instantiation. 90 SMLoc InstantiationLoc; 91 92 /// The buffer where parsing should resume upon instantiation completion. 93 unsigned ExitBuffer; 94 95 /// The location where parsing should resume upon instantiation completion. 96 SMLoc ExitLoc; 97 98 /// The depth of TheCondStack at the start of the instantiation. 99 size_t CondStackDepth; 100 }; 101 102 struct ParseStatementInfo { 103 /// The parsed operands from the last parsed statement. 104 SmallVector<std::unique_ptr<MCParsedAsmOperand>, 8> ParsedOperands; 105 106 /// The opcode from the last parsed instruction. 107 unsigned Opcode = ~0U; 108 109 /// Was there an error parsing the inline assembly? 110 bool ParseError = false; 111 112 SmallVectorImpl<AsmRewrite> *AsmRewrites = nullptr; 113 114 ParseStatementInfo() = delete; 115 ParseStatementInfo(SmallVectorImpl<AsmRewrite> *rewrites) 116 : AsmRewrites(rewrites) {} 117 }; 118 119 /// The concrete assembly parser instance. 120 class AsmParser : public MCAsmParser { 121 private: 122 AsmLexer Lexer; 123 MCContext &Ctx; 124 MCStreamer &Out; 125 const MCAsmInfo &MAI; 126 SourceMgr &SrcMgr; 127 SourceMgr::DiagHandlerTy SavedDiagHandler; 128 void *SavedDiagContext; 129 std::unique_ptr<MCAsmParserExtension> PlatformParser; 130 SMLoc StartTokLoc; 131 132 /// This is the current buffer index we're lexing from as managed by the 133 /// SourceMgr object. 134 unsigned CurBuffer; 135 136 AsmCond TheCondState; 137 std::vector<AsmCond> TheCondStack; 138 139 /// maps directive names to handler methods in parser 140 /// extensions. Extensions register themselves in this map by calling 141 /// addDirectiveHandler. 142 StringMap<ExtensionDirectiveHandler> ExtensionDirectiveMap; 143 144 /// Stack of active macro instantiations. 145 std::vector<MacroInstantiation*> ActiveMacros; 146 147 /// List of bodies of anonymous macros. 148 std::deque<MCAsmMacro> MacroLikeBodies; 149 150 /// Boolean tracking whether macro substitution is enabled. 151 unsigned MacrosEnabledFlag : 1; 152 153 /// Keeps track of how many .macro's have been instantiated. 154 unsigned NumOfMacroInstantiations; 155 156 /// The values from the last parsed cpp hash file line comment if any. 157 struct CppHashInfoTy { 158 StringRef Filename; 159 int64_t LineNumber; 160 SMLoc Loc; 161 unsigned Buf; 162 CppHashInfoTy() : Filename(), LineNumber(0), Loc(), Buf(0) {} 163 }; 164 CppHashInfoTy CppHashInfo; 165 166 /// The filename from the first cpp hash file line comment, if any. 167 StringRef FirstCppHashFilename; 168 169 /// List of forward directional labels for diagnosis at the end. 170 SmallVector<std::tuple<SMLoc, CppHashInfoTy, MCSymbol *>, 4> DirLabels; 171 172 SmallSet<StringRef, 2> LTODiscardSymbols; 173 174 /// AssemblerDialect. ~OU means unset value and use value provided by MAI. 175 unsigned AssemblerDialect = ~0U; 176 177 /// is Darwin compatibility enabled? 178 bool IsDarwin = false; 179 180 /// Are we parsing ms-style inline assembly? 181 bool ParsingMSInlineAsm = false; 182 183 /// Did we already inform the user about inconsistent MD5 usage? 184 bool ReportedInconsistentMD5 = false; 185 186 // Is alt macro mode enabled. 187 bool AltMacroMode = false; 188 189 protected: 190 virtual bool parseStatement(ParseStatementInfo &Info, 191 MCAsmParserSemaCallback *SI); 192 193 /// This routine uses the target specific ParseInstruction function to 194 /// parse an instruction into Operands, and then call the target specific 195 /// MatchAndEmit function to match and emit the instruction. 196 bool parseAndMatchAndEmitTargetInstruction(ParseStatementInfo &Info, 197 StringRef IDVal, AsmToken ID, 198 SMLoc IDLoc); 199 200 public: 201 AsmParser(SourceMgr &SM, MCContext &Ctx, MCStreamer &Out, 202 const MCAsmInfo &MAI, unsigned CB); 203 AsmParser(const AsmParser &) = delete; 204 AsmParser &operator=(const AsmParser &) = delete; 205 ~AsmParser() override; 206 207 bool Run(bool NoInitialTextSection, bool NoFinalize = false) override; 208 209 void addDirectiveHandler(StringRef Directive, 210 ExtensionDirectiveHandler Handler) override { 211 ExtensionDirectiveMap[Directive] = Handler; 212 } 213 214 void addAliasForDirective(StringRef Directive, StringRef Alias) override { 215 DirectiveKindMap[Directive.lower()] = DirectiveKindMap[Alias.lower()]; 216 } 217 218 /// @name MCAsmParser Interface 219 /// { 220 221 SourceMgr &getSourceManager() override { return SrcMgr; } 222 MCAsmLexer &getLexer() override { return Lexer; } 223 MCContext &getContext() override { return Ctx; } 224 MCStreamer &getStreamer() override { return Out; } 225 226 CodeViewContext &getCVContext() { return Ctx.getCVContext(); } 227 228 unsigned getAssemblerDialect() override { 229 if (AssemblerDialect == ~0U) 230 return MAI.getAssemblerDialect(); 231 else 232 return AssemblerDialect; 233 } 234 void setAssemblerDialect(unsigned i) override { 235 AssemblerDialect = i; 236 } 237 238 void Note(SMLoc L, const Twine &Msg, SMRange Range = None) override; 239 bool Warning(SMLoc L, const Twine &Msg, SMRange Range = None) override; 240 bool printError(SMLoc L, const Twine &Msg, SMRange Range = None) override; 241 242 const AsmToken &Lex() override; 243 244 void setParsingMSInlineAsm(bool V) override { 245 ParsingMSInlineAsm = V; 246 // When parsing MS inline asm, we must lex 0b1101 and 0ABCH as binary and 247 // hex integer literals. 248 Lexer.setLexMasmIntegers(V); 249 } 250 bool isParsingMSInlineAsm() override { return ParsingMSInlineAsm; } 251 252 bool discardLTOSymbol(StringRef Name) const override { 253 return LTODiscardSymbols.contains(Name); 254 } 255 256 bool parseMSInlineAsm(void *AsmLoc, std::string &AsmString, 257 unsigned &NumOutputs, unsigned &NumInputs, 258 SmallVectorImpl<std::pair<void *,bool>> &OpDecls, 259 SmallVectorImpl<std::string> &Constraints, 260 SmallVectorImpl<std::string> &Clobbers, 261 const MCInstrInfo *MII, const MCInstPrinter *IP, 262 MCAsmParserSemaCallback &SI) override; 263 264 bool parseExpression(const MCExpr *&Res); 265 bool parseExpression(const MCExpr *&Res, SMLoc &EndLoc) override; 266 bool parsePrimaryExpr(const MCExpr *&Res, SMLoc &EndLoc, 267 AsmTypeInfo *TypeInfo) override; 268 bool parseParenExpression(const MCExpr *&Res, SMLoc &EndLoc) override; 269 bool parseParenExprOfDepth(unsigned ParenDepth, const MCExpr *&Res, 270 SMLoc &EndLoc) override; 271 bool parseAbsoluteExpression(int64_t &Res) override; 272 273 /// Parse a floating point expression using the float \p Semantics 274 /// and set \p Res to the value. 275 bool parseRealValue(const fltSemantics &Semantics, APInt &Res); 276 277 /// Parse an identifier or string (as a quoted identifier) 278 /// and set \p Res to the identifier contents. 279 bool parseIdentifier(StringRef &Res) override; 280 void eatToEndOfStatement() override; 281 282 bool checkForValidSection() override; 283 284 /// } 285 286 private: 287 bool parseCurlyBlockScope(SmallVectorImpl<AsmRewrite>& AsmStrRewrites); 288 bool parseCppHashLineFilenameComment(SMLoc L, bool SaveLocInfo = true); 289 290 void checkForBadMacro(SMLoc DirectiveLoc, StringRef Name, StringRef Body, 291 ArrayRef<MCAsmMacroParameter> Parameters); 292 bool expandMacro(raw_svector_ostream &OS, StringRef Body, 293 ArrayRef<MCAsmMacroParameter> Parameters, 294 ArrayRef<MCAsmMacroArgument> A, bool EnableAtPseudoVariable, 295 SMLoc L); 296 297 /// Are macros enabled in the parser? 298 bool areMacrosEnabled() {return MacrosEnabledFlag;} 299 300 /// Control a flag in the parser that enables or disables macros. 301 void setMacrosEnabled(bool Flag) {MacrosEnabledFlag = Flag;} 302 303 /// Are we inside a macro instantiation? 304 bool isInsideMacroInstantiation() {return !ActiveMacros.empty();} 305 306 /// Handle entry to macro instantiation. 307 /// 308 /// \param M The macro. 309 /// \param NameLoc Instantiation location. 310 bool handleMacroEntry(const MCAsmMacro *M, SMLoc NameLoc); 311 312 /// Handle exit from macro instantiation. 313 void handleMacroExit(); 314 315 /// Extract AsmTokens for a macro argument. 316 bool parseMacroArgument(MCAsmMacroArgument &MA, bool Vararg); 317 318 /// Parse all macro arguments for a given macro. 319 bool parseMacroArguments(const MCAsmMacro *M, MCAsmMacroArguments &A); 320 321 void printMacroInstantiations(); 322 void printMessage(SMLoc Loc, SourceMgr::DiagKind Kind, const Twine &Msg, 323 SMRange Range = None) const { 324 ArrayRef<SMRange> Ranges(Range); 325 SrcMgr.PrintMessage(Loc, Kind, Msg, Ranges); 326 } 327 static void DiagHandler(const SMDiagnostic &Diag, void *Context); 328 329 /// Should we emit DWARF describing this assembler source? (Returns false if 330 /// the source has .file directives, which means we don't want to generate 331 /// info describing the assembler source itself.) 332 bool enabledGenDwarfForAssembly(); 333 334 /// Enter the specified file. This returns true on failure. 335 bool enterIncludeFile(const std::string &Filename); 336 337 /// Process the specified file for the .incbin directive. 338 /// This returns true on failure. 339 bool processIncbinFile(const std::string &Filename, int64_t Skip = 0, 340 const MCExpr *Count = nullptr, SMLoc Loc = SMLoc()); 341 342 /// Reset the current lexer position to that given by \p Loc. The 343 /// current token is not set; clients should ensure Lex() is called 344 /// subsequently. 345 /// 346 /// \param InBuffer If not 0, should be the known buffer id that contains the 347 /// location. 348 void jumpToLoc(SMLoc Loc, unsigned InBuffer = 0); 349 350 /// Parse up to the end of statement and a return the contents from the 351 /// current token until the end of the statement; the current token on exit 352 /// will be either the EndOfStatement or EOF. 353 StringRef parseStringToEndOfStatement() override; 354 355 /// Parse until the end of a statement or a comma is encountered, 356 /// return the contents from the current token up to the end or comma. 357 StringRef parseStringToComma(); 358 359 bool parseAssignment(StringRef Name, bool allow_redef, 360 bool NoDeadStrip = false); 361 362 unsigned getBinOpPrecedence(AsmToken::TokenKind K, 363 MCBinaryExpr::Opcode &Kind); 364 365 bool parseBinOpRHS(unsigned Precedence, const MCExpr *&Res, SMLoc &EndLoc); 366 bool parseParenExpr(const MCExpr *&Res, SMLoc &EndLoc); 367 bool parseBracketExpr(const MCExpr *&Res, SMLoc &EndLoc); 368 369 bool parseRegisterOrRegisterNumber(int64_t &Register, SMLoc DirectiveLoc); 370 371 bool parseCVFunctionId(int64_t &FunctionId, StringRef DirectiveName); 372 bool parseCVFileId(int64_t &FileId, StringRef DirectiveName); 373 374 // Generic (target and platform independent) directive parsing. 375 enum DirectiveKind { 376 DK_NO_DIRECTIVE, // Placeholder 377 DK_SET, 378 DK_EQU, 379 DK_EQUIV, 380 DK_ASCII, 381 DK_ASCIZ, 382 DK_STRING, 383 DK_BYTE, 384 DK_SHORT, 385 DK_RELOC, 386 DK_VALUE, 387 DK_2BYTE, 388 DK_LONG, 389 DK_INT, 390 DK_4BYTE, 391 DK_QUAD, 392 DK_8BYTE, 393 DK_OCTA, 394 DK_DC, 395 DK_DC_A, 396 DK_DC_B, 397 DK_DC_D, 398 DK_DC_L, 399 DK_DC_S, 400 DK_DC_W, 401 DK_DC_X, 402 DK_DCB, 403 DK_DCB_B, 404 DK_DCB_D, 405 DK_DCB_L, 406 DK_DCB_S, 407 DK_DCB_W, 408 DK_DCB_X, 409 DK_DS, 410 DK_DS_B, 411 DK_DS_D, 412 DK_DS_L, 413 DK_DS_P, 414 DK_DS_S, 415 DK_DS_W, 416 DK_DS_X, 417 DK_SINGLE, 418 DK_FLOAT, 419 DK_DOUBLE, 420 DK_ALIGN, 421 DK_ALIGN32, 422 DK_BALIGN, 423 DK_BALIGNW, 424 DK_BALIGNL, 425 DK_P2ALIGN, 426 DK_P2ALIGNW, 427 DK_P2ALIGNL, 428 DK_ORG, 429 DK_FILL, 430 DK_ENDR, 431 DK_BUNDLE_ALIGN_MODE, 432 DK_BUNDLE_LOCK, 433 DK_BUNDLE_UNLOCK, 434 DK_ZERO, 435 DK_EXTERN, 436 DK_GLOBL, 437 DK_GLOBAL, 438 DK_LAZY_REFERENCE, 439 DK_NO_DEAD_STRIP, 440 DK_SYMBOL_RESOLVER, 441 DK_PRIVATE_EXTERN, 442 DK_REFERENCE, 443 DK_WEAK_DEFINITION, 444 DK_WEAK_REFERENCE, 445 DK_WEAK_DEF_CAN_BE_HIDDEN, 446 DK_COLD, 447 DK_COMM, 448 DK_COMMON, 449 DK_LCOMM, 450 DK_ABORT, 451 DK_INCLUDE, 452 DK_INCBIN, 453 DK_CODE16, 454 DK_CODE16GCC, 455 DK_REPT, 456 DK_IRP, 457 DK_IRPC, 458 DK_IF, 459 DK_IFEQ, 460 DK_IFGE, 461 DK_IFGT, 462 DK_IFLE, 463 DK_IFLT, 464 DK_IFNE, 465 DK_IFB, 466 DK_IFNB, 467 DK_IFC, 468 DK_IFEQS, 469 DK_IFNC, 470 DK_IFNES, 471 DK_IFDEF, 472 DK_IFNDEF, 473 DK_IFNOTDEF, 474 DK_ELSEIF, 475 DK_ELSE, 476 DK_ENDIF, 477 DK_SPACE, 478 DK_SKIP, 479 DK_FILE, 480 DK_LINE, 481 DK_LOC, 482 DK_STABS, 483 DK_CV_FILE, 484 DK_CV_FUNC_ID, 485 DK_CV_INLINE_SITE_ID, 486 DK_CV_LOC, 487 DK_CV_LINETABLE, 488 DK_CV_INLINE_LINETABLE, 489 DK_CV_DEF_RANGE, 490 DK_CV_STRINGTABLE, 491 DK_CV_STRING, 492 DK_CV_FILECHECKSUMS, 493 DK_CV_FILECHECKSUM_OFFSET, 494 DK_CV_FPO_DATA, 495 DK_CFI_SECTIONS, 496 DK_CFI_STARTPROC, 497 DK_CFI_ENDPROC, 498 DK_CFI_DEF_CFA, 499 DK_CFI_DEF_CFA_OFFSET, 500 DK_CFI_ADJUST_CFA_OFFSET, 501 DK_CFI_DEF_CFA_REGISTER, 502 DK_CFI_OFFSET, 503 DK_CFI_REL_OFFSET, 504 DK_CFI_PERSONALITY, 505 DK_CFI_LSDA, 506 DK_CFI_REMEMBER_STATE, 507 DK_CFI_RESTORE_STATE, 508 DK_CFI_SAME_VALUE, 509 DK_CFI_RESTORE, 510 DK_CFI_ESCAPE, 511 DK_CFI_RETURN_COLUMN, 512 DK_CFI_SIGNAL_FRAME, 513 DK_CFI_UNDEFINED, 514 DK_CFI_REGISTER, 515 DK_CFI_WINDOW_SAVE, 516 DK_CFI_B_KEY_FRAME, 517 DK_MACROS_ON, 518 DK_MACROS_OFF, 519 DK_ALTMACRO, 520 DK_NOALTMACRO, 521 DK_MACRO, 522 DK_EXITM, 523 DK_ENDM, 524 DK_ENDMACRO, 525 DK_PURGEM, 526 DK_SLEB128, 527 DK_ULEB128, 528 DK_ERR, 529 DK_ERROR, 530 DK_WARNING, 531 DK_PRINT, 532 DK_ADDRSIG, 533 DK_ADDRSIG_SYM, 534 DK_PSEUDO_PROBE, 535 DK_LTO_DISCARD, 536 DK_END 537 }; 538 539 /// Maps directive name --> DirectiveKind enum, for 540 /// directives parsed by this class. 541 StringMap<DirectiveKind> DirectiveKindMap; 542 543 // Codeview def_range type parsing. 544 enum CVDefRangeType { 545 CVDR_DEFRANGE = 0, // Placeholder 546 CVDR_DEFRANGE_REGISTER, 547 CVDR_DEFRANGE_FRAMEPOINTER_REL, 548 CVDR_DEFRANGE_SUBFIELD_REGISTER, 549 CVDR_DEFRANGE_REGISTER_REL 550 }; 551 552 /// Maps Codeview def_range types --> CVDefRangeType enum, for 553 /// Codeview def_range types parsed by this class. 554 StringMap<CVDefRangeType> CVDefRangeTypeMap; 555 556 // ".ascii", ".asciz", ".string" 557 bool parseDirectiveAscii(StringRef IDVal, bool ZeroTerminated); 558 bool parseDirectiveReloc(SMLoc DirectiveLoc); // ".reloc" 559 bool parseDirectiveValue(StringRef IDVal, 560 unsigned Size); // ".byte", ".long", ... 561 bool parseDirectiveOctaValue(StringRef IDVal); // ".octa", ... 562 bool parseDirectiveRealValue(StringRef IDVal, 563 const fltSemantics &); // ".single", ... 564 bool parseDirectiveFill(); // ".fill" 565 bool parseDirectiveZero(); // ".zero" 566 // ".set", ".equ", ".equiv" 567 bool parseDirectiveSet(StringRef IDVal, bool allow_redef); 568 bool parseDirectiveOrg(); // ".org" 569 // ".align{,32}", ".p2align{,w,l}" 570 bool parseDirectiveAlign(bool IsPow2, unsigned ValueSize); 571 572 // ".file", ".line", ".loc", ".stabs" 573 bool parseDirectiveFile(SMLoc DirectiveLoc); 574 bool parseDirectiveLine(); 575 bool parseDirectiveLoc(); 576 bool parseDirectiveStabs(); 577 578 // ".cv_file", ".cv_func_id", ".cv_inline_site_id", ".cv_loc", ".cv_linetable", 579 // ".cv_inline_linetable", ".cv_def_range", ".cv_string" 580 bool parseDirectiveCVFile(); 581 bool parseDirectiveCVFuncId(); 582 bool parseDirectiveCVInlineSiteId(); 583 bool parseDirectiveCVLoc(); 584 bool parseDirectiveCVLinetable(); 585 bool parseDirectiveCVInlineLinetable(); 586 bool parseDirectiveCVDefRange(); 587 bool parseDirectiveCVString(); 588 bool parseDirectiveCVStringTable(); 589 bool parseDirectiveCVFileChecksums(); 590 bool parseDirectiveCVFileChecksumOffset(); 591 bool parseDirectiveCVFPOData(); 592 593 // .cfi directives 594 bool parseDirectiveCFIRegister(SMLoc DirectiveLoc); 595 bool parseDirectiveCFIWindowSave(); 596 bool parseDirectiveCFISections(); 597 bool parseDirectiveCFIStartProc(); 598 bool parseDirectiveCFIEndProc(); 599 bool parseDirectiveCFIDefCfaOffset(); 600 bool parseDirectiveCFIDefCfa(SMLoc DirectiveLoc); 601 bool parseDirectiveCFIAdjustCfaOffset(); 602 bool parseDirectiveCFIDefCfaRegister(SMLoc DirectiveLoc); 603 bool parseDirectiveCFIOffset(SMLoc DirectiveLoc); 604 bool parseDirectiveCFIRelOffset(SMLoc DirectiveLoc); 605 bool parseDirectiveCFIPersonalityOrLsda(bool IsPersonality); 606 bool parseDirectiveCFIRememberState(); 607 bool parseDirectiveCFIRestoreState(); 608 bool parseDirectiveCFISameValue(SMLoc DirectiveLoc); 609 bool parseDirectiveCFIRestore(SMLoc DirectiveLoc); 610 bool parseDirectiveCFIEscape(); 611 bool parseDirectiveCFIReturnColumn(SMLoc DirectiveLoc); 612 bool parseDirectiveCFISignalFrame(); 613 bool parseDirectiveCFIUndefined(SMLoc DirectiveLoc); 614 615 // macro directives 616 bool parseDirectivePurgeMacro(SMLoc DirectiveLoc); 617 bool parseDirectiveExitMacro(StringRef Directive); 618 bool parseDirectiveEndMacro(StringRef Directive); 619 bool parseDirectiveMacro(SMLoc DirectiveLoc); 620 bool parseDirectiveMacrosOnOff(StringRef Directive); 621 // alternate macro mode directives 622 bool parseDirectiveAltmacro(StringRef Directive); 623 // ".bundle_align_mode" 624 bool parseDirectiveBundleAlignMode(); 625 // ".bundle_lock" 626 bool parseDirectiveBundleLock(); 627 // ".bundle_unlock" 628 bool parseDirectiveBundleUnlock(); 629 630 // ".space", ".skip" 631 bool parseDirectiveSpace(StringRef IDVal); 632 633 // ".dcb" 634 bool parseDirectiveDCB(StringRef IDVal, unsigned Size); 635 bool parseDirectiveRealDCB(StringRef IDVal, const fltSemantics &); 636 // ".ds" 637 bool parseDirectiveDS(StringRef IDVal, unsigned Size); 638 639 // .sleb128 (Signed=true) and .uleb128 (Signed=false) 640 bool parseDirectiveLEB128(bool Signed); 641 642 /// Parse a directive like ".globl" which 643 /// accepts a single symbol (which should be a label or an external). 644 bool parseDirectiveSymbolAttribute(MCSymbolAttr Attr); 645 646 bool parseDirectiveComm(bool IsLocal); // ".comm" and ".lcomm" 647 648 bool parseDirectiveAbort(); // ".abort" 649 bool parseDirectiveInclude(); // ".include" 650 bool parseDirectiveIncbin(); // ".incbin" 651 652 // ".if", ".ifeq", ".ifge", ".ifgt" , ".ifle", ".iflt" or ".ifne" 653 bool parseDirectiveIf(SMLoc DirectiveLoc, DirectiveKind DirKind); 654 // ".ifb" or ".ifnb", depending on ExpectBlank. 655 bool parseDirectiveIfb(SMLoc DirectiveLoc, bool ExpectBlank); 656 // ".ifc" or ".ifnc", depending on ExpectEqual. 657 bool parseDirectiveIfc(SMLoc DirectiveLoc, bool ExpectEqual); 658 // ".ifeqs" or ".ifnes", depending on ExpectEqual. 659 bool parseDirectiveIfeqs(SMLoc DirectiveLoc, bool ExpectEqual); 660 // ".ifdef" or ".ifndef", depending on expect_defined 661 bool parseDirectiveIfdef(SMLoc DirectiveLoc, bool expect_defined); 662 bool parseDirectiveElseIf(SMLoc DirectiveLoc); // ".elseif" 663 bool parseDirectiveElse(SMLoc DirectiveLoc); // ".else" 664 bool parseDirectiveEndIf(SMLoc DirectiveLoc); // .endif 665 bool parseEscapedString(std::string &Data) override; 666 bool parseAngleBracketString(std::string &Data) override; 667 668 const MCExpr *applyModifierToExpr(const MCExpr *E, 669 MCSymbolRefExpr::VariantKind Variant); 670 671 // Macro-like directives 672 MCAsmMacro *parseMacroLikeBody(SMLoc DirectiveLoc); 673 void instantiateMacroLikeBody(MCAsmMacro *M, SMLoc DirectiveLoc, 674 raw_svector_ostream &OS); 675 bool parseDirectiveRept(SMLoc DirectiveLoc, StringRef Directive); 676 bool parseDirectiveIrp(SMLoc DirectiveLoc); // ".irp" 677 bool parseDirectiveIrpc(SMLoc DirectiveLoc); // ".irpc" 678 bool parseDirectiveEndr(SMLoc DirectiveLoc); // ".endr" 679 680 // "_emit" or "__emit" 681 bool parseDirectiveMSEmit(SMLoc DirectiveLoc, ParseStatementInfo &Info, 682 size_t Len); 683 684 // "align" 685 bool parseDirectiveMSAlign(SMLoc DirectiveLoc, ParseStatementInfo &Info); 686 687 // "end" 688 bool parseDirectiveEnd(SMLoc DirectiveLoc); 689 690 // ".err" or ".error" 691 bool parseDirectiveError(SMLoc DirectiveLoc, bool WithMessage); 692 693 // ".warning" 694 bool parseDirectiveWarning(SMLoc DirectiveLoc); 695 696 // .print <double-quotes-string> 697 bool parseDirectivePrint(SMLoc DirectiveLoc); 698 699 // .pseudoprobe 700 bool parseDirectivePseudoProbe(); 701 702 // ".lto_discard" 703 bool parseDirectiveLTODiscard(); 704 705 // Directives to support address-significance tables. 706 bool parseDirectiveAddrsig(); 707 bool parseDirectiveAddrsigSym(); 708 709 void initializeDirectiveKindMap(); 710 void initializeCVDefRangeTypeMap(); 711 }; 712 713 class HLASMAsmParser final : public AsmParser { 714 private: 715 MCAsmLexer &Lexer; 716 MCStreamer &Out; 717 718 void lexLeadingSpaces() { 719 while (Lexer.is(AsmToken::Space)) 720 Lexer.Lex(); 721 } 722 723 bool parseAsHLASMLabel(ParseStatementInfo &Info, MCAsmParserSemaCallback *SI); 724 bool parseAsMachineInstruction(ParseStatementInfo &Info, 725 MCAsmParserSemaCallback *SI); 726 727 public: 728 HLASMAsmParser(SourceMgr &SM, MCContext &Ctx, MCStreamer &Out, 729 const MCAsmInfo &MAI, unsigned CB = 0) 730 : AsmParser(SM, Ctx, Out, MAI, CB), Lexer(getLexer()), Out(Out) { 731 Lexer.setSkipSpace(false); 732 Lexer.setAllowHashInIdentifier(true); 733 Lexer.setLexHLASMIntegers(true); 734 Lexer.setLexHLASMStrings(true); 735 } 736 737 ~HLASMAsmParser() { Lexer.setSkipSpace(true); } 738 739 bool parseStatement(ParseStatementInfo &Info, 740 MCAsmParserSemaCallback *SI) override; 741 }; 742 743 } // end anonymous namespace 744 745 namespace llvm { 746 747 extern MCAsmParserExtension *createDarwinAsmParser(); 748 extern MCAsmParserExtension *createELFAsmParser(); 749 extern MCAsmParserExtension *createCOFFAsmParser(); 750 extern MCAsmParserExtension *createWasmAsmParser(); 751 752 } // end namespace llvm 753 754 enum { DEFAULT_ADDRSPACE = 0 }; 755 756 AsmParser::AsmParser(SourceMgr &SM, MCContext &Ctx, MCStreamer &Out, 757 const MCAsmInfo &MAI, unsigned CB = 0) 758 : Lexer(MAI), Ctx(Ctx), Out(Out), MAI(MAI), SrcMgr(SM), 759 CurBuffer(CB ? CB : SM.getMainFileID()), MacrosEnabledFlag(true) { 760 HadError = false; 761 // Save the old handler. 762 SavedDiagHandler = SrcMgr.getDiagHandler(); 763 SavedDiagContext = SrcMgr.getDiagContext(); 764 // Set our own handler which calls the saved handler. 765 SrcMgr.setDiagHandler(DiagHandler, this); 766 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer()); 767 // Make MCStreamer aware of the StartTokLoc for locations in diagnostics. 768 Out.setStartTokLocPtr(&StartTokLoc); 769 770 // Initialize the platform / file format parser. 771 switch (Ctx.getObjectFileType()) { 772 case MCContext::IsCOFF: 773 PlatformParser.reset(createCOFFAsmParser()); 774 break; 775 case MCContext::IsMachO: 776 PlatformParser.reset(createDarwinAsmParser()); 777 IsDarwin = true; 778 break; 779 case MCContext::IsELF: 780 PlatformParser.reset(createELFAsmParser()); 781 break; 782 case MCContext::IsWasm: 783 PlatformParser.reset(createWasmAsmParser()); 784 break; 785 case MCContext::IsXCOFF: 786 report_fatal_error( 787 "Need to implement createXCOFFAsmParser for XCOFF format."); 788 break; 789 } 790 791 PlatformParser->Initialize(*this); 792 initializeDirectiveKindMap(); 793 initializeCVDefRangeTypeMap(); 794 795 NumOfMacroInstantiations = 0; 796 } 797 798 AsmParser::~AsmParser() { 799 assert((HadError || ActiveMacros.empty()) && 800 "Unexpected active macro instantiation!"); 801 802 // Remove MCStreamer's reference to the parser SMLoc. 803 Out.setStartTokLocPtr(nullptr); 804 // Restore the saved diagnostics handler and context for use during 805 // finalization. 806 SrcMgr.setDiagHandler(SavedDiagHandler, SavedDiagContext); 807 } 808 809 void AsmParser::printMacroInstantiations() { 810 // Print the active macro instantiation stack. 811 for (std::vector<MacroInstantiation *>::const_reverse_iterator 812 it = ActiveMacros.rbegin(), 813 ie = ActiveMacros.rend(); 814 it != ie; ++it) 815 printMessage((*it)->InstantiationLoc, SourceMgr::DK_Note, 816 "while in macro instantiation"); 817 } 818 819 void AsmParser::Note(SMLoc L, const Twine &Msg, SMRange Range) { 820 printPendingErrors(); 821 printMessage(L, SourceMgr::DK_Note, Msg, Range); 822 printMacroInstantiations(); 823 } 824 825 bool AsmParser::Warning(SMLoc L, const Twine &Msg, SMRange Range) { 826 if(getTargetParser().getTargetOptions().MCNoWarn) 827 return false; 828 if (getTargetParser().getTargetOptions().MCFatalWarnings) 829 return Error(L, Msg, Range); 830 printMessage(L, SourceMgr::DK_Warning, Msg, Range); 831 printMacroInstantiations(); 832 return false; 833 } 834 835 bool AsmParser::printError(SMLoc L, const Twine &Msg, SMRange Range) { 836 HadError = true; 837 printMessage(L, SourceMgr::DK_Error, Msg, Range); 838 printMacroInstantiations(); 839 return true; 840 } 841 842 bool AsmParser::enterIncludeFile(const std::string &Filename) { 843 std::string IncludedFile; 844 unsigned NewBuf = 845 SrcMgr.AddIncludeFile(Filename, Lexer.getLoc(), IncludedFile); 846 if (!NewBuf) 847 return true; 848 849 CurBuffer = NewBuf; 850 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer()); 851 return false; 852 } 853 854 /// Process the specified .incbin file by searching for it in the include paths 855 /// then just emitting the byte contents of the file to the streamer. This 856 /// returns true on failure. 857 bool AsmParser::processIncbinFile(const std::string &Filename, int64_t Skip, 858 const MCExpr *Count, SMLoc Loc) { 859 std::string IncludedFile; 860 unsigned NewBuf = 861 SrcMgr.AddIncludeFile(Filename, Lexer.getLoc(), IncludedFile); 862 if (!NewBuf) 863 return true; 864 865 // Pick up the bytes from the file and emit them. 866 StringRef Bytes = SrcMgr.getMemoryBuffer(NewBuf)->getBuffer(); 867 Bytes = Bytes.drop_front(Skip); 868 if (Count) { 869 int64_t Res; 870 if (!Count->evaluateAsAbsolute(Res, getStreamer().getAssemblerPtr())) 871 return Error(Loc, "expected absolute expression"); 872 if (Res < 0) 873 return Warning(Loc, "negative count has no effect"); 874 Bytes = Bytes.take_front(Res); 875 } 876 getStreamer().emitBytes(Bytes); 877 return false; 878 } 879 880 void AsmParser::jumpToLoc(SMLoc Loc, unsigned InBuffer) { 881 CurBuffer = InBuffer ? InBuffer : SrcMgr.FindBufferContainingLoc(Loc); 882 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer(), 883 Loc.getPointer()); 884 } 885 886 const AsmToken &AsmParser::Lex() { 887 if (Lexer.getTok().is(AsmToken::Error)) 888 Error(Lexer.getErrLoc(), Lexer.getErr()); 889 890 // if it's a end of statement with a comment in it 891 if (getTok().is(AsmToken::EndOfStatement)) { 892 // if this is a line comment output it. 893 if (!getTok().getString().empty() && getTok().getString().front() != '\n' && 894 getTok().getString().front() != '\r' && MAI.preserveAsmComments()) 895 Out.addExplicitComment(Twine(getTok().getString())); 896 } 897 898 const AsmToken *tok = &Lexer.Lex(); 899 900 // Parse comments here to be deferred until end of next statement. 901 while (tok->is(AsmToken::Comment)) { 902 if (MAI.preserveAsmComments()) 903 Out.addExplicitComment(Twine(tok->getString())); 904 tok = &Lexer.Lex(); 905 } 906 907 if (tok->is(AsmToken::Eof)) { 908 // If this is the end of an included file, pop the parent file off the 909 // include stack. 910 SMLoc ParentIncludeLoc = SrcMgr.getParentIncludeLoc(CurBuffer); 911 if (ParentIncludeLoc != SMLoc()) { 912 jumpToLoc(ParentIncludeLoc); 913 return Lex(); 914 } 915 } 916 917 return *tok; 918 } 919 920 bool AsmParser::enabledGenDwarfForAssembly() { 921 // Check whether the user specified -g. 922 if (!getContext().getGenDwarfForAssembly()) 923 return false; 924 // If we haven't encountered any .file directives (which would imply that 925 // the assembler source was produced with debug info already) then emit one 926 // describing the assembler source file itself. 927 if (getContext().getGenDwarfFileNumber() == 0) { 928 // Use the first #line directive for this, if any. It's preprocessed, so 929 // there is no checksum, and of course no source directive. 930 if (!FirstCppHashFilename.empty()) 931 getContext().setMCLineTableRootFile(/*CUID=*/0, 932 getContext().getCompilationDir(), 933 FirstCppHashFilename, 934 /*Cksum=*/None, /*Source=*/None); 935 const MCDwarfFile &RootFile = 936 getContext().getMCDwarfLineTable(/*CUID=*/0).getRootFile(); 937 getContext().setGenDwarfFileNumber(getStreamer().emitDwarfFileDirective( 938 /*CUID=*/0, getContext().getCompilationDir(), RootFile.Name, 939 RootFile.Checksum, RootFile.Source)); 940 } 941 return true; 942 } 943 944 bool AsmParser::Run(bool NoInitialTextSection, bool NoFinalize) { 945 LTODiscardSymbols.clear(); 946 947 // Create the initial section, if requested. 948 if (!NoInitialTextSection) 949 Out.InitSections(false); 950 951 // Prime the lexer. 952 Lex(); 953 954 HadError = false; 955 AsmCond StartingCondState = TheCondState; 956 SmallVector<AsmRewrite, 4> AsmStrRewrites; 957 958 // If we are generating dwarf for assembly source files save the initial text 959 // section. (Don't use enabledGenDwarfForAssembly() here, as we aren't 960 // emitting any actual debug info yet and haven't had a chance to parse any 961 // embedded .file directives.) 962 if (getContext().getGenDwarfForAssembly()) { 963 MCSection *Sec = getStreamer().getCurrentSectionOnly(); 964 if (!Sec->getBeginSymbol()) { 965 MCSymbol *SectionStartSym = getContext().createTempSymbol(); 966 getStreamer().emitLabel(SectionStartSym); 967 Sec->setBeginSymbol(SectionStartSym); 968 } 969 bool InsertResult = getContext().addGenDwarfSection(Sec); 970 assert(InsertResult && ".text section should not have debug info yet"); 971 (void)InsertResult; 972 } 973 974 getTargetParser().onBeginOfFile(); 975 976 // While we have input, parse each statement. 977 while (Lexer.isNot(AsmToken::Eof)) { 978 ParseStatementInfo Info(&AsmStrRewrites); 979 bool Parsed = parseStatement(Info, nullptr); 980 981 // If we have a Lexer Error we are on an Error Token. Load in Lexer Error 982 // for printing ErrMsg via Lex() only if no (presumably better) parser error 983 // exists. 984 if (Parsed && !hasPendingError() && Lexer.getTok().is(AsmToken::Error)) { 985 Lex(); 986 } 987 988 // parseStatement returned true so may need to emit an error. 989 printPendingErrors(); 990 991 // Skipping to the next line if needed. 992 if (Parsed && !getLexer().isAtStartOfStatement()) 993 eatToEndOfStatement(); 994 } 995 996 getTargetParser().onEndOfFile(); 997 printPendingErrors(); 998 999 // All errors should have been emitted. 1000 assert(!hasPendingError() && "unexpected error from parseStatement"); 1001 1002 getTargetParser().flushPendingInstructions(getStreamer()); 1003 1004 if (TheCondState.TheCond != StartingCondState.TheCond || 1005 TheCondState.Ignore != StartingCondState.Ignore) 1006 printError(getTok().getLoc(), "unmatched .ifs or .elses"); 1007 // Check to see there are no empty DwarfFile slots. 1008 const auto &LineTables = getContext().getMCDwarfLineTables(); 1009 if (!LineTables.empty()) { 1010 unsigned Index = 0; 1011 for (const auto &File : LineTables.begin()->second.getMCDwarfFiles()) { 1012 if (File.Name.empty() && Index != 0) 1013 printError(getTok().getLoc(), "unassigned file number: " + 1014 Twine(Index) + 1015 " for .file directives"); 1016 ++Index; 1017 } 1018 } 1019 1020 // Check to see that all assembler local symbols were actually defined. 1021 // Targets that don't do subsections via symbols may not want this, though, 1022 // so conservatively exclude them. Only do this if we're finalizing, though, 1023 // as otherwise we won't necessarilly have seen everything yet. 1024 if (!NoFinalize) { 1025 if (MAI.hasSubsectionsViaSymbols()) { 1026 for (const auto &TableEntry : getContext().getSymbols()) { 1027 MCSymbol *Sym = TableEntry.getValue(); 1028 // Variable symbols may not be marked as defined, so check those 1029 // explicitly. If we know it's a variable, we have a definition for 1030 // the purposes of this check. 1031 if (Sym->isTemporary() && !Sym->isVariable() && !Sym->isDefined()) 1032 // FIXME: We would really like to refer back to where the symbol was 1033 // first referenced for a source location. We need to add something 1034 // to track that. Currently, we just point to the end of the file. 1035 printError(getTok().getLoc(), "assembler local symbol '" + 1036 Sym->getName() + "' not defined"); 1037 } 1038 } 1039 1040 // Temporary symbols like the ones for directional jumps don't go in the 1041 // symbol table. They also need to be diagnosed in all (final) cases. 1042 for (std::tuple<SMLoc, CppHashInfoTy, MCSymbol *> &LocSym : DirLabels) { 1043 if (std::get<2>(LocSym)->isUndefined()) { 1044 // Reset the state of any "# line file" directives we've seen to the 1045 // context as it was at the diagnostic site. 1046 CppHashInfo = std::get<1>(LocSym); 1047 printError(std::get<0>(LocSym), "directional label undefined"); 1048 } 1049 } 1050 } 1051 1052 // Finalize the output stream if there are no errors and if the client wants 1053 // us to. 1054 if (!HadError && !NoFinalize) 1055 Out.Finish(Lexer.getLoc()); 1056 1057 return HadError || getContext().hadError(); 1058 } 1059 1060 bool AsmParser::checkForValidSection() { 1061 if (!ParsingMSInlineAsm && !getStreamer().getCurrentSectionOnly()) { 1062 Out.InitSections(false); 1063 return Error(getTok().getLoc(), 1064 "expected section directive before assembly directive"); 1065 } 1066 return false; 1067 } 1068 1069 /// Throw away the rest of the line for testing purposes. 1070 void AsmParser::eatToEndOfStatement() { 1071 while (Lexer.isNot(AsmToken::EndOfStatement) && Lexer.isNot(AsmToken::Eof)) 1072 Lexer.Lex(); 1073 1074 // Eat EOL. 1075 if (Lexer.is(AsmToken::EndOfStatement)) 1076 Lexer.Lex(); 1077 } 1078 1079 StringRef AsmParser::parseStringToEndOfStatement() { 1080 const char *Start = getTok().getLoc().getPointer(); 1081 1082 while (Lexer.isNot(AsmToken::EndOfStatement) && Lexer.isNot(AsmToken::Eof)) 1083 Lexer.Lex(); 1084 1085 const char *End = getTok().getLoc().getPointer(); 1086 return StringRef(Start, End - Start); 1087 } 1088 1089 StringRef AsmParser::parseStringToComma() { 1090 const char *Start = getTok().getLoc().getPointer(); 1091 1092 while (Lexer.isNot(AsmToken::EndOfStatement) && 1093 Lexer.isNot(AsmToken::Comma) && Lexer.isNot(AsmToken::Eof)) 1094 Lexer.Lex(); 1095 1096 const char *End = getTok().getLoc().getPointer(); 1097 return StringRef(Start, End - Start); 1098 } 1099 1100 /// Parse a paren expression and return it. 1101 /// NOTE: This assumes the leading '(' has already been consumed. 1102 /// 1103 /// parenexpr ::= expr) 1104 /// 1105 bool AsmParser::parseParenExpr(const MCExpr *&Res, SMLoc &EndLoc) { 1106 if (parseExpression(Res)) 1107 return true; 1108 if (Lexer.isNot(AsmToken::RParen)) 1109 return TokError("expected ')' in parentheses expression"); 1110 EndLoc = Lexer.getTok().getEndLoc(); 1111 Lex(); 1112 return false; 1113 } 1114 1115 /// Parse a bracket expression and return it. 1116 /// NOTE: This assumes the leading '[' has already been consumed. 1117 /// 1118 /// bracketexpr ::= expr] 1119 /// 1120 bool AsmParser::parseBracketExpr(const MCExpr *&Res, SMLoc &EndLoc) { 1121 if (parseExpression(Res)) 1122 return true; 1123 EndLoc = getTok().getEndLoc(); 1124 if (parseToken(AsmToken::RBrac, "expected ']' in brackets expression")) 1125 return true; 1126 return false; 1127 } 1128 1129 /// Parse a primary expression and return it. 1130 /// primaryexpr ::= (parenexpr 1131 /// primaryexpr ::= symbol 1132 /// primaryexpr ::= number 1133 /// primaryexpr ::= '.' 1134 /// primaryexpr ::= ~,+,- primaryexpr 1135 bool AsmParser::parsePrimaryExpr(const MCExpr *&Res, SMLoc &EndLoc, 1136 AsmTypeInfo *TypeInfo) { 1137 SMLoc FirstTokenLoc = getLexer().getLoc(); 1138 AsmToken::TokenKind FirstTokenKind = Lexer.getKind(); 1139 switch (FirstTokenKind) { 1140 default: 1141 return TokError("unknown token in expression"); 1142 // If we have an error assume that we've already handled it. 1143 case AsmToken::Error: 1144 return true; 1145 case AsmToken::Exclaim: 1146 Lex(); // Eat the operator. 1147 if (parsePrimaryExpr(Res, EndLoc, TypeInfo)) 1148 return true; 1149 Res = MCUnaryExpr::createLNot(Res, getContext(), FirstTokenLoc); 1150 return false; 1151 case AsmToken::Dollar: 1152 case AsmToken::Star: 1153 case AsmToken::At: 1154 case AsmToken::String: 1155 case AsmToken::Identifier: { 1156 StringRef Identifier; 1157 if (parseIdentifier(Identifier)) { 1158 // We may have failed but '$'|'*' may be a valid token in context of 1159 // the current PC. 1160 if (getTok().is(AsmToken::Dollar) || getTok().is(AsmToken::Star)) { 1161 bool ShouldGenerateTempSymbol = false; 1162 if ((getTok().is(AsmToken::Dollar) && MAI.getDollarIsPC()) || 1163 (getTok().is(AsmToken::Star) && MAI.getStarIsPC())) 1164 ShouldGenerateTempSymbol = true; 1165 1166 if (!ShouldGenerateTempSymbol) 1167 return Error(FirstTokenLoc, "invalid token in expression"); 1168 1169 // Eat the '$'|'*' token. 1170 Lex(); 1171 // This is either a '$'|'*' reference, which references the current PC. 1172 // Emit a temporary label to the streamer and refer to it. 1173 MCSymbol *Sym = Ctx.createTempSymbol(); 1174 Out.emitLabel(Sym); 1175 Res = MCSymbolRefExpr::create(Sym, MCSymbolRefExpr::VK_None, 1176 getContext()); 1177 EndLoc = FirstTokenLoc; 1178 return false; 1179 } 1180 } 1181 // Parse symbol variant 1182 std::pair<StringRef, StringRef> Split; 1183 if (!MAI.useParensForSymbolVariant()) { 1184 if (FirstTokenKind == AsmToken::String) { 1185 if (Lexer.is(AsmToken::At)) { 1186 Lex(); // eat @ 1187 SMLoc AtLoc = getLexer().getLoc(); 1188 StringRef VName; 1189 if (parseIdentifier(VName)) 1190 return Error(AtLoc, "expected symbol variant after '@'"); 1191 1192 Split = std::make_pair(Identifier, VName); 1193 } 1194 } else { 1195 Split = Identifier.split('@'); 1196 } 1197 } else if (Lexer.is(AsmToken::LParen)) { 1198 Lex(); // eat '('. 1199 StringRef VName; 1200 parseIdentifier(VName); 1201 // eat ')'. 1202 if (parseToken(AsmToken::RParen, 1203 "unexpected token in variant, expected ')'")) 1204 return true; 1205 Split = std::make_pair(Identifier, VName); 1206 } 1207 1208 EndLoc = SMLoc::getFromPointer(Identifier.end()); 1209 1210 // This is a symbol reference. 1211 StringRef SymbolName = Identifier; 1212 if (SymbolName.empty()) 1213 return Error(getLexer().getLoc(), "expected a symbol reference"); 1214 1215 MCSymbolRefExpr::VariantKind Variant = MCSymbolRefExpr::VK_None; 1216 1217 // Lookup the symbol variant if used. 1218 if (!Split.second.empty()) { 1219 Variant = MCSymbolRefExpr::getVariantKindForName(Split.second); 1220 if (Variant != MCSymbolRefExpr::VK_Invalid) { 1221 SymbolName = Split.first; 1222 } else if (MAI.doesAllowAtInName() && !MAI.useParensForSymbolVariant()) { 1223 Variant = MCSymbolRefExpr::VK_None; 1224 } else { 1225 return Error(SMLoc::getFromPointer(Split.second.begin()), 1226 "invalid variant '" + Split.second + "'"); 1227 } 1228 } 1229 1230 MCSymbol *Sym = getContext().getInlineAsmLabel(SymbolName); 1231 if (!Sym) 1232 Sym = getContext().getOrCreateSymbol(SymbolName); 1233 1234 // If this is an absolute variable reference, substitute it now to preserve 1235 // semantics in the face of reassignment. 1236 if (Sym->isVariable()) { 1237 auto V = Sym->getVariableValue(/*SetUsed*/ false); 1238 bool DoInline = isa<MCConstantExpr>(V) && !Variant; 1239 if (auto TV = dyn_cast<MCTargetExpr>(V)) 1240 DoInline = TV->inlineAssignedExpr(); 1241 if (DoInline) { 1242 if (Variant) 1243 return Error(EndLoc, "unexpected modifier on variable reference"); 1244 Res = Sym->getVariableValue(/*SetUsed*/ false); 1245 return false; 1246 } 1247 } 1248 1249 // Otherwise create a symbol ref. 1250 Res = MCSymbolRefExpr::create(Sym, Variant, getContext(), FirstTokenLoc); 1251 return false; 1252 } 1253 case AsmToken::BigNum: 1254 return TokError("literal value out of range for directive"); 1255 case AsmToken::Integer: { 1256 SMLoc Loc = getTok().getLoc(); 1257 int64_t IntVal = getTok().getIntVal(); 1258 Res = MCConstantExpr::create(IntVal, getContext()); 1259 EndLoc = Lexer.getTok().getEndLoc(); 1260 Lex(); // Eat token. 1261 // Look for 'b' or 'f' following an Integer as a directional label 1262 if (Lexer.getKind() == AsmToken::Identifier) { 1263 StringRef IDVal = getTok().getString(); 1264 // Lookup the symbol variant if used. 1265 std::pair<StringRef, StringRef> Split = IDVal.split('@'); 1266 MCSymbolRefExpr::VariantKind Variant = MCSymbolRefExpr::VK_None; 1267 if (Split.first.size() != IDVal.size()) { 1268 Variant = MCSymbolRefExpr::getVariantKindForName(Split.second); 1269 if (Variant == MCSymbolRefExpr::VK_Invalid) 1270 return TokError("invalid variant '" + Split.second + "'"); 1271 IDVal = Split.first; 1272 } 1273 if (IDVal == "f" || IDVal == "b") { 1274 MCSymbol *Sym = 1275 Ctx.getDirectionalLocalSymbol(IntVal, IDVal == "b"); 1276 Res = MCSymbolRefExpr::create(Sym, Variant, getContext()); 1277 if (IDVal == "b" && Sym->isUndefined()) 1278 return Error(Loc, "directional label undefined"); 1279 DirLabels.push_back(std::make_tuple(Loc, CppHashInfo, Sym)); 1280 EndLoc = Lexer.getTok().getEndLoc(); 1281 Lex(); // Eat identifier. 1282 } 1283 } 1284 return false; 1285 } 1286 case AsmToken::Real: { 1287 APFloat RealVal(APFloat::IEEEdouble(), getTok().getString()); 1288 uint64_t IntVal = RealVal.bitcastToAPInt().getZExtValue(); 1289 Res = MCConstantExpr::create(IntVal, getContext()); 1290 EndLoc = Lexer.getTok().getEndLoc(); 1291 Lex(); // Eat token. 1292 return false; 1293 } 1294 case AsmToken::Dot: { 1295 if (!MAI.getDotIsPC()) 1296 return TokError("cannot use . as current PC"); 1297 1298 // This is a '.' reference, which references the current PC. Emit a 1299 // temporary label to the streamer and refer to it. 1300 MCSymbol *Sym = Ctx.createTempSymbol(); 1301 Out.emitLabel(Sym); 1302 Res = MCSymbolRefExpr::create(Sym, MCSymbolRefExpr::VK_None, getContext()); 1303 EndLoc = Lexer.getTok().getEndLoc(); 1304 Lex(); // Eat identifier. 1305 return false; 1306 } 1307 case AsmToken::LParen: 1308 Lex(); // Eat the '('. 1309 return parseParenExpr(Res, EndLoc); 1310 case AsmToken::LBrac: 1311 if (!PlatformParser->HasBracketExpressions()) 1312 return TokError("brackets expression not supported on this target"); 1313 Lex(); // Eat the '['. 1314 return parseBracketExpr(Res, EndLoc); 1315 case AsmToken::Minus: 1316 Lex(); // Eat the operator. 1317 if (parsePrimaryExpr(Res, EndLoc, TypeInfo)) 1318 return true; 1319 Res = MCUnaryExpr::createMinus(Res, getContext(), FirstTokenLoc); 1320 return false; 1321 case AsmToken::Plus: 1322 Lex(); // Eat the operator. 1323 if (parsePrimaryExpr(Res, EndLoc, TypeInfo)) 1324 return true; 1325 Res = MCUnaryExpr::createPlus(Res, getContext(), FirstTokenLoc); 1326 return false; 1327 case AsmToken::Tilde: 1328 Lex(); // Eat the operator. 1329 if (parsePrimaryExpr(Res, EndLoc, TypeInfo)) 1330 return true; 1331 Res = MCUnaryExpr::createNot(Res, getContext(), FirstTokenLoc); 1332 return false; 1333 // MIPS unary expression operators. The lexer won't generate these tokens if 1334 // MCAsmInfo::HasMipsExpressions is false for the target. 1335 case AsmToken::PercentCall16: 1336 case AsmToken::PercentCall_Hi: 1337 case AsmToken::PercentCall_Lo: 1338 case AsmToken::PercentDtprel_Hi: 1339 case AsmToken::PercentDtprel_Lo: 1340 case AsmToken::PercentGot: 1341 case AsmToken::PercentGot_Disp: 1342 case AsmToken::PercentGot_Hi: 1343 case AsmToken::PercentGot_Lo: 1344 case AsmToken::PercentGot_Ofst: 1345 case AsmToken::PercentGot_Page: 1346 case AsmToken::PercentGottprel: 1347 case AsmToken::PercentGp_Rel: 1348 case AsmToken::PercentHi: 1349 case AsmToken::PercentHigher: 1350 case AsmToken::PercentHighest: 1351 case AsmToken::PercentLo: 1352 case AsmToken::PercentNeg: 1353 case AsmToken::PercentPcrel_Hi: 1354 case AsmToken::PercentPcrel_Lo: 1355 case AsmToken::PercentTlsgd: 1356 case AsmToken::PercentTlsldm: 1357 case AsmToken::PercentTprel_Hi: 1358 case AsmToken::PercentTprel_Lo: 1359 Lex(); // Eat the operator. 1360 if (Lexer.isNot(AsmToken::LParen)) 1361 return TokError("expected '(' after operator"); 1362 Lex(); // Eat the operator. 1363 if (parseExpression(Res, EndLoc)) 1364 return true; 1365 if (Lexer.isNot(AsmToken::RParen)) 1366 return TokError("expected ')'"); 1367 Lex(); // Eat the operator. 1368 Res = getTargetParser().createTargetUnaryExpr(Res, FirstTokenKind, Ctx); 1369 return !Res; 1370 } 1371 } 1372 1373 bool AsmParser::parseExpression(const MCExpr *&Res) { 1374 SMLoc EndLoc; 1375 return parseExpression(Res, EndLoc); 1376 } 1377 1378 const MCExpr * 1379 AsmParser::applyModifierToExpr(const MCExpr *E, 1380 MCSymbolRefExpr::VariantKind Variant) { 1381 // Ask the target implementation about this expression first. 1382 const MCExpr *NewE = getTargetParser().applyModifierToExpr(E, Variant, Ctx); 1383 if (NewE) 1384 return NewE; 1385 // Recurse over the given expression, rebuilding it to apply the given variant 1386 // if there is exactly one symbol. 1387 switch (E->getKind()) { 1388 case MCExpr::Target: 1389 case MCExpr::Constant: 1390 return nullptr; 1391 1392 case MCExpr::SymbolRef: { 1393 const MCSymbolRefExpr *SRE = cast<MCSymbolRefExpr>(E); 1394 1395 if (SRE->getKind() != MCSymbolRefExpr::VK_None) { 1396 TokError("invalid variant on expression '" + getTok().getIdentifier() + 1397 "' (already modified)"); 1398 return E; 1399 } 1400 1401 return MCSymbolRefExpr::create(&SRE->getSymbol(), Variant, getContext()); 1402 } 1403 1404 case MCExpr::Unary: { 1405 const MCUnaryExpr *UE = cast<MCUnaryExpr>(E); 1406 const MCExpr *Sub = applyModifierToExpr(UE->getSubExpr(), Variant); 1407 if (!Sub) 1408 return nullptr; 1409 return MCUnaryExpr::create(UE->getOpcode(), Sub, getContext()); 1410 } 1411 1412 case MCExpr::Binary: { 1413 const MCBinaryExpr *BE = cast<MCBinaryExpr>(E); 1414 const MCExpr *LHS = applyModifierToExpr(BE->getLHS(), Variant); 1415 const MCExpr *RHS = applyModifierToExpr(BE->getRHS(), Variant); 1416 1417 if (!LHS && !RHS) 1418 return nullptr; 1419 1420 if (!LHS) 1421 LHS = BE->getLHS(); 1422 if (!RHS) 1423 RHS = BE->getRHS(); 1424 1425 return MCBinaryExpr::create(BE->getOpcode(), LHS, RHS, getContext()); 1426 } 1427 } 1428 1429 llvm_unreachable("Invalid expression kind!"); 1430 } 1431 1432 /// This function checks if the next token is <string> type or arithmetic. 1433 /// string that begin with character '<' must end with character '>'. 1434 /// otherwise it is arithmetics. 1435 /// If the function returns a 'true' value, 1436 /// the End argument will be filled with the last location pointed to the '>' 1437 /// character. 1438 1439 /// There is a gap between the AltMacro's documentation and the single quote 1440 /// implementation. GCC does not fully support this feature and so we will not 1441 /// support it. 1442 /// TODO: Adding single quote as a string. 1443 static bool isAngleBracketString(SMLoc &StrLoc, SMLoc &EndLoc) { 1444 assert((StrLoc.getPointer() != nullptr) && 1445 "Argument to the function cannot be a NULL value"); 1446 const char *CharPtr = StrLoc.getPointer(); 1447 while ((*CharPtr != '>') && (*CharPtr != '\n') && (*CharPtr != '\r') && 1448 (*CharPtr != '\0')) { 1449 if (*CharPtr == '!') 1450 CharPtr++; 1451 CharPtr++; 1452 } 1453 if (*CharPtr == '>') { 1454 EndLoc = StrLoc.getFromPointer(CharPtr + 1); 1455 return true; 1456 } 1457 return false; 1458 } 1459 1460 /// creating a string without the escape characters '!'. 1461 static std::string angleBracketString(StringRef AltMacroStr) { 1462 std::string Res; 1463 for (size_t Pos = 0; Pos < AltMacroStr.size(); Pos++) { 1464 if (AltMacroStr[Pos] == '!') 1465 Pos++; 1466 Res += AltMacroStr[Pos]; 1467 } 1468 return Res; 1469 } 1470 1471 /// Parse an expression and return it. 1472 /// 1473 /// expr ::= expr &&,|| expr -> lowest. 1474 /// expr ::= expr |,^,&,! expr 1475 /// expr ::= expr ==,!=,<>,<,<=,>,>= expr 1476 /// expr ::= expr <<,>> expr 1477 /// expr ::= expr +,- expr 1478 /// expr ::= expr *,/,% expr -> highest. 1479 /// expr ::= primaryexpr 1480 /// 1481 bool AsmParser::parseExpression(const MCExpr *&Res, SMLoc &EndLoc) { 1482 // Parse the expression. 1483 Res = nullptr; 1484 if (getTargetParser().parsePrimaryExpr(Res, EndLoc) || 1485 parseBinOpRHS(1, Res, EndLoc)) 1486 return true; 1487 1488 // As a special case, we support 'a op b @ modifier' by rewriting the 1489 // expression to include the modifier. This is inefficient, but in general we 1490 // expect users to use 'a@modifier op b'. 1491 if (Lexer.getKind() == AsmToken::At) { 1492 Lex(); 1493 1494 if (Lexer.isNot(AsmToken::Identifier)) 1495 return TokError("unexpected symbol modifier following '@'"); 1496 1497 MCSymbolRefExpr::VariantKind Variant = 1498 MCSymbolRefExpr::getVariantKindForName(getTok().getIdentifier()); 1499 if (Variant == MCSymbolRefExpr::VK_Invalid) 1500 return TokError("invalid variant '" + getTok().getIdentifier() + "'"); 1501 1502 const MCExpr *ModifiedRes = applyModifierToExpr(Res, Variant); 1503 if (!ModifiedRes) { 1504 return TokError("invalid modifier '" + getTok().getIdentifier() + 1505 "' (no symbols present)"); 1506 } 1507 1508 Res = ModifiedRes; 1509 Lex(); 1510 } 1511 1512 // Try to constant fold it up front, if possible. Do not exploit 1513 // assembler here. 1514 int64_t Value; 1515 if (Res->evaluateAsAbsolute(Value)) 1516 Res = MCConstantExpr::create(Value, getContext()); 1517 1518 return false; 1519 } 1520 1521 bool AsmParser::parseParenExpression(const MCExpr *&Res, SMLoc &EndLoc) { 1522 Res = nullptr; 1523 return parseParenExpr(Res, EndLoc) || parseBinOpRHS(1, Res, EndLoc); 1524 } 1525 1526 bool AsmParser::parseParenExprOfDepth(unsigned ParenDepth, const MCExpr *&Res, 1527 SMLoc &EndLoc) { 1528 if (parseParenExpr(Res, EndLoc)) 1529 return true; 1530 1531 for (; ParenDepth > 0; --ParenDepth) { 1532 if (parseBinOpRHS(1, Res, EndLoc)) 1533 return true; 1534 1535 // We don't Lex() the last RParen. 1536 // This is the same behavior as parseParenExpression(). 1537 if (ParenDepth - 1 > 0) { 1538 EndLoc = getTok().getEndLoc(); 1539 if (parseToken(AsmToken::RParen, 1540 "expected ')' in parentheses expression")) 1541 return true; 1542 } 1543 } 1544 return false; 1545 } 1546 1547 bool AsmParser::parseAbsoluteExpression(int64_t &Res) { 1548 const MCExpr *Expr; 1549 1550 SMLoc StartLoc = Lexer.getLoc(); 1551 if (parseExpression(Expr)) 1552 return true; 1553 1554 if (!Expr->evaluateAsAbsolute(Res, getStreamer().getAssemblerPtr())) 1555 return Error(StartLoc, "expected absolute expression"); 1556 1557 return false; 1558 } 1559 1560 static unsigned getDarwinBinOpPrecedence(AsmToken::TokenKind K, 1561 MCBinaryExpr::Opcode &Kind, 1562 bool ShouldUseLogicalShr) { 1563 switch (K) { 1564 default: 1565 return 0; // not a binop. 1566 1567 // Lowest Precedence: &&, || 1568 case AsmToken::AmpAmp: 1569 Kind = MCBinaryExpr::LAnd; 1570 return 1; 1571 case AsmToken::PipePipe: 1572 Kind = MCBinaryExpr::LOr; 1573 return 1; 1574 1575 // Low Precedence: |, &, ^ 1576 case AsmToken::Pipe: 1577 Kind = MCBinaryExpr::Or; 1578 return 2; 1579 case AsmToken::Caret: 1580 Kind = MCBinaryExpr::Xor; 1581 return 2; 1582 case AsmToken::Amp: 1583 Kind = MCBinaryExpr::And; 1584 return 2; 1585 1586 // Low Intermediate Precedence: ==, !=, <>, <, <=, >, >= 1587 case AsmToken::EqualEqual: 1588 Kind = MCBinaryExpr::EQ; 1589 return 3; 1590 case AsmToken::ExclaimEqual: 1591 case AsmToken::LessGreater: 1592 Kind = MCBinaryExpr::NE; 1593 return 3; 1594 case AsmToken::Less: 1595 Kind = MCBinaryExpr::LT; 1596 return 3; 1597 case AsmToken::LessEqual: 1598 Kind = MCBinaryExpr::LTE; 1599 return 3; 1600 case AsmToken::Greater: 1601 Kind = MCBinaryExpr::GT; 1602 return 3; 1603 case AsmToken::GreaterEqual: 1604 Kind = MCBinaryExpr::GTE; 1605 return 3; 1606 1607 // Intermediate Precedence: <<, >> 1608 case AsmToken::LessLess: 1609 Kind = MCBinaryExpr::Shl; 1610 return 4; 1611 case AsmToken::GreaterGreater: 1612 Kind = ShouldUseLogicalShr ? MCBinaryExpr::LShr : MCBinaryExpr::AShr; 1613 return 4; 1614 1615 // High Intermediate Precedence: +, - 1616 case AsmToken::Plus: 1617 Kind = MCBinaryExpr::Add; 1618 return 5; 1619 case AsmToken::Minus: 1620 Kind = MCBinaryExpr::Sub; 1621 return 5; 1622 1623 // Highest Precedence: *, /, % 1624 case AsmToken::Star: 1625 Kind = MCBinaryExpr::Mul; 1626 return 6; 1627 case AsmToken::Slash: 1628 Kind = MCBinaryExpr::Div; 1629 return 6; 1630 case AsmToken::Percent: 1631 Kind = MCBinaryExpr::Mod; 1632 return 6; 1633 } 1634 } 1635 1636 static unsigned getGNUBinOpPrecedence(const MCAsmInfo &MAI, 1637 AsmToken::TokenKind K, 1638 MCBinaryExpr::Opcode &Kind, 1639 bool ShouldUseLogicalShr) { 1640 switch (K) { 1641 default: 1642 return 0; // not a binop. 1643 1644 // Lowest Precedence: &&, || 1645 case AsmToken::AmpAmp: 1646 Kind = MCBinaryExpr::LAnd; 1647 return 2; 1648 case AsmToken::PipePipe: 1649 Kind = MCBinaryExpr::LOr; 1650 return 1; 1651 1652 // Low Precedence: ==, !=, <>, <, <=, >, >= 1653 case AsmToken::EqualEqual: 1654 Kind = MCBinaryExpr::EQ; 1655 return 3; 1656 case AsmToken::ExclaimEqual: 1657 case AsmToken::LessGreater: 1658 Kind = MCBinaryExpr::NE; 1659 return 3; 1660 case AsmToken::Less: 1661 Kind = MCBinaryExpr::LT; 1662 return 3; 1663 case AsmToken::LessEqual: 1664 Kind = MCBinaryExpr::LTE; 1665 return 3; 1666 case AsmToken::Greater: 1667 Kind = MCBinaryExpr::GT; 1668 return 3; 1669 case AsmToken::GreaterEqual: 1670 Kind = MCBinaryExpr::GTE; 1671 return 3; 1672 1673 // Low Intermediate Precedence: +, - 1674 case AsmToken::Plus: 1675 Kind = MCBinaryExpr::Add; 1676 return 4; 1677 case AsmToken::Minus: 1678 Kind = MCBinaryExpr::Sub; 1679 return 4; 1680 1681 // High Intermediate Precedence: |, !, &, ^ 1682 // 1683 case AsmToken::Pipe: 1684 Kind = MCBinaryExpr::Or; 1685 return 5; 1686 case AsmToken::Exclaim: 1687 // Hack to support ARM compatible aliases (implied 'sp' operand in 'srs*' 1688 // instructions like 'srsda #31!') and not parse ! as an infix operator. 1689 if (MAI.getCommentString() == "@") 1690 return 0; 1691 Kind = MCBinaryExpr::OrNot; 1692 return 5; 1693 case AsmToken::Caret: 1694 Kind = MCBinaryExpr::Xor; 1695 return 5; 1696 case AsmToken::Amp: 1697 Kind = MCBinaryExpr::And; 1698 return 5; 1699 1700 // Highest Precedence: *, /, %, <<, >> 1701 case AsmToken::Star: 1702 Kind = MCBinaryExpr::Mul; 1703 return 6; 1704 case AsmToken::Slash: 1705 Kind = MCBinaryExpr::Div; 1706 return 6; 1707 case AsmToken::Percent: 1708 Kind = MCBinaryExpr::Mod; 1709 return 6; 1710 case AsmToken::LessLess: 1711 Kind = MCBinaryExpr::Shl; 1712 return 6; 1713 case AsmToken::GreaterGreater: 1714 Kind = ShouldUseLogicalShr ? MCBinaryExpr::LShr : MCBinaryExpr::AShr; 1715 return 6; 1716 } 1717 } 1718 1719 unsigned AsmParser::getBinOpPrecedence(AsmToken::TokenKind K, 1720 MCBinaryExpr::Opcode &Kind) { 1721 bool ShouldUseLogicalShr = MAI.shouldUseLogicalShr(); 1722 return IsDarwin ? getDarwinBinOpPrecedence(K, Kind, ShouldUseLogicalShr) 1723 : getGNUBinOpPrecedence(MAI, K, Kind, ShouldUseLogicalShr); 1724 } 1725 1726 /// Parse all binary operators with precedence >= 'Precedence'. 1727 /// Res contains the LHS of the expression on input. 1728 bool AsmParser::parseBinOpRHS(unsigned Precedence, const MCExpr *&Res, 1729 SMLoc &EndLoc) { 1730 SMLoc StartLoc = Lexer.getLoc(); 1731 while (true) { 1732 MCBinaryExpr::Opcode Kind = MCBinaryExpr::Add; 1733 unsigned TokPrec = getBinOpPrecedence(Lexer.getKind(), Kind); 1734 1735 // If the next token is lower precedence than we are allowed to eat, return 1736 // successfully with what we ate already. 1737 if (TokPrec < Precedence) 1738 return false; 1739 1740 Lex(); 1741 1742 // Eat the next primary expression. 1743 const MCExpr *RHS; 1744 if (getTargetParser().parsePrimaryExpr(RHS, EndLoc)) 1745 return true; 1746 1747 // If BinOp binds less tightly with RHS than the operator after RHS, let 1748 // the pending operator take RHS as its LHS. 1749 MCBinaryExpr::Opcode Dummy; 1750 unsigned NextTokPrec = getBinOpPrecedence(Lexer.getKind(), Dummy); 1751 if (TokPrec < NextTokPrec && parseBinOpRHS(TokPrec + 1, RHS, EndLoc)) 1752 return true; 1753 1754 // Merge LHS and RHS according to operator. 1755 Res = MCBinaryExpr::create(Kind, Res, RHS, getContext(), StartLoc); 1756 } 1757 } 1758 1759 /// ParseStatement: 1760 /// ::= EndOfStatement 1761 /// ::= Label* Directive ...Operands... EndOfStatement 1762 /// ::= Label* Identifier OperandList* EndOfStatement 1763 bool AsmParser::parseStatement(ParseStatementInfo &Info, 1764 MCAsmParserSemaCallback *SI) { 1765 assert(!hasPendingError() && "parseStatement started with pending error"); 1766 // Eat initial spaces and comments 1767 while (Lexer.is(AsmToken::Space)) 1768 Lex(); 1769 if (Lexer.is(AsmToken::EndOfStatement)) { 1770 // if this is a line comment we can drop it safely 1771 if (getTok().getString().empty() || getTok().getString().front() == '\r' || 1772 getTok().getString().front() == '\n') 1773 Out.AddBlankLine(); 1774 Lex(); 1775 return false; 1776 } 1777 // Statements always start with an identifier. 1778 AsmToken ID = getTok(); 1779 SMLoc IDLoc = ID.getLoc(); 1780 StringRef IDVal; 1781 int64_t LocalLabelVal = -1; 1782 StartTokLoc = ID.getLoc(); 1783 if (Lexer.is(AsmToken::HashDirective)) 1784 return parseCppHashLineFilenameComment(IDLoc, 1785 !isInsideMacroInstantiation()); 1786 1787 // Allow an integer followed by a ':' as a directional local label. 1788 if (Lexer.is(AsmToken::Integer)) { 1789 LocalLabelVal = getTok().getIntVal(); 1790 if (LocalLabelVal < 0) { 1791 if (!TheCondState.Ignore) { 1792 Lex(); // always eat a token 1793 return Error(IDLoc, "unexpected token at start of statement"); 1794 } 1795 IDVal = ""; 1796 } else { 1797 IDVal = getTok().getString(); 1798 Lex(); // Consume the integer token to be used as an identifier token. 1799 if (Lexer.getKind() != AsmToken::Colon) { 1800 if (!TheCondState.Ignore) { 1801 Lex(); // always eat a token 1802 return Error(IDLoc, "unexpected token at start of statement"); 1803 } 1804 } 1805 } 1806 } else if (Lexer.is(AsmToken::Dot)) { 1807 // Treat '.' as a valid identifier in this context. 1808 Lex(); 1809 IDVal = "."; 1810 } else if (Lexer.is(AsmToken::LCurly)) { 1811 // Treat '{' as a valid identifier in this context. 1812 Lex(); 1813 IDVal = "{"; 1814 1815 } else if (Lexer.is(AsmToken::RCurly)) { 1816 // Treat '}' as a valid identifier in this context. 1817 Lex(); 1818 IDVal = "}"; 1819 } else if (Lexer.is(AsmToken::Star) && 1820 getTargetParser().starIsStartOfStatement()) { 1821 // Accept '*' as a valid start of statement. 1822 Lex(); 1823 IDVal = "*"; 1824 } else if (parseIdentifier(IDVal)) { 1825 if (!TheCondState.Ignore) { 1826 Lex(); // always eat a token 1827 return Error(IDLoc, "unexpected token at start of statement"); 1828 } 1829 IDVal = ""; 1830 } 1831 1832 // Handle conditional assembly here before checking for skipping. We 1833 // have to do this so that .endif isn't skipped in a ".if 0" block for 1834 // example. 1835 StringMap<DirectiveKind>::const_iterator DirKindIt = 1836 DirectiveKindMap.find(IDVal.lower()); 1837 DirectiveKind DirKind = (DirKindIt == DirectiveKindMap.end()) 1838 ? DK_NO_DIRECTIVE 1839 : DirKindIt->getValue(); 1840 switch (DirKind) { 1841 default: 1842 break; 1843 case DK_IF: 1844 case DK_IFEQ: 1845 case DK_IFGE: 1846 case DK_IFGT: 1847 case DK_IFLE: 1848 case DK_IFLT: 1849 case DK_IFNE: 1850 return parseDirectiveIf(IDLoc, DirKind); 1851 case DK_IFB: 1852 return parseDirectiveIfb(IDLoc, true); 1853 case DK_IFNB: 1854 return parseDirectiveIfb(IDLoc, false); 1855 case DK_IFC: 1856 return parseDirectiveIfc(IDLoc, true); 1857 case DK_IFEQS: 1858 return parseDirectiveIfeqs(IDLoc, true); 1859 case DK_IFNC: 1860 return parseDirectiveIfc(IDLoc, false); 1861 case DK_IFNES: 1862 return parseDirectiveIfeqs(IDLoc, false); 1863 case DK_IFDEF: 1864 return parseDirectiveIfdef(IDLoc, true); 1865 case DK_IFNDEF: 1866 case DK_IFNOTDEF: 1867 return parseDirectiveIfdef(IDLoc, false); 1868 case DK_ELSEIF: 1869 return parseDirectiveElseIf(IDLoc); 1870 case DK_ELSE: 1871 return parseDirectiveElse(IDLoc); 1872 case DK_ENDIF: 1873 return parseDirectiveEndIf(IDLoc); 1874 } 1875 1876 // Ignore the statement if in the middle of inactive conditional 1877 // (e.g. ".if 0"). 1878 if (TheCondState.Ignore) { 1879 eatToEndOfStatement(); 1880 return false; 1881 } 1882 1883 // FIXME: Recurse on local labels? 1884 1885 // See what kind of statement we have. 1886 switch (Lexer.getKind()) { 1887 case AsmToken::Colon: { 1888 if (!getTargetParser().isLabel(ID)) 1889 break; 1890 if (checkForValidSection()) 1891 return true; 1892 1893 // identifier ':' -> Label. 1894 Lex(); 1895 1896 // Diagnose attempt to use '.' as a label. 1897 if (IDVal == ".") 1898 return Error(IDLoc, "invalid use of pseudo-symbol '.' as a label"); 1899 1900 // Diagnose attempt to use a variable as a label. 1901 // 1902 // FIXME: Diagnostics. Note the location of the definition as a label. 1903 // FIXME: This doesn't diagnose assignment to a symbol which has been 1904 // implicitly marked as external. 1905 MCSymbol *Sym; 1906 if (LocalLabelVal == -1) { 1907 if (ParsingMSInlineAsm && SI) { 1908 StringRef RewrittenLabel = 1909 SI->LookupInlineAsmLabel(IDVal, getSourceManager(), IDLoc, true); 1910 assert(!RewrittenLabel.empty() && 1911 "We should have an internal name here."); 1912 Info.AsmRewrites->emplace_back(AOK_Label, IDLoc, IDVal.size(), 1913 RewrittenLabel); 1914 IDVal = RewrittenLabel; 1915 } 1916 Sym = getContext().getOrCreateSymbol(IDVal); 1917 } else 1918 Sym = Ctx.createDirectionalLocalSymbol(LocalLabelVal); 1919 // End of Labels should be treated as end of line for lexing 1920 // purposes but that information is not available to the Lexer who 1921 // does not understand Labels. This may cause us to see a Hash 1922 // here instead of a preprocessor line comment. 1923 if (getTok().is(AsmToken::Hash)) { 1924 StringRef CommentStr = parseStringToEndOfStatement(); 1925 Lexer.Lex(); 1926 Lexer.UnLex(AsmToken(AsmToken::EndOfStatement, CommentStr)); 1927 } 1928 1929 // Consume any end of statement token, if present, to avoid spurious 1930 // AddBlankLine calls(). 1931 if (getTok().is(AsmToken::EndOfStatement)) { 1932 Lex(); 1933 } 1934 1935 if (discardLTOSymbol(IDVal)) 1936 return false; 1937 1938 getTargetParser().doBeforeLabelEmit(Sym); 1939 1940 // Emit the label. 1941 if (!getTargetParser().isParsingMSInlineAsm()) 1942 Out.emitLabel(Sym, IDLoc); 1943 1944 // If we are generating dwarf for assembly source files then gather the 1945 // info to make a dwarf label entry for this label if needed. 1946 if (enabledGenDwarfForAssembly()) 1947 MCGenDwarfLabelEntry::Make(Sym, &getStreamer(), getSourceManager(), 1948 IDLoc); 1949 1950 getTargetParser().onLabelParsed(Sym); 1951 1952 return false; 1953 } 1954 1955 case AsmToken::Equal: 1956 if (!getTargetParser().equalIsAsmAssignment()) 1957 break; 1958 // identifier '=' ... -> assignment statement 1959 Lex(); 1960 1961 return parseAssignment(IDVal, true); 1962 1963 default: // Normal instruction or directive. 1964 break; 1965 } 1966 1967 // If macros are enabled, check to see if this is a macro instantiation. 1968 if (areMacrosEnabled()) 1969 if (const MCAsmMacro *M = getContext().lookupMacro(IDVal)) { 1970 return handleMacroEntry(M, IDLoc); 1971 } 1972 1973 // Otherwise, we have a normal instruction or directive. 1974 1975 // Directives start with "." 1976 if (IDVal.startswith(".") && IDVal != ".") { 1977 // There are several entities interested in parsing directives: 1978 // 1979 // 1. The target-specific assembly parser. Some directives are target 1980 // specific or may potentially behave differently on certain targets. 1981 // 2. Asm parser extensions. For example, platform-specific parsers 1982 // (like the ELF parser) register themselves as extensions. 1983 // 3. The generic directive parser implemented by this class. These are 1984 // all the directives that behave in a target and platform independent 1985 // manner, or at least have a default behavior that's shared between 1986 // all targets and platforms. 1987 1988 getTargetParser().flushPendingInstructions(getStreamer()); 1989 1990 SMLoc StartTokLoc = getTok().getLoc(); 1991 bool TPDirectiveReturn = getTargetParser().ParseDirective(ID); 1992 1993 if (hasPendingError()) 1994 return true; 1995 // Currently the return value should be true if we are 1996 // uninterested but as this is at odds with the standard parsing 1997 // convention (return true = error) we have instances of a parsed 1998 // directive that fails returning true as an error. Catch these 1999 // cases as best as possible errors here. 2000 if (TPDirectiveReturn && StartTokLoc != getTok().getLoc()) 2001 return true; 2002 // Return if we did some parsing or believe we succeeded. 2003 if (!TPDirectiveReturn || StartTokLoc != getTok().getLoc()) 2004 return false; 2005 2006 // Next, check the extension directive map to see if any extension has 2007 // registered itself to parse this directive. 2008 std::pair<MCAsmParserExtension *, DirectiveHandler> Handler = 2009 ExtensionDirectiveMap.lookup(IDVal); 2010 if (Handler.first) 2011 return (*Handler.second)(Handler.first, IDVal, IDLoc); 2012 2013 // Finally, if no one else is interested in this directive, it must be 2014 // generic and familiar to this class. 2015 switch (DirKind) { 2016 default: 2017 break; 2018 case DK_SET: 2019 case DK_EQU: 2020 return parseDirectiveSet(IDVal, true); 2021 case DK_EQUIV: 2022 return parseDirectiveSet(IDVal, false); 2023 case DK_ASCII: 2024 return parseDirectiveAscii(IDVal, false); 2025 case DK_ASCIZ: 2026 case DK_STRING: 2027 return parseDirectiveAscii(IDVal, true); 2028 case DK_BYTE: 2029 case DK_DC_B: 2030 return parseDirectiveValue(IDVal, 1); 2031 case DK_DC: 2032 case DK_DC_W: 2033 case DK_SHORT: 2034 case DK_VALUE: 2035 case DK_2BYTE: 2036 return parseDirectiveValue(IDVal, 2); 2037 case DK_LONG: 2038 case DK_INT: 2039 case DK_4BYTE: 2040 case DK_DC_L: 2041 return parseDirectiveValue(IDVal, 4); 2042 case DK_QUAD: 2043 case DK_8BYTE: 2044 return parseDirectiveValue(IDVal, 8); 2045 case DK_DC_A: 2046 return parseDirectiveValue( 2047 IDVal, getContext().getAsmInfo()->getCodePointerSize()); 2048 case DK_OCTA: 2049 return parseDirectiveOctaValue(IDVal); 2050 case DK_SINGLE: 2051 case DK_FLOAT: 2052 case DK_DC_S: 2053 return parseDirectiveRealValue(IDVal, APFloat::IEEEsingle()); 2054 case DK_DOUBLE: 2055 case DK_DC_D: 2056 return parseDirectiveRealValue(IDVal, APFloat::IEEEdouble()); 2057 case DK_ALIGN: { 2058 bool IsPow2 = !getContext().getAsmInfo()->getAlignmentIsInBytes(); 2059 return parseDirectiveAlign(IsPow2, /*ExprSize=*/1); 2060 } 2061 case DK_ALIGN32: { 2062 bool IsPow2 = !getContext().getAsmInfo()->getAlignmentIsInBytes(); 2063 return parseDirectiveAlign(IsPow2, /*ExprSize=*/4); 2064 } 2065 case DK_BALIGN: 2066 return parseDirectiveAlign(/*IsPow2=*/false, /*ExprSize=*/1); 2067 case DK_BALIGNW: 2068 return parseDirectiveAlign(/*IsPow2=*/false, /*ExprSize=*/2); 2069 case DK_BALIGNL: 2070 return parseDirectiveAlign(/*IsPow2=*/false, /*ExprSize=*/4); 2071 case DK_P2ALIGN: 2072 return parseDirectiveAlign(/*IsPow2=*/true, /*ExprSize=*/1); 2073 case DK_P2ALIGNW: 2074 return parseDirectiveAlign(/*IsPow2=*/true, /*ExprSize=*/2); 2075 case DK_P2ALIGNL: 2076 return parseDirectiveAlign(/*IsPow2=*/true, /*ExprSize=*/4); 2077 case DK_ORG: 2078 return parseDirectiveOrg(); 2079 case DK_FILL: 2080 return parseDirectiveFill(); 2081 case DK_ZERO: 2082 return parseDirectiveZero(); 2083 case DK_EXTERN: 2084 eatToEndOfStatement(); // .extern is the default, ignore it. 2085 return false; 2086 case DK_GLOBL: 2087 case DK_GLOBAL: 2088 return parseDirectiveSymbolAttribute(MCSA_Global); 2089 case DK_LAZY_REFERENCE: 2090 return parseDirectiveSymbolAttribute(MCSA_LazyReference); 2091 case DK_NO_DEAD_STRIP: 2092 return parseDirectiveSymbolAttribute(MCSA_NoDeadStrip); 2093 case DK_SYMBOL_RESOLVER: 2094 return parseDirectiveSymbolAttribute(MCSA_SymbolResolver); 2095 case DK_PRIVATE_EXTERN: 2096 return parseDirectiveSymbolAttribute(MCSA_PrivateExtern); 2097 case DK_REFERENCE: 2098 return parseDirectiveSymbolAttribute(MCSA_Reference); 2099 case DK_WEAK_DEFINITION: 2100 return parseDirectiveSymbolAttribute(MCSA_WeakDefinition); 2101 case DK_WEAK_REFERENCE: 2102 return parseDirectiveSymbolAttribute(MCSA_WeakReference); 2103 case DK_WEAK_DEF_CAN_BE_HIDDEN: 2104 return parseDirectiveSymbolAttribute(MCSA_WeakDefAutoPrivate); 2105 case DK_COLD: 2106 return parseDirectiveSymbolAttribute(MCSA_Cold); 2107 case DK_COMM: 2108 case DK_COMMON: 2109 return parseDirectiveComm(/*IsLocal=*/false); 2110 case DK_LCOMM: 2111 return parseDirectiveComm(/*IsLocal=*/true); 2112 case DK_ABORT: 2113 return parseDirectiveAbort(); 2114 case DK_INCLUDE: 2115 return parseDirectiveInclude(); 2116 case DK_INCBIN: 2117 return parseDirectiveIncbin(); 2118 case DK_CODE16: 2119 case DK_CODE16GCC: 2120 return TokError(Twine(IDVal) + 2121 " not currently supported for this target"); 2122 case DK_REPT: 2123 return parseDirectiveRept(IDLoc, IDVal); 2124 case DK_IRP: 2125 return parseDirectiveIrp(IDLoc); 2126 case DK_IRPC: 2127 return parseDirectiveIrpc(IDLoc); 2128 case DK_ENDR: 2129 return parseDirectiveEndr(IDLoc); 2130 case DK_BUNDLE_ALIGN_MODE: 2131 return parseDirectiveBundleAlignMode(); 2132 case DK_BUNDLE_LOCK: 2133 return parseDirectiveBundleLock(); 2134 case DK_BUNDLE_UNLOCK: 2135 return parseDirectiveBundleUnlock(); 2136 case DK_SLEB128: 2137 return parseDirectiveLEB128(true); 2138 case DK_ULEB128: 2139 return parseDirectiveLEB128(false); 2140 case DK_SPACE: 2141 case DK_SKIP: 2142 return parseDirectiveSpace(IDVal); 2143 case DK_FILE: 2144 return parseDirectiveFile(IDLoc); 2145 case DK_LINE: 2146 return parseDirectiveLine(); 2147 case DK_LOC: 2148 return parseDirectiveLoc(); 2149 case DK_STABS: 2150 return parseDirectiveStabs(); 2151 case DK_CV_FILE: 2152 return parseDirectiveCVFile(); 2153 case DK_CV_FUNC_ID: 2154 return parseDirectiveCVFuncId(); 2155 case DK_CV_INLINE_SITE_ID: 2156 return parseDirectiveCVInlineSiteId(); 2157 case DK_CV_LOC: 2158 return parseDirectiveCVLoc(); 2159 case DK_CV_LINETABLE: 2160 return parseDirectiveCVLinetable(); 2161 case DK_CV_INLINE_LINETABLE: 2162 return parseDirectiveCVInlineLinetable(); 2163 case DK_CV_DEF_RANGE: 2164 return parseDirectiveCVDefRange(); 2165 case DK_CV_STRING: 2166 return parseDirectiveCVString(); 2167 case DK_CV_STRINGTABLE: 2168 return parseDirectiveCVStringTable(); 2169 case DK_CV_FILECHECKSUMS: 2170 return parseDirectiveCVFileChecksums(); 2171 case DK_CV_FILECHECKSUM_OFFSET: 2172 return parseDirectiveCVFileChecksumOffset(); 2173 case DK_CV_FPO_DATA: 2174 return parseDirectiveCVFPOData(); 2175 case DK_CFI_SECTIONS: 2176 return parseDirectiveCFISections(); 2177 case DK_CFI_STARTPROC: 2178 return parseDirectiveCFIStartProc(); 2179 case DK_CFI_ENDPROC: 2180 return parseDirectiveCFIEndProc(); 2181 case DK_CFI_DEF_CFA: 2182 return parseDirectiveCFIDefCfa(IDLoc); 2183 case DK_CFI_DEF_CFA_OFFSET: 2184 return parseDirectiveCFIDefCfaOffset(); 2185 case DK_CFI_ADJUST_CFA_OFFSET: 2186 return parseDirectiveCFIAdjustCfaOffset(); 2187 case DK_CFI_DEF_CFA_REGISTER: 2188 return parseDirectiveCFIDefCfaRegister(IDLoc); 2189 case DK_CFI_OFFSET: 2190 return parseDirectiveCFIOffset(IDLoc); 2191 case DK_CFI_REL_OFFSET: 2192 return parseDirectiveCFIRelOffset(IDLoc); 2193 case DK_CFI_PERSONALITY: 2194 return parseDirectiveCFIPersonalityOrLsda(true); 2195 case DK_CFI_LSDA: 2196 return parseDirectiveCFIPersonalityOrLsda(false); 2197 case DK_CFI_REMEMBER_STATE: 2198 return parseDirectiveCFIRememberState(); 2199 case DK_CFI_RESTORE_STATE: 2200 return parseDirectiveCFIRestoreState(); 2201 case DK_CFI_SAME_VALUE: 2202 return parseDirectiveCFISameValue(IDLoc); 2203 case DK_CFI_RESTORE: 2204 return parseDirectiveCFIRestore(IDLoc); 2205 case DK_CFI_ESCAPE: 2206 return parseDirectiveCFIEscape(); 2207 case DK_CFI_RETURN_COLUMN: 2208 return parseDirectiveCFIReturnColumn(IDLoc); 2209 case DK_CFI_SIGNAL_FRAME: 2210 return parseDirectiveCFISignalFrame(); 2211 case DK_CFI_UNDEFINED: 2212 return parseDirectiveCFIUndefined(IDLoc); 2213 case DK_CFI_REGISTER: 2214 return parseDirectiveCFIRegister(IDLoc); 2215 case DK_CFI_WINDOW_SAVE: 2216 return parseDirectiveCFIWindowSave(); 2217 case DK_MACROS_ON: 2218 case DK_MACROS_OFF: 2219 return parseDirectiveMacrosOnOff(IDVal); 2220 case DK_MACRO: 2221 return parseDirectiveMacro(IDLoc); 2222 case DK_ALTMACRO: 2223 case DK_NOALTMACRO: 2224 return parseDirectiveAltmacro(IDVal); 2225 case DK_EXITM: 2226 return parseDirectiveExitMacro(IDVal); 2227 case DK_ENDM: 2228 case DK_ENDMACRO: 2229 return parseDirectiveEndMacro(IDVal); 2230 case DK_PURGEM: 2231 return parseDirectivePurgeMacro(IDLoc); 2232 case DK_END: 2233 return parseDirectiveEnd(IDLoc); 2234 case DK_ERR: 2235 return parseDirectiveError(IDLoc, false); 2236 case DK_ERROR: 2237 return parseDirectiveError(IDLoc, true); 2238 case DK_WARNING: 2239 return parseDirectiveWarning(IDLoc); 2240 case DK_RELOC: 2241 return parseDirectiveReloc(IDLoc); 2242 case DK_DCB: 2243 case DK_DCB_W: 2244 return parseDirectiveDCB(IDVal, 2); 2245 case DK_DCB_B: 2246 return parseDirectiveDCB(IDVal, 1); 2247 case DK_DCB_D: 2248 return parseDirectiveRealDCB(IDVal, APFloat::IEEEdouble()); 2249 case DK_DCB_L: 2250 return parseDirectiveDCB(IDVal, 4); 2251 case DK_DCB_S: 2252 return parseDirectiveRealDCB(IDVal, APFloat::IEEEsingle()); 2253 case DK_DC_X: 2254 case DK_DCB_X: 2255 return TokError(Twine(IDVal) + 2256 " not currently supported for this target"); 2257 case DK_DS: 2258 case DK_DS_W: 2259 return parseDirectiveDS(IDVal, 2); 2260 case DK_DS_B: 2261 return parseDirectiveDS(IDVal, 1); 2262 case DK_DS_D: 2263 return parseDirectiveDS(IDVal, 8); 2264 case DK_DS_L: 2265 case DK_DS_S: 2266 return parseDirectiveDS(IDVal, 4); 2267 case DK_DS_P: 2268 case DK_DS_X: 2269 return parseDirectiveDS(IDVal, 12); 2270 case DK_PRINT: 2271 return parseDirectivePrint(IDLoc); 2272 case DK_ADDRSIG: 2273 return parseDirectiveAddrsig(); 2274 case DK_ADDRSIG_SYM: 2275 return parseDirectiveAddrsigSym(); 2276 case DK_PSEUDO_PROBE: 2277 return parseDirectivePseudoProbe(); 2278 case DK_LTO_DISCARD: 2279 return parseDirectiveLTODiscard(); 2280 } 2281 2282 return Error(IDLoc, "unknown directive"); 2283 } 2284 2285 // __asm _emit or __asm __emit 2286 if (ParsingMSInlineAsm && (IDVal == "_emit" || IDVal == "__emit" || 2287 IDVal == "_EMIT" || IDVal == "__EMIT")) 2288 return parseDirectiveMSEmit(IDLoc, Info, IDVal.size()); 2289 2290 // __asm align 2291 if (ParsingMSInlineAsm && (IDVal == "align" || IDVal == "ALIGN")) 2292 return parseDirectiveMSAlign(IDLoc, Info); 2293 2294 if (ParsingMSInlineAsm && (IDVal == "even" || IDVal == "EVEN")) 2295 Info.AsmRewrites->emplace_back(AOK_EVEN, IDLoc, 4); 2296 if (checkForValidSection()) 2297 return true; 2298 2299 return parseAndMatchAndEmitTargetInstruction(Info, IDVal, ID, IDLoc); 2300 } 2301 2302 bool AsmParser::parseAndMatchAndEmitTargetInstruction(ParseStatementInfo &Info, 2303 StringRef IDVal, 2304 AsmToken ID, 2305 SMLoc IDLoc) { 2306 // Canonicalize the opcode to lower case. 2307 std::string OpcodeStr = IDVal.lower(); 2308 ParseInstructionInfo IInfo(Info.AsmRewrites); 2309 bool ParseHadError = getTargetParser().ParseInstruction(IInfo, OpcodeStr, ID, 2310 Info.ParsedOperands); 2311 Info.ParseError = ParseHadError; 2312 2313 // Dump the parsed representation, if requested. 2314 if (getShowParsedOperands()) { 2315 SmallString<256> Str; 2316 raw_svector_ostream OS(Str); 2317 OS << "parsed instruction: ["; 2318 for (unsigned i = 0; i != Info.ParsedOperands.size(); ++i) { 2319 if (i != 0) 2320 OS << ", "; 2321 Info.ParsedOperands[i]->print(OS); 2322 } 2323 OS << "]"; 2324 2325 printMessage(IDLoc, SourceMgr::DK_Note, OS.str()); 2326 } 2327 2328 // Fail even if ParseInstruction erroneously returns false. 2329 if (hasPendingError() || ParseHadError) 2330 return true; 2331 2332 // If we are generating dwarf for the current section then generate a .loc 2333 // directive for the instruction. 2334 if (!ParseHadError && enabledGenDwarfForAssembly() && 2335 getContext().getGenDwarfSectionSyms().count( 2336 getStreamer().getCurrentSectionOnly())) { 2337 unsigned Line; 2338 if (ActiveMacros.empty()) 2339 Line = SrcMgr.FindLineNumber(IDLoc, CurBuffer); 2340 else 2341 Line = SrcMgr.FindLineNumber(ActiveMacros.front()->InstantiationLoc, 2342 ActiveMacros.front()->ExitBuffer); 2343 2344 // If we previously parsed a cpp hash file line comment then make sure the 2345 // current Dwarf File is for the CppHashFilename if not then emit the 2346 // Dwarf File table for it and adjust the line number for the .loc. 2347 if (!CppHashInfo.Filename.empty()) { 2348 unsigned FileNumber = getStreamer().emitDwarfFileDirective( 2349 0, StringRef(), CppHashInfo.Filename); 2350 getContext().setGenDwarfFileNumber(FileNumber); 2351 2352 unsigned CppHashLocLineNo = 2353 SrcMgr.FindLineNumber(CppHashInfo.Loc, CppHashInfo.Buf); 2354 Line = CppHashInfo.LineNumber - 1 + (Line - CppHashLocLineNo); 2355 } 2356 2357 getStreamer().emitDwarfLocDirective( 2358 getContext().getGenDwarfFileNumber(), Line, 0, 2359 DWARF2_LINE_DEFAULT_IS_STMT ? DWARF2_FLAG_IS_STMT : 0, 0, 0, 2360 StringRef()); 2361 } 2362 2363 // If parsing succeeded, match the instruction. 2364 if (!ParseHadError) { 2365 uint64_t ErrorInfo; 2366 if (getTargetParser().MatchAndEmitInstruction( 2367 IDLoc, Info.Opcode, Info.ParsedOperands, Out, ErrorInfo, 2368 getTargetParser().isParsingMSInlineAsm())) 2369 return true; 2370 } 2371 return false; 2372 } 2373 2374 // Parse and erase curly braces marking block start/end 2375 bool 2376 AsmParser::parseCurlyBlockScope(SmallVectorImpl<AsmRewrite> &AsmStrRewrites) { 2377 // Identify curly brace marking block start/end 2378 if (Lexer.isNot(AsmToken::LCurly) && Lexer.isNot(AsmToken::RCurly)) 2379 return false; 2380 2381 SMLoc StartLoc = Lexer.getLoc(); 2382 Lex(); // Eat the brace 2383 if (Lexer.is(AsmToken::EndOfStatement)) 2384 Lex(); // Eat EndOfStatement following the brace 2385 2386 // Erase the block start/end brace from the output asm string 2387 AsmStrRewrites.emplace_back(AOK_Skip, StartLoc, Lexer.getLoc().getPointer() - 2388 StartLoc.getPointer()); 2389 return true; 2390 } 2391 2392 /// parseCppHashLineFilenameComment as this: 2393 /// ::= # number "filename" 2394 bool AsmParser::parseCppHashLineFilenameComment(SMLoc L, bool SaveLocInfo) { 2395 Lex(); // Eat the hash token. 2396 // Lexer only ever emits HashDirective if it fully formed if it's 2397 // done the checking already so this is an internal error. 2398 assert(getTok().is(AsmToken::Integer) && 2399 "Lexing Cpp line comment: Expected Integer"); 2400 int64_t LineNumber = getTok().getIntVal(); 2401 Lex(); 2402 assert(getTok().is(AsmToken::String) && 2403 "Lexing Cpp line comment: Expected String"); 2404 StringRef Filename = getTok().getString(); 2405 Lex(); 2406 2407 if (!SaveLocInfo) 2408 return false; 2409 2410 // Get rid of the enclosing quotes. 2411 Filename = Filename.substr(1, Filename.size() - 2); 2412 2413 // Save the SMLoc, Filename and LineNumber for later use by diagnostics 2414 // and possibly DWARF file info. 2415 CppHashInfo.Loc = L; 2416 CppHashInfo.Filename = Filename; 2417 CppHashInfo.LineNumber = LineNumber; 2418 CppHashInfo.Buf = CurBuffer; 2419 if (FirstCppHashFilename.empty()) 2420 FirstCppHashFilename = Filename; 2421 return false; 2422 } 2423 2424 /// will use the last parsed cpp hash line filename comment 2425 /// for the Filename and LineNo if any in the diagnostic. 2426 void AsmParser::DiagHandler(const SMDiagnostic &Diag, void *Context) { 2427 auto *Parser = static_cast<AsmParser *>(Context); 2428 raw_ostream &OS = errs(); 2429 2430 const SourceMgr &DiagSrcMgr = *Diag.getSourceMgr(); 2431 SMLoc DiagLoc = Diag.getLoc(); 2432 unsigned DiagBuf = DiagSrcMgr.FindBufferContainingLoc(DiagLoc); 2433 unsigned CppHashBuf = 2434 Parser->SrcMgr.FindBufferContainingLoc(Parser->CppHashInfo.Loc); 2435 2436 // Like SourceMgr::printMessage() we need to print the include stack if any 2437 // before printing the message. 2438 unsigned DiagCurBuffer = DiagSrcMgr.FindBufferContainingLoc(DiagLoc); 2439 if (!Parser->SavedDiagHandler && DiagCurBuffer && 2440 DiagCurBuffer != DiagSrcMgr.getMainFileID()) { 2441 SMLoc ParentIncludeLoc = DiagSrcMgr.getParentIncludeLoc(DiagCurBuffer); 2442 DiagSrcMgr.PrintIncludeStack(ParentIncludeLoc, OS); 2443 } 2444 2445 // If we have not parsed a cpp hash line filename comment or the source 2446 // manager changed or buffer changed (like in a nested include) then just 2447 // print the normal diagnostic using its Filename and LineNo. 2448 if (!Parser->CppHashInfo.LineNumber || DiagBuf != CppHashBuf) { 2449 if (Parser->SavedDiagHandler) 2450 Parser->SavedDiagHandler(Diag, Parser->SavedDiagContext); 2451 else 2452 Parser->getContext().diagnose(Diag); 2453 return; 2454 } 2455 2456 // Use the CppHashFilename and calculate a line number based on the 2457 // CppHashInfo.Loc and CppHashInfo.LineNumber relative to this Diag's SMLoc 2458 // for the diagnostic. 2459 const std::string &Filename = std::string(Parser->CppHashInfo.Filename); 2460 2461 int DiagLocLineNo = DiagSrcMgr.FindLineNumber(DiagLoc, DiagBuf); 2462 int CppHashLocLineNo = 2463 Parser->SrcMgr.FindLineNumber(Parser->CppHashInfo.Loc, CppHashBuf); 2464 int LineNo = 2465 Parser->CppHashInfo.LineNumber - 1 + (DiagLocLineNo - CppHashLocLineNo); 2466 2467 SMDiagnostic NewDiag(*Diag.getSourceMgr(), Diag.getLoc(), Filename, LineNo, 2468 Diag.getColumnNo(), Diag.getKind(), Diag.getMessage(), 2469 Diag.getLineContents(), Diag.getRanges()); 2470 2471 if (Parser->SavedDiagHandler) 2472 Parser->SavedDiagHandler(Diag, Parser->SavedDiagContext); 2473 else 2474 Parser->getContext().diagnose(NewDiag); 2475 } 2476 2477 // FIXME: This is mostly duplicated from the function in AsmLexer.cpp. The 2478 // difference being that that function accepts '@' as part of identifiers and 2479 // we can't do that. AsmLexer.cpp should probably be changed to handle 2480 // '@' as a special case when needed. 2481 static bool isIdentifierChar(char c) { 2482 return isalnum(static_cast<unsigned char>(c)) || c == '_' || c == '$' || 2483 c == '.'; 2484 } 2485 2486 bool AsmParser::expandMacro(raw_svector_ostream &OS, StringRef Body, 2487 ArrayRef<MCAsmMacroParameter> Parameters, 2488 ArrayRef<MCAsmMacroArgument> A, 2489 bool EnableAtPseudoVariable, SMLoc L) { 2490 unsigned NParameters = Parameters.size(); 2491 bool HasVararg = NParameters ? Parameters.back().Vararg : false; 2492 if ((!IsDarwin || NParameters != 0) && NParameters != A.size()) 2493 return Error(L, "Wrong number of arguments"); 2494 2495 // A macro without parameters is handled differently on Darwin: 2496 // gas accepts no arguments and does no substitutions 2497 while (!Body.empty()) { 2498 // Scan for the next substitution. 2499 std::size_t End = Body.size(), Pos = 0; 2500 for (; Pos != End; ++Pos) { 2501 // Check for a substitution or escape. 2502 if (IsDarwin && !NParameters) { 2503 // This macro has no parameters, look for $0, $1, etc. 2504 if (Body[Pos] != '$' || Pos + 1 == End) 2505 continue; 2506 2507 char Next = Body[Pos + 1]; 2508 if (Next == '$' || Next == 'n' || 2509 isdigit(static_cast<unsigned char>(Next))) 2510 break; 2511 } else { 2512 // This macro has parameters, look for \foo, \bar, etc. 2513 if (Body[Pos] == '\\' && Pos + 1 != End) 2514 break; 2515 } 2516 } 2517 2518 // Add the prefix. 2519 OS << Body.slice(0, Pos); 2520 2521 // Check if we reached the end. 2522 if (Pos == End) 2523 break; 2524 2525 if (IsDarwin && !NParameters) { 2526 switch (Body[Pos + 1]) { 2527 // $$ => $ 2528 case '$': 2529 OS << '$'; 2530 break; 2531 2532 // $n => number of arguments 2533 case 'n': 2534 OS << A.size(); 2535 break; 2536 2537 // $[0-9] => argument 2538 default: { 2539 // Missing arguments are ignored. 2540 unsigned Index = Body[Pos + 1] - '0'; 2541 if (Index >= A.size()) 2542 break; 2543 2544 // Otherwise substitute with the token values, with spaces eliminated. 2545 for (const AsmToken &Token : A[Index]) 2546 OS << Token.getString(); 2547 break; 2548 } 2549 } 2550 Pos += 2; 2551 } else { 2552 unsigned I = Pos + 1; 2553 2554 // Check for the \@ pseudo-variable. 2555 if (EnableAtPseudoVariable && Body[I] == '@' && I + 1 != End) 2556 ++I; 2557 else 2558 while (isIdentifierChar(Body[I]) && I + 1 != End) 2559 ++I; 2560 2561 const char *Begin = Body.data() + Pos + 1; 2562 StringRef Argument(Begin, I - (Pos + 1)); 2563 unsigned Index = 0; 2564 2565 if (Argument == "@") { 2566 OS << NumOfMacroInstantiations; 2567 Pos += 2; 2568 } else { 2569 for (; Index < NParameters; ++Index) 2570 if (Parameters[Index].Name == Argument) 2571 break; 2572 2573 if (Index == NParameters) { 2574 if (Body[Pos + 1] == '(' && Body[Pos + 2] == ')') 2575 Pos += 3; 2576 else { 2577 OS << '\\' << Argument; 2578 Pos = I; 2579 } 2580 } else { 2581 bool VarargParameter = HasVararg && Index == (NParameters - 1); 2582 for (const AsmToken &Token : A[Index]) 2583 // For altmacro mode, you can write '%expr'. 2584 // The prefix '%' evaluates the expression 'expr' 2585 // and uses the result as a string (e.g. replace %(1+2) with the 2586 // string "3"). 2587 // Here, we identify the integer token which is the result of the 2588 // absolute expression evaluation and replace it with its string 2589 // representation. 2590 if (AltMacroMode && Token.getString().front() == '%' && 2591 Token.is(AsmToken::Integer)) 2592 // Emit an integer value to the buffer. 2593 OS << Token.getIntVal(); 2594 // Only Token that was validated as a string and begins with '<' 2595 // is considered altMacroString!!! 2596 else if (AltMacroMode && Token.getString().front() == '<' && 2597 Token.is(AsmToken::String)) { 2598 OS << angleBracketString(Token.getStringContents()); 2599 } 2600 // We expect no quotes around the string's contents when 2601 // parsing for varargs. 2602 else if (Token.isNot(AsmToken::String) || VarargParameter) 2603 OS << Token.getString(); 2604 else 2605 OS << Token.getStringContents(); 2606 2607 Pos += 1 + Argument.size(); 2608 } 2609 } 2610 } 2611 // Update the scan point. 2612 Body = Body.substr(Pos); 2613 } 2614 2615 return false; 2616 } 2617 2618 static bool isOperator(AsmToken::TokenKind kind) { 2619 switch (kind) { 2620 default: 2621 return false; 2622 case AsmToken::Plus: 2623 case AsmToken::Minus: 2624 case AsmToken::Tilde: 2625 case AsmToken::Slash: 2626 case AsmToken::Star: 2627 case AsmToken::Dot: 2628 case AsmToken::Equal: 2629 case AsmToken::EqualEqual: 2630 case AsmToken::Pipe: 2631 case AsmToken::PipePipe: 2632 case AsmToken::Caret: 2633 case AsmToken::Amp: 2634 case AsmToken::AmpAmp: 2635 case AsmToken::Exclaim: 2636 case AsmToken::ExclaimEqual: 2637 case AsmToken::Less: 2638 case AsmToken::LessEqual: 2639 case AsmToken::LessLess: 2640 case AsmToken::LessGreater: 2641 case AsmToken::Greater: 2642 case AsmToken::GreaterEqual: 2643 case AsmToken::GreaterGreater: 2644 return true; 2645 } 2646 } 2647 2648 namespace { 2649 2650 class AsmLexerSkipSpaceRAII { 2651 public: 2652 AsmLexerSkipSpaceRAII(AsmLexer &Lexer, bool SkipSpace) : Lexer(Lexer) { 2653 Lexer.setSkipSpace(SkipSpace); 2654 } 2655 2656 ~AsmLexerSkipSpaceRAII() { 2657 Lexer.setSkipSpace(true); 2658 } 2659 2660 private: 2661 AsmLexer &Lexer; 2662 }; 2663 2664 } // end anonymous namespace 2665 2666 bool AsmParser::parseMacroArgument(MCAsmMacroArgument &MA, bool Vararg) { 2667 2668 if (Vararg) { 2669 if (Lexer.isNot(AsmToken::EndOfStatement)) { 2670 StringRef Str = parseStringToEndOfStatement(); 2671 MA.emplace_back(AsmToken::String, Str); 2672 } 2673 return false; 2674 } 2675 2676 unsigned ParenLevel = 0; 2677 2678 // Darwin doesn't use spaces to delmit arguments. 2679 AsmLexerSkipSpaceRAII ScopedSkipSpace(Lexer, IsDarwin); 2680 2681 bool SpaceEaten; 2682 2683 while (true) { 2684 SpaceEaten = false; 2685 if (Lexer.is(AsmToken::Eof) || Lexer.is(AsmToken::Equal)) 2686 return TokError("unexpected token in macro instantiation"); 2687 2688 if (ParenLevel == 0) { 2689 2690 if (Lexer.is(AsmToken::Comma)) 2691 break; 2692 2693 if (Lexer.is(AsmToken::Space)) { 2694 SpaceEaten = true; 2695 Lexer.Lex(); // Eat spaces 2696 } 2697 2698 // Spaces can delimit parameters, but could also be part an expression. 2699 // If the token after a space is an operator, add the token and the next 2700 // one into this argument 2701 if (!IsDarwin) { 2702 if (isOperator(Lexer.getKind())) { 2703 MA.push_back(getTok()); 2704 Lexer.Lex(); 2705 2706 // Whitespace after an operator can be ignored. 2707 if (Lexer.is(AsmToken::Space)) 2708 Lexer.Lex(); 2709 2710 continue; 2711 } 2712 } 2713 if (SpaceEaten) 2714 break; 2715 } 2716 2717 // handleMacroEntry relies on not advancing the lexer here 2718 // to be able to fill in the remaining default parameter values 2719 if (Lexer.is(AsmToken::EndOfStatement)) 2720 break; 2721 2722 // Adjust the current parentheses level. 2723 if (Lexer.is(AsmToken::LParen)) 2724 ++ParenLevel; 2725 else if (Lexer.is(AsmToken::RParen) && ParenLevel) 2726 --ParenLevel; 2727 2728 // Append the token to the current argument list. 2729 MA.push_back(getTok()); 2730 Lexer.Lex(); 2731 } 2732 2733 if (ParenLevel != 0) 2734 return TokError("unbalanced parentheses in macro argument"); 2735 return false; 2736 } 2737 2738 // Parse the macro instantiation arguments. 2739 bool AsmParser::parseMacroArguments(const MCAsmMacro *M, 2740 MCAsmMacroArguments &A) { 2741 const unsigned NParameters = M ? M->Parameters.size() : 0; 2742 bool NamedParametersFound = false; 2743 SmallVector<SMLoc, 4> FALocs; 2744 2745 A.resize(NParameters); 2746 FALocs.resize(NParameters); 2747 2748 // Parse two kinds of macro invocations: 2749 // - macros defined without any parameters accept an arbitrary number of them 2750 // - macros defined with parameters accept at most that many of them 2751 bool HasVararg = NParameters ? M->Parameters.back().Vararg : false; 2752 for (unsigned Parameter = 0; !NParameters || Parameter < NParameters; 2753 ++Parameter) { 2754 SMLoc IDLoc = Lexer.getLoc(); 2755 MCAsmMacroParameter FA; 2756 2757 if (Lexer.is(AsmToken::Identifier) && Lexer.peekTok().is(AsmToken::Equal)) { 2758 if (parseIdentifier(FA.Name)) 2759 return Error(IDLoc, "invalid argument identifier for formal argument"); 2760 2761 if (Lexer.isNot(AsmToken::Equal)) 2762 return TokError("expected '=' after formal parameter identifier"); 2763 2764 Lex(); 2765 2766 NamedParametersFound = true; 2767 } 2768 bool Vararg = HasVararg && Parameter == (NParameters - 1); 2769 2770 if (NamedParametersFound && FA.Name.empty()) 2771 return Error(IDLoc, "cannot mix positional and keyword arguments"); 2772 2773 SMLoc StrLoc = Lexer.getLoc(); 2774 SMLoc EndLoc; 2775 if (AltMacroMode && Lexer.is(AsmToken::Percent)) { 2776 const MCExpr *AbsoluteExp; 2777 int64_t Value; 2778 /// Eat '%' 2779 Lex(); 2780 if (parseExpression(AbsoluteExp, EndLoc)) 2781 return false; 2782 if (!AbsoluteExp->evaluateAsAbsolute(Value, 2783 getStreamer().getAssemblerPtr())) 2784 return Error(StrLoc, "expected absolute expression"); 2785 const char *StrChar = StrLoc.getPointer(); 2786 const char *EndChar = EndLoc.getPointer(); 2787 AsmToken newToken(AsmToken::Integer, 2788 StringRef(StrChar, EndChar - StrChar), Value); 2789 FA.Value.push_back(newToken); 2790 } else if (AltMacroMode && Lexer.is(AsmToken::Less) && 2791 isAngleBracketString(StrLoc, EndLoc)) { 2792 const char *StrChar = StrLoc.getPointer(); 2793 const char *EndChar = EndLoc.getPointer(); 2794 jumpToLoc(EndLoc, CurBuffer); 2795 /// Eat from '<' to '>' 2796 Lex(); 2797 AsmToken newToken(AsmToken::String, 2798 StringRef(StrChar, EndChar - StrChar)); 2799 FA.Value.push_back(newToken); 2800 } else if(parseMacroArgument(FA.Value, Vararg)) 2801 return true; 2802 2803 unsigned PI = Parameter; 2804 if (!FA.Name.empty()) { 2805 unsigned FAI = 0; 2806 for (FAI = 0; FAI < NParameters; ++FAI) 2807 if (M->Parameters[FAI].Name == FA.Name) 2808 break; 2809 2810 if (FAI >= NParameters) { 2811 assert(M && "expected macro to be defined"); 2812 return Error(IDLoc, "parameter named '" + FA.Name + 2813 "' does not exist for macro '" + M->Name + "'"); 2814 } 2815 PI = FAI; 2816 } 2817 2818 if (!FA.Value.empty()) { 2819 if (A.size() <= PI) 2820 A.resize(PI + 1); 2821 A[PI] = FA.Value; 2822 2823 if (FALocs.size() <= PI) 2824 FALocs.resize(PI + 1); 2825 2826 FALocs[PI] = Lexer.getLoc(); 2827 } 2828 2829 // At the end of the statement, fill in remaining arguments that have 2830 // default values. If there aren't any, then the next argument is 2831 // required but missing 2832 if (Lexer.is(AsmToken::EndOfStatement)) { 2833 bool Failure = false; 2834 for (unsigned FAI = 0; FAI < NParameters; ++FAI) { 2835 if (A[FAI].empty()) { 2836 if (M->Parameters[FAI].Required) { 2837 Error(FALocs[FAI].isValid() ? FALocs[FAI] : Lexer.getLoc(), 2838 "missing value for required parameter " 2839 "'" + M->Parameters[FAI].Name + "' in macro '" + M->Name + "'"); 2840 Failure = true; 2841 } 2842 2843 if (!M->Parameters[FAI].Value.empty()) 2844 A[FAI] = M->Parameters[FAI].Value; 2845 } 2846 } 2847 return Failure; 2848 } 2849 2850 if (Lexer.is(AsmToken::Comma)) 2851 Lex(); 2852 } 2853 2854 return TokError("too many positional arguments"); 2855 } 2856 2857 bool AsmParser::handleMacroEntry(const MCAsmMacro *M, SMLoc NameLoc) { 2858 // Arbitrarily limit macro nesting depth (default matches 'as'). We can 2859 // eliminate this, although we should protect against infinite loops. 2860 unsigned MaxNestingDepth = AsmMacroMaxNestingDepth; 2861 if (ActiveMacros.size() == MaxNestingDepth) { 2862 std::ostringstream MaxNestingDepthError; 2863 MaxNestingDepthError << "macros cannot be nested more than " 2864 << MaxNestingDepth << " levels deep." 2865 << " Use -asm-macro-max-nesting-depth to increase " 2866 "this limit."; 2867 return TokError(MaxNestingDepthError.str()); 2868 } 2869 2870 MCAsmMacroArguments A; 2871 if (parseMacroArguments(M, A)) 2872 return true; 2873 2874 // Macro instantiation is lexical, unfortunately. We construct a new buffer 2875 // to hold the macro body with substitutions. 2876 SmallString<256> Buf; 2877 StringRef Body = M->Body; 2878 raw_svector_ostream OS(Buf); 2879 2880 if (expandMacro(OS, Body, M->Parameters, A, true, getTok().getLoc())) 2881 return true; 2882 2883 // We include the .endmacro in the buffer as our cue to exit the macro 2884 // instantiation. 2885 OS << ".endmacro\n"; 2886 2887 std::unique_ptr<MemoryBuffer> Instantiation = 2888 MemoryBuffer::getMemBufferCopy(OS.str(), "<instantiation>"); 2889 2890 // Create the macro instantiation object and add to the current macro 2891 // instantiation stack. 2892 MacroInstantiation *MI = new MacroInstantiation{ 2893 NameLoc, CurBuffer, getTok().getLoc(), TheCondStack.size()}; 2894 ActiveMacros.push_back(MI); 2895 2896 ++NumOfMacroInstantiations; 2897 2898 // Jump to the macro instantiation and prime the lexer. 2899 CurBuffer = SrcMgr.AddNewSourceBuffer(std::move(Instantiation), SMLoc()); 2900 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer()); 2901 Lex(); 2902 2903 return false; 2904 } 2905 2906 void AsmParser::handleMacroExit() { 2907 // Jump to the EndOfStatement we should return to, and consume it. 2908 jumpToLoc(ActiveMacros.back()->ExitLoc, ActiveMacros.back()->ExitBuffer); 2909 Lex(); 2910 2911 // Pop the instantiation entry. 2912 delete ActiveMacros.back(); 2913 ActiveMacros.pop_back(); 2914 } 2915 2916 bool AsmParser::parseAssignment(StringRef Name, bool allow_redef, 2917 bool NoDeadStrip) { 2918 MCSymbol *Sym; 2919 const MCExpr *Value; 2920 if (MCParserUtils::parseAssignmentExpression(Name, allow_redef, *this, Sym, 2921 Value)) 2922 return true; 2923 2924 if (!Sym) { 2925 // In the case where we parse an expression starting with a '.', we will 2926 // not generate an error, nor will we create a symbol. In this case we 2927 // should just return out. 2928 return false; 2929 } 2930 2931 if (discardLTOSymbol(Name)) 2932 return false; 2933 2934 // Do the assignment. 2935 Out.emitAssignment(Sym, Value); 2936 if (NoDeadStrip) 2937 Out.emitSymbolAttribute(Sym, MCSA_NoDeadStrip); 2938 2939 return false; 2940 } 2941 2942 /// parseIdentifier: 2943 /// ::= identifier 2944 /// ::= string 2945 bool AsmParser::parseIdentifier(StringRef &Res) { 2946 // The assembler has relaxed rules for accepting identifiers, in particular we 2947 // allow things like '.globl $foo' and '.def @feat.00', which would normally be 2948 // separate tokens. At this level, we have already lexed so we cannot (currently) 2949 // handle this as a context dependent token, instead we detect adjacent tokens 2950 // and return the combined identifier. 2951 if (Lexer.is(AsmToken::Dollar) || Lexer.is(AsmToken::At)) { 2952 SMLoc PrefixLoc = getLexer().getLoc(); 2953 2954 // Consume the prefix character, and check for a following identifier. 2955 2956 AsmToken Buf[1]; 2957 Lexer.peekTokens(Buf, false); 2958 2959 if (Buf[0].isNot(AsmToken::Identifier) && Buf[0].isNot(AsmToken::Integer)) 2960 return true; 2961 2962 // We have a '$' or '@' followed by an identifier or integer token, make 2963 // sure they are adjacent. 2964 if (PrefixLoc.getPointer() + 1 != Buf[0].getLoc().getPointer()) 2965 return true; 2966 2967 // eat $ or @ 2968 Lexer.Lex(); // Lexer's Lex guarantees consecutive token. 2969 // Construct the joined identifier and consume the token. 2970 Res = StringRef(PrefixLoc.getPointer(), getTok().getString().size() + 1); 2971 Lex(); // Parser Lex to maintain invariants. 2972 return false; 2973 } 2974 2975 if (Lexer.isNot(AsmToken::Identifier) && Lexer.isNot(AsmToken::String)) 2976 return true; 2977 2978 Res = getTok().getIdentifier(); 2979 2980 Lex(); // Consume the identifier token. 2981 2982 return false; 2983 } 2984 2985 /// parseDirectiveSet: 2986 /// ::= .equ identifier ',' expression 2987 /// ::= .equiv identifier ',' expression 2988 /// ::= .set identifier ',' expression 2989 bool AsmParser::parseDirectiveSet(StringRef IDVal, bool allow_redef) { 2990 StringRef Name; 2991 if (check(parseIdentifier(Name), "expected identifier") || parseComma() || 2992 parseAssignment(Name, allow_redef, true)) 2993 return true; 2994 return false; 2995 } 2996 2997 bool AsmParser::parseEscapedString(std::string &Data) { 2998 if (check(getTok().isNot(AsmToken::String), "expected string")) 2999 return true; 3000 3001 Data = ""; 3002 StringRef Str = getTok().getStringContents(); 3003 for (unsigned i = 0, e = Str.size(); i != e; ++i) { 3004 if (Str[i] != '\\') { 3005 Data += Str[i]; 3006 continue; 3007 } 3008 3009 // Recognize escaped characters. Note that this escape semantics currently 3010 // loosely follows Darwin 'as'. 3011 ++i; 3012 if (i == e) 3013 return TokError("unexpected backslash at end of string"); 3014 3015 // Recognize hex sequences similarly to GNU 'as'. 3016 if (Str[i] == 'x' || Str[i] == 'X') { 3017 size_t length = Str.size(); 3018 if (i + 1 >= length || !isHexDigit(Str[i + 1])) 3019 return TokError("invalid hexadecimal escape sequence"); 3020 3021 // Consume hex characters. GNU 'as' reads all hexadecimal characters and 3022 // then truncates to the lower 16 bits. Seems reasonable. 3023 unsigned Value = 0; 3024 while (i + 1 < length && isHexDigit(Str[i + 1])) 3025 Value = Value * 16 + hexDigitValue(Str[++i]); 3026 3027 Data += (unsigned char)(Value & 0xFF); 3028 continue; 3029 } 3030 3031 // Recognize octal sequences. 3032 if ((unsigned)(Str[i] - '0') <= 7) { 3033 // Consume up to three octal characters. 3034 unsigned Value = Str[i] - '0'; 3035 3036 if (i + 1 != e && ((unsigned)(Str[i + 1] - '0')) <= 7) { 3037 ++i; 3038 Value = Value * 8 + (Str[i] - '0'); 3039 3040 if (i + 1 != e && ((unsigned)(Str[i + 1] - '0')) <= 7) { 3041 ++i; 3042 Value = Value * 8 + (Str[i] - '0'); 3043 } 3044 } 3045 3046 if (Value > 255) 3047 return TokError("invalid octal escape sequence (out of range)"); 3048 3049 Data += (unsigned char)Value; 3050 continue; 3051 } 3052 3053 // Otherwise recognize individual escapes. 3054 switch (Str[i]) { 3055 default: 3056 // Just reject invalid escape sequences for now. 3057 return TokError("invalid escape sequence (unrecognized character)"); 3058 3059 case 'b': Data += '\b'; break; 3060 case 'f': Data += '\f'; break; 3061 case 'n': Data += '\n'; break; 3062 case 'r': Data += '\r'; break; 3063 case 't': Data += '\t'; break; 3064 case '"': Data += '"'; break; 3065 case '\\': Data += '\\'; break; 3066 } 3067 } 3068 3069 Lex(); 3070 return false; 3071 } 3072 3073 bool AsmParser::parseAngleBracketString(std::string &Data) { 3074 SMLoc EndLoc, StartLoc = getTok().getLoc(); 3075 if (isAngleBracketString(StartLoc, EndLoc)) { 3076 const char *StartChar = StartLoc.getPointer() + 1; 3077 const char *EndChar = EndLoc.getPointer() - 1; 3078 jumpToLoc(EndLoc, CurBuffer); 3079 /// Eat from '<' to '>' 3080 Lex(); 3081 3082 Data = angleBracketString(StringRef(StartChar, EndChar - StartChar)); 3083 return false; 3084 } 3085 return true; 3086 } 3087 3088 /// parseDirectiveAscii: 3089 // ::= .ascii [ "string"+ ( , "string"+ )* ] 3090 /// ::= ( .asciz | .string ) [ "string" ( , "string" )* ] 3091 bool AsmParser::parseDirectiveAscii(StringRef IDVal, bool ZeroTerminated) { 3092 auto parseOp = [&]() -> bool { 3093 std::string Data; 3094 if (checkForValidSection()) 3095 return true; 3096 // Only support spaces as separators for .ascii directive for now. See the 3097 // discusssion at https://reviews.llvm.org/D91460 for more details. 3098 do { 3099 if (parseEscapedString(Data)) 3100 return true; 3101 getStreamer().emitBytes(Data); 3102 } while (!ZeroTerminated && getTok().is(AsmToken::String)); 3103 if (ZeroTerminated) 3104 getStreamer().emitBytes(StringRef("\0", 1)); 3105 return false; 3106 }; 3107 3108 return parseMany(parseOp); 3109 } 3110 3111 /// parseDirectiveReloc 3112 /// ::= .reloc expression , identifier [ , expression ] 3113 bool AsmParser::parseDirectiveReloc(SMLoc DirectiveLoc) { 3114 const MCExpr *Offset; 3115 const MCExpr *Expr = nullptr; 3116 SMLoc OffsetLoc = Lexer.getTok().getLoc(); 3117 3118 if (parseExpression(Offset)) 3119 return true; 3120 if (parseComma() || 3121 check(getTok().isNot(AsmToken::Identifier), "expected relocation name")) 3122 return true; 3123 3124 SMLoc NameLoc = Lexer.getTok().getLoc(); 3125 StringRef Name = Lexer.getTok().getIdentifier(); 3126 Lex(); 3127 3128 if (Lexer.is(AsmToken::Comma)) { 3129 Lex(); 3130 SMLoc ExprLoc = Lexer.getLoc(); 3131 if (parseExpression(Expr)) 3132 return true; 3133 3134 MCValue Value; 3135 if (!Expr->evaluateAsRelocatable(Value, nullptr, nullptr)) 3136 return Error(ExprLoc, "expression must be relocatable"); 3137 } 3138 3139 if (parseEOL()) 3140 return true; 3141 3142 const MCTargetAsmParser &MCT = getTargetParser(); 3143 const MCSubtargetInfo &STI = MCT.getSTI(); 3144 if (Optional<std::pair<bool, std::string>> Err = 3145 getStreamer().emitRelocDirective(*Offset, Name, Expr, DirectiveLoc, 3146 STI)) 3147 return Error(Err->first ? NameLoc : OffsetLoc, Err->second); 3148 3149 return false; 3150 } 3151 3152 /// parseDirectiveValue 3153 /// ::= (.byte | .short | ... ) [ expression (, expression)* ] 3154 bool AsmParser::parseDirectiveValue(StringRef IDVal, unsigned Size) { 3155 auto parseOp = [&]() -> bool { 3156 const MCExpr *Value; 3157 SMLoc ExprLoc = getLexer().getLoc(); 3158 if (checkForValidSection() || parseExpression(Value)) 3159 return true; 3160 // Special case constant expressions to match code generator. 3161 if (const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value)) { 3162 assert(Size <= 8 && "Invalid size"); 3163 uint64_t IntValue = MCE->getValue(); 3164 if (!isUIntN(8 * Size, IntValue) && !isIntN(8 * Size, IntValue)) 3165 return Error(ExprLoc, "out of range literal value"); 3166 getStreamer().emitIntValue(IntValue, Size); 3167 } else 3168 getStreamer().emitValue(Value, Size, ExprLoc); 3169 return false; 3170 }; 3171 3172 return parseMany(parseOp); 3173 } 3174 3175 static bool parseHexOcta(AsmParser &Asm, uint64_t &hi, uint64_t &lo) { 3176 if (Asm.getTok().isNot(AsmToken::Integer) && 3177 Asm.getTok().isNot(AsmToken::BigNum)) 3178 return Asm.TokError("unknown token in expression"); 3179 SMLoc ExprLoc = Asm.getTok().getLoc(); 3180 APInt IntValue = Asm.getTok().getAPIntVal(); 3181 Asm.Lex(); 3182 if (!IntValue.isIntN(128)) 3183 return Asm.Error(ExprLoc, "out of range literal value"); 3184 if (!IntValue.isIntN(64)) { 3185 hi = IntValue.getHiBits(IntValue.getBitWidth() - 64).getZExtValue(); 3186 lo = IntValue.getLoBits(64).getZExtValue(); 3187 } else { 3188 hi = 0; 3189 lo = IntValue.getZExtValue(); 3190 } 3191 return false; 3192 } 3193 3194 /// ParseDirectiveOctaValue 3195 /// ::= .octa [ hexconstant (, hexconstant)* ] 3196 3197 bool AsmParser::parseDirectiveOctaValue(StringRef IDVal) { 3198 auto parseOp = [&]() -> bool { 3199 if (checkForValidSection()) 3200 return true; 3201 uint64_t hi, lo; 3202 if (parseHexOcta(*this, hi, lo)) 3203 return true; 3204 if (MAI.isLittleEndian()) { 3205 getStreamer().emitInt64(lo); 3206 getStreamer().emitInt64(hi); 3207 } else { 3208 getStreamer().emitInt64(hi); 3209 getStreamer().emitInt64(lo); 3210 } 3211 return false; 3212 }; 3213 3214 return parseMany(parseOp); 3215 } 3216 3217 bool AsmParser::parseRealValue(const fltSemantics &Semantics, APInt &Res) { 3218 // We don't truly support arithmetic on floating point expressions, so we 3219 // have to manually parse unary prefixes. 3220 bool IsNeg = false; 3221 if (getLexer().is(AsmToken::Minus)) { 3222 Lexer.Lex(); 3223 IsNeg = true; 3224 } else if (getLexer().is(AsmToken::Plus)) 3225 Lexer.Lex(); 3226 3227 if (Lexer.is(AsmToken::Error)) 3228 return TokError(Lexer.getErr()); 3229 if (Lexer.isNot(AsmToken::Integer) && Lexer.isNot(AsmToken::Real) && 3230 Lexer.isNot(AsmToken::Identifier)) 3231 return TokError("unexpected token in directive"); 3232 3233 // Convert to an APFloat. 3234 APFloat Value(Semantics); 3235 StringRef IDVal = getTok().getString(); 3236 if (getLexer().is(AsmToken::Identifier)) { 3237 if (!IDVal.compare_lower("infinity") || !IDVal.compare_lower("inf")) 3238 Value = APFloat::getInf(Semantics); 3239 else if (!IDVal.compare_lower("nan")) 3240 Value = APFloat::getNaN(Semantics, false, ~0); 3241 else 3242 return TokError("invalid floating point literal"); 3243 } else if (errorToBool( 3244 Value.convertFromString(IDVal, APFloat::rmNearestTiesToEven) 3245 .takeError())) 3246 return TokError("invalid floating point literal"); 3247 if (IsNeg) 3248 Value.changeSign(); 3249 3250 // Consume the numeric token. 3251 Lex(); 3252 3253 Res = Value.bitcastToAPInt(); 3254 3255 return false; 3256 } 3257 3258 /// parseDirectiveRealValue 3259 /// ::= (.single | .double) [ expression (, expression)* ] 3260 bool AsmParser::parseDirectiveRealValue(StringRef IDVal, 3261 const fltSemantics &Semantics) { 3262 auto parseOp = [&]() -> bool { 3263 APInt AsInt; 3264 if (checkForValidSection() || parseRealValue(Semantics, AsInt)) 3265 return true; 3266 getStreamer().emitIntValue(AsInt.getLimitedValue(), 3267 AsInt.getBitWidth() / 8); 3268 return false; 3269 }; 3270 3271 return parseMany(parseOp); 3272 } 3273 3274 /// parseDirectiveZero 3275 /// ::= .zero expression 3276 bool AsmParser::parseDirectiveZero() { 3277 SMLoc NumBytesLoc = Lexer.getLoc(); 3278 const MCExpr *NumBytes; 3279 if (checkForValidSection() || parseExpression(NumBytes)) 3280 return true; 3281 3282 int64_t Val = 0; 3283 if (getLexer().is(AsmToken::Comma)) { 3284 Lex(); 3285 if (parseAbsoluteExpression(Val)) 3286 return true; 3287 } 3288 3289 if (parseEOL()) 3290 return true; 3291 getStreamer().emitFill(*NumBytes, Val, NumBytesLoc); 3292 3293 return false; 3294 } 3295 3296 /// parseDirectiveFill 3297 /// ::= .fill expression [ , expression [ , expression ] ] 3298 bool AsmParser::parseDirectiveFill() { 3299 SMLoc NumValuesLoc = Lexer.getLoc(); 3300 const MCExpr *NumValues; 3301 if (checkForValidSection() || parseExpression(NumValues)) 3302 return true; 3303 3304 int64_t FillSize = 1; 3305 int64_t FillExpr = 0; 3306 3307 SMLoc SizeLoc, ExprLoc; 3308 3309 if (parseOptionalToken(AsmToken::Comma)) { 3310 SizeLoc = getTok().getLoc(); 3311 if (parseAbsoluteExpression(FillSize)) 3312 return true; 3313 if (parseOptionalToken(AsmToken::Comma)) { 3314 ExprLoc = getTok().getLoc(); 3315 if (parseAbsoluteExpression(FillExpr)) 3316 return true; 3317 } 3318 } 3319 if (parseEOL()) 3320 return true; 3321 3322 if (FillSize < 0) { 3323 Warning(SizeLoc, "'.fill' directive with negative size has no effect"); 3324 return false; 3325 } 3326 if (FillSize > 8) { 3327 Warning(SizeLoc, "'.fill' directive with size greater than 8 has been truncated to 8"); 3328 FillSize = 8; 3329 } 3330 3331 if (!isUInt<32>(FillExpr) && FillSize > 4) 3332 Warning(ExprLoc, "'.fill' directive pattern has been truncated to 32-bits"); 3333 3334 getStreamer().emitFill(*NumValues, FillSize, FillExpr, NumValuesLoc); 3335 3336 return false; 3337 } 3338 3339 /// parseDirectiveOrg 3340 /// ::= .org expression [ , expression ] 3341 bool AsmParser::parseDirectiveOrg() { 3342 const MCExpr *Offset; 3343 SMLoc OffsetLoc = Lexer.getLoc(); 3344 if (checkForValidSection() || parseExpression(Offset)) 3345 return true; 3346 3347 // Parse optional fill expression. 3348 int64_t FillExpr = 0; 3349 if (parseOptionalToken(AsmToken::Comma)) 3350 if (parseAbsoluteExpression(FillExpr)) 3351 return true; 3352 if (parseEOL()) 3353 return true; 3354 3355 getStreamer().emitValueToOffset(Offset, FillExpr, OffsetLoc); 3356 return false; 3357 } 3358 3359 /// parseDirectiveAlign 3360 /// ::= {.align, ...} expression [ , expression [ , expression ]] 3361 bool AsmParser::parseDirectiveAlign(bool IsPow2, unsigned ValueSize) { 3362 SMLoc AlignmentLoc = getLexer().getLoc(); 3363 int64_t Alignment; 3364 SMLoc MaxBytesLoc; 3365 bool HasFillExpr = false; 3366 int64_t FillExpr = 0; 3367 int64_t MaxBytesToFill = 0; 3368 3369 auto parseAlign = [&]() -> bool { 3370 if (parseAbsoluteExpression(Alignment)) 3371 return true; 3372 if (parseOptionalToken(AsmToken::Comma)) { 3373 // The fill expression can be omitted while specifying a maximum number of 3374 // alignment bytes, e.g: 3375 // .align 3,,4 3376 if (getTok().isNot(AsmToken::Comma)) { 3377 HasFillExpr = true; 3378 if (parseAbsoluteExpression(FillExpr)) 3379 return true; 3380 } 3381 if (parseOptionalToken(AsmToken::Comma)) 3382 if (parseTokenLoc(MaxBytesLoc) || 3383 parseAbsoluteExpression(MaxBytesToFill)) 3384 return true; 3385 } 3386 return parseEOL(); 3387 }; 3388 3389 if (checkForValidSection()) 3390 return true; 3391 // Ignore empty '.p2align' directives for GNU-as compatibility 3392 if (IsPow2 && (ValueSize == 1) && getTok().is(AsmToken::EndOfStatement)) { 3393 Warning(AlignmentLoc, "p2align directive with no operand(s) is ignored"); 3394 return parseEOL(); 3395 } 3396 if (parseAlign()) 3397 return true; 3398 3399 // Always emit an alignment here even if we thrown an error. 3400 bool ReturnVal = false; 3401 3402 // Compute alignment in bytes. 3403 if (IsPow2) { 3404 // FIXME: Diagnose overflow. 3405 if (Alignment >= 32) { 3406 ReturnVal |= Error(AlignmentLoc, "invalid alignment value"); 3407 Alignment = 31; 3408 } 3409 3410 Alignment = 1ULL << Alignment; 3411 } else { 3412 // Reject alignments that aren't either a power of two or zero, 3413 // for gas compatibility. Alignment of zero is silently rounded 3414 // up to one. 3415 if (Alignment == 0) 3416 Alignment = 1; 3417 if (!isPowerOf2_64(Alignment)) 3418 ReturnVal |= Error(AlignmentLoc, "alignment must be a power of 2"); 3419 if (!isUInt<32>(Alignment)) 3420 ReturnVal |= Error(AlignmentLoc, "alignment must be smaller than 2**32"); 3421 } 3422 3423 // Diagnose non-sensical max bytes to align. 3424 if (MaxBytesLoc.isValid()) { 3425 if (MaxBytesToFill < 1) { 3426 ReturnVal |= Error(MaxBytesLoc, 3427 "alignment directive can never be satisfied in this " 3428 "many bytes, ignoring maximum bytes expression"); 3429 MaxBytesToFill = 0; 3430 } 3431 3432 if (MaxBytesToFill >= Alignment) { 3433 Warning(MaxBytesLoc, "maximum bytes expression exceeds alignment and " 3434 "has no effect"); 3435 MaxBytesToFill = 0; 3436 } 3437 } 3438 3439 // Check whether we should use optimal code alignment for this .align 3440 // directive. 3441 const MCSection *Section = getStreamer().getCurrentSectionOnly(); 3442 assert(Section && "must have section to emit alignment"); 3443 bool UseCodeAlign = Section->UseCodeAlign(); 3444 if ((!HasFillExpr || Lexer.getMAI().getTextAlignFillValue() == FillExpr) && 3445 ValueSize == 1 && UseCodeAlign) { 3446 getStreamer().emitCodeAlignment(Alignment, MaxBytesToFill); 3447 } else { 3448 // FIXME: Target specific behavior about how the "extra" bytes are filled. 3449 getStreamer().emitValueToAlignment(Alignment, FillExpr, ValueSize, 3450 MaxBytesToFill); 3451 } 3452 3453 return ReturnVal; 3454 } 3455 3456 /// parseDirectiveFile 3457 /// ::= .file filename 3458 /// ::= .file number [directory] filename [md5 checksum] [source source-text] 3459 bool AsmParser::parseDirectiveFile(SMLoc DirectiveLoc) { 3460 // FIXME: I'm not sure what this is. 3461 int64_t FileNumber = -1; 3462 if (getLexer().is(AsmToken::Integer)) { 3463 FileNumber = getTok().getIntVal(); 3464 Lex(); 3465 3466 if (FileNumber < 0) 3467 return TokError("negative file number"); 3468 } 3469 3470 std::string Path; 3471 3472 // Usually the directory and filename together, otherwise just the directory. 3473 // Allow the strings to have escaped octal character sequence. 3474 if (parseEscapedString(Path)) 3475 return true; 3476 3477 StringRef Directory; 3478 StringRef Filename; 3479 std::string FilenameData; 3480 if (getLexer().is(AsmToken::String)) { 3481 if (check(FileNumber == -1, 3482 "explicit path specified, but no file number") || 3483 parseEscapedString(FilenameData)) 3484 return true; 3485 Filename = FilenameData; 3486 Directory = Path; 3487 } else { 3488 Filename = Path; 3489 } 3490 3491 uint64_t MD5Hi, MD5Lo; 3492 bool HasMD5 = false; 3493 3494 Optional<StringRef> Source; 3495 bool HasSource = false; 3496 std::string SourceString; 3497 3498 while (!parseOptionalToken(AsmToken::EndOfStatement)) { 3499 StringRef Keyword; 3500 if (check(getTok().isNot(AsmToken::Identifier), 3501 "unexpected token in '.file' directive") || 3502 parseIdentifier(Keyword)) 3503 return true; 3504 if (Keyword == "md5") { 3505 HasMD5 = true; 3506 if (check(FileNumber == -1, 3507 "MD5 checksum specified, but no file number") || 3508 parseHexOcta(*this, MD5Hi, MD5Lo)) 3509 return true; 3510 } else if (Keyword == "source") { 3511 HasSource = true; 3512 if (check(FileNumber == -1, 3513 "source specified, but no file number") || 3514 check(getTok().isNot(AsmToken::String), 3515 "unexpected token in '.file' directive") || 3516 parseEscapedString(SourceString)) 3517 return true; 3518 } else { 3519 return TokError("unexpected token in '.file' directive"); 3520 } 3521 } 3522 3523 if (FileNumber == -1) { 3524 // Ignore the directive if there is no number and the target doesn't support 3525 // numberless .file directives. This allows some portability of assembler 3526 // between different object file formats. 3527 if (getContext().getAsmInfo()->hasSingleParameterDotFile()) 3528 getStreamer().emitFileDirective(Filename); 3529 } else { 3530 // In case there is a -g option as well as debug info from directive .file, 3531 // we turn off the -g option, directly use the existing debug info instead. 3532 // Throw away any implicit file table for the assembler source. 3533 if (Ctx.getGenDwarfForAssembly()) { 3534 Ctx.getMCDwarfLineTable(0).resetFileTable(); 3535 Ctx.setGenDwarfForAssembly(false); 3536 } 3537 3538 Optional<MD5::MD5Result> CKMem; 3539 if (HasMD5) { 3540 MD5::MD5Result Sum; 3541 for (unsigned i = 0; i != 8; ++i) { 3542 Sum.Bytes[i] = uint8_t(MD5Hi >> ((7 - i) * 8)); 3543 Sum.Bytes[i + 8] = uint8_t(MD5Lo >> ((7 - i) * 8)); 3544 } 3545 CKMem = Sum; 3546 } 3547 if (HasSource) { 3548 char *SourceBuf = static_cast<char *>(Ctx.allocate(SourceString.size())); 3549 memcpy(SourceBuf, SourceString.data(), SourceString.size()); 3550 Source = StringRef(SourceBuf, SourceString.size()); 3551 } 3552 if (FileNumber == 0) { 3553 // Upgrade to Version 5 for assembly actions like clang -c a.s. 3554 if (Ctx.getDwarfVersion() < 5) 3555 Ctx.setDwarfVersion(5); 3556 getStreamer().emitDwarfFile0Directive(Directory, Filename, CKMem, Source); 3557 } else { 3558 Expected<unsigned> FileNumOrErr = getStreamer().tryEmitDwarfFileDirective( 3559 FileNumber, Directory, Filename, CKMem, Source); 3560 if (!FileNumOrErr) 3561 return Error(DirectiveLoc, toString(FileNumOrErr.takeError())); 3562 } 3563 // Alert the user if there are some .file directives with MD5 and some not. 3564 // But only do that once. 3565 if (!ReportedInconsistentMD5 && !Ctx.isDwarfMD5UsageConsistent(0)) { 3566 ReportedInconsistentMD5 = true; 3567 return Warning(DirectiveLoc, "inconsistent use of MD5 checksums"); 3568 } 3569 } 3570 3571 return false; 3572 } 3573 3574 /// parseDirectiveLine 3575 /// ::= .line [number] 3576 bool AsmParser::parseDirectiveLine() { 3577 int64_t LineNumber; 3578 if (getLexer().is(AsmToken::Integer)) { 3579 if (parseIntToken(LineNumber, "unexpected token in '.line' directive")) 3580 return true; 3581 (void)LineNumber; 3582 // FIXME: Do something with the .line. 3583 } 3584 return parseEOL(); 3585 } 3586 3587 /// parseDirectiveLoc 3588 /// ::= .loc FileNumber [LineNumber] [ColumnPos] [basic_block] [prologue_end] 3589 /// [epilogue_begin] [is_stmt VALUE] [isa VALUE] 3590 /// The first number is a file number, must have been previously assigned with 3591 /// a .file directive, the second number is the line number and optionally the 3592 /// third number is a column position (zero if not specified). The remaining 3593 /// optional items are .loc sub-directives. 3594 bool AsmParser::parseDirectiveLoc() { 3595 int64_t FileNumber = 0, LineNumber = 0; 3596 SMLoc Loc = getTok().getLoc(); 3597 if (parseIntToken(FileNumber, "unexpected token in '.loc' directive") || 3598 check(FileNumber < 1 && Ctx.getDwarfVersion() < 5, Loc, 3599 "file number less than one in '.loc' directive") || 3600 check(!getContext().isValidDwarfFileNumber(FileNumber), Loc, 3601 "unassigned file number in '.loc' directive")) 3602 return true; 3603 3604 // optional 3605 if (getLexer().is(AsmToken::Integer)) { 3606 LineNumber = getTok().getIntVal(); 3607 if (LineNumber < 0) 3608 return TokError("line number less than zero in '.loc' directive"); 3609 Lex(); 3610 } 3611 3612 int64_t ColumnPos = 0; 3613 if (getLexer().is(AsmToken::Integer)) { 3614 ColumnPos = getTok().getIntVal(); 3615 if (ColumnPos < 0) 3616 return TokError("column position less than zero in '.loc' directive"); 3617 Lex(); 3618 } 3619 3620 auto PrevFlags = getContext().getCurrentDwarfLoc().getFlags(); 3621 unsigned Flags = PrevFlags & DWARF2_FLAG_IS_STMT; 3622 unsigned Isa = 0; 3623 int64_t Discriminator = 0; 3624 3625 auto parseLocOp = [&]() -> bool { 3626 StringRef Name; 3627 SMLoc Loc = getTok().getLoc(); 3628 if (parseIdentifier(Name)) 3629 return TokError("unexpected token in '.loc' directive"); 3630 3631 if (Name == "basic_block") 3632 Flags |= DWARF2_FLAG_BASIC_BLOCK; 3633 else if (Name == "prologue_end") 3634 Flags |= DWARF2_FLAG_PROLOGUE_END; 3635 else if (Name == "epilogue_begin") 3636 Flags |= DWARF2_FLAG_EPILOGUE_BEGIN; 3637 else if (Name == "is_stmt") { 3638 Loc = getTok().getLoc(); 3639 const MCExpr *Value; 3640 if (parseExpression(Value)) 3641 return true; 3642 // The expression must be the constant 0 or 1. 3643 if (const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value)) { 3644 int Value = MCE->getValue(); 3645 if (Value == 0) 3646 Flags &= ~DWARF2_FLAG_IS_STMT; 3647 else if (Value == 1) 3648 Flags |= DWARF2_FLAG_IS_STMT; 3649 else 3650 return Error(Loc, "is_stmt value not 0 or 1"); 3651 } else { 3652 return Error(Loc, "is_stmt value not the constant value of 0 or 1"); 3653 } 3654 } else if (Name == "isa") { 3655 Loc = getTok().getLoc(); 3656 const MCExpr *Value; 3657 if (parseExpression(Value)) 3658 return true; 3659 // The expression must be a constant greater or equal to 0. 3660 if (const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value)) { 3661 int Value = MCE->getValue(); 3662 if (Value < 0) 3663 return Error(Loc, "isa number less than zero"); 3664 Isa = Value; 3665 } else { 3666 return Error(Loc, "isa number not a constant value"); 3667 } 3668 } else if (Name == "discriminator") { 3669 if (parseAbsoluteExpression(Discriminator)) 3670 return true; 3671 } else { 3672 return Error(Loc, "unknown sub-directive in '.loc' directive"); 3673 } 3674 return false; 3675 }; 3676 3677 if (parseMany(parseLocOp, false /*hasComma*/)) 3678 return true; 3679 3680 getStreamer().emitDwarfLocDirective(FileNumber, LineNumber, ColumnPos, Flags, 3681 Isa, Discriminator, StringRef()); 3682 3683 return false; 3684 } 3685 3686 /// parseDirectiveStabs 3687 /// ::= .stabs string, number, number, number 3688 bool AsmParser::parseDirectiveStabs() { 3689 return TokError("unsupported directive '.stabs'"); 3690 } 3691 3692 /// parseDirectiveCVFile 3693 /// ::= .cv_file number filename [checksum] [checksumkind] 3694 bool AsmParser::parseDirectiveCVFile() { 3695 SMLoc FileNumberLoc = getTok().getLoc(); 3696 int64_t FileNumber; 3697 std::string Filename; 3698 std::string Checksum; 3699 int64_t ChecksumKind = 0; 3700 3701 if (parseIntToken(FileNumber, 3702 "expected file number in '.cv_file' directive") || 3703 check(FileNumber < 1, FileNumberLoc, "file number less than one") || 3704 check(getTok().isNot(AsmToken::String), 3705 "unexpected token in '.cv_file' directive") || 3706 parseEscapedString(Filename)) 3707 return true; 3708 if (!parseOptionalToken(AsmToken::EndOfStatement)) { 3709 if (check(getTok().isNot(AsmToken::String), 3710 "unexpected token in '.cv_file' directive") || 3711 parseEscapedString(Checksum) || 3712 parseIntToken(ChecksumKind, 3713 "expected checksum kind in '.cv_file' directive") || 3714 parseToken(AsmToken::EndOfStatement, 3715 "unexpected token in '.cv_file' directive")) 3716 return true; 3717 } 3718 3719 Checksum = fromHex(Checksum); 3720 void *CKMem = Ctx.allocate(Checksum.size(), 1); 3721 memcpy(CKMem, Checksum.data(), Checksum.size()); 3722 ArrayRef<uint8_t> ChecksumAsBytes(reinterpret_cast<const uint8_t *>(CKMem), 3723 Checksum.size()); 3724 3725 if (!getStreamer().EmitCVFileDirective(FileNumber, Filename, ChecksumAsBytes, 3726 static_cast<uint8_t>(ChecksumKind))) 3727 return Error(FileNumberLoc, "file number already allocated"); 3728 3729 return false; 3730 } 3731 3732 bool AsmParser::parseCVFunctionId(int64_t &FunctionId, 3733 StringRef DirectiveName) { 3734 SMLoc Loc; 3735 return parseTokenLoc(Loc) || 3736 parseIntToken(FunctionId, "expected function id in '" + DirectiveName + 3737 "' directive") || 3738 check(FunctionId < 0 || FunctionId >= UINT_MAX, Loc, 3739 "expected function id within range [0, UINT_MAX)"); 3740 } 3741 3742 bool AsmParser::parseCVFileId(int64_t &FileNumber, StringRef DirectiveName) { 3743 SMLoc Loc; 3744 return parseTokenLoc(Loc) || 3745 parseIntToken(FileNumber, "expected integer in '" + DirectiveName + 3746 "' directive") || 3747 check(FileNumber < 1, Loc, "file number less than one in '" + 3748 DirectiveName + "' directive") || 3749 check(!getCVContext().isValidFileNumber(FileNumber), Loc, 3750 "unassigned file number in '" + DirectiveName + "' directive"); 3751 } 3752 3753 /// parseDirectiveCVFuncId 3754 /// ::= .cv_func_id FunctionId 3755 /// 3756 /// Introduces a function ID that can be used with .cv_loc. 3757 bool AsmParser::parseDirectiveCVFuncId() { 3758 SMLoc FunctionIdLoc = getTok().getLoc(); 3759 int64_t FunctionId; 3760 3761 if (parseCVFunctionId(FunctionId, ".cv_func_id") || 3762 parseToken(AsmToken::EndOfStatement, 3763 "unexpected token in '.cv_func_id' directive")) 3764 return true; 3765 3766 if (!getStreamer().EmitCVFuncIdDirective(FunctionId)) 3767 return Error(FunctionIdLoc, "function id already allocated"); 3768 3769 return false; 3770 } 3771 3772 /// parseDirectiveCVInlineSiteId 3773 /// ::= .cv_inline_site_id FunctionId 3774 /// "within" IAFunc 3775 /// "inlined_at" IAFile IALine [IACol] 3776 /// 3777 /// Introduces a function ID that can be used with .cv_loc. Includes "inlined 3778 /// at" source location information for use in the line table of the caller, 3779 /// whether the caller is a real function or another inlined call site. 3780 bool AsmParser::parseDirectiveCVInlineSiteId() { 3781 SMLoc FunctionIdLoc = getTok().getLoc(); 3782 int64_t FunctionId; 3783 int64_t IAFunc; 3784 int64_t IAFile; 3785 int64_t IALine; 3786 int64_t IACol = 0; 3787 3788 // FunctionId 3789 if (parseCVFunctionId(FunctionId, ".cv_inline_site_id")) 3790 return true; 3791 3792 // "within" 3793 if (check((getLexer().isNot(AsmToken::Identifier) || 3794 getTok().getIdentifier() != "within"), 3795 "expected 'within' identifier in '.cv_inline_site_id' directive")) 3796 return true; 3797 Lex(); 3798 3799 // IAFunc 3800 if (parseCVFunctionId(IAFunc, ".cv_inline_site_id")) 3801 return true; 3802 3803 // "inlined_at" 3804 if (check((getLexer().isNot(AsmToken::Identifier) || 3805 getTok().getIdentifier() != "inlined_at"), 3806 "expected 'inlined_at' identifier in '.cv_inline_site_id' " 3807 "directive") ) 3808 return true; 3809 Lex(); 3810 3811 // IAFile IALine 3812 if (parseCVFileId(IAFile, ".cv_inline_site_id") || 3813 parseIntToken(IALine, "expected line number after 'inlined_at'")) 3814 return true; 3815 3816 // [IACol] 3817 if (getLexer().is(AsmToken::Integer)) { 3818 IACol = getTok().getIntVal(); 3819 Lex(); 3820 } 3821 3822 if (parseToken(AsmToken::EndOfStatement, 3823 "unexpected token in '.cv_inline_site_id' directive")) 3824 return true; 3825 3826 if (!getStreamer().EmitCVInlineSiteIdDirective(FunctionId, IAFunc, IAFile, 3827 IALine, IACol, FunctionIdLoc)) 3828 return Error(FunctionIdLoc, "function id already allocated"); 3829 3830 return false; 3831 } 3832 3833 /// parseDirectiveCVLoc 3834 /// ::= .cv_loc FunctionId FileNumber [LineNumber] [ColumnPos] [prologue_end] 3835 /// [is_stmt VALUE] 3836 /// The first number is a file number, must have been previously assigned with 3837 /// a .file directive, the second number is the line number and optionally the 3838 /// third number is a column position (zero if not specified). The remaining 3839 /// optional items are .loc sub-directives. 3840 bool AsmParser::parseDirectiveCVLoc() { 3841 SMLoc DirectiveLoc = getTok().getLoc(); 3842 int64_t FunctionId, FileNumber; 3843 if (parseCVFunctionId(FunctionId, ".cv_loc") || 3844 parseCVFileId(FileNumber, ".cv_loc")) 3845 return true; 3846 3847 int64_t LineNumber = 0; 3848 if (getLexer().is(AsmToken::Integer)) { 3849 LineNumber = getTok().getIntVal(); 3850 if (LineNumber < 0) 3851 return TokError("line number less than zero in '.cv_loc' directive"); 3852 Lex(); 3853 } 3854 3855 int64_t ColumnPos = 0; 3856 if (getLexer().is(AsmToken::Integer)) { 3857 ColumnPos = getTok().getIntVal(); 3858 if (ColumnPos < 0) 3859 return TokError("column position less than zero in '.cv_loc' directive"); 3860 Lex(); 3861 } 3862 3863 bool PrologueEnd = false; 3864 uint64_t IsStmt = 0; 3865 3866 auto parseOp = [&]() -> bool { 3867 StringRef Name; 3868 SMLoc Loc = getTok().getLoc(); 3869 if (parseIdentifier(Name)) 3870 return TokError("unexpected token in '.cv_loc' directive"); 3871 if (Name == "prologue_end") 3872 PrologueEnd = true; 3873 else if (Name == "is_stmt") { 3874 Loc = getTok().getLoc(); 3875 const MCExpr *Value; 3876 if (parseExpression(Value)) 3877 return true; 3878 // The expression must be the constant 0 or 1. 3879 IsStmt = ~0ULL; 3880 if (const auto *MCE = dyn_cast<MCConstantExpr>(Value)) 3881 IsStmt = MCE->getValue(); 3882 3883 if (IsStmt > 1) 3884 return Error(Loc, "is_stmt value not 0 or 1"); 3885 } else { 3886 return Error(Loc, "unknown sub-directive in '.cv_loc' directive"); 3887 } 3888 return false; 3889 }; 3890 3891 if (parseMany(parseOp, false /*hasComma*/)) 3892 return true; 3893 3894 getStreamer().emitCVLocDirective(FunctionId, FileNumber, LineNumber, 3895 ColumnPos, PrologueEnd, IsStmt, StringRef(), 3896 DirectiveLoc); 3897 return false; 3898 } 3899 3900 /// parseDirectiveCVLinetable 3901 /// ::= .cv_linetable FunctionId, FnStart, FnEnd 3902 bool AsmParser::parseDirectiveCVLinetable() { 3903 int64_t FunctionId; 3904 StringRef FnStartName, FnEndName; 3905 SMLoc Loc = getTok().getLoc(); 3906 if (parseCVFunctionId(FunctionId, ".cv_linetable") || parseComma() || 3907 parseTokenLoc(Loc) || 3908 check(parseIdentifier(FnStartName), Loc, 3909 "expected identifier in directive") || 3910 parseComma() || parseTokenLoc(Loc) || 3911 check(parseIdentifier(FnEndName), Loc, 3912 "expected identifier in directive")) 3913 return true; 3914 3915 MCSymbol *FnStartSym = getContext().getOrCreateSymbol(FnStartName); 3916 MCSymbol *FnEndSym = getContext().getOrCreateSymbol(FnEndName); 3917 3918 getStreamer().emitCVLinetableDirective(FunctionId, FnStartSym, FnEndSym); 3919 return false; 3920 } 3921 3922 /// parseDirectiveCVInlineLinetable 3923 /// ::= .cv_inline_linetable PrimaryFunctionId FileId LineNum FnStart FnEnd 3924 bool AsmParser::parseDirectiveCVInlineLinetable() { 3925 int64_t PrimaryFunctionId, SourceFileId, SourceLineNum; 3926 StringRef FnStartName, FnEndName; 3927 SMLoc Loc = getTok().getLoc(); 3928 if (parseCVFunctionId(PrimaryFunctionId, ".cv_inline_linetable") || 3929 parseTokenLoc(Loc) || 3930 parseIntToken( 3931 SourceFileId, 3932 "expected SourceField in '.cv_inline_linetable' directive") || 3933 check(SourceFileId <= 0, Loc, 3934 "File id less than zero in '.cv_inline_linetable' directive") || 3935 parseTokenLoc(Loc) || 3936 parseIntToken( 3937 SourceLineNum, 3938 "expected SourceLineNum in '.cv_inline_linetable' directive") || 3939 check(SourceLineNum < 0, Loc, 3940 "Line number less than zero in '.cv_inline_linetable' directive") || 3941 parseTokenLoc(Loc) || check(parseIdentifier(FnStartName), Loc, 3942 "expected identifier in directive") || 3943 parseTokenLoc(Loc) || check(parseIdentifier(FnEndName), Loc, 3944 "expected identifier in directive")) 3945 return true; 3946 3947 if (parseToken(AsmToken::EndOfStatement, "Expected End of Statement")) 3948 return true; 3949 3950 MCSymbol *FnStartSym = getContext().getOrCreateSymbol(FnStartName); 3951 MCSymbol *FnEndSym = getContext().getOrCreateSymbol(FnEndName); 3952 getStreamer().emitCVInlineLinetableDirective(PrimaryFunctionId, SourceFileId, 3953 SourceLineNum, FnStartSym, 3954 FnEndSym); 3955 return false; 3956 } 3957 3958 void AsmParser::initializeCVDefRangeTypeMap() { 3959 CVDefRangeTypeMap["reg"] = CVDR_DEFRANGE_REGISTER; 3960 CVDefRangeTypeMap["frame_ptr_rel"] = CVDR_DEFRANGE_FRAMEPOINTER_REL; 3961 CVDefRangeTypeMap["subfield_reg"] = CVDR_DEFRANGE_SUBFIELD_REGISTER; 3962 CVDefRangeTypeMap["reg_rel"] = CVDR_DEFRANGE_REGISTER_REL; 3963 } 3964 3965 /// parseDirectiveCVDefRange 3966 /// ::= .cv_def_range RangeStart RangeEnd (GapStart GapEnd)*, bytes* 3967 bool AsmParser::parseDirectiveCVDefRange() { 3968 SMLoc Loc; 3969 std::vector<std::pair<const MCSymbol *, const MCSymbol *>> Ranges; 3970 while (getLexer().is(AsmToken::Identifier)) { 3971 Loc = getLexer().getLoc(); 3972 StringRef GapStartName; 3973 if (parseIdentifier(GapStartName)) 3974 return Error(Loc, "expected identifier in directive"); 3975 MCSymbol *GapStartSym = getContext().getOrCreateSymbol(GapStartName); 3976 3977 Loc = getLexer().getLoc(); 3978 StringRef GapEndName; 3979 if (parseIdentifier(GapEndName)) 3980 return Error(Loc, "expected identifier in directive"); 3981 MCSymbol *GapEndSym = getContext().getOrCreateSymbol(GapEndName); 3982 3983 Ranges.push_back({GapStartSym, GapEndSym}); 3984 } 3985 3986 StringRef CVDefRangeTypeStr; 3987 if (parseToken( 3988 AsmToken::Comma, 3989 "expected comma before def_range type in .cv_def_range directive") || 3990 parseIdentifier(CVDefRangeTypeStr)) 3991 return Error(Loc, "expected def_range type in directive"); 3992 3993 StringMap<CVDefRangeType>::const_iterator CVTypeIt = 3994 CVDefRangeTypeMap.find(CVDefRangeTypeStr); 3995 CVDefRangeType CVDRType = (CVTypeIt == CVDefRangeTypeMap.end()) 3996 ? CVDR_DEFRANGE 3997 : CVTypeIt->getValue(); 3998 switch (CVDRType) { 3999 case CVDR_DEFRANGE_REGISTER: { 4000 int64_t DRRegister; 4001 if (parseToken(AsmToken::Comma, "expected comma before register number in " 4002 ".cv_def_range directive") || 4003 parseAbsoluteExpression(DRRegister)) 4004 return Error(Loc, "expected register number"); 4005 4006 codeview::DefRangeRegisterHeader DRHdr; 4007 DRHdr.Register = DRRegister; 4008 DRHdr.MayHaveNoName = 0; 4009 getStreamer().emitCVDefRangeDirective(Ranges, DRHdr); 4010 break; 4011 } 4012 case CVDR_DEFRANGE_FRAMEPOINTER_REL: { 4013 int64_t DROffset; 4014 if (parseToken(AsmToken::Comma, 4015 "expected comma before offset in .cv_def_range directive") || 4016 parseAbsoluteExpression(DROffset)) 4017 return Error(Loc, "expected offset value"); 4018 4019 codeview::DefRangeFramePointerRelHeader DRHdr; 4020 DRHdr.Offset = DROffset; 4021 getStreamer().emitCVDefRangeDirective(Ranges, DRHdr); 4022 break; 4023 } 4024 case CVDR_DEFRANGE_SUBFIELD_REGISTER: { 4025 int64_t DRRegister; 4026 int64_t DROffsetInParent; 4027 if (parseToken(AsmToken::Comma, "expected comma before register number in " 4028 ".cv_def_range directive") || 4029 parseAbsoluteExpression(DRRegister)) 4030 return Error(Loc, "expected register number"); 4031 if (parseToken(AsmToken::Comma, 4032 "expected comma before offset in .cv_def_range directive") || 4033 parseAbsoluteExpression(DROffsetInParent)) 4034 return Error(Loc, "expected offset value"); 4035 4036 codeview::DefRangeSubfieldRegisterHeader DRHdr; 4037 DRHdr.Register = DRRegister; 4038 DRHdr.MayHaveNoName = 0; 4039 DRHdr.OffsetInParent = DROffsetInParent; 4040 getStreamer().emitCVDefRangeDirective(Ranges, DRHdr); 4041 break; 4042 } 4043 case CVDR_DEFRANGE_REGISTER_REL: { 4044 int64_t DRRegister; 4045 int64_t DRFlags; 4046 int64_t DRBasePointerOffset; 4047 if (parseToken(AsmToken::Comma, "expected comma before register number in " 4048 ".cv_def_range directive") || 4049 parseAbsoluteExpression(DRRegister)) 4050 return Error(Loc, "expected register value"); 4051 if (parseToken( 4052 AsmToken::Comma, 4053 "expected comma before flag value in .cv_def_range directive") || 4054 parseAbsoluteExpression(DRFlags)) 4055 return Error(Loc, "expected flag value"); 4056 if (parseToken(AsmToken::Comma, "expected comma before base pointer offset " 4057 "in .cv_def_range directive") || 4058 parseAbsoluteExpression(DRBasePointerOffset)) 4059 return Error(Loc, "expected base pointer offset value"); 4060 4061 codeview::DefRangeRegisterRelHeader DRHdr; 4062 DRHdr.Register = DRRegister; 4063 DRHdr.Flags = DRFlags; 4064 DRHdr.BasePointerOffset = DRBasePointerOffset; 4065 getStreamer().emitCVDefRangeDirective(Ranges, DRHdr); 4066 break; 4067 } 4068 default: 4069 return Error(Loc, "unexpected def_range type in .cv_def_range directive"); 4070 } 4071 return true; 4072 } 4073 4074 /// parseDirectiveCVString 4075 /// ::= .cv_stringtable "string" 4076 bool AsmParser::parseDirectiveCVString() { 4077 std::string Data; 4078 if (checkForValidSection() || parseEscapedString(Data)) 4079 return true; 4080 4081 // Put the string in the table and emit the offset. 4082 std::pair<StringRef, unsigned> Insertion = 4083 getCVContext().addToStringTable(Data); 4084 getStreamer().emitInt32(Insertion.second); 4085 return false; 4086 } 4087 4088 /// parseDirectiveCVStringTable 4089 /// ::= .cv_stringtable 4090 bool AsmParser::parseDirectiveCVStringTable() { 4091 getStreamer().emitCVStringTableDirective(); 4092 return false; 4093 } 4094 4095 /// parseDirectiveCVFileChecksums 4096 /// ::= .cv_filechecksums 4097 bool AsmParser::parseDirectiveCVFileChecksums() { 4098 getStreamer().emitCVFileChecksumsDirective(); 4099 return false; 4100 } 4101 4102 /// parseDirectiveCVFileChecksumOffset 4103 /// ::= .cv_filechecksumoffset fileno 4104 bool AsmParser::parseDirectiveCVFileChecksumOffset() { 4105 int64_t FileNo; 4106 if (parseIntToken(FileNo, "expected identifier in directive")) 4107 return true; 4108 if (parseToken(AsmToken::EndOfStatement, "Expected End of Statement")) 4109 return true; 4110 getStreamer().emitCVFileChecksumOffsetDirective(FileNo); 4111 return false; 4112 } 4113 4114 /// parseDirectiveCVFPOData 4115 /// ::= .cv_fpo_data procsym 4116 bool AsmParser::parseDirectiveCVFPOData() { 4117 SMLoc DirLoc = getLexer().getLoc(); 4118 StringRef ProcName; 4119 if (parseIdentifier(ProcName)) 4120 return TokError("expected symbol name"); 4121 if (parseEOL()) 4122 return true; 4123 MCSymbol *ProcSym = getContext().getOrCreateSymbol(ProcName); 4124 getStreamer().EmitCVFPOData(ProcSym, DirLoc); 4125 return false; 4126 } 4127 4128 /// parseDirectiveCFISections 4129 /// ::= .cfi_sections section [, section] 4130 bool AsmParser::parseDirectiveCFISections() { 4131 StringRef Name; 4132 bool EH = false; 4133 bool Debug = false; 4134 4135 if (!parseOptionalToken(AsmToken::EndOfStatement)) { 4136 for (;;) { 4137 if (parseIdentifier(Name)) 4138 return TokError("expected .eh_frame or .debug_frame"); 4139 if (Name == ".eh_frame") 4140 EH = true; 4141 else if (Name == ".debug_frame") 4142 Debug = true; 4143 if (parseOptionalToken(AsmToken::EndOfStatement)) 4144 break; 4145 if (parseComma()) 4146 return true; 4147 } 4148 } 4149 getStreamer().emitCFISections(EH, Debug); 4150 return false; 4151 } 4152 4153 /// parseDirectiveCFIStartProc 4154 /// ::= .cfi_startproc [simple] 4155 bool AsmParser::parseDirectiveCFIStartProc() { 4156 StringRef Simple; 4157 if (!parseOptionalToken(AsmToken::EndOfStatement)) { 4158 if (check(parseIdentifier(Simple) || Simple != "simple", 4159 "unexpected token") || 4160 parseEOL()) 4161 return true; 4162 } 4163 4164 // TODO(kristina): Deal with a corner case of incorrect diagnostic context 4165 // being produced if this directive is emitted as part of preprocessor macro 4166 // expansion which can *ONLY* happen if Clang's cc1as is the API consumer. 4167 // Tools like llvm-mc on the other hand are not affected by it, and report 4168 // correct context information. 4169 getStreamer().emitCFIStartProc(!Simple.empty(), Lexer.getLoc()); 4170 return false; 4171 } 4172 4173 /// parseDirectiveCFIEndProc 4174 /// ::= .cfi_endproc 4175 bool AsmParser::parseDirectiveCFIEndProc() { 4176 if (parseEOL()) 4177 return true; 4178 getStreamer().emitCFIEndProc(); 4179 return false; 4180 } 4181 4182 /// parse register name or number. 4183 bool AsmParser::parseRegisterOrRegisterNumber(int64_t &Register, 4184 SMLoc DirectiveLoc) { 4185 unsigned RegNo; 4186 4187 if (getLexer().isNot(AsmToken::Integer)) { 4188 if (getTargetParser().ParseRegister(RegNo, DirectiveLoc, DirectiveLoc)) 4189 return true; 4190 Register = getContext().getRegisterInfo()->getDwarfRegNum(RegNo, true); 4191 } else 4192 return parseAbsoluteExpression(Register); 4193 4194 return false; 4195 } 4196 4197 /// parseDirectiveCFIDefCfa 4198 /// ::= .cfi_def_cfa register, offset 4199 bool AsmParser::parseDirectiveCFIDefCfa(SMLoc DirectiveLoc) { 4200 int64_t Register = 0, Offset = 0; 4201 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc) || parseComma() || 4202 parseAbsoluteExpression(Offset) || parseEOL()) 4203 return true; 4204 4205 getStreamer().emitCFIDefCfa(Register, Offset); 4206 return false; 4207 } 4208 4209 /// parseDirectiveCFIDefCfaOffset 4210 /// ::= .cfi_def_cfa_offset offset 4211 bool AsmParser::parseDirectiveCFIDefCfaOffset() { 4212 int64_t Offset = 0; 4213 if (parseAbsoluteExpression(Offset) || parseEOL()) 4214 return true; 4215 4216 getStreamer().emitCFIDefCfaOffset(Offset); 4217 return false; 4218 } 4219 4220 /// parseDirectiveCFIRegister 4221 /// ::= .cfi_register register, register 4222 bool AsmParser::parseDirectiveCFIRegister(SMLoc DirectiveLoc) { 4223 int64_t Register1 = 0, Register2 = 0; 4224 if (parseRegisterOrRegisterNumber(Register1, DirectiveLoc) || parseComma() || 4225 parseRegisterOrRegisterNumber(Register2, DirectiveLoc) || parseEOL()) 4226 return true; 4227 4228 getStreamer().emitCFIRegister(Register1, Register2); 4229 return false; 4230 } 4231 4232 /// parseDirectiveCFIWindowSave 4233 /// ::= .cfi_window_save 4234 bool AsmParser::parseDirectiveCFIWindowSave() { 4235 if (parseEOL()) 4236 return true; 4237 getStreamer().emitCFIWindowSave(); 4238 return false; 4239 } 4240 4241 /// parseDirectiveCFIAdjustCfaOffset 4242 /// ::= .cfi_adjust_cfa_offset adjustment 4243 bool AsmParser::parseDirectiveCFIAdjustCfaOffset() { 4244 int64_t Adjustment = 0; 4245 if (parseAbsoluteExpression(Adjustment) || parseEOL()) 4246 return true; 4247 4248 getStreamer().emitCFIAdjustCfaOffset(Adjustment); 4249 return false; 4250 } 4251 4252 /// parseDirectiveCFIDefCfaRegister 4253 /// ::= .cfi_def_cfa_register register 4254 bool AsmParser::parseDirectiveCFIDefCfaRegister(SMLoc DirectiveLoc) { 4255 int64_t Register = 0; 4256 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc) || parseEOL()) 4257 return true; 4258 4259 getStreamer().emitCFIDefCfaRegister(Register); 4260 return false; 4261 } 4262 4263 /// parseDirectiveCFIOffset 4264 /// ::= .cfi_offset register, offset 4265 bool AsmParser::parseDirectiveCFIOffset(SMLoc DirectiveLoc) { 4266 int64_t Register = 0; 4267 int64_t Offset = 0; 4268 4269 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc) || parseComma() || 4270 parseAbsoluteExpression(Offset) || parseEOL()) 4271 return true; 4272 4273 getStreamer().emitCFIOffset(Register, Offset); 4274 return false; 4275 } 4276 4277 /// parseDirectiveCFIRelOffset 4278 /// ::= .cfi_rel_offset register, offset 4279 bool AsmParser::parseDirectiveCFIRelOffset(SMLoc DirectiveLoc) { 4280 int64_t Register = 0, Offset = 0; 4281 4282 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc) || parseComma() || 4283 parseAbsoluteExpression(Offset) || parseEOL()) 4284 return true; 4285 4286 getStreamer().emitCFIRelOffset(Register, Offset); 4287 return false; 4288 } 4289 4290 static bool isValidEncoding(int64_t Encoding) { 4291 if (Encoding & ~0xff) 4292 return false; 4293 4294 if (Encoding == dwarf::DW_EH_PE_omit) 4295 return true; 4296 4297 const unsigned Format = Encoding & 0xf; 4298 if (Format != dwarf::DW_EH_PE_absptr && Format != dwarf::DW_EH_PE_udata2 && 4299 Format != dwarf::DW_EH_PE_udata4 && Format != dwarf::DW_EH_PE_udata8 && 4300 Format != dwarf::DW_EH_PE_sdata2 && Format != dwarf::DW_EH_PE_sdata4 && 4301 Format != dwarf::DW_EH_PE_sdata8 && Format != dwarf::DW_EH_PE_signed) 4302 return false; 4303 4304 const unsigned Application = Encoding & 0x70; 4305 if (Application != dwarf::DW_EH_PE_absptr && 4306 Application != dwarf::DW_EH_PE_pcrel) 4307 return false; 4308 4309 return true; 4310 } 4311 4312 /// parseDirectiveCFIPersonalityOrLsda 4313 /// IsPersonality true for cfi_personality, false for cfi_lsda 4314 /// ::= .cfi_personality encoding, [symbol_name] 4315 /// ::= .cfi_lsda encoding, [symbol_name] 4316 bool AsmParser::parseDirectiveCFIPersonalityOrLsda(bool IsPersonality) { 4317 int64_t Encoding = 0; 4318 if (parseAbsoluteExpression(Encoding)) 4319 return true; 4320 if (Encoding == dwarf::DW_EH_PE_omit) 4321 return false; 4322 4323 StringRef Name; 4324 if (check(!isValidEncoding(Encoding), "unsupported encoding.") || 4325 parseComma() || 4326 check(parseIdentifier(Name), "expected identifier in directive") || 4327 parseEOL()) 4328 return true; 4329 4330 MCSymbol *Sym = getContext().getOrCreateSymbol(Name); 4331 4332 if (IsPersonality) 4333 getStreamer().emitCFIPersonality(Sym, Encoding); 4334 else 4335 getStreamer().emitCFILsda(Sym, Encoding); 4336 return false; 4337 } 4338 4339 /// parseDirectiveCFIRememberState 4340 /// ::= .cfi_remember_state 4341 bool AsmParser::parseDirectiveCFIRememberState() { 4342 if (parseEOL()) 4343 return true; 4344 getStreamer().emitCFIRememberState(); 4345 return false; 4346 } 4347 4348 /// parseDirectiveCFIRestoreState 4349 /// ::= .cfi_remember_state 4350 bool AsmParser::parseDirectiveCFIRestoreState() { 4351 if (parseEOL()) 4352 return true; 4353 getStreamer().emitCFIRestoreState(); 4354 return false; 4355 } 4356 4357 /// parseDirectiveCFISameValue 4358 /// ::= .cfi_same_value register 4359 bool AsmParser::parseDirectiveCFISameValue(SMLoc DirectiveLoc) { 4360 int64_t Register = 0; 4361 4362 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc) || parseEOL()) 4363 return true; 4364 4365 getStreamer().emitCFISameValue(Register); 4366 return false; 4367 } 4368 4369 /// parseDirectiveCFIRestore 4370 /// ::= .cfi_restore register 4371 bool AsmParser::parseDirectiveCFIRestore(SMLoc DirectiveLoc) { 4372 int64_t Register = 0; 4373 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc) || parseEOL()) 4374 return true; 4375 4376 getStreamer().emitCFIRestore(Register); 4377 return false; 4378 } 4379 4380 /// parseDirectiveCFIEscape 4381 /// ::= .cfi_escape expression[,...] 4382 bool AsmParser::parseDirectiveCFIEscape() { 4383 std::string Values; 4384 int64_t CurrValue; 4385 if (parseAbsoluteExpression(CurrValue)) 4386 return true; 4387 4388 Values.push_back((uint8_t)CurrValue); 4389 4390 while (getLexer().is(AsmToken::Comma)) { 4391 Lex(); 4392 4393 if (parseAbsoluteExpression(CurrValue)) 4394 return true; 4395 4396 Values.push_back((uint8_t)CurrValue); 4397 } 4398 4399 getStreamer().emitCFIEscape(Values); 4400 return false; 4401 } 4402 4403 /// parseDirectiveCFIReturnColumn 4404 /// ::= .cfi_return_column register 4405 bool AsmParser::parseDirectiveCFIReturnColumn(SMLoc DirectiveLoc) { 4406 int64_t Register = 0; 4407 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc) || parseEOL()) 4408 return true; 4409 getStreamer().emitCFIReturnColumn(Register); 4410 return false; 4411 } 4412 4413 /// parseDirectiveCFISignalFrame 4414 /// ::= .cfi_signal_frame 4415 bool AsmParser::parseDirectiveCFISignalFrame() { 4416 if (parseEOL()) 4417 return true; 4418 4419 getStreamer().emitCFISignalFrame(); 4420 return false; 4421 } 4422 4423 /// parseDirectiveCFIUndefined 4424 /// ::= .cfi_undefined register 4425 bool AsmParser::parseDirectiveCFIUndefined(SMLoc DirectiveLoc) { 4426 int64_t Register = 0; 4427 4428 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc) || parseEOL()) 4429 return true; 4430 4431 getStreamer().emitCFIUndefined(Register); 4432 return false; 4433 } 4434 4435 /// parseDirectiveAltmacro 4436 /// ::= .altmacro 4437 /// ::= .noaltmacro 4438 bool AsmParser::parseDirectiveAltmacro(StringRef Directive) { 4439 if (parseEOL()) 4440 return true; 4441 AltMacroMode = (Directive == ".altmacro"); 4442 return false; 4443 } 4444 4445 /// parseDirectiveMacrosOnOff 4446 /// ::= .macros_on 4447 /// ::= .macros_off 4448 bool AsmParser::parseDirectiveMacrosOnOff(StringRef Directive) { 4449 if (parseEOL()) 4450 return true; 4451 setMacrosEnabled(Directive == ".macros_on"); 4452 return false; 4453 } 4454 4455 /// parseDirectiveMacro 4456 /// ::= .macro name[,] [parameters] 4457 bool AsmParser::parseDirectiveMacro(SMLoc DirectiveLoc) { 4458 StringRef Name; 4459 if (parseIdentifier(Name)) 4460 return TokError("expected identifier in '.macro' directive"); 4461 4462 if (getLexer().is(AsmToken::Comma)) 4463 Lex(); 4464 4465 MCAsmMacroParameters Parameters; 4466 while (getLexer().isNot(AsmToken::EndOfStatement)) { 4467 4468 if (!Parameters.empty() && Parameters.back().Vararg) 4469 return Error(Lexer.getLoc(), "vararg parameter '" + 4470 Parameters.back().Name + 4471 "' should be the last parameter"); 4472 4473 MCAsmMacroParameter Parameter; 4474 if (parseIdentifier(Parameter.Name)) 4475 return TokError("expected identifier in '.macro' directive"); 4476 4477 // Emit an error if two (or more) named parameters share the same name 4478 for (const MCAsmMacroParameter& CurrParam : Parameters) 4479 if (CurrParam.Name.equals(Parameter.Name)) 4480 return TokError("macro '" + Name + "' has multiple parameters" 4481 " named '" + Parameter.Name + "'"); 4482 4483 if (Lexer.is(AsmToken::Colon)) { 4484 Lex(); // consume ':' 4485 4486 SMLoc QualLoc; 4487 StringRef Qualifier; 4488 4489 QualLoc = Lexer.getLoc(); 4490 if (parseIdentifier(Qualifier)) 4491 return Error(QualLoc, "missing parameter qualifier for " 4492 "'" + Parameter.Name + "' in macro '" + Name + "'"); 4493 4494 if (Qualifier == "req") 4495 Parameter.Required = true; 4496 else if (Qualifier == "vararg") 4497 Parameter.Vararg = true; 4498 else 4499 return Error(QualLoc, Qualifier + " is not a valid parameter qualifier " 4500 "for '" + Parameter.Name + "' in macro '" + Name + "'"); 4501 } 4502 4503 if (getLexer().is(AsmToken::Equal)) { 4504 Lex(); 4505 4506 SMLoc ParamLoc; 4507 4508 ParamLoc = Lexer.getLoc(); 4509 if (parseMacroArgument(Parameter.Value, /*Vararg=*/false )) 4510 return true; 4511 4512 if (Parameter.Required) 4513 Warning(ParamLoc, "pointless default value for required parameter " 4514 "'" + Parameter.Name + "' in macro '" + Name + "'"); 4515 } 4516 4517 Parameters.push_back(std::move(Parameter)); 4518 4519 if (getLexer().is(AsmToken::Comma)) 4520 Lex(); 4521 } 4522 4523 // Eat just the end of statement. 4524 Lexer.Lex(); 4525 4526 // Consuming deferred text, so use Lexer.Lex to ignore Lexing Errors 4527 AsmToken EndToken, StartToken = getTok(); 4528 unsigned MacroDepth = 0; 4529 // Lex the macro definition. 4530 while (true) { 4531 // Ignore Lexing errors in macros. 4532 while (Lexer.is(AsmToken::Error)) { 4533 Lexer.Lex(); 4534 } 4535 4536 // Check whether we have reached the end of the file. 4537 if (getLexer().is(AsmToken::Eof)) 4538 return Error(DirectiveLoc, "no matching '.endmacro' in definition"); 4539 4540 // Otherwise, check whether we have reach the .endmacro or the start of a 4541 // preprocessor line marker. 4542 if (getLexer().is(AsmToken::Identifier)) { 4543 if (getTok().getIdentifier() == ".endm" || 4544 getTok().getIdentifier() == ".endmacro") { 4545 if (MacroDepth == 0) { // Outermost macro. 4546 EndToken = getTok(); 4547 Lexer.Lex(); 4548 if (getLexer().isNot(AsmToken::EndOfStatement)) 4549 return TokError("unexpected token in '" + EndToken.getIdentifier() + 4550 "' directive"); 4551 break; 4552 } else { 4553 // Otherwise we just found the end of an inner macro. 4554 --MacroDepth; 4555 } 4556 } else if (getTok().getIdentifier() == ".macro") { 4557 // We allow nested macros. Those aren't instantiated until the outermost 4558 // macro is expanded so just ignore them for now. 4559 ++MacroDepth; 4560 } 4561 } else if (Lexer.is(AsmToken::HashDirective)) { 4562 (void)parseCppHashLineFilenameComment(getLexer().getLoc()); 4563 } 4564 4565 // Otherwise, scan til the end of the statement. 4566 eatToEndOfStatement(); 4567 } 4568 4569 if (getContext().lookupMacro(Name)) { 4570 return Error(DirectiveLoc, "macro '" + Name + "' is already defined"); 4571 } 4572 4573 const char *BodyStart = StartToken.getLoc().getPointer(); 4574 const char *BodyEnd = EndToken.getLoc().getPointer(); 4575 StringRef Body = StringRef(BodyStart, BodyEnd - BodyStart); 4576 checkForBadMacro(DirectiveLoc, Name, Body, Parameters); 4577 MCAsmMacro Macro(Name, Body, std::move(Parameters)); 4578 DEBUG_WITH_TYPE("asm-macros", dbgs() << "Defining new macro:\n"; 4579 Macro.dump()); 4580 getContext().defineMacro(Name, std::move(Macro)); 4581 return false; 4582 } 4583 4584 /// checkForBadMacro 4585 /// 4586 /// With the support added for named parameters there may be code out there that 4587 /// is transitioning from positional parameters. In versions of gas that did 4588 /// not support named parameters they would be ignored on the macro definition. 4589 /// But to support both styles of parameters this is not possible so if a macro 4590 /// definition has named parameters but does not use them and has what appears 4591 /// to be positional parameters, strings like $1, $2, ... and $n, then issue a 4592 /// warning that the positional parameter found in body which have no effect. 4593 /// Hoping the developer will either remove the named parameters from the macro 4594 /// definition so the positional parameters get used if that was what was 4595 /// intended or change the macro to use the named parameters. It is possible 4596 /// this warning will trigger when the none of the named parameters are used 4597 /// and the strings like $1 are infact to simply to be passed trough unchanged. 4598 void AsmParser::checkForBadMacro(SMLoc DirectiveLoc, StringRef Name, 4599 StringRef Body, 4600 ArrayRef<MCAsmMacroParameter> Parameters) { 4601 // If this macro is not defined with named parameters the warning we are 4602 // checking for here doesn't apply. 4603 unsigned NParameters = Parameters.size(); 4604 if (NParameters == 0) 4605 return; 4606 4607 bool NamedParametersFound = false; 4608 bool PositionalParametersFound = false; 4609 4610 // Look at the body of the macro for use of both the named parameters and what 4611 // are likely to be positional parameters. This is what expandMacro() is 4612 // doing when it finds the parameters in the body. 4613 while (!Body.empty()) { 4614 // Scan for the next possible parameter. 4615 std::size_t End = Body.size(), Pos = 0; 4616 for (; Pos != End; ++Pos) { 4617 // Check for a substitution or escape. 4618 // This macro is defined with parameters, look for \foo, \bar, etc. 4619 if (Body[Pos] == '\\' && Pos + 1 != End) 4620 break; 4621 4622 // This macro should have parameters, but look for $0, $1, ..., $n too. 4623 if (Body[Pos] != '$' || Pos + 1 == End) 4624 continue; 4625 char Next = Body[Pos + 1]; 4626 if (Next == '$' || Next == 'n' || 4627 isdigit(static_cast<unsigned char>(Next))) 4628 break; 4629 } 4630 4631 // Check if we reached the end. 4632 if (Pos == End) 4633 break; 4634 4635 if (Body[Pos] == '$') { 4636 switch (Body[Pos + 1]) { 4637 // $$ => $ 4638 case '$': 4639 break; 4640 4641 // $n => number of arguments 4642 case 'n': 4643 PositionalParametersFound = true; 4644 break; 4645 4646 // $[0-9] => argument 4647 default: { 4648 PositionalParametersFound = true; 4649 break; 4650 } 4651 } 4652 Pos += 2; 4653 } else { 4654 unsigned I = Pos + 1; 4655 while (isIdentifierChar(Body[I]) && I + 1 != End) 4656 ++I; 4657 4658 const char *Begin = Body.data() + Pos + 1; 4659 StringRef Argument(Begin, I - (Pos + 1)); 4660 unsigned Index = 0; 4661 for (; Index < NParameters; ++Index) 4662 if (Parameters[Index].Name == Argument) 4663 break; 4664 4665 if (Index == NParameters) { 4666 if (Body[Pos + 1] == '(' && Body[Pos + 2] == ')') 4667 Pos += 3; 4668 else { 4669 Pos = I; 4670 } 4671 } else { 4672 NamedParametersFound = true; 4673 Pos += 1 + Argument.size(); 4674 } 4675 } 4676 // Update the scan point. 4677 Body = Body.substr(Pos); 4678 } 4679 4680 if (!NamedParametersFound && PositionalParametersFound) 4681 Warning(DirectiveLoc, "macro defined with named parameters which are not " 4682 "used in macro body, possible positional parameter " 4683 "found in body which will have no effect"); 4684 } 4685 4686 /// parseDirectiveExitMacro 4687 /// ::= .exitm 4688 bool AsmParser::parseDirectiveExitMacro(StringRef Directive) { 4689 if (parseEOL()) 4690 return true; 4691 4692 if (!isInsideMacroInstantiation()) 4693 return TokError("unexpected '" + Directive + "' in file, " 4694 "no current macro definition"); 4695 4696 // Exit all conditionals that are active in the current macro. 4697 while (TheCondStack.size() != ActiveMacros.back()->CondStackDepth) { 4698 TheCondState = TheCondStack.back(); 4699 TheCondStack.pop_back(); 4700 } 4701 4702 handleMacroExit(); 4703 return false; 4704 } 4705 4706 /// parseDirectiveEndMacro 4707 /// ::= .endm 4708 /// ::= .endmacro 4709 bool AsmParser::parseDirectiveEndMacro(StringRef Directive) { 4710 if (getLexer().isNot(AsmToken::EndOfStatement)) 4711 return TokError("unexpected token in '" + Directive + "' directive"); 4712 4713 // If we are inside a macro instantiation, terminate the current 4714 // instantiation. 4715 if (isInsideMacroInstantiation()) { 4716 handleMacroExit(); 4717 return false; 4718 } 4719 4720 // Otherwise, this .endmacro is a stray entry in the file; well formed 4721 // .endmacro directives are handled during the macro definition parsing. 4722 return TokError("unexpected '" + Directive + "' in file, " 4723 "no current macro definition"); 4724 } 4725 4726 /// parseDirectivePurgeMacro 4727 /// ::= .purgem name 4728 bool AsmParser::parseDirectivePurgeMacro(SMLoc DirectiveLoc) { 4729 StringRef Name; 4730 SMLoc Loc; 4731 if (parseTokenLoc(Loc) || 4732 check(parseIdentifier(Name), Loc, 4733 "expected identifier in '.purgem' directive") || 4734 parseEOL()) 4735 return true; 4736 4737 if (!getContext().lookupMacro(Name)) 4738 return Error(DirectiveLoc, "macro '" + Name + "' is not defined"); 4739 4740 getContext().undefineMacro(Name); 4741 DEBUG_WITH_TYPE("asm-macros", dbgs() 4742 << "Un-defining macro: " << Name << "\n"); 4743 return false; 4744 } 4745 4746 /// parseDirectiveBundleAlignMode 4747 /// ::= {.bundle_align_mode} expression 4748 bool AsmParser::parseDirectiveBundleAlignMode() { 4749 // Expect a single argument: an expression that evaluates to a constant 4750 // in the inclusive range 0-30. 4751 SMLoc ExprLoc = getLexer().getLoc(); 4752 int64_t AlignSizePow2; 4753 if (checkForValidSection() || parseAbsoluteExpression(AlignSizePow2) || 4754 parseEOL() || 4755 check(AlignSizePow2 < 0 || AlignSizePow2 > 30, ExprLoc, 4756 "invalid bundle alignment size (expected between 0 and 30)")) 4757 return true; 4758 4759 // Because of AlignSizePow2's verified range we can safely truncate it to 4760 // unsigned. 4761 getStreamer().emitBundleAlignMode(static_cast<unsigned>(AlignSizePow2)); 4762 return false; 4763 } 4764 4765 /// parseDirectiveBundleLock 4766 /// ::= {.bundle_lock} [align_to_end] 4767 bool AsmParser::parseDirectiveBundleLock() { 4768 if (checkForValidSection()) 4769 return true; 4770 bool AlignToEnd = false; 4771 4772 StringRef Option; 4773 SMLoc Loc = getTok().getLoc(); 4774 const char *kInvalidOptionError = 4775 "invalid option for '.bundle_lock' directive"; 4776 4777 if (!parseOptionalToken(AsmToken::EndOfStatement)) { 4778 if (check(parseIdentifier(Option), Loc, kInvalidOptionError) || 4779 check(Option != "align_to_end", Loc, kInvalidOptionError) || parseEOL()) 4780 return true; 4781 AlignToEnd = true; 4782 } 4783 4784 getStreamer().emitBundleLock(AlignToEnd); 4785 return false; 4786 } 4787 4788 /// parseDirectiveBundleLock 4789 /// ::= {.bundle_lock} 4790 bool AsmParser::parseDirectiveBundleUnlock() { 4791 if (checkForValidSection() || parseEOL()) 4792 return true; 4793 4794 getStreamer().emitBundleUnlock(); 4795 return false; 4796 } 4797 4798 /// parseDirectiveSpace 4799 /// ::= (.skip | .space) expression [ , expression ] 4800 bool AsmParser::parseDirectiveSpace(StringRef IDVal) { 4801 SMLoc NumBytesLoc = Lexer.getLoc(); 4802 const MCExpr *NumBytes; 4803 if (checkForValidSection() || parseExpression(NumBytes)) 4804 return true; 4805 4806 int64_t FillExpr = 0; 4807 if (parseOptionalToken(AsmToken::Comma)) 4808 if (parseAbsoluteExpression(FillExpr)) 4809 return true; 4810 if (parseEOL()) 4811 return true; 4812 4813 // FIXME: Sometimes the fill expr is 'nop' if it isn't supplied, instead of 0. 4814 getStreamer().emitFill(*NumBytes, FillExpr, NumBytesLoc); 4815 4816 return false; 4817 } 4818 4819 /// parseDirectiveDCB 4820 /// ::= .dcb.{b, l, w} expression, expression 4821 bool AsmParser::parseDirectiveDCB(StringRef IDVal, unsigned Size) { 4822 SMLoc NumValuesLoc = Lexer.getLoc(); 4823 int64_t NumValues; 4824 if (checkForValidSection() || parseAbsoluteExpression(NumValues)) 4825 return true; 4826 4827 if (NumValues < 0) { 4828 Warning(NumValuesLoc, "'" + Twine(IDVal) + "' directive with negative repeat count has no effect"); 4829 return false; 4830 } 4831 4832 if (parseComma()) 4833 return true; 4834 4835 const MCExpr *Value; 4836 SMLoc ExprLoc = getLexer().getLoc(); 4837 if (parseExpression(Value)) 4838 return true; 4839 4840 // Special case constant expressions to match code generator. 4841 if (const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value)) { 4842 assert(Size <= 8 && "Invalid size"); 4843 uint64_t IntValue = MCE->getValue(); 4844 if (!isUIntN(8 * Size, IntValue) && !isIntN(8 * Size, IntValue)) 4845 return Error(ExprLoc, "literal value out of range for directive"); 4846 for (uint64_t i = 0, e = NumValues; i != e; ++i) 4847 getStreamer().emitIntValue(IntValue, Size); 4848 } else { 4849 for (uint64_t i = 0, e = NumValues; i != e; ++i) 4850 getStreamer().emitValue(Value, Size, ExprLoc); 4851 } 4852 4853 return parseEOL(); 4854 } 4855 4856 /// parseDirectiveRealDCB 4857 /// ::= .dcb.{d, s} expression, expression 4858 bool AsmParser::parseDirectiveRealDCB(StringRef IDVal, const fltSemantics &Semantics) { 4859 SMLoc NumValuesLoc = Lexer.getLoc(); 4860 int64_t NumValues; 4861 if (checkForValidSection() || parseAbsoluteExpression(NumValues)) 4862 return true; 4863 4864 if (NumValues < 0) { 4865 Warning(NumValuesLoc, "'" + Twine(IDVal) + "' directive with negative repeat count has no effect"); 4866 return false; 4867 } 4868 4869 if (parseComma()) 4870 return true; 4871 4872 APInt AsInt; 4873 if (parseRealValue(Semantics, AsInt) || parseEOL()) 4874 return true; 4875 4876 for (uint64_t i = 0, e = NumValues; i != e; ++i) 4877 getStreamer().emitIntValue(AsInt.getLimitedValue(), 4878 AsInt.getBitWidth() / 8); 4879 4880 return false; 4881 } 4882 4883 /// parseDirectiveDS 4884 /// ::= .ds.{b, d, l, p, s, w, x} expression 4885 bool AsmParser::parseDirectiveDS(StringRef IDVal, unsigned Size) { 4886 SMLoc NumValuesLoc = Lexer.getLoc(); 4887 int64_t NumValues; 4888 if (checkForValidSection() || parseAbsoluteExpression(NumValues) || 4889 parseEOL()) 4890 return true; 4891 4892 if (NumValues < 0) { 4893 Warning(NumValuesLoc, "'" + Twine(IDVal) + "' directive with negative repeat count has no effect"); 4894 return false; 4895 } 4896 4897 for (uint64_t i = 0, e = NumValues; i != e; ++i) 4898 getStreamer().emitFill(Size, 0); 4899 4900 return false; 4901 } 4902 4903 /// parseDirectiveLEB128 4904 /// ::= (.sleb128 | .uleb128) [ expression (, expression)* ] 4905 bool AsmParser::parseDirectiveLEB128(bool Signed) { 4906 if (checkForValidSection()) 4907 return true; 4908 4909 auto parseOp = [&]() -> bool { 4910 const MCExpr *Value; 4911 if (parseExpression(Value)) 4912 return true; 4913 if (Signed) 4914 getStreamer().emitSLEB128Value(Value); 4915 else 4916 getStreamer().emitULEB128Value(Value); 4917 return false; 4918 }; 4919 4920 return parseMany(parseOp); 4921 } 4922 4923 /// parseDirectiveSymbolAttribute 4924 /// ::= { ".globl", ".weak", ... } [ identifier ( , identifier )* ] 4925 bool AsmParser::parseDirectiveSymbolAttribute(MCSymbolAttr Attr) { 4926 auto parseOp = [&]() -> bool { 4927 StringRef Name; 4928 SMLoc Loc = getTok().getLoc(); 4929 if (parseIdentifier(Name)) 4930 return Error(Loc, "expected identifier"); 4931 4932 if (discardLTOSymbol(Name)) 4933 return false; 4934 4935 MCSymbol *Sym = getContext().getOrCreateSymbol(Name); 4936 4937 // Assembler local symbols don't make any sense here. Complain loudly. 4938 if (Sym->isTemporary()) 4939 return Error(Loc, "non-local symbol required"); 4940 4941 if (!getStreamer().emitSymbolAttribute(Sym, Attr)) 4942 return Error(Loc, "unable to emit symbol attribute"); 4943 return false; 4944 }; 4945 4946 return parseMany(parseOp); 4947 } 4948 4949 /// parseDirectiveComm 4950 /// ::= ( .comm | .lcomm ) identifier , size_expression [ , align_expression ] 4951 bool AsmParser::parseDirectiveComm(bool IsLocal) { 4952 if (checkForValidSection()) 4953 return true; 4954 4955 SMLoc IDLoc = getLexer().getLoc(); 4956 StringRef Name; 4957 if (parseIdentifier(Name)) 4958 return TokError("expected identifier in directive"); 4959 4960 // Handle the identifier as the key symbol. 4961 MCSymbol *Sym = getContext().getOrCreateSymbol(Name); 4962 4963 if (parseComma()) 4964 return true; 4965 4966 int64_t Size; 4967 SMLoc SizeLoc = getLexer().getLoc(); 4968 if (parseAbsoluteExpression(Size)) 4969 return true; 4970 4971 int64_t Pow2Alignment = 0; 4972 SMLoc Pow2AlignmentLoc; 4973 if (getLexer().is(AsmToken::Comma)) { 4974 Lex(); 4975 Pow2AlignmentLoc = getLexer().getLoc(); 4976 if (parseAbsoluteExpression(Pow2Alignment)) 4977 return true; 4978 4979 LCOMM::LCOMMType LCOMM = Lexer.getMAI().getLCOMMDirectiveAlignmentType(); 4980 if (IsLocal && LCOMM == LCOMM::NoAlignment) 4981 return Error(Pow2AlignmentLoc, "alignment not supported on this target"); 4982 4983 // If this target takes alignments in bytes (not log) validate and convert. 4984 if ((!IsLocal && Lexer.getMAI().getCOMMDirectiveAlignmentIsInBytes()) || 4985 (IsLocal && LCOMM == LCOMM::ByteAlignment)) { 4986 if (!isPowerOf2_64(Pow2Alignment)) 4987 return Error(Pow2AlignmentLoc, "alignment must be a power of 2"); 4988 Pow2Alignment = Log2_64(Pow2Alignment); 4989 } 4990 } 4991 4992 if (parseEOL()) 4993 return true; 4994 4995 // NOTE: a size of zero for a .comm should create a undefined symbol 4996 // but a size of .lcomm creates a bss symbol of size zero. 4997 if (Size < 0) 4998 return Error(SizeLoc, "invalid '.comm' or '.lcomm' directive size, can't " 4999 "be less than zero"); 5000 5001 // NOTE: The alignment in the directive is a power of 2 value, the assembler 5002 // may internally end up wanting an alignment in bytes. 5003 // FIXME: Diagnose overflow. 5004 if (Pow2Alignment < 0) 5005 return Error(Pow2AlignmentLoc, "invalid '.comm' or '.lcomm' directive " 5006 "alignment, can't be less than zero"); 5007 5008 Sym->redefineIfPossible(); 5009 if (!Sym->isUndefined()) 5010 return Error(IDLoc, "invalid symbol redefinition"); 5011 5012 // Create the Symbol as a common or local common with Size and Pow2Alignment 5013 if (IsLocal) { 5014 getStreamer().emitLocalCommonSymbol(Sym, Size, 1 << Pow2Alignment); 5015 return false; 5016 } 5017 5018 getStreamer().emitCommonSymbol(Sym, Size, 1 << Pow2Alignment); 5019 return false; 5020 } 5021 5022 /// parseDirectiveAbort 5023 /// ::= .abort [... message ...] 5024 bool AsmParser::parseDirectiveAbort() { 5025 // FIXME: Use loc from directive. 5026 SMLoc Loc = getLexer().getLoc(); 5027 5028 StringRef Str = parseStringToEndOfStatement(); 5029 if (parseEOL()) 5030 return true; 5031 5032 if (Str.empty()) 5033 return Error(Loc, ".abort detected. Assembly stopping."); 5034 else 5035 return Error(Loc, ".abort '" + Str + "' detected. Assembly stopping."); 5036 // FIXME: Actually abort assembly here. 5037 5038 return false; 5039 } 5040 5041 /// parseDirectiveInclude 5042 /// ::= .include "filename" 5043 bool AsmParser::parseDirectiveInclude() { 5044 // Allow the strings to have escaped octal character sequence. 5045 std::string Filename; 5046 SMLoc IncludeLoc = getTok().getLoc(); 5047 5048 if (check(getTok().isNot(AsmToken::String), 5049 "expected string in '.include' directive") || 5050 parseEscapedString(Filename) || 5051 check(getTok().isNot(AsmToken::EndOfStatement), 5052 "unexpected token in '.include' directive") || 5053 // Attempt to switch the lexer to the included file before consuming the 5054 // end of statement to avoid losing it when we switch. 5055 check(enterIncludeFile(Filename), IncludeLoc, 5056 "Could not find include file '" + Filename + "'")) 5057 return true; 5058 5059 return false; 5060 } 5061 5062 /// parseDirectiveIncbin 5063 /// ::= .incbin "filename" [ , skip [ , count ] ] 5064 bool AsmParser::parseDirectiveIncbin() { 5065 // Allow the strings to have escaped octal character sequence. 5066 std::string Filename; 5067 SMLoc IncbinLoc = getTok().getLoc(); 5068 if (check(getTok().isNot(AsmToken::String), 5069 "expected string in '.incbin' directive") || 5070 parseEscapedString(Filename)) 5071 return true; 5072 5073 int64_t Skip = 0; 5074 const MCExpr *Count = nullptr; 5075 SMLoc SkipLoc, CountLoc; 5076 if (parseOptionalToken(AsmToken::Comma)) { 5077 // The skip expression can be omitted while specifying the count, e.g: 5078 // .incbin "filename",,4 5079 if (getTok().isNot(AsmToken::Comma)) { 5080 if (parseTokenLoc(SkipLoc) || parseAbsoluteExpression(Skip)) 5081 return true; 5082 } 5083 if (parseOptionalToken(AsmToken::Comma)) { 5084 CountLoc = getTok().getLoc(); 5085 if (parseExpression(Count)) 5086 return true; 5087 } 5088 } 5089 5090 if (parseEOL()) 5091 return true; 5092 5093 if (check(Skip < 0, SkipLoc, "skip is negative")) 5094 return true; 5095 5096 // Attempt to process the included file. 5097 if (processIncbinFile(Filename, Skip, Count, CountLoc)) 5098 return Error(IncbinLoc, "Could not find incbin file '" + Filename + "'"); 5099 return false; 5100 } 5101 5102 /// parseDirectiveIf 5103 /// ::= .if{,eq,ge,gt,le,lt,ne} expression 5104 bool AsmParser::parseDirectiveIf(SMLoc DirectiveLoc, DirectiveKind DirKind) { 5105 TheCondStack.push_back(TheCondState); 5106 TheCondState.TheCond = AsmCond::IfCond; 5107 if (TheCondState.Ignore) { 5108 eatToEndOfStatement(); 5109 } else { 5110 int64_t ExprValue; 5111 if (parseAbsoluteExpression(ExprValue) || parseEOL()) 5112 return true; 5113 5114 switch (DirKind) { 5115 default: 5116 llvm_unreachable("unsupported directive"); 5117 case DK_IF: 5118 case DK_IFNE: 5119 break; 5120 case DK_IFEQ: 5121 ExprValue = ExprValue == 0; 5122 break; 5123 case DK_IFGE: 5124 ExprValue = ExprValue >= 0; 5125 break; 5126 case DK_IFGT: 5127 ExprValue = ExprValue > 0; 5128 break; 5129 case DK_IFLE: 5130 ExprValue = ExprValue <= 0; 5131 break; 5132 case DK_IFLT: 5133 ExprValue = ExprValue < 0; 5134 break; 5135 } 5136 5137 TheCondState.CondMet = ExprValue; 5138 TheCondState.Ignore = !TheCondState.CondMet; 5139 } 5140 5141 return false; 5142 } 5143 5144 /// parseDirectiveIfb 5145 /// ::= .ifb string 5146 bool AsmParser::parseDirectiveIfb(SMLoc DirectiveLoc, bool ExpectBlank) { 5147 TheCondStack.push_back(TheCondState); 5148 TheCondState.TheCond = AsmCond::IfCond; 5149 5150 if (TheCondState.Ignore) { 5151 eatToEndOfStatement(); 5152 } else { 5153 StringRef Str = parseStringToEndOfStatement(); 5154 5155 if (parseEOL()) 5156 return true; 5157 5158 TheCondState.CondMet = ExpectBlank == Str.empty(); 5159 TheCondState.Ignore = !TheCondState.CondMet; 5160 } 5161 5162 return false; 5163 } 5164 5165 /// parseDirectiveIfc 5166 /// ::= .ifc string1, string2 5167 /// ::= .ifnc string1, string2 5168 bool AsmParser::parseDirectiveIfc(SMLoc DirectiveLoc, bool ExpectEqual) { 5169 TheCondStack.push_back(TheCondState); 5170 TheCondState.TheCond = AsmCond::IfCond; 5171 5172 if (TheCondState.Ignore) { 5173 eatToEndOfStatement(); 5174 } else { 5175 StringRef Str1 = parseStringToComma(); 5176 5177 if (parseComma()) 5178 return true; 5179 5180 StringRef Str2 = parseStringToEndOfStatement(); 5181 5182 if (parseEOL()) 5183 return true; 5184 5185 TheCondState.CondMet = ExpectEqual == (Str1.trim() == Str2.trim()); 5186 TheCondState.Ignore = !TheCondState.CondMet; 5187 } 5188 5189 return false; 5190 } 5191 5192 /// parseDirectiveIfeqs 5193 /// ::= .ifeqs string1, string2 5194 bool AsmParser::parseDirectiveIfeqs(SMLoc DirectiveLoc, bool ExpectEqual) { 5195 if (Lexer.isNot(AsmToken::String)) { 5196 if (ExpectEqual) 5197 return TokError("expected string parameter for '.ifeqs' directive"); 5198 return TokError("expected string parameter for '.ifnes' directive"); 5199 } 5200 5201 StringRef String1 = getTok().getStringContents(); 5202 Lex(); 5203 5204 if (Lexer.isNot(AsmToken::Comma)) { 5205 if (ExpectEqual) 5206 return TokError( 5207 "expected comma after first string for '.ifeqs' directive"); 5208 return TokError("expected comma after first string for '.ifnes' directive"); 5209 } 5210 5211 Lex(); 5212 5213 if (Lexer.isNot(AsmToken::String)) { 5214 if (ExpectEqual) 5215 return TokError("expected string parameter for '.ifeqs' directive"); 5216 return TokError("expected string parameter for '.ifnes' directive"); 5217 } 5218 5219 StringRef String2 = getTok().getStringContents(); 5220 Lex(); 5221 5222 TheCondStack.push_back(TheCondState); 5223 TheCondState.TheCond = AsmCond::IfCond; 5224 TheCondState.CondMet = ExpectEqual == (String1 == String2); 5225 TheCondState.Ignore = !TheCondState.CondMet; 5226 5227 return false; 5228 } 5229 5230 /// parseDirectiveIfdef 5231 /// ::= .ifdef symbol 5232 bool AsmParser::parseDirectiveIfdef(SMLoc DirectiveLoc, bool expect_defined) { 5233 StringRef Name; 5234 TheCondStack.push_back(TheCondState); 5235 TheCondState.TheCond = AsmCond::IfCond; 5236 5237 if (TheCondState.Ignore) { 5238 eatToEndOfStatement(); 5239 } else { 5240 if (check(parseIdentifier(Name), "expected identifier after '.ifdef'") || 5241 parseEOL()) 5242 return true; 5243 5244 MCSymbol *Sym = getContext().lookupSymbol(Name); 5245 5246 if (expect_defined) 5247 TheCondState.CondMet = (Sym && !Sym->isUndefined(false)); 5248 else 5249 TheCondState.CondMet = (!Sym || Sym->isUndefined(false)); 5250 TheCondState.Ignore = !TheCondState.CondMet; 5251 } 5252 5253 return false; 5254 } 5255 5256 /// parseDirectiveElseIf 5257 /// ::= .elseif expression 5258 bool AsmParser::parseDirectiveElseIf(SMLoc DirectiveLoc) { 5259 if (TheCondState.TheCond != AsmCond::IfCond && 5260 TheCondState.TheCond != AsmCond::ElseIfCond) 5261 return Error(DirectiveLoc, "Encountered a .elseif that doesn't follow an" 5262 " .if or an .elseif"); 5263 TheCondState.TheCond = AsmCond::ElseIfCond; 5264 5265 bool LastIgnoreState = false; 5266 if (!TheCondStack.empty()) 5267 LastIgnoreState = TheCondStack.back().Ignore; 5268 if (LastIgnoreState || TheCondState.CondMet) { 5269 TheCondState.Ignore = true; 5270 eatToEndOfStatement(); 5271 } else { 5272 int64_t ExprValue; 5273 if (parseAbsoluteExpression(ExprValue)) 5274 return true; 5275 5276 if (parseEOL()) 5277 return true; 5278 5279 TheCondState.CondMet = ExprValue; 5280 TheCondState.Ignore = !TheCondState.CondMet; 5281 } 5282 5283 return false; 5284 } 5285 5286 /// parseDirectiveElse 5287 /// ::= .else 5288 bool AsmParser::parseDirectiveElse(SMLoc DirectiveLoc) { 5289 if (parseEOL()) 5290 return true; 5291 5292 if (TheCondState.TheCond != AsmCond::IfCond && 5293 TheCondState.TheCond != AsmCond::ElseIfCond) 5294 return Error(DirectiveLoc, "Encountered a .else that doesn't follow " 5295 " an .if or an .elseif"); 5296 TheCondState.TheCond = AsmCond::ElseCond; 5297 bool LastIgnoreState = false; 5298 if (!TheCondStack.empty()) 5299 LastIgnoreState = TheCondStack.back().Ignore; 5300 if (LastIgnoreState || TheCondState.CondMet) 5301 TheCondState.Ignore = true; 5302 else 5303 TheCondState.Ignore = false; 5304 5305 return false; 5306 } 5307 5308 /// parseDirectiveEnd 5309 /// ::= .end 5310 bool AsmParser::parseDirectiveEnd(SMLoc DirectiveLoc) { 5311 if (parseEOL()) 5312 return true; 5313 5314 while (Lexer.isNot(AsmToken::Eof)) 5315 Lexer.Lex(); 5316 5317 return false; 5318 } 5319 5320 /// parseDirectiveError 5321 /// ::= .err 5322 /// ::= .error [string] 5323 bool AsmParser::parseDirectiveError(SMLoc L, bool WithMessage) { 5324 if (!TheCondStack.empty()) { 5325 if (TheCondStack.back().Ignore) { 5326 eatToEndOfStatement(); 5327 return false; 5328 } 5329 } 5330 5331 if (!WithMessage) 5332 return Error(L, ".err encountered"); 5333 5334 StringRef Message = ".error directive invoked in source file"; 5335 if (Lexer.isNot(AsmToken::EndOfStatement)) { 5336 if (Lexer.isNot(AsmToken::String)) 5337 return TokError(".error argument must be a string"); 5338 5339 Message = getTok().getStringContents(); 5340 Lex(); 5341 } 5342 5343 return Error(L, Message); 5344 } 5345 5346 /// parseDirectiveWarning 5347 /// ::= .warning [string] 5348 bool AsmParser::parseDirectiveWarning(SMLoc L) { 5349 if (!TheCondStack.empty()) { 5350 if (TheCondStack.back().Ignore) { 5351 eatToEndOfStatement(); 5352 return false; 5353 } 5354 } 5355 5356 StringRef Message = ".warning directive invoked in source file"; 5357 5358 if (!parseOptionalToken(AsmToken::EndOfStatement)) { 5359 if (Lexer.isNot(AsmToken::String)) 5360 return TokError(".warning argument must be a string"); 5361 5362 Message = getTok().getStringContents(); 5363 Lex(); 5364 if (parseEOL()) 5365 return true; 5366 } 5367 5368 return Warning(L, Message); 5369 } 5370 5371 /// parseDirectiveEndIf 5372 /// ::= .endif 5373 bool AsmParser::parseDirectiveEndIf(SMLoc DirectiveLoc) { 5374 if (parseEOL()) 5375 return true; 5376 5377 if ((TheCondState.TheCond == AsmCond::NoCond) || TheCondStack.empty()) 5378 return Error(DirectiveLoc, "Encountered a .endif that doesn't follow " 5379 "an .if or .else"); 5380 if (!TheCondStack.empty()) { 5381 TheCondState = TheCondStack.back(); 5382 TheCondStack.pop_back(); 5383 } 5384 5385 return false; 5386 } 5387 5388 void AsmParser::initializeDirectiveKindMap() { 5389 /* Lookup will be done with the directive 5390 * converted to lower case, so all these 5391 * keys should be lower case. 5392 * (target specific directives are handled 5393 * elsewhere) 5394 */ 5395 DirectiveKindMap[".set"] = DK_SET; 5396 DirectiveKindMap[".equ"] = DK_EQU; 5397 DirectiveKindMap[".equiv"] = DK_EQUIV; 5398 DirectiveKindMap[".ascii"] = DK_ASCII; 5399 DirectiveKindMap[".asciz"] = DK_ASCIZ; 5400 DirectiveKindMap[".string"] = DK_STRING; 5401 DirectiveKindMap[".byte"] = DK_BYTE; 5402 DirectiveKindMap[".short"] = DK_SHORT; 5403 DirectiveKindMap[".value"] = DK_VALUE; 5404 DirectiveKindMap[".2byte"] = DK_2BYTE; 5405 DirectiveKindMap[".long"] = DK_LONG; 5406 DirectiveKindMap[".int"] = DK_INT; 5407 DirectiveKindMap[".4byte"] = DK_4BYTE; 5408 DirectiveKindMap[".quad"] = DK_QUAD; 5409 DirectiveKindMap[".8byte"] = DK_8BYTE; 5410 DirectiveKindMap[".octa"] = DK_OCTA; 5411 DirectiveKindMap[".single"] = DK_SINGLE; 5412 DirectiveKindMap[".float"] = DK_FLOAT; 5413 DirectiveKindMap[".double"] = DK_DOUBLE; 5414 DirectiveKindMap[".align"] = DK_ALIGN; 5415 DirectiveKindMap[".align32"] = DK_ALIGN32; 5416 DirectiveKindMap[".balign"] = DK_BALIGN; 5417 DirectiveKindMap[".balignw"] = DK_BALIGNW; 5418 DirectiveKindMap[".balignl"] = DK_BALIGNL; 5419 DirectiveKindMap[".p2align"] = DK_P2ALIGN; 5420 DirectiveKindMap[".p2alignw"] = DK_P2ALIGNW; 5421 DirectiveKindMap[".p2alignl"] = DK_P2ALIGNL; 5422 DirectiveKindMap[".org"] = DK_ORG; 5423 DirectiveKindMap[".fill"] = DK_FILL; 5424 DirectiveKindMap[".zero"] = DK_ZERO; 5425 DirectiveKindMap[".extern"] = DK_EXTERN; 5426 DirectiveKindMap[".globl"] = DK_GLOBL; 5427 DirectiveKindMap[".global"] = DK_GLOBAL; 5428 DirectiveKindMap[".lazy_reference"] = DK_LAZY_REFERENCE; 5429 DirectiveKindMap[".no_dead_strip"] = DK_NO_DEAD_STRIP; 5430 DirectiveKindMap[".symbol_resolver"] = DK_SYMBOL_RESOLVER; 5431 DirectiveKindMap[".private_extern"] = DK_PRIVATE_EXTERN; 5432 DirectiveKindMap[".reference"] = DK_REFERENCE; 5433 DirectiveKindMap[".weak_definition"] = DK_WEAK_DEFINITION; 5434 DirectiveKindMap[".weak_reference"] = DK_WEAK_REFERENCE; 5435 DirectiveKindMap[".weak_def_can_be_hidden"] = DK_WEAK_DEF_CAN_BE_HIDDEN; 5436 DirectiveKindMap[".cold"] = DK_COLD; 5437 DirectiveKindMap[".comm"] = DK_COMM; 5438 DirectiveKindMap[".common"] = DK_COMMON; 5439 DirectiveKindMap[".lcomm"] = DK_LCOMM; 5440 DirectiveKindMap[".abort"] = DK_ABORT; 5441 DirectiveKindMap[".include"] = DK_INCLUDE; 5442 DirectiveKindMap[".incbin"] = DK_INCBIN; 5443 DirectiveKindMap[".code16"] = DK_CODE16; 5444 DirectiveKindMap[".code16gcc"] = DK_CODE16GCC; 5445 DirectiveKindMap[".rept"] = DK_REPT; 5446 DirectiveKindMap[".rep"] = DK_REPT; 5447 DirectiveKindMap[".irp"] = DK_IRP; 5448 DirectiveKindMap[".irpc"] = DK_IRPC; 5449 DirectiveKindMap[".endr"] = DK_ENDR; 5450 DirectiveKindMap[".bundle_align_mode"] = DK_BUNDLE_ALIGN_MODE; 5451 DirectiveKindMap[".bundle_lock"] = DK_BUNDLE_LOCK; 5452 DirectiveKindMap[".bundle_unlock"] = DK_BUNDLE_UNLOCK; 5453 DirectiveKindMap[".if"] = DK_IF; 5454 DirectiveKindMap[".ifeq"] = DK_IFEQ; 5455 DirectiveKindMap[".ifge"] = DK_IFGE; 5456 DirectiveKindMap[".ifgt"] = DK_IFGT; 5457 DirectiveKindMap[".ifle"] = DK_IFLE; 5458 DirectiveKindMap[".iflt"] = DK_IFLT; 5459 DirectiveKindMap[".ifne"] = DK_IFNE; 5460 DirectiveKindMap[".ifb"] = DK_IFB; 5461 DirectiveKindMap[".ifnb"] = DK_IFNB; 5462 DirectiveKindMap[".ifc"] = DK_IFC; 5463 DirectiveKindMap[".ifeqs"] = DK_IFEQS; 5464 DirectiveKindMap[".ifnc"] = DK_IFNC; 5465 DirectiveKindMap[".ifnes"] = DK_IFNES; 5466 DirectiveKindMap[".ifdef"] = DK_IFDEF; 5467 DirectiveKindMap[".ifndef"] = DK_IFNDEF; 5468 DirectiveKindMap[".ifnotdef"] = DK_IFNOTDEF; 5469 DirectiveKindMap[".elseif"] = DK_ELSEIF; 5470 DirectiveKindMap[".else"] = DK_ELSE; 5471 DirectiveKindMap[".end"] = DK_END; 5472 DirectiveKindMap[".endif"] = DK_ENDIF; 5473 DirectiveKindMap[".skip"] = DK_SKIP; 5474 DirectiveKindMap[".space"] = DK_SPACE; 5475 DirectiveKindMap[".file"] = DK_FILE; 5476 DirectiveKindMap[".line"] = DK_LINE; 5477 DirectiveKindMap[".loc"] = DK_LOC; 5478 DirectiveKindMap[".stabs"] = DK_STABS; 5479 DirectiveKindMap[".cv_file"] = DK_CV_FILE; 5480 DirectiveKindMap[".cv_func_id"] = DK_CV_FUNC_ID; 5481 DirectiveKindMap[".cv_loc"] = DK_CV_LOC; 5482 DirectiveKindMap[".cv_linetable"] = DK_CV_LINETABLE; 5483 DirectiveKindMap[".cv_inline_linetable"] = DK_CV_INLINE_LINETABLE; 5484 DirectiveKindMap[".cv_inline_site_id"] = DK_CV_INLINE_SITE_ID; 5485 DirectiveKindMap[".cv_def_range"] = DK_CV_DEF_RANGE; 5486 DirectiveKindMap[".cv_string"] = DK_CV_STRING; 5487 DirectiveKindMap[".cv_stringtable"] = DK_CV_STRINGTABLE; 5488 DirectiveKindMap[".cv_filechecksums"] = DK_CV_FILECHECKSUMS; 5489 DirectiveKindMap[".cv_filechecksumoffset"] = DK_CV_FILECHECKSUM_OFFSET; 5490 DirectiveKindMap[".cv_fpo_data"] = DK_CV_FPO_DATA; 5491 DirectiveKindMap[".sleb128"] = DK_SLEB128; 5492 DirectiveKindMap[".uleb128"] = DK_ULEB128; 5493 DirectiveKindMap[".cfi_sections"] = DK_CFI_SECTIONS; 5494 DirectiveKindMap[".cfi_startproc"] = DK_CFI_STARTPROC; 5495 DirectiveKindMap[".cfi_endproc"] = DK_CFI_ENDPROC; 5496 DirectiveKindMap[".cfi_def_cfa"] = DK_CFI_DEF_CFA; 5497 DirectiveKindMap[".cfi_def_cfa_offset"] = DK_CFI_DEF_CFA_OFFSET; 5498 DirectiveKindMap[".cfi_adjust_cfa_offset"] = DK_CFI_ADJUST_CFA_OFFSET; 5499 DirectiveKindMap[".cfi_def_cfa_register"] = DK_CFI_DEF_CFA_REGISTER; 5500 DirectiveKindMap[".cfi_offset"] = DK_CFI_OFFSET; 5501 DirectiveKindMap[".cfi_rel_offset"] = DK_CFI_REL_OFFSET; 5502 DirectiveKindMap[".cfi_personality"] = DK_CFI_PERSONALITY; 5503 DirectiveKindMap[".cfi_lsda"] = DK_CFI_LSDA; 5504 DirectiveKindMap[".cfi_remember_state"] = DK_CFI_REMEMBER_STATE; 5505 DirectiveKindMap[".cfi_restore_state"] = DK_CFI_RESTORE_STATE; 5506 DirectiveKindMap[".cfi_same_value"] = DK_CFI_SAME_VALUE; 5507 DirectiveKindMap[".cfi_restore"] = DK_CFI_RESTORE; 5508 DirectiveKindMap[".cfi_escape"] = DK_CFI_ESCAPE; 5509 DirectiveKindMap[".cfi_return_column"] = DK_CFI_RETURN_COLUMN; 5510 DirectiveKindMap[".cfi_signal_frame"] = DK_CFI_SIGNAL_FRAME; 5511 DirectiveKindMap[".cfi_undefined"] = DK_CFI_UNDEFINED; 5512 DirectiveKindMap[".cfi_register"] = DK_CFI_REGISTER; 5513 DirectiveKindMap[".cfi_window_save"] = DK_CFI_WINDOW_SAVE; 5514 DirectiveKindMap[".cfi_b_key_frame"] = DK_CFI_B_KEY_FRAME; 5515 DirectiveKindMap[".macros_on"] = DK_MACROS_ON; 5516 DirectiveKindMap[".macros_off"] = DK_MACROS_OFF; 5517 DirectiveKindMap[".macro"] = DK_MACRO; 5518 DirectiveKindMap[".exitm"] = DK_EXITM; 5519 DirectiveKindMap[".endm"] = DK_ENDM; 5520 DirectiveKindMap[".endmacro"] = DK_ENDMACRO; 5521 DirectiveKindMap[".purgem"] = DK_PURGEM; 5522 DirectiveKindMap[".err"] = DK_ERR; 5523 DirectiveKindMap[".error"] = DK_ERROR; 5524 DirectiveKindMap[".warning"] = DK_WARNING; 5525 DirectiveKindMap[".altmacro"] = DK_ALTMACRO; 5526 DirectiveKindMap[".noaltmacro"] = DK_NOALTMACRO; 5527 DirectiveKindMap[".reloc"] = DK_RELOC; 5528 DirectiveKindMap[".dc"] = DK_DC; 5529 DirectiveKindMap[".dc.a"] = DK_DC_A; 5530 DirectiveKindMap[".dc.b"] = DK_DC_B; 5531 DirectiveKindMap[".dc.d"] = DK_DC_D; 5532 DirectiveKindMap[".dc.l"] = DK_DC_L; 5533 DirectiveKindMap[".dc.s"] = DK_DC_S; 5534 DirectiveKindMap[".dc.w"] = DK_DC_W; 5535 DirectiveKindMap[".dc.x"] = DK_DC_X; 5536 DirectiveKindMap[".dcb"] = DK_DCB; 5537 DirectiveKindMap[".dcb.b"] = DK_DCB_B; 5538 DirectiveKindMap[".dcb.d"] = DK_DCB_D; 5539 DirectiveKindMap[".dcb.l"] = DK_DCB_L; 5540 DirectiveKindMap[".dcb.s"] = DK_DCB_S; 5541 DirectiveKindMap[".dcb.w"] = DK_DCB_W; 5542 DirectiveKindMap[".dcb.x"] = DK_DCB_X; 5543 DirectiveKindMap[".ds"] = DK_DS; 5544 DirectiveKindMap[".ds.b"] = DK_DS_B; 5545 DirectiveKindMap[".ds.d"] = DK_DS_D; 5546 DirectiveKindMap[".ds.l"] = DK_DS_L; 5547 DirectiveKindMap[".ds.p"] = DK_DS_P; 5548 DirectiveKindMap[".ds.s"] = DK_DS_S; 5549 DirectiveKindMap[".ds.w"] = DK_DS_W; 5550 DirectiveKindMap[".ds.x"] = DK_DS_X; 5551 DirectiveKindMap[".print"] = DK_PRINT; 5552 DirectiveKindMap[".addrsig"] = DK_ADDRSIG; 5553 DirectiveKindMap[".addrsig_sym"] = DK_ADDRSIG_SYM; 5554 DirectiveKindMap[".pseudoprobe"] = DK_PSEUDO_PROBE; 5555 DirectiveKindMap[".lto_discard"] = DK_LTO_DISCARD; 5556 } 5557 5558 MCAsmMacro *AsmParser::parseMacroLikeBody(SMLoc DirectiveLoc) { 5559 AsmToken EndToken, StartToken = getTok(); 5560 5561 unsigned NestLevel = 0; 5562 while (true) { 5563 // Check whether we have reached the end of the file. 5564 if (getLexer().is(AsmToken::Eof)) { 5565 printError(DirectiveLoc, "no matching '.endr' in definition"); 5566 return nullptr; 5567 } 5568 5569 if (Lexer.is(AsmToken::Identifier) && 5570 (getTok().getIdentifier() == ".rep" || 5571 getTok().getIdentifier() == ".rept" || 5572 getTok().getIdentifier() == ".irp" || 5573 getTok().getIdentifier() == ".irpc")) { 5574 ++NestLevel; 5575 } 5576 5577 // Otherwise, check whether we have reached the .endr. 5578 if (Lexer.is(AsmToken::Identifier) && getTok().getIdentifier() == ".endr") { 5579 if (NestLevel == 0) { 5580 EndToken = getTok(); 5581 Lex(); 5582 if (Lexer.isNot(AsmToken::EndOfStatement)) { 5583 printError(getTok().getLoc(), 5584 "unexpected token in '.endr' directive"); 5585 return nullptr; 5586 } 5587 break; 5588 } 5589 --NestLevel; 5590 } 5591 5592 // Otherwise, scan till the end of the statement. 5593 eatToEndOfStatement(); 5594 } 5595 5596 const char *BodyStart = StartToken.getLoc().getPointer(); 5597 const char *BodyEnd = EndToken.getLoc().getPointer(); 5598 StringRef Body = StringRef(BodyStart, BodyEnd - BodyStart); 5599 5600 // We Are Anonymous. 5601 MacroLikeBodies.emplace_back(StringRef(), Body, MCAsmMacroParameters()); 5602 return &MacroLikeBodies.back(); 5603 } 5604 5605 void AsmParser::instantiateMacroLikeBody(MCAsmMacro *M, SMLoc DirectiveLoc, 5606 raw_svector_ostream &OS) { 5607 OS << ".endr\n"; 5608 5609 std::unique_ptr<MemoryBuffer> Instantiation = 5610 MemoryBuffer::getMemBufferCopy(OS.str(), "<instantiation>"); 5611 5612 // Create the macro instantiation object and add to the current macro 5613 // instantiation stack. 5614 MacroInstantiation *MI = new MacroInstantiation{ 5615 DirectiveLoc, CurBuffer, getTok().getLoc(), TheCondStack.size()}; 5616 ActiveMacros.push_back(MI); 5617 5618 // Jump to the macro instantiation and prime the lexer. 5619 CurBuffer = SrcMgr.AddNewSourceBuffer(std::move(Instantiation), SMLoc()); 5620 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer()); 5621 Lex(); 5622 } 5623 5624 /// parseDirectiveRept 5625 /// ::= .rep | .rept count 5626 bool AsmParser::parseDirectiveRept(SMLoc DirectiveLoc, StringRef Dir) { 5627 const MCExpr *CountExpr; 5628 SMLoc CountLoc = getTok().getLoc(); 5629 if (parseExpression(CountExpr)) 5630 return true; 5631 5632 int64_t Count; 5633 if (!CountExpr->evaluateAsAbsolute(Count, getStreamer().getAssemblerPtr())) { 5634 return Error(CountLoc, "unexpected token in '" + Dir + "' directive"); 5635 } 5636 5637 if (check(Count < 0, CountLoc, "Count is negative") || parseEOL()) 5638 return true; 5639 5640 // Lex the rept definition. 5641 MCAsmMacro *M = parseMacroLikeBody(DirectiveLoc); 5642 if (!M) 5643 return true; 5644 5645 // Macro instantiation is lexical, unfortunately. We construct a new buffer 5646 // to hold the macro body with substitutions. 5647 SmallString<256> Buf; 5648 raw_svector_ostream OS(Buf); 5649 while (Count--) { 5650 // Note that the AtPseudoVariable is disabled for instantiations of .rep(t). 5651 if (expandMacro(OS, M->Body, None, None, false, getTok().getLoc())) 5652 return true; 5653 } 5654 instantiateMacroLikeBody(M, DirectiveLoc, OS); 5655 5656 return false; 5657 } 5658 5659 /// parseDirectiveIrp 5660 /// ::= .irp symbol,values 5661 bool AsmParser::parseDirectiveIrp(SMLoc DirectiveLoc) { 5662 MCAsmMacroParameter Parameter; 5663 MCAsmMacroArguments A; 5664 if (check(parseIdentifier(Parameter.Name), 5665 "expected identifier in '.irp' directive") || 5666 parseComma() || parseMacroArguments(nullptr, A) || parseEOL()) 5667 return true; 5668 5669 // Lex the irp definition. 5670 MCAsmMacro *M = parseMacroLikeBody(DirectiveLoc); 5671 if (!M) 5672 return true; 5673 5674 // Macro instantiation is lexical, unfortunately. We construct a new buffer 5675 // to hold the macro body with substitutions. 5676 SmallString<256> Buf; 5677 raw_svector_ostream OS(Buf); 5678 5679 for (const MCAsmMacroArgument &Arg : A) { 5680 // Note that the AtPseudoVariable is enabled for instantiations of .irp. 5681 // This is undocumented, but GAS seems to support it. 5682 if (expandMacro(OS, M->Body, Parameter, Arg, true, getTok().getLoc())) 5683 return true; 5684 } 5685 5686 instantiateMacroLikeBody(M, DirectiveLoc, OS); 5687 5688 return false; 5689 } 5690 5691 /// parseDirectiveIrpc 5692 /// ::= .irpc symbol,values 5693 bool AsmParser::parseDirectiveIrpc(SMLoc DirectiveLoc) { 5694 MCAsmMacroParameter Parameter; 5695 MCAsmMacroArguments A; 5696 5697 if (check(parseIdentifier(Parameter.Name), 5698 "expected identifier in '.irpc' directive") || 5699 parseComma() || parseMacroArguments(nullptr, A)) 5700 return true; 5701 5702 if (A.size() != 1 || A.front().size() != 1) 5703 return TokError("unexpected token in '.irpc' directive"); 5704 if (parseEOL()) 5705 return true; 5706 5707 // Lex the irpc definition. 5708 MCAsmMacro *M = parseMacroLikeBody(DirectiveLoc); 5709 if (!M) 5710 return true; 5711 5712 // Macro instantiation is lexical, unfortunately. We construct a new buffer 5713 // to hold the macro body with substitutions. 5714 SmallString<256> Buf; 5715 raw_svector_ostream OS(Buf); 5716 5717 StringRef Values = A.front().front().getString(); 5718 for (std::size_t I = 0, End = Values.size(); I != End; ++I) { 5719 MCAsmMacroArgument Arg; 5720 Arg.emplace_back(AsmToken::Identifier, Values.slice(I, I + 1)); 5721 5722 // Note that the AtPseudoVariable is enabled for instantiations of .irpc. 5723 // This is undocumented, but GAS seems to support it. 5724 if (expandMacro(OS, M->Body, Parameter, Arg, true, getTok().getLoc())) 5725 return true; 5726 } 5727 5728 instantiateMacroLikeBody(M, DirectiveLoc, OS); 5729 5730 return false; 5731 } 5732 5733 bool AsmParser::parseDirectiveEndr(SMLoc DirectiveLoc) { 5734 if (ActiveMacros.empty()) 5735 return TokError("unmatched '.endr' directive"); 5736 5737 // The only .repl that should get here are the ones created by 5738 // instantiateMacroLikeBody. 5739 assert(getLexer().is(AsmToken::EndOfStatement)); 5740 5741 handleMacroExit(); 5742 return false; 5743 } 5744 5745 bool AsmParser::parseDirectiveMSEmit(SMLoc IDLoc, ParseStatementInfo &Info, 5746 size_t Len) { 5747 const MCExpr *Value; 5748 SMLoc ExprLoc = getLexer().getLoc(); 5749 if (parseExpression(Value)) 5750 return true; 5751 const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value); 5752 if (!MCE) 5753 return Error(ExprLoc, "unexpected expression in _emit"); 5754 uint64_t IntValue = MCE->getValue(); 5755 if (!isUInt<8>(IntValue) && !isInt<8>(IntValue)) 5756 return Error(ExprLoc, "literal value out of range for directive"); 5757 5758 Info.AsmRewrites->emplace_back(AOK_Emit, IDLoc, Len); 5759 return false; 5760 } 5761 5762 bool AsmParser::parseDirectiveMSAlign(SMLoc IDLoc, ParseStatementInfo &Info) { 5763 const MCExpr *Value; 5764 SMLoc ExprLoc = getLexer().getLoc(); 5765 if (parseExpression(Value)) 5766 return true; 5767 const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value); 5768 if (!MCE) 5769 return Error(ExprLoc, "unexpected expression in align"); 5770 uint64_t IntValue = MCE->getValue(); 5771 if (!isPowerOf2_64(IntValue)) 5772 return Error(ExprLoc, "literal value not a power of two greater then zero"); 5773 5774 Info.AsmRewrites->emplace_back(AOK_Align, IDLoc, 5, Log2_64(IntValue)); 5775 return false; 5776 } 5777 5778 bool AsmParser::parseDirectivePrint(SMLoc DirectiveLoc) { 5779 const AsmToken StrTok = getTok(); 5780 Lex(); 5781 if (StrTok.isNot(AsmToken::String) || StrTok.getString().front() != '"') 5782 return Error(DirectiveLoc, "expected double quoted string after .print"); 5783 if (parseEOL()) 5784 return true; 5785 llvm::outs() << StrTok.getStringContents() << '\n'; 5786 return false; 5787 } 5788 5789 bool AsmParser::parseDirectiveAddrsig() { 5790 if (parseEOL()) 5791 return true; 5792 getStreamer().emitAddrsig(); 5793 return false; 5794 } 5795 5796 bool AsmParser::parseDirectiveAddrsigSym() { 5797 StringRef Name; 5798 if (check(parseIdentifier(Name), "expected identifier") || parseEOL()) 5799 return true; 5800 MCSymbol *Sym = getContext().getOrCreateSymbol(Name); 5801 getStreamer().emitAddrsigSym(Sym); 5802 return false; 5803 } 5804 5805 bool AsmParser::parseDirectivePseudoProbe() { 5806 int64_t Guid; 5807 int64_t Index; 5808 int64_t Type; 5809 int64_t Attr; 5810 5811 if (getLexer().is(AsmToken::Integer)) { 5812 if (parseIntToken(Guid, "unexpected token in '.pseudoprobe' directive")) 5813 return true; 5814 } 5815 5816 if (getLexer().is(AsmToken::Integer)) { 5817 if (parseIntToken(Index, "unexpected token in '.pseudoprobe' directive")) 5818 return true; 5819 } 5820 5821 if (getLexer().is(AsmToken::Integer)) { 5822 if (parseIntToken(Type, "unexpected token in '.pseudoprobe' directive")) 5823 return true; 5824 } 5825 5826 if (getLexer().is(AsmToken::Integer)) { 5827 if (parseIntToken(Attr, "unexpected token in '.pseudoprobe' directive")) 5828 return true; 5829 } 5830 5831 // Parse inline stack like @ GUID:11:12 @ GUID:1:11 @ GUID:3:21 5832 MCPseudoProbeInlineStack InlineStack; 5833 5834 while (getLexer().is(AsmToken::At)) { 5835 // eat @ 5836 Lex(); 5837 5838 int64_t CallerGuid = 0; 5839 if (getLexer().is(AsmToken::Integer)) { 5840 if (parseIntToken(CallerGuid, 5841 "unexpected token in '.pseudoprobe' directive")) 5842 return true; 5843 } 5844 5845 // eat colon 5846 if (getLexer().is(AsmToken::Colon)) 5847 Lex(); 5848 5849 int64_t CallerProbeId = 0; 5850 if (getLexer().is(AsmToken::Integer)) { 5851 if (parseIntToken(CallerProbeId, 5852 "unexpected token in '.pseudoprobe' directive")) 5853 return true; 5854 } 5855 5856 InlineSite Site(CallerGuid, CallerProbeId); 5857 InlineStack.push_back(Site); 5858 } 5859 5860 if (parseEOL()) 5861 return true; 5862 5863 getStreamer().emitPseudoProbe(Guid, Index, Type, Attr, InlineStack); 5864 return false; 5865 } 5866 5867 /// parseDirectiveLTODiscard 5868 /// ::= ".lto_discard" [ identifier ( , identifier )* ] 5869 /// The LTO library emits this directive to discard non-prevailing symbols. 5870 /// We ignore symbol assignments and attribute changes for the specified 5871 /// symbols. 5872 bool AsmParser::parseDirectiveLTODiscard() { 5873 auto ParseOp = [&]() -> bool { 5874 StringRef Name; 5875 SMLoc Loc = getTok().getLoc(); 5876 if (parseIdentifier(Name)) 5877 return Error(Loc, "expected identifier"); 5878 LTODiscardSymbols.insert(Name); 5879 return false; 5880 }; 5881 5882 LTODiscardSymbols.clear(); 5883 return parseMany(ParseOp); 5884 } 5885 5886 // We are comparing pointers, but the pointers are relative to a single string. 5887 // Thus, this should always be deterministic. 5888 static int rewritesSort(const AsmRewrite *AsmRewriteA, 5889 const AsmRewrite *AsmRewriteB) { 5890 if (AsmRewriteA->Loc.getPointer() < AsmRewriteB->Loc.getPointer()) 5891 return -1; 5892 if (AsmRewriteB->Loc.getPointer() < AsmRewriteA->Loc.getPointer()) 5893 return 1; 5894 5895 // It's possible to have a SizeDirective, Imm/ImmPrefix and an Input/Output 5896 // rewrite to the same location. Make sure the SizeDirective rewrite is 5897 // performed first, then the Imm/ImmPrefix and finally the Input/Output. This 5898 // ensures the sort algorithm is stable. 5899 if (AsmRewritePrecedence[AsmRewriteA->Kind] > 5900 AsmRewritePrecedence[AsmRewriteB->Kind]) 5901 return -1; 5902 5903 if (AsmRewritePrecedence[AsmRewriteA->Kind] < 5904 AsmRewritePrecedence[AsmRewriteB->Kind]) 5905 return 1; 5906 llvm_unreachable("Unstable rewrite sort."); 5907 } 5908 5909 bool AsmParser::parseMSInlineAsm( 5910 void *AsmLoc, std::string &AsmString, unsigned &NumOutputs, 5911 unsigned &NumInputs, SmallVectorImpl<std::pair<void *, bool>> &OpDecls, 5912 SmallVectorImpl<std::string> &Constraints, 5913 SmallVectorImpl<std::string> &Clobbers, const MCInstrInfo *MII, 5914 const MCInstPrinter *IP, MCAsmParserSemaCallback &SI) { 5915 SmallVector<void *, 4> InputDecls; 5916 SmallVector<void *, 4> OutputDecls; 5917 SmallVector<bool, 4> InputDeclsAddressOf; 5918 SmallVector<bool, 4> OutputDeclsAddressOf; 5919 SmallVector<std::string, 4> InputConstraints; 5920 SmallVector<std::string, 4> OutputConstraints; 5921 SmallVector<unsigned, 4> ClobberRegs; 5922 5923 SmallVector<AsmRewrite, 4> AsmStrRewrites; 5924 5925 // Prime the lexer. 5926 Lex(); 5927 5928 // While we have input, parse each statement. 5929 unsigned InputIdx = 0; 5930 unsigned OutputIdx = 0; 5931 while (getLexer().isNot(AsmToken::Eof)) { 5932 // Parse curly braces marking block start/end 5933 if (parseCurlyBlockScope(AsmStrRewrites)) 5934 continue; 5935 5936 ParseStatementInfo Info(&AsmStrRewrites); 5937 bool StatementErr = parseStatement(Info, &SI); 5938 5939 if (StatementErr || Info.ParseError) { 5940 // Emit pending errors if any exist. 5941 printPendingErrors(); 5942 return true; 5943 } 5944 5945 // No pending error should exist here. 5946 assert(!hasPendingError() && "unexpected error from parseStatement"); 5947 5948 if (Info.Opcode == ~0U) 5949 continue; 5950 5951 const MCInstrDesc &Desc = MII->get(Info.Opcode); 5952 5953 // Build the list of clobbers, outputs and inputs. 5954 for (unsigned i = 1, e = Info.ParsedOperands.size(); i != e; ++i) { 5955 MCParsedAsmOperand &Operand = *Info.ParsedOperands[i]; 5956 5957 // Register operand. 5958 if (Operand.isReg() && !Operand.needAddressOf() && 5959 !getTargetParser().OmitRegisterFromClobberLists(Operand.getReg())) { 5960 unsigned NumDefs = Desc.getNumDefs(); 5961 // Clobber. 5962 if (NumDefs && Operand.getMCOperandNum() < NumDefs) 5963 ClobberRegs.push_back(Operand.getReg()); 5964 continue; 5965 } 5966 5967 // Expr/Input or Output. 5968 StringRef SymName = Operand.getSymName(); 5969 if (SymName.empty()) 5970 continue; 5971 5972 void *OpDecl = Operand.getOpDecl(); 5973 if (!OpDecl) 5974 continue; 5975 5976 StringRef Constraint = Operand.getConstraint(); 5977 if (Operand.isImm()) { 5978 // Offset as immediate 5979 if (Operand.isOffsetOfLocal()) 5980 Constraint = "r"; 5981 else 5982 Constraint = "i"; 5983 } 5984 5985 bool isOutput = (i == 1) && Desc.mayStore(); 5986 SMLoc Start = SMLoc::getFromPointer(SymName.data()); 5987 if (isOutput) { 5988 ++InputIdx; 5989 OutputDecls.push_back(OpDecl); 5990 OutputDeclsAddressOf.push_back(Operand.needAddressOf()); 5991 OutputConstraints.push_back(("=" + Constraint).str()); 5992 AsmStrRewrites.emplace_back(AOK_Output, Start, SymName.size()); 5993 } else { 5994 InputDecls.push_back(OpDecl); 5995 InputDeclsAddressOf.push_back(Operand.needAddressOf()); 5996 InputConstraints.push_back(Constraint.str()); 5997 if (Desc.OpInfo[i - 1].isBranchTarget()) 5998 AsmStrRewrites.emplace_back(AOK_CallInput, Start, SymName.size()); 5999 else 6000 AsmStrRewrites.emplace_back(AOK_Input, Start, SymName.size()); 6001 } 6002 } 6003 6004 // Consider implicit defs to be clobbers. Think of cpuid and push. 6005 ArrayRef<MCPhysReg> ImpDefs(Desc.getImplicitDefs(), 6006 Desc.getNumImplicitDefs()); 6007 llvm::append_range(ClobberRegs, ImpDefs); 6008 } 6009 6010 // Set the number of Outputs and Inputs. 6011 NumOutputs = OutputDecls.size(); 6012 NumInputs = InputDecls.size(); 6013 6014 // Set the unique clobbers. 6015 array_pod_sort(ClobberRegs.begin(), ClobberRegs.end()); 6016 ClobberRegs.erase(std::unique(ClobberRegs.begin(), ClobberRegs.end()), 6017 ClobberRegs.end()); 6018 Clobbers.assign(ClobberRegs.size(), std::string()); 6019 for (unsigned I = 0, E = ClobberRegs.size(); I != E; ++I) { 6020 raw_string_ostream OS(Clobbers[I]); 6021 IP->printRegName(OS, ClobberRegs[I]); 6022 } 6023 6024 // Merge the various outputs and inputs. Output are expected first. 6025 if (NumOutputs || NumInputs) { 6026 unsigned NumExprs = NumOutputs + NumInputs; 6027 OpDecls.resize(NumExprs); 6028 Constraints.resize(NumExprs); 6029 for (unsigned i = 0; i < NumOutputs; ++i) { 6030 OpDecls[i] = std::make_pair(OutputDecls[i], OutputDeclsAddressOf[i]); 6031 Constraints[i] = OutputConstraints[i]; 6032 } 6033 for (unsigned i = 0, j = NumOutputs; i < NumInputs; ++i, ++j) { 6034 OpDecls[j] = std::make_pair(InputDecls[i], InputDeclsAddressOf[i]); 6035 Constraints[j] = InputConstraints[i]; 6036 } 6037 } 6038 6039 // Build the IR assembly string. 6040 std::string AsmStringIR; 6041 raw_string_ostream OS(AsmStringIR); 6042 StringRef ASMString = 6043 SrcMgr.getMemoryBuffer(SrcMgr.getMainFileID())->getBuffer(); 6044 const char *AsmStart = ASMString.begin(); 6045 const char *AsmEnd = ASMString.end(); 6046 array_pod_sort(AsmStrRewrites.begin(), AsmStrRewrites.end(), rewritesSort); 6047 for (auto it = AsmStrRewrites.begin(); it != AsmStrRewrites.end(); ++it) { 6048 const AsmRewrite &AR = *it; 6049 // Check if this has already been covered by another rewrite... 6050 if (AR.Done) 6051 continue; 6052 AsmRewriteKind Kind = AR.Kind; 6053 6054 const char *Loc = AR.Loc.getPointer(); 6055 assert(Loc >= AsmStart && "Expected Loc to be at or after Start!"); 6056 6057 // Emit everything up to the immediate/expression. 6058 if (unsigned Len = Loc - AsmStart) 6059 OS << StringRef(AsmStart, Len); 6060 6061 // Skip the original expression. 6062 if (Kind == AOK_Skip) { 6063 AsmStart = Loc + AR.Len; 6064 continue; 6065 } 6066 6067 unsigned AdditionalSkip = 0; 6068 // Rewrite expressions in $N notation. 6069 switch (Kind) { 6070 default: 6071 break; 6072 case AOK_IntelExpr: 6073 assert(AR.IntelExp.isValid() && "cannot write invalid intel expression"); 6074 if (AR.IntelExp.NeedBracs) 6075 OS << "["; 6076 if (AR.IntelExp.hasBaseReg()) 6077 OS << AR.IntelExp.BaseReg; 6078 if (AR.IntelExp.hasIndexReg()) 6079 OS << (AR.IntelExp.hasBaseReg() ? " + " : "") 6080 << AR.IntelExp.IndexReg; 6081 if (AR.IntelExp.Scale > 1) 6082 OS << " * $$" << AR.IntelExp.Scale; 6083 if (AR.IntelExp.hasOffset()) { 6084 if (AR.IntelExp.hasRegs()) 6085 OS << " + "; 6086 // Fuse this rewrite with a rewrite of the offset name, if present. 6087 StringRef OffsetName = AR.IntelExp.OffsetName; 6088 SMLoc OffsetLoc = SMLoc::getFromPointer(AR.IntelExp.OffsetName.data()); 6089 size_t OffsetLen = OffsetName.size(); 6090 auto rewrite_it = std::find_if( 6091 it, AsmStrRewrites.end(), [&](const AsmRewrite &FusingAR) { 6092 return FusingAR.Loc == OffsetLoc && FusingAR.Len == OffsetLen && 6093 (FusingAR.Kind == AOK_Input || 6094 FusingAR.Kind == AOK_CallInput); 6095 }); 6096 if (rewrite_it == AsmStrRewrites.end()) { 6097 OS << "offset " << OffsetName; 6098 } else if (rewrite_it->Kind == AOK_CallInput) { 6099 OS << "${" << InputIdx++ << ":P}"; 6100 rewrite_it->Done = true; 6101 } else { 6102 OS << '$' << InputIdx++; 6103 rewrite_it->Done = true; 6104 } 6105 } 6106 if (AR.IntelExp.Imm || AR.IntelExp.emitImm()) 6107 OS << (AR.IntelExp.emitImm() ? "$$" : " + $$") << AR.IntelExp.Imm; 6108 if (AR.IntelExp.NeedBracs) 6109 OS << "]"; 6110 break; 6111 case AOK_Label: 6112 OS << Ctx.getAsmInfo()->getPrivateLabelPrefix() << AR.Label; 6113 break; 6114 case AOK_Input: 6115 OS << '$' << InputIdx++; 6116 break; 6117 case AOK_CallInput: 6118 OS << "${" << InputIdx++ << ":P}"; 6119 break; 6120 case AOK_Output: 6121 OS << '$' << OutputIdx++; 6122 break; 6123 case AOK_SizeDirective: 6124 switch (AR.Val) { 6125 default: break; 6126 case 8: OS << "byte ptr "; break; 6127 case 16: OS << "word ptr "; break; 6128 case 32: OS << "dword ptr "; break; 6129 case 64: OS << "qword ptr "; break; 6130 case 80: OS << "xword ptr "; break; 6131 case 128: OS << "xmmword ptr "; break; 6132 case 256: OS << "ymmword ptr "; break; 6133 } 6134 break; 6135 case AOK_Emit: 6136 OS << ".byte"; 6137 break; 6138 case AOK_Align: { 6139 // MS alignment directives are measured in bytes. If the native assembler 6140 // measures alignment in bytes, we can pass it straight through. 6141 OS << ".align"; 6142 if (getContext().getAsmInfo()->getAlignmentIsInBytes()) 6143 break; 6144 6145 // Alignment is in log2 form, so print that instead and skip the original 6146 // immediate. 6147 unsigned Val = AR.Val; 6148 OS << ' ' << Val; 6149 assert(Val < 10 && "Expected alignment less then 2^10."); 6150 AdditionalSkip = (Val < 4) ? 2 : Val < 7 ? 3 : 4; 6151 break; 6152 } 6153 case AOK_EVEN: 6154 OS << ".even"; 6155 break; 6156 case AOK_EndOfStatement: 6157 OS << "\n\t"; 6158 break; 6159 } 6160 6161 // Skip the original expression. 6162 AsmStart = Loc + AR.Len + AdditionalSkip; 6163 } 6164 6165 // Emit the remainder of the asm string. 6166 if (AsmStart != AsmEnd) 6167 OS << StringRef(AsmStart, AsmEnd - AsmStart); 6168 6169 AsmString = OS.str(); 6170 return false; 6171 } 6172 6173 bool HLASMAsmParser::parseAsMachineInstruction(ParseStatementInfo &Info, 6174 MCAsmParserSemaCallback *SI) { 6175 AsmToken OperationEntryTok = Lexer.getTok(); 6176 SMLoc OperationEntryLoc = OperationEntryTok.getLoc(); 6177 StringRef OperationEntryVal; 6178 6179 // If we see a new line or carriage return, emit the new line 6180 // and lex it. 6181 if (OperationEntryTok.is(AsmToken::EndOfStatement)) { 6182 if (getTok().getString().front() == '\n' || 6183 getTok().getString().front() == '\r') { 6184 Out.AddBlankLine(); 6185 Lex(); 6186 return false; 6187 } 6188 } 6189 6190 // Attempt to parse the first token as an Identifier 6191 if (parseIdentifier(OperationEntryVal)) 6192 return Error(OperationEntryLoc, "unexpected token at start of statement"); 6193 6194 // Once we've parsed the operation entry successfully, lex 6195 // any spaces to get to the OperandEntries. 6196 lexLeadingSpaces(); 6197 6198 return parseAndMatchAndEmitTargetInstruction( 6199 Info, OperationEntryVal, OperationEntryTok, OperationEntryLoc); 6200 } 6201 6202 bool HLASMAsmParser::parseStatement(ParseStatementInfo &Info, 6203 MCAsmParserSemaCallback *SI) { 6204 assert(!hasPendingError() && "parseStatement started with pending error"); 6205 6206 // Should the first token be interpreted as a machine instruction. 6207 bool ShouldParseAsMachineInstruction = false; 6208 6209 // If a Name Entry exists, it should occur at the very 6210 // start of the string. In this case, we should parse the 6211 // first non-space token as a Label. 6212 // If the Name entry is missing (i.e. there's some other 6213 // token), then we attempt to parse the first non-space 6214 // token as a Machine Instruction. 6215 if (getTok().is(AsmToken::Space)) 6216 ShouldParseAsMachineInstruction = true; 6217 6218 // If we have an EndOfStatement (which includes the target's comment 6219 // string) we can appropriately lex it early on) 6220 if (Lexer.is(AsmToken::EndOfStatement)) { 6221 // if this is a line comment we can drop it safely 6222 if (getTok().getString().empty() || getTok().getString().front() == '\r' || 6223 getTok().getString().front() == '\n') 6224 Out.AddBlankLine(); 6225 Lex(); 6226 return false; 6227 } 6228 6229 // We have established how to parse the inline asm statement. 6230 // Now we can safely lex any leading spaces to get to the 6231 // first token. 6232 lexLeadingSpaces(); 6233 6234 if (ShouldParseAsMachineInstruction) 6235 return parseAsMachineInstruction(Info, SI); 6236 6237 // Label parsing support isn't implemented completely (yet). 6238 SMLoc Loc = getTok().getLoc(); 6239 eatToEndOfStatement(); 6240 return Error(Loc, "HLASM Label parsing support not yet implemented"); 6241 } 6242 6243 namespace llvm { 6244 namespace MCParserUtils { 6245 6246 /// Returns whether the given symbol is used anywhere in the given expression, 6247 /// or subexpressions. 6248 static bool isSymbolUsedInExpression(const MCSymbol *Sym, const MCExpr *Value) { 6249 switch (Value->getKind()) { 6250 case MCExpr::Binary: { 6251 const MCBinaryExpr *BE = static_cast<const MCBinaryExpr *>(Value); 6252 return isSymbolUsedInExpression(Sym, BE->getLHS()) || 6253 isSymbolUsedInExpression(Sym, BE->getRHS()); 6254 } 6255 case MCExpr::Target: 6256 case MCExpr::Constant: 6257 return false; 6258 case MCExpr::SymbolRef: { 6259 const MCSymbol &S = 6260 static_cast<const MCSymbolRefExpr *>(Value)->getSymbol(); 6261 if (S.isVariable()) 6262 return isSymbolUsedInExpression(Sym, S.getVariableValue()); 6263 return &S == Sym; 6264 } 6265 case MCExpr::Unary: 6266 return isSymbolUsedInExpression( 6267 Sym, static_cast<const MCUnaryExpr *>(Value)->getSubExpr()); 6268 } 6269 6270 llvm_unreachable("Unknown expr kind!"); 6271 } 6272 6273 bool parseAssignmentExpression(StringRef Name, bool allow_redef, 6274 MCAsmParser &Parser, MCSymbol *&Sym, 6275 const MCExpr *&Value) { 6276 6277 // FIXME: Use better location, we should use proper tokens. 6278 SMLoc EqualLoc = Parser.getTok().getLoc(); 6279 if (Parser.parseExpression(Value)) 6280 return Parser.TokError("missing expression"); 6281 6282 // Note: we don't count b as used in "a = b". This is to allow 6283 // a = b 6284 // b = c 6285 6286 if (Parser.parseEOL()) 6287 return true; 6288 6289 // Validate that the LHS is allowed to be a variable (either it has not been 6290 // used as a symbol, or it is an absolute symbol). 6291 Sym = Parser.getContext().lookupSymbol(Name); 6292 if (Sym) { 6293 // Diagnose assignment to a label. 6294 // 6295 // FIXME: Diagnostics. Note the location of the definition as a label. 6296 // FIXME: Diagnose assignment to protected identifier (e.g., register name). 6297 if (isSymbolUsedInExpression(Sym, Value)) 6298 return Parser.Error(EqualLoc, "Recursive use of '" + Name + "'"); 6299 else if (Sym->isUndefined(/*SetUsed*/ false) && !Sym->isUsed() && 6300 !Sym->isVariable()) 6301 ; // Allow redefinitions of undefined symbols only used in directives. 6302 else if (Sym->isVariable() && !Sym->isUsed() && allow_redef) 6303 ; // Allow redefinitions of variables that haven't yet been used. 6304 else if (!Sym->isUndefined() && (!Sym->isVariable() || !allow_redef)) 6305 return Parser.Error(EqualLoc, "redefinition of '" + Name + "'"); 6306 else if (!Sym->isVariable()) 6307 return Parser.Error(EqualLoc, "invalid assignment to '" + Name + "'"); 6308 else if (!isa<MCConstantExpr>(Sym->getVariableValue())) 6309 return Parser.Error(EqualLoc, 6310 "invalid reassignment of non-absolute variable '" + 6311 Name + "'"); 6312 } else if (Name == ".") { 6313 Parser.getStreamer().emitValueToOffset(Value, 0, EqualLoc); 6314 return false; 6315 } else 6316 Sym = Parser.getContext().getOrCreateSymbol(Name); 6317 6318 Sym->setRedefinable(allow_redef); 6319 6320 return false; 6321 } 6322 6323 } // end namespace MCParserUtils 6324 } // end namespace llvm 6325 6326 /// Create an MCAsmParser instance. 6327 MCAsmParser *llvm::createMCAsmParser(SourceMgr &SM, MCContext &C, 6328 MCStreamer &Out, const MCAsmInfo &MAI, 6329 unsigned CB) { 6330 if (C.getTargetTriple().isOSzOS()) 6331 return new HLASMAsmParser(SM, C, Out, MAI, CB); 6332 6333 return new AsmParser(SM, C, Out, MAI, CB); 6334 } 6335
Indexes created Tue Jun 16 00:25:01 UTC 2026