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/STLExtras.h" 17 #include "llvm/ADT/SmallSet.h" 18 #include "llvm/ADT/SmallString.h" 19 #include "llvm/ADT/SmallVector.h" 20 #include "llvm/ADT/StringExtras.h" 21 #include "llvm/ADT/StringMap.h" 22 #include "llvm/ADT/StringRef.h" 23 #include "llvm/ADT/Twine.h" 24 #include "llvm/BinaryFormat/Dwarf.h" 25 #include "llvm/DebugInfo/CodeView/SymbolRecord.h" 26 #include "llvm/MC/MCAsmInfo.h" 27 #include "llvm/MC/MCCodeView.h" 28 #include "llvm/MC/MCContext.h" 29 #include "llvm/MC/MCDirectives.h" 30 #include "llvm/MC/MCDwarf.h" 31 #include "llvm/MC/MCExpr.h" 32 #include "llvm/MC/MCInstPrinter.h" 33 #include "llvm/MC/MCInstrDesc.h" 34 #include "llvm/MC/MCInstrInfo.h" 35 #include "llvm/MC/MCParser/AsmCond.h" 36 #include "llvm/MC/MCParser/AsmLexer.h" 37 #include "llvm/MC/MCParser/MCAsmLexer.h" 38 #include "llvm/MC/MCParser/MCAsmParser.h" 39 #include "llvm/MC/MCParser/MCAsmParserExtension.h" 40 #include "llvm/MC/MCParser/MCAsmParserUtils.h" 41 #include "llvm/MC/MCParser/MCParsedAsmOperand.h" 42 #include "llvm/MC/MCParser/MCTargetAsmParser.h" 43 #include "llvm/MC/MCRegisterInfo.h" 44 #include "llvm/MC/MCSection.h" 45 #include "llvm/MC/MCStreamer.h" 46 #include "llvm/MC/MCSymbol.h" 47 #include "llvm/MC/MCTargetOptions.h" 48 #include "llvm/MC/MCValue.h" 49 #include "llvm/Support/Casting.h" 50 #include "llvm/Support/CommandLine.h" 51 #include "llvm/Support/ErrorHandling.h" 52 #include "llvm/Support/MD5.h" 53 #include "llvm/Support/MathExtras.h" 54 #include "llvm/Support/MemoryBuffer.h" 55 #include "llvm/Support/SMLoc.h" 56 #include "llvm/Support/SourceMgr.h" 57 #include "llvm/Support/raw_ostream.h" 58 #include <algorithm> 59 #include <cassert> 60 #include <cctype> 61 #include <climits> 62 #include <cstddef> 63 #include <cstdint> 64 #include <deque> 65 #include <memory> 66 #include <optional> 67 #include <sstream> 68 #include <string> 69 #include <tuple> 70 #include <utility> 71 #include <vector> 72 73 using namespace llvm; 74 75 MCAsmParserSemaCallback::~MCAsmParserSemaCallback() = default; 76 77 namespace { 78 79 /// Helper types for tracking macro definitions. 80 typedef std::vector<AsmToken> MCAsmMacroArgument; 81 typedef std::vector<MCAsmMacroArgument> MCAsmMacroArguments; 82 83 /// Helper class for storing information about an active macro 84 /// instantiation. 85 struct MacroInstantiation { 86 /// The location of the instantiation. 87 SMLoc InstantiationLoc; 88 89 /// The buffer where parsing should resume upon instantiation completion. 90 unsigned ExitBuffer; 91 92 /// The location where parsing should resume upon instantiation completion. 93 SMLoc ExitLoc; 94 95 /// The depth of TheCondStack at the start of the instantiation. 96 size_t CondStackDepth; 97 }; 98 99 struct ParseStatementInfo { 100 /// The parsed operands from the last parsed statement. 101 SmallVector<std::unique_ptr<MCParsedAsmOperand>, 8> ParsedOperands; 102 103 /// The opcode from the last parsed instruction. 104 unsigned Opcode = ~0U; 105 106 /// Was there an error parsing the inline assembly? 107 bool ParseError = false; 108 109 SmallVectorImpl<AsmRewrite> *AsmRewrites = nullptr; 110 111 ParseStatementInfo() = delete; 112 ParseStatementInfo(SmallVectorImpl<AsmRewrite> *rewrites) 113 : AsmRewrites(rewrites) {} 114 }; 115 116 /// The concrete assembly parser instance. 117 class AsmParser : public MCAsmParser { 118 private: 119 AsmLexer Lexer; 120 MCContext &Ctx; 121 MCStreamer &Out; 122 const MCAsmInfo &MAI; 123 SourceMgr &SrcMgr; 124 SourceMgr::DiagHandlerTy SavedDiagHandler; 125 void *SavedDiagContext; 126 std::unique_ptr<MCAsmParserExtension> PlatformParser; 127 SMLoc StartTokLoc; 128 129 /// This is the current buffer index we're lexing from as managed by the 130 /// SourceMgr object. 131 unsigned CurBuffer; 132 133 AsmCond TheCondState; 134 std::vector<AsmCond> TheCondStack; 135 136 /// maps directive names to handler methods in parser 137 /// extensions. Extensions register themselves in this map by calling 138 /// addDirectiveHandler. 139 StringMap<ExtensionDirectiveHandler> ExtensionDirectiveMap; 140 141 /// Stack of active macro instantiations. 142 std::vector<MacroInstantiation*> ActiveMacros; 143 144 /// List of bodies of anonymous macros. 145 std::deque<MCAsmMacro> MacroLikeBodies; 146 147 /// Boolean tracking whether macro substitution is enabled. 148 unsigned MacrosEnabledFlag : 1; 149 150 /// Keeps track of how many .macro's have been instantiated. 151 unsigned NumOfMacroInstantiations; 152 153 /// The values from the last parsed cpp hash file line comment if any. 154 struct CppHashInfoTy { 155 StringRef Filename; 156 int64_t LineNumber; 157 SMLoc Loc; 158 unsigned Buf; 159 CppHashInfoTy() : LineNumber(0), Buf(0) {} 160 }; 161 CppHashInfoTy CppHashInfo; 162 163 /// The filename from the first cpp hash file line comment, if any. 164 StringRef FirstCppHashFilename; 165 166 /// List of forward directional labels for diagnosis at the end. 167 SmallVector<std::tuple<SMLoc, CppHashInfoTy, MCSymbol *>, 4> DirLabels; 168 169 SmallSet<StringRef, 2> LTODiscardSymbols; 170 171 /// AssemblerDialect. ~OU means unset value and use value provided by MAI. 172 unsigned AssemblerDialect = ~0U; 173 174 /// is Darwin compatibility enabled? 175 bool IsDarwin = false; 176 177 /// Are we parsing ms-style inline assembly? 178 bool ParsingMSInlineAsm = false; 179 180 /// Did we already inform the user about inconsistent MD5 usage? 181 bool ReportedInconsistentMD5 = false; 182 183 // Is alt macro mode enabled. 184 bool AltMacroMode = false; 185 186 protected: 187 virtual bool parseStatement(ParseStatementInfo &Info, 188 MCAsmParserSemaCallback *SI); 189 190 /// This routine uses the target specific ParseInstruction function to 191 /// parse an instruction into Operands, and then call the target specific 192 /// MatchAndEmit function to match and emit the instruction. 193 bool parseAndMatchAndEmitTargetInstruction(ParseStatementInfo &Info, 194 StringRef IDVal, AsmToken ID, 195 SMLoc IDLoc); 196 197 /// Should we emit DWARF describing this assembler source? (Returns false if 198 /// the source has .file directives, which means we don't want to generate 199 /// info describing the assembler source itself.) 200 bool enabledGenDwarfForAssembly(); 201 202 public: 203 AsmParser(SourceMgr &SM, MCContext &Ctx, MCStreamer &Out, 204 const MCAsmInfo &MAI, unsigned CB); 205 AsmParser(const AsmParser &) = delete; 206 AsmParser &operator=(const AsmParser &) = delete; 207 ~AsmParser() override; 208 209 bool Run(bool NoInitialTextSection, bool NoFinalize = false) override; 210 211 void addDirectiveHandler(StringRef Directive, 212 ExtensionDirectiveHandler Handler) override { 213 ExtensionDirectiveMap[Directive] = Handler; 214 } 215 216 void addAliasForDirective(StringRef Directive, StringRef Alias) override { 217 DirectiveKindMap[Directive.lower()] = DirectiveKindMap[Alias.lower()]; 218 } 219 220 /// @name MCAsmParser Interface 221 /// { 222 223 SourceMgr &getSourceManager() override { return SrcMgr; } 224 MCAsmLexer &getLexer() override { return Lexer; } 225 MCContext &getContext() override { return Ctx; } 226 MCStreamer &getStreamer() override { return Out; } 227 228 CodeViewContext &getCVContext() { return Ctx.getCVContext(); } 229 230 unsigned getAssemblerDialect() override { 231 if (AssemblerDialect == ~0U) 232 return MAI.getAssemblerDialect(); 233 else 234 return AssemblerDialect; 235 } 236 void setAssemblerDialect(unsigned i) override { 237 AssemblerDialect = i; 238 } 239 240 void Note(SMLoc L, const Twine &Msg, SMRange Range = std::nullopt) override; 241 bool Warning(SMLoc L, const Twine &Msg, 242 SMRange Range = std::nullopt) override; 243 bool printError(SMLoc L, const Twine &Msg, 244 SMRange Range = std::nullopt) override; 245 246 const AsmToken &Lex() override; 247 248 void setParsingMSInlineAsm(bool V) override { 249 ParsingMSInlineAsm = V; 250 // When parsing MS inline asm, we must lex 0b1101 and 0ABCH as binary and 251 // hex integer literals. 252 Lexer.setLexMasmIntegers(V); 253 } 254 bool isParsingMSInlineAsm() override { return ParsingMSInlineAsm; } 255 256 bool discardLTOSymbol(StringRef Name) const override { 257 return LTODiscardSymbols.contains(Name); 258 } 259 260 bool parseMSInlineAsm(std::string &AsmString, unsigned &NumOutputs, 261 unsigned &NumInputs, 262 SmallVectorImpl<std::pair<void *, bool>> &OpDecls, 263 SmallVectorImpl<std::string> &Constraints, 264 SmallVectorImpl<std::string> &Clobbers, 265 const MCInstrInfo *MII, const MCInstPrinter *IP, 266 MCAsmParserSemaCallback &SI) override; 267 268 bool parseExpression(const MCExpr *&Res); 269 bool parseExpression(const MCExpr *&Res, SMLoc &EndLoc) override; 270 bool parsePrimaryExpr(const MCExpr *&Res, SMLoc &EndLoc, 271 AsmTypeInfo *TypeInfo) override; 272 bool parseParenExpression(const MCExpr *&Res, SMLoc &EndLoc) override; 273 bool parseParenExprOfDepth(unsigned ParenDepth, const MCExpr *&Res, 274 SMLoc &EndLoc) override; 275 bool parseAbsoluteExpression(int64_t &Res) override; 276 277 /// Parse a floating point expression using the float \p Semantics 278 /// and set \p Res to the value. 279 bool parseRealValue(const fltSemantics &Semantics, APInt &Res); 280 281 /// Parse an identifier or string (as a quoted identifier) 282 /// and set \p Res to the identifier contents. 283 bool parseIdentifier(StringRef &Res) override; 284 void eatToEndOfStatement() override; 285 286 bool checkForValidSection() override; 287 288 /// } 289 290 private: 291 bool parseCurlyBlockScope(SmallVectorImpl<AsmRewrite>& AsmStrRewrites); 292 bool parseCppHashLineFilenameComment(SMLoc L, bool SaveLocInfo = true); 293 294 void checkForBadMacro(SMLoc DirectiveLoc, StringRef Name, StringRef Body, 295 ArrayRef<MCAsmMacroParameter> Parameters); 296 bool expandMacro(raw_svector_ostream &OS, StringRef Body, 297 ArrayRef<MCAsmMacroParameter> Parameters, 298 ArrayRef<MCAsmMacroArgument> A, bool EnableAtPseudoVariable, 299 SMLoc L); 300 301 /// Are macros enabled in the parser? 302 bool areMacrosEnabled() {return MacrosEnabledFlag;} 303 304 /// Control a flag in the parser that enables or disables macros. 305 void setMacrosEnabled(bool Flag) {MacrosEnabledFlag = Flag;} 306 307 /// Are we inside a macro instantiation? 308 bool isInsideMacroInstantiation() {return !ActiveMacros.empty();} 309 310 /// Handle entry to macro instantiation. 311 /// 312 /// \param M The macro. 313 /// \param NameLoc Instantiation location. 314 bool handleMacroEntry(const MCAsmMacro *M, SMLoc NameLoc); 315 316 /// Handle exit from macro instantiation. 317 void handleMacroExit(); 318 319 /// Extract AsmTokens for a macro argument. 320 bool parseMacroArgument(MCAsmMacroArgument &MA, bool Vararg); 321 322 /// Parse all macro arguments for a given macro. 323 bool parseMacroArguments(const MCAsmMacro *M, MCAsmMacroArguments &A); 324 325 void printMacroInstantiations(); 326 void printMessage(SMLoc Loc, SourceMgr::DiagKind Kind, const Twine &Msg, 327 SMRange Range = std::nullopt) const { 328 ArrayRef<SMRange> Ranges(Range); 329 SrcMgr.PrintMessage(Loc, Kind, Msg, Ranges); 330 } 331 static void DiagHandler(const SMDiagnostic &Diag, void *Context); 332 333 /// Enter the specified file. This returns true on failure. 334 bool enterIncludeFile(const std::string &Filename); 335 336 /// Process the specified file for the .incbin directive. 337 /// This returns true on failure. 338 bool processIncbinFile(const std::string &Filename, int64_t Skip = 0, 339 const MCExpr *Count = nullptr, SMLoc Loc = SMLoc()); 340 341 /// Reset the current lexer position to that given by \p Loc. The 342 /// current token is not set; clients should ensure Lex() is called 343 /// subsequently. 344 /// 345 /// \param InBuffer If not 0, should be the known buffer id that contains the 346 /// location. 347 void jumpToLoc(SMLoc Loc, unsigned InBuffer = 0); 348 349 /// Parse up to the end of statement and a return the contents from the 350 /// current token until the end of the statement; the current token on exit 351 /// will be either the EndOfStatement or EOF. 352 StringRef parseStringToEndOfStatement() override; 353 354 /// Parse until the end of a statement or a comma is encountered, 355 /// return the contents from the current token up to the end or comma. 356 StringRef parseStringToComma(); 357 358 enum class AssignmentKind { 359 Set, 360 Equiv, 361 Equal, 362 LTOSetConditional, 363 }; 364 365 bool parseAssignment(StringRef Name, AssignmentKind Kind); 366 367 unsigned getBinOpPrecedence(AsmToken::TokenKind K, 368 MCBinaryExpr::Opcode &Kind); 369 370 bool parseBinOpRHS(unsigned Precedence, const MCExpr *&Res, SMLoc &EndLoc); 371 bool parseParenExpr(const MCExpr *&Res, SMLoc &EndLoc); 372 bool parseBracketExpr(const MCExpr *&Res, SMLoc &EndLoc); 373 374 bool parseRegisterOrRegisterNumber(int64_t &Register, SMLoc DirectiveLoc); 375 376 bool parseCVFunctionId(int64_t &FunctionId, StringRef DirectiveName); 377 bool parseCVFileId(int64_t &FileId, StringRef DirectiveName); 378 379 // Generic (target and platform independent) directive parsing. 380 enum DirectiveKind { 381 DK_NO_DIRECTIVE, // Placeholder 382 DK_SET, 383 DK_EQU, 384 DK_EQUIV, 385 DK_ASCII, 386 DK_ASCIZ, 387 DK_STRING, 388 DK_BYTE, 389 DK_SHORT, 390 DK_RELOC, 391 DK_VALUE, 392 DK_2BYTE, 393 DK_LONG, 394 DK_INT, 395 DK_4BYTE, 396 DK_QUAD, 397 DK_8BYTE, 398 DK_OCTA, 399 DK_DC, 400 DK_DC_A, 401 DK_DC_B, 402 DK_DC_D, 403 DK_DC_L, 404 DK_DC_S, 405 DK_DC_W, 406 DK_DC_X, 407 DK_DCB, 408 DK_DCB_B, 409 DK_DCB_D, 410 DK_DCB_L, 411 DK_DCB_S, 412 DK_DCB_W, 413 DK_DCB_X, 414 DK_DS, 415 DK_DS_B, 416 DK_DS_D, 417 DK_DS_L, 418 DK_DS_P, 419 DK_DS_S, 420 DK_DS_W, 421 DK_DS_X, 422 DK_SINGLE, 423 DK_FLOAT, 424 DK_DOUBLE, 425 DK_ALIGN, 426 DK_ALIGN32, 427 DK_BALIGN, 428 DK_BALIGNW, 429 DK_BALIGNL, 430 DK_P2ALIGN, 431 DK_P2ALIGNW, 432 DK_P2ALIGNL, 433 DK_ORG, 434 DK_FILL, 435 DK_ENDR, 436 DK_BUNDLE_ALIGN_MODE, 437 DK_BUNDLE_LOCK, 438 DK_BUNDLE_UNLOCK, 439 DK_ZERO, 440 DK_EXTERN, 441 DK_GLOBL, 442 DK_GLOBAL, 443 DK_LAZY_REFERENCE, 444 DK_NO_DEAD_STRIP, 445 DK_SYMBOL_RESOLVER, 446 DK_PRIVATE_EXTERN, 447 DK_REFERENCE, 448 DK_WEAK_DEFINITION, 449 DK_WEAK_REFERENCE, 450 DK_WEAK_DEF_CAN_BE_HIDDEN, 451 DK_COLD, 452 DK_COMM, 453 DK_COMMON, 454 DK_LCOMM, 455 DK_ABORT, 456 DK_INCLUDE, 457 DK_INCBIN, 458 DK_CODE16, 459 DK_CODE16GCC, 460 DK_REPT, 461 DK_IRP, 462 DK_IRPC, 463 DK_IF, 464 DK_IFEQ, 465 DK_IFGE, 466 DK_IFGT, 467 DK_IFLE, 468 DK_IFLT, 469 DK_IFNE, 470 DK_IFB, 471 DK_IFNB, 472 DK_IFC, 473 DK_IFEQS, 474 DK_IFNC, 475 DK_IFNES, 476 DK_IFDEF, 477 DK_IFNDEF, 478 DK_IFNOTDEF, 479 DK_ELSEIF, 480 DK_ELSE, 481 DK_ENDIF, 482 DK_SPACE, 483 DK_SKIP, 484 DK_FILE, 485 DK_LINE, 486 DK_LOC, 487 DK_STABS, 488 DK_CV_FILE, 489 DK_CV_FUNC_ID, 490 DK_CV_INLINE_SITE_ID, 491 DK_CV_LOC, 492 DK_CV_LINETABLE, 493 DK_CV_INLINE_LINETABLE, 494 DK_CV_DEF_RANGE, 495 DK_CV_STRINGTABLE, 496 DK_CV_STRING, 497 DK_CV_FILECHECKSUMS, 498 DK_CV_FILECHECKSUM_OFFSET, 499 DK_CV_FPO_DATA, 500 DK_CFI_SECTIONS, 501 DK_CFI_STARTPROC, 502 DK_CFI_ENDPROC, 503 DK_CFI_DEF_CFA, 504 DK_CFI_DEF_CFA_OFFSET, 505 DK_CFI_ADJUST_CFA_OFFSET, 506 DK_CFI_DEF_CFA_REGISTER, 507 DK_CFI_LLVM_DEF_ASPACE_CFA, 508 DK_CFI_OFFSET, 509 DK_CFI_REL_OFFSET, 510 DK_CFI_PERSONALITY, 511 DK_CFI_LSDA, 512 DK_CFI_REMEMBER_STATE, 513 DK_CFI_RESTORE_STATE, 514 DK_CFI_SAME_VALUE, 515 DK_CFI_RESTORE, 516 DK_CFI_ESCAPE, 517 DK_CFI_RETURN_COLUMN, 518 DK_CFI_SIGNAL_FRAME, 519 DK_CFI_UNDEFINED, 520 DK_CFI_REGISTER, 521 DK_CFI_WINDOW_SAVE, 522 DK_CFI_B_KEY_FRAME, 523 DK_MACROS_ON, 524 DK_MACROS_OFF, 525 DK_ALTMACRO, 526 DK_NOALTMACRO, 527 DK_MACRO, 528 DK_EXITM, 529 DK_ENDM, 530 DK_ENDMACRO, 531 DK_PURGEM, 532 DK_SLEB128, 533 DK_ULEB128, 534 DK_ERR, 535 DK_ERROR, 536 DK_WARNING, 537 DK_PRINT, 538 DK_ADDRSIG, 539 DK_ADDRSIG_SYM, 540 DK_PSEUDO_PROBE, 541 DK_LTO_DISCARD, 542 DK_LTO_SET_CONDITIONAL, 543 DK_CFI_MTE_TAGGED_FRAME, 544 DK_MEMTAG, 545 DK_END 546 }; 547 548 /// Maps directive name --> DirectiveKind enum, for 549 /// directives parsed by this class. 550 StringMap<DirectiveKind> DirectiveKindMap; 551 552 // Codeview def_range type parsing. 553 enum CVDefRangeType { 554 CVDR_DEFRANGE = 0, // Placeholder 555 CVDR_DEFRANGE_REGISTER, 556 CVDR_DEFRANGE_FRAMEPOINTER_REL, 557 CVDR_DEFRANGE_SUBFIELD_REGISTER, 558 CVDR_DEFRANGE_REGISTER_REL 559 }; 560 561 /// Maps Codeview def_range types --> CVDefRangeType enum, for 562 /// Codeview def_range types parsed by this class. 563 StringMap<CVDefRangeType> CVDefRangeTypeMap; 564 565 // ".ascii", ".asciz", ".string" 566 bool parseDirectiveAscii(StringRef IDVal, bool ZeroTerminated); 567 bool parseDirectiveReloc(SMLoc DirectiveLoc); // ".reloc" 568 bool parseDirectiveValue(StringRef IDVal, 569 unsigned Size); // ".byte", ".long", ... 570 bool parseDirectiveOctaValue(StringRef IDVal); // ".octa", ... 571 bool parseDirectiveRealValue(StringRef IDVal, 572 const fltSemantics &); // ".single", ... 573 bool parseDirectiveFill(); // ".fill" 574 bool parseDirectiveZero(); // ".zero" 575 // ".set", ".equ", ".equiv", ".lto_set_conditional" 576 bool parseDirectiveSet(StringRef IDVal, AssignmentKind Kind); 577 bool parseDirectiveOrg(); // ".org" 578 // ".align{,32}", ".p2align{,w,l}" 579 bool parseDirectiveAlign(bool IsPow2, unsigned ValueSize); 580 581 // ".file", ".line", ".loc", ".stabs" 582 bool parseDirectiveFile(SMLoc DirectiveLoc); 583 bool parseDirectiveLine(); 584 bool parseDirectiveLoc(); 585 bool parseDirectiveStabs(); 586 587 // ".cv_file", ".cv_func_id", ".cv_inline_site_id", ".cv_loc", ".cv_linetable", 588 // ".cv_inline_linetable", ".cv_def_range", ".cv_string" 589 bool parseDirectiveCVFile(); 590 bool parseDirectiveCVFuncId(); 591 bool parseDirectiveCVInlineSiteId(); 592 bool parseDirectiveCVLoc(); 593 bool parseDirectiveCVLinetable(); 594 bool parseDirectiveCVInlineLinetable(); 595 bool parseDirectiveCVDefRange(); 596 bool parseDirectiveCVString(); 597 bool parseDirectiveCVStringTable(); 598 bool parseDirectiveCVFileChecksums(); 599 bool parseDirectiveCVFileChecksumOffset(); 600 bool parseDirectiveCVFPOData(); 601 602 // .cfi directives 603 bool parseDirectiveCFIRegister(SMLoc DirectiveLoc); 604 bool parseDirectiveCFIWindowSave(); 605 bool parseDirectiveCFISections(); 606 bool parseDirectiveCFIStartProc(); 607 bool parseDirectiveCFIEndProc(); 608 bool parseDirectiveCFIDefCfaOffset(); 609 bool parseDirectiveCFIDefCfa(SMLoc DirectiveLoc); 610 bool parseDirectiveCFIAdjustCfaOffset(); 611 bool parseDirectiveCFIDefCfaRegister(SMLoc DirectiveLoc); 612 bool parseDirectiveCFILLVMDefAspaceCfa(SMLoc DirectiveLoc); 613 bool parseDirectiveCFIOffset(SMLoc DirectiveLoc); 614 bool parseDirectiveCFIRelOffset(SMLoc DirectiveLoc); 615 bool parseDirectiveCFIPersonalityOrLsda(bool IsPersonality); 616 bool parseDirectiveCFIRememberState(); 617 bool parseDirectiveCFIRestoreState(); 618 bool parseDirectiveCFISameValue(SMLoc DirectiveLoc); 619 bool parseDirectiveCFIRestore(SMLoc DirectiveLoc); 620 bool parseDirectiveCFIEscape(); 621 bool parseDirectiveCFIReturnColumn(SMLoc DirectiveLoc); 622 bool parseDirectiveCFISignalFrame(); 623 bool parseDirectiveCFIUndefined(SMLoc DirectiveLoc); 624 625 // macro directives 626 bool parseDirectivePurgeMacro(SMLoc DirectiveLoc); 627 bool parseDirectiveExitMacro(StringRef Directive); 628 bool parseDirectiveEndMacro(StringRef Directive); 629 bool parseDirectiveMacro(SMLoc DirectiveLoc); 630 bool parseDirectiveMacrosOnOff(StringRef Directive); 631 // alternate macro mode directives 632 bool parseDirectiveAltmacro(StringRef Directive); 633 // ".bundle_align_mode" 634 bool parseDirectiveBundleAlignMode(); 635 // ".bundle_lock" 636 bool parseDirectiveBundleLock(); 637 // ".bundle_unlock" 638 bool parseDirectiveBundleUnlock(); 639 640 // ".space", ".skip" 641 bool parseDirectiveSpace(StringRef IDVal); 642 643 // ".dcb" 644 bool parseDirectiveDCB(StringRef IDVal, unsigned Size); 645 bool parseDirectiveRealDCB(StringRef IDVal, const fltSemantics &); 646 // ".ds" 647 bool parseDirectiveDS(StringRef IDVal, unsigned Size); 648 649 // .sleb128 (Signed=true) and .uleb128 (Signed=false) 650 bool parseDirectiveLEB128(bool Signed); 651 652 /// Parse a directive like ".globl" which 653 /// accepts a single symbol (which should be a label or an external). 654 bool parseDirectiveSymbolAttribute(MCSymbolAttr Attr); 655 656 bool parseDirectiveComm(bool IsLocal); // ".comm" and ".lcomm" 657 658 bool parseDirectiveAbort(); // ".abort" 659 bool parseDirectiveInclude(); // ".include" 660 bool parseDirectiveIncbin(); // ".incbin" 661 662 // ".if", ".ifeq", ".ifge", ".ifgt" , ".ifle", ".iflt" or ".ifne" 663 bool parseDirectiveIf(SMLoc DirectiveLoc, DirectiveKind DirKind); 664 // ".ifb" or ".ifnb", depending on ExpectBlank. 665 bool parseDirectiveIfb(SMLoc DirectiveLoc, bool ExpectBlank); 666 // ".ifc" or ".ifnc", depending on ExpectEqual. 667 bool parseDirectiveIfc(SMLoc DirectiveLoc, bool ExpectEqual); 668 // ".ifeqs" or ".ifnes", depending on ExpectEqual. 669 bool parseDirectiveIfeqs(SMLoc DirectiveLoc, bool ExpectEqual); 670 // ".ifdef" or ".ifndef", depending on expect_defined 671 bool parseDirectiveIfdef(SMLoc DirectiveLoc, bool expect_defined); 672 bool parseDirectiveElseIf(SMLoc DirectiveLoc); // ".elseif" 673 bool parseDirectiveElse(SMLoc DirectiveLoc); // ".else" 674 bool parseDirectiveEndIf(SMLoc DirectiveLoc); // .endif 675 bool parseEscapedString(std::string &Data) override; 676 bool parseAngleBracketString(std::string &Data) override; 677 678 const MCExpr *applyModifierToExpr(const MCExpr *E, 679 MCSymbolRefExpr::VariantKind Variant); 680 681 // Macro-like directives 682 MCAsmMacro *parseMacroLikeBody(SMLoc DirectiveLoc); 683 void instantiateMacroLikeBody(MCAsmMacro *M, SMLoc DirectiveLoc, 684 raw_svector_ostream &OS); 685 bool parseDirectiveRept(SMLoc DirectiveLoc, StringRef Directive); 686 bool parseDirectiveIrp(SMLoc DirectiveLoc); // ".irp" 687 bool parseDirectiveIrpc(SMLoc DirectiveLoc); // ".irpc" 688 bool parseDirectiveEndr(SMLoc DirectiveLoc); // ".endr" 689 690 // "_emit" or "__emit" 691 bool parseDirectiveMSEmit(SMLoc DirectiveLoc, ParseStatementInfo &Info, 692 size_t Len); 693 694 // "align" 695 bool parseDirectiveMSAlign(SMLoc DirectiveLoc, ParseStatementInfo &Info); 696 697 // "end" 698 bool parseDirectiveEnd(SMLoc DirectiveLoc); 699 700 // ".err" or ".error" 701 bool parseDirectiveError(SMLoc DirectiveLoc, bool WithMessage); 702 703 // ".warning" 704 bool parseDirectiveWarning(SMLoc DirectiveLoc); 705 706 // .print <double-quotes-string> 707 bool parseDirectivePrint(SMLoc DirectiveLoc); 708 709 // .pseudoprobe 710 bool parseDirectivePseudoProbe(); 711 712 // ".lto_discard" 713 bool parseDirectiveLTODiscard(); 714 715 // Directives to support address-significance tables. 716 bool parseDirectiveAddrsig(); 717 bool parseDirectiveAddrsigSym(); 718 719 void initializeDirectiveKindMap(); 720 void initializeCVDefRangeTypeMap(); 721 }; 722 723 class HLASMAsmParser final : public AsmParser { 724 private: 725 MCAsmLexer &Lexer; 726 MCStreamer &Out; 727 728 void lexLeadingSpaces() { 729 while (Lexer.is(AsmToken::Space)) 730 Lexer.Lex(); 731 } 732 733 bool parseAsHLASMLabel(ParseStatementInfo &Info, MCAsmParserSemaCallback *SI); 734 bool parseAsMachineInstruction(ParseStatementInfo &Info, 735 MCAsmParserSemaCallback *SI); 736 737 public: 738 HLASMAsmParser(SourceMgr &SM, MCContext &Ctx, MCStreamer &Out, 739 const MCAsmInfo &MAI, unsigned CB = 0) 740 : AsmParser(SM, Ctx, Out, MAI, CB), Lexer(getLexer()), Out(Out) { 741 Lexer.setSkipSpace(false); 742 Lexer.setAllowHashInIdentifier(true); 743 Lexer.setLexHLASMIntegers(true); 744 Lexer.setLexHLASMStrings(true); 745 } 746 747 ~HLASMAsmParser() { Lexer.setSkipSpace(true); } 748 749 bool parseStatement(ParseStatementInfo &Info, 750 MCAsmParserSemaCallback *SI) override; 751 }; 752 753 } // end anonymous namespace 754 755 namespace llvm { 756 757 extern cl::opt<unsigned> AsmMacroMaxNestingDepth; 758 759 extern MCAsmParserExtension *createDarwinAsmParser(); 760 extern MCAsmParserExtension *createELFAsmParser(); 761 extern MCAsmParserExtension *createCOFFAsmParser(); 762 extern MCAsmParserExtension *createGOFFAsmParser(); 763 extern MCAsmParserExtension *createXCOFFAsmParser(); 764 extern MCAsmParserExtension *createWasmAsmParser(); 765 766 } // end namespace llvm 767 768 enum { DEFAULT_ADDRSPACE = 0 }; 769 770 AsmParser::AsmParser(SourceMgr &SM, MCContext &Ctx, MCStreamer &Out, 771 const MCAsmInfo &MAI, unsigned CB = 0) 772 : Lexer(MAI), Ctx(Ctx), Out(Out), MAI(MAI), SrcMgr(SM), 773 CurBuffer(CB ? CB : SM.getMainFileID()), MacrosEnabledFlag(true) { 774 HadError = false; 775 // Save the old handler. 776 SavedDiagHandler = SrcMgr.getDiagHandler(); 777 SavedDiagContext = SrcMgr.getDiagContext(); 778 // Set our own handler which calls the saved handler. 779 SrcMgr.setDiagHandler(DiagHandler, this); 780 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer()); 781 // Make MCStreamer aware of the StartTokLoc for locations in diagnostics. 782 Out.setStartTokLocPtr(&StartTokLoc); 783 784 // Initialize the platform / file format parser. 785 switch (Ctx.getObjectFileType()) { 786 case MCContext::IsCOFF: 787 PlatformParser.reset(createCOFFAsmParser()); 788 break; 789 case MCContext::IsMachO: 790 PlatformParser.reset(createDarwinAsmParser()); 791 IsDarwin = true; 792 break; 793 case MCContext::IsELF: 794 PlatformParser.reset(createELFAsmParser()); 795 break; 796 case MCContext::IsGOFF: 797 PlatformParser.reset(createGOFFAsmParser()); 798 break; 799 case MCContext::IsSPIRV: 800 report_fatal_error( 801 "Need to implement createSPIRVAsmParser for SPIRV format."); 802 break; 803 case MCContext::IsWasm: 804 PlatformParser.reset(createWasmAsmParser()); 805 break; 806 case MCContext::IsXCOFF: 807 PlatformParser.reset(createXCOFFAsmParser()); 808 break; 809 case MCContext::IsDXContainer: 810 llvm_unreachable("DXContainer is not supported yet"); 811 break; 812 } 813 814 PlatformParser->Initialize(*this); 815 initializeDirectiveKindMap(); 816 initializeCVDefRangeTypeMap(); 817 818 NumOfMacroInstantiations = 0; 819 } 820 821 AsmParser::~AsmParser() { 822 assert((HadError || ActiveMacros.empty()) && 823 "Unexpected active macro instantiation!"); 824 825 // Remove MCStreamer's reference to the parser SMLoc. 826 Out.setStartTokLocPtr(nullptr); 827 // Restore the saved diagnostics handler and context for use during 828 // finalization. 829 SrcMgr.setDiagHandler(SavedDiagHandler, SavedDiagContext); 830 } 831 832 void AsmParser::printMacroInstantiations() { 833 // Print the active macro instantiation stack. 834 for (std::vector<MacroInstantiation *>::const_reverse_iterator 835 it = ActiveMacros.rbegin(), 836 ie = ActiveMacros.rend(); 837 it != ie; ++it) 838 printMessage((*it)->InstantiationLoc, SourceMgr::DK_Note, 839 "while in macro instantiation"); 840 } 841 842 void AsmParser::Note(SMLoc L, const Twine &Msg, SMRange Range) { 843 printPendingErrors(); 844 printMessage(L, SourceMgr::DK_Note, Msg, Range); 845 printMacroInstantiations(); 846 } 847 848 bool AsmParser::Warning(SMLoc L, const Twine &Msg, SMRange Range) { 849 if(getTargetParser().getTargetOptions().MCNoWarn) 850 return false; 851 if (getTargetParser().getTargetOptions().MCFatalWarnings) 852 return Error(L, Msg, Range); 853 printMessage(L, SourceMgr::DK_Warning, Msg, Range); 854 printMacroInstantiations(); 855 return false; 856 } 857 858 bool AsmParser::printError(SMLoc L, const Twine &Msg, SMRange Range) { 859 HadError = true; 860 printMessage(L, SourceMgr::DK_Error, Msg, Range); 861 printMacroInstantiations(); 862 return true; 863 } 864 865 bool AsmParser::enterIncludeFile(const std::string &Filename) { 866 std::string IncludedFile; 867 unsigned NewBuf = 868 SrcMgr.AddIncludeFile(Filename, Lexer.getLoc(), IncludedFile); 869 if (!NewBuf) 870 return true; 871 872 CurBuffer = NewBuf; 873 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer()); 874 return false; 875 } 876 877 /// Process the specified .incbin file by searching for it in the include paths 878 /// then just emitting the byte contents of the file to the streamer. This 879 /// returns true on failure. 880 bool AsmParser::processIncbinFile(const std::string &Filename, int64_t Skip, 881 const MCExpr *Count, SMLoc Loc) { 882 std::string IncludedFile; 883 unsigned NewBuf = 884 SrcMgr.AddIncludeFile(Filename, Lexer.getLoc(), IncludedFile); 885 if (!NewBuf) 886 return true; 887 888 // Pick up the bytes from the file and emit them. 889 StringRef Bytes = SrcMgr.getMemoryBuffer(NewBuf)->getBuffer(); 890 Bytes = Bytes.drop_front(Skip); 891 if (Count) { 892 int64_t Res; 893 if (!Count->evaluateAsAbsolute(Res, getStreamer().getAssemblerPtr())) 894 return Error(Loc, "expected absolute expression"); 895 if (Res < 0) 896 return Warning(Loc, "negative count has no effect"); 897 Bytes = Bytes.take_front(Res); 898 } 899 getStreamer().emitBytes(Bytes); 900 return false; 901 } 902 903 void AsmParser::jumpToLoc(SMLoc Loc, unsigned InBuffer) { 904 CurBuffer = InBuffer ? InBuffer : SrcMgr.FindBufferContainingLoc(Loc); 905 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer(), 906 Loc.getPointer()); 907 } 908 909 const AsmToken &AsmParser::Lex() { 910 if (Lexer.getTok().is(AsmToken::Error)) 911 Error(Lexer.getErrLoc(), Lexer.getErr()); 912 913 // if it's a end of statement with a comment in it 914 if (getTok().is(AsmToken::EndOfStatement)) { 915 // if this is a line comment output it. 916 if (!getTok().getString().empty() && getTok().getString().front() != '\n' && 917 getTok().getString().front() != '\r' && MAI.preserveAsmComments()) 918 Out.addExplicitComment(Twine(getTok().getString())); 919 } 920 921 const AsmToken *tok = &Lexer.Lex(); 922 923 // Parse comments here to be deferred until end of next statement. 924 while (tok->is(AsmToken::Comment)) { 925 if (MAI.preserveAsmComments()) 926 Out.addExplicitComment(Twine(tok->getString())); 927 tok = &Lexer.Lex(); 928 } 929 930 if (tok->is(AsmToken::Eof)) { 931 // If this is the end of an included file, pop the parent file off the 932 // include stack. 933 SMLoc ParentIncludeLoc = SrcMgr.getParentIncludeLoc(CurBuffer); 934 if (ParentIncludeLoc != SMLoc()) { 935 jumpToLoc(ParentIncludeLoc); 936 return Lex(); 937 } 938 } 939 940 return *tok; 941 } 942 943 bool AsmParser::enabledGenDwarfForAssembly() { 944 // Check whether the user specified -g. 945 if (!getContext().getGenDwarfForAssembly()) 946 return false; 947 // If we haven't encountered any .file directives (which would imply that 948 // the assembler source was produced with debug info already) then emit one 949 // describing the assembler source file itself. 950 if (getContext().getGenDwarfFileNumber() == 0) { 951 // Use the first #line directive for this, if any. It's preprocessed, so 952 // there is no checksum, and of course no source directive. 953 if (!FirstCppHashFilename.empty()) 954 getContext().setMCLineTableRootFile( 955 /*CUID=*/0, getContext().getCompilationDir(), FirstCppHashFilename, 956 /*Cksum=*/std::nullopt, /*Source=*/std::nullopt); 957 const MCDwarfFile &RootFile = 958 getContext().getMCDwarfLineTable(/*CUID=*/0).getRootFile(); 959 getContext().setGenDwarfFileNumber(getStreamer().emitDwarfFileDirective( 960 /*CUID=*/0, getContext().getCompilationDir(), RootFile.Name, 961 RootFile.Checksum, RootFile.Source)); 962 } 963 return true; 964 } 965 966 bool AsmParser::Run(bool NoInitialTextSection, bool NoFinalize) { 967 LTODiscardSymbols.clear(); 968 969 // Create the initial section, if requested. 970 if (!NoInitialTextSection) 971 Out.initSections(false, getTargetParser().getSTI()); 972 973 // Prime the lexer. 974 Lex(); 975 976 HadError = false; 977 AsmCond StartingCondState = TheCondState; 978 SmallVector<AsmRewrite, 4> AsmStrRewrites; 979 980 // If we are generating dwarf for assembly source files save the initial text 981 // section. (Don't use enabledGenDwarfForAssembly() here, as we aren't 982 // emitting any actual debug info yet and haven't had a chance to parse any 983 // embedded .file directives.) 984 if (getContext().getGenDwarfForAssembly()) { 985 MCSection *Sec = getStreamer().getCurrentSectionOnly(); 986 if (!Sec->getBeginSymbol()) { 987 MCSymbol *SectionStartSym = getContext().createTempSymbol(); 988 getStreamer().emitLabel(SectionStartSym); 989 Sec->setBeginSymbol(SectionStartSym); 990 } 991 bool InsertResult = getContext().addGenDwarfSection(Sec); 992 assert(InsertResult && ".text section should not have debug info yet"); 993 (void)InsertResult; 994 } 995 996 StringRef Filename = getContext().getMainFileName(); 997 if (!Filename.empty() && (Filename.compare(StringRef("-")) != 0)) 998 Out.emitFileDirective(Filename); 999 1000 getTargetParser().onBeginOfFile(); 1001 1002 // While we have input, parse each statement. 1003 while (Lexer.isNot(AsmToken::Eof)) { 1004 ParseStatementInfo Info(&AsmStrRewrites); 1005 bool Parsed = parseStatement(Info, nullptr); 1006 1007 // If we have a Lexer Error we are on an Error Token. Load in Lexer Error 1008 // for printing ErrMsg via Lex() only if no (presumably better) parser error 1009 // exists. 1010 if (Parsed && !hasPendingError() && Lexer.getTok().is(AsmToken::Error)) { 1011 Lex(); 1012 } 1013 1014 // parseStatement returned true so may need to emit an error. 1015 printPendingErrors(); 1016 1017 // Skipping to the next line if needed. 1018 if (Parsed && !getLexer().isAtStartOfStatement()) 1019 eatToEndOfStatement(); 1020 } 1021 1022 getTargetParser().onEndOfFile(); 1023 printPendingErrors(); 1024 1025 // All errors should have been emitted. 1026 assert(!hasPendingError() && "unexpected error from parseStatement"); 1027 1028 getTargetParser().flushPendingInstructions(getStreamer()); 1029 1030 if (TheCondState.TheCond != StartingCondState.TheCond || 1031 TheCondState.Ignore != StartingCondState.Ignore) 1032 printError(getTok().getLoc(), "unmatched .ifs or .elses"); 1033 // Check to see there are no empty DwarfFile slots. 1034 const auto &LineTables = getContext().getMCDwarfLineTables(); 1035 if (!LineTables.empty()) { 1036 unsigned Index = 0; 1037 for (const auto &File : LineTables.begin()->second.getMCDwarfFiles()) { 1038 if (File.Name.empty() && Index != 0) 1039 printError(getTok().getLoc(), "unassigned file number: " + 1040 Twine(Index) + 1041 " for .file directives"); 1042 ++Index; 1043 } 1044 } 1045 1046 // Check to see that all assembler local symbols were actually defined. 1047 // Targets that don't do subsections via symbols may not want this, though, 1048 // so conservatively exclude them. Only do this if we're finalizing, though, 1049 // as otherwise we won't necessarilly have seen everything yet. 1050 if (!NoFinalize) { 1051 if (MAI.hasSubsectionsViaSymbols()) { 1052 for (const auto &TableEntry : getContext().getSymbols()) { 1053 MCSymbol *Sym = TableEntry.getValue(); 1054 // Variable symbols may not be marked as defined, so check those 1055 // explicitly. If we know it's a variable, we have a definition for 1056 // the purposes of this check. 1057 if (Sym->isTemporary() && !Sym->isVariable() && !Sym->isDefined()) 1058 // FIXME: We would really like to refer back to where the symbol was 1059 // first referenced for a source location. We need to add something 1060 // to track that. Currently, we just point to the end of the file. 1061 printError(getTok().getLoc(), "assembler local symbol '" + 1062 Sym->getName() + "' not defined"); 1063 } 1064 } 1065 1066 // Temporary symbols like the ones for directional jumps don't go in the 1067 // symbol table. They also need to be diagnosed in all (final) cases. 1068 for (std::tuple<SMLoc, CppHashInfoTy, MCSymbol *> &LocSym : DirLabels) { 1069 if (std::get<2>(LocSym)->isUndefined()) { 1070 // Reset the state of any "# line file" directives we've seen to the 1071 // context as it was at the diagnostic site. 1072 CppHashInfo = std::get<1>(LocSym); 1073 printError(std::get<0>(LocSym), "directional label undefined"); 1074 } 1075 } 1076 } 1077 // Finalize the output stream if there are no errors and if the client wants 1078 // us to. 1079 if (!HadError && !NoFinalize) { 1080 if (auto *TS = Out.getTargetStreamer()) 1081 TS->emitConstantPools(); 1082 1083 Out.finish(Lexer.getLoc()); 1084 } 1085 1086 return HadError || getContext().hadError(); 1087 } 1088 1089 bool AsmParser::checkForValidSection() { 1090 if (!ParsingMSInlineAsm && !getStreamer().getCurrentSectionOnly()) { 1091 Out.initSections(false, getTargetParser().getSTI()); 1092 return Error(getTok().getLoc(), 1093 "expected section directive before assembly directive"); 1094 } 1095 return false; 1096 } 1097 1098 /// Throw away the rest of the line for testing purposes. 1099 void AsmParser::eatToEndOfStatement() { 1100 while (Lexer.isNot(AsmToken::EndOfStatement) && Lexer.isNot(AsmToken::Eof)) 1101 Lexer.Lex(); 1102 1103 // Eat EOL. 1104 if (Lexer.is(AsmToken::EndOfStatement)) 1105 Lexer.Lex(); 1106 } 1107 1108 StringRef AsmParser::parseStringToEndOfStatement() { 1109 const char *Start = getTok().getLoc().getPointer(); 1110 1111 while (Lexer.isNot(AsmToken::EndOfStatement) && Lexer.isNot(AsmToken::Eof)) 1112 Lexer.Lex(); 1113 1114 const char *End = getTok().getLoc().getPointer(); 1115 return StringRef(Start, End - Start); 1116 } 1117 1118 StringRef AsmParser::parseStringToComma() { 1119 const char *Start = getTok().getLoc().getPointer(); 1120 1121 while (Lexer.isNot(AsmToken::EndOfStatement) && 1122 Lexer.isNot(AsmToken::Comma) && Lexer.isNot(AsmToken::Eof)) 1123 Lexer.Lex(); 1124 1125 const char *End = getTok().getLoc().getPointer(); 1126 return StringRef(Start, End - Start); 1127 } 1128 1129 /// Parse a paren expression and return it. 1130 /// NOTE: This assumes the leading '(' has already been consumed. 1131 /// 1132 /// parenexpr ::= expr) 1133 /// 1134 bool AsmParser::parseParenExpr(const MCExpr *&Res, SMLoc &EndLoc) { 1135 if (parseExpression(Res)) 1136 return true; 1137 EndLoc = Lexer.getTok().getEndLoc(); 1138 return parseRParen(); 1139 } 1140 1141 /// Parse a bracket expression and return it. 1142 /// NOTE: This assumes the leading '[' has already been consumed. 1143 /// 1144 /// bracketexpr ::= expr] 1145 /// 1146 bool AsmParser::parseBracketExpr(const MCExpr *&Res, SMLoc &EndLoc) { 1147 if (parseExpression(Res)) 1148 return true; 1149 EndLoc = getTok().getEndLoc(); 1150 if (parseToken(AsmToken::RBrac, "expected ']' in brackets expression")) 1151 return true; 1152 return false; 1153 } 1154 1155 /// Parse a primary expression and return it. 1156 /// primaryexpr ::= (parenexpr 1157 /// primaryexpr ::= symbol 1158 /// primaryexpr ::= number 1159 /// primaryexpr ::= '.' 1160 /// primaryexpr ::= ~,+,- primaryexpr 1161 bool AsmParser::parsePrimaryExpr(const MCExpr *&Res, SMLoc &EndLoc, 1162 AsmTypeInfo *TypeInfo) { 1163 SMLoc FirstTokenLoc = getLexer().getLoc(); 1164 AsmToken::TokenKind FirstTokenKind = Lexer.getKind(); 1165 switch (FirstTokenKind) { 1166 default: 1167 return TokError("unknown token in expression"); 1168 // If we have an error assume that we've already handled it. 1169 case AsmToken::Error: 1170 return true; 1171 case AsmToken::Exclaim: 1172 Lex(); // Eat the operator. 1173 if (parsePrimaryExpr(Res, EndLoc, TypeInfo)) 1174 return true; 1175 Res = MCUnaryExpr::createLNot(Res, getContext(), FirstTokenLoc); 1176 return false; 1177 case AsmToken::Dollar: 1178 case AsmToken::Star: 1179 case AsmToken::At: 1180 case AsmToken::String: 1181 case AsmToken::Identifier: { 1182 StringRef Identifier; 1183 if (parseIdentifier(Identifier)) { 1184 // We may have failed but '$'|'*' may be a valid token in context of 1185 // the current PC. 1186 if (getTok().is(AsmToken::Dollar) || getTok().is(AsmToken::Star)) { 1187 bool ShouldGenerateTempSymbol = false; 1188 if ((getTok().is(AsmToken::Dollar) && MAI.getDollarIsPC()) || 1189 (getTok().is(AsmToken::Star) && MAI.getStarIsPC())) 1190 ShouldGenerateTempSymbol = true; 1191 1192 if (!ShouldGenerateTempSymbol) 1193 return Error(FirstTokenLoc, "invalid token in expression"); 1194 1195 // Eat the '$'|'*' token. 1196 Lex(); 1197 // This is either a '$'|'*' reference, which references the current PC. 1198 // Emit a temporary label to the streamer and refer to it. 1199 MCSymbol *Sym = Ctx.createTempSymbol(); 1200 Out.emitLabel(Sym); 1201 Res = MCSymbolRefExpr::create(Sym, MCSymbolRefExpr::VK_None, 1202 getContext()); 1203 EndLoc = FirstTokenLoc; 1204 return false; 1205 } 1206 } 1207 // Parse symbol variant 1208 std::pair<StringRef, StringRef> Split; 1209 if (!MAI.useParensForSymbolVariant()) { 1210 if (FirstTokenKind == AsmToken::String) { 1211 if (Lexer.is(AsmToken::At)) { 1212 Lex(); // eat @ 1213 SMLoc AtLoc = getLexer().getLoc(); 1214 StringRef VName; 1215 if (parseIdentifier(VName)) 1216 return Error(AtLoc, "expected symbol variant after '@'"); 1217 1218 Split = std::make_pair(Identifier, VName); 1219 } 1220 } else { 1221 Split = Identifier.split('@'); 1222 } 1223 } else if (Lexer.is(AsmToken::LParen)) { 1224 Lex(); // eat '('. 1225 StringRef VName; 1226 parseIdentifier(VName); 1227 if (parseRParen()) 1228 return true; 1229 Split = std::make_pair(Identifier, VName); 1230 } 1231 1232 EndLoc = SMLoc::getFromPointer(Identifier.end()); 1233 1234 // This is a symbol reference. 1235 StringRef SymbolName = Identifier; 1236 if (SymbolName.empty()) 1237 return Error(getLexer().getLoc(), "expected a symbol reference"); 1238 1239 MCSymbolRefExpr::VariantKind Variant = MCSymbolRefExpr::VK_None; 1240 1241 // Lookup the symbol variant if used. 1242 if (!Split.second.empty()) { 1243 Variant = MCSymbolRefExpr::getVariantKindForName(Split.second); 1244 if (Variant != MCSymbolRefExpr::VK_Invalid) { 1245 SymbolName = Split.first; 1246 } else if (MAI.doesAllowAtInName() && !MAI.useParensForSymbolVariant()) { 1247 Variant = MCSymbolRefExpr::VK_None; 1248 } else { 1249 return Error(SMLoc::getFromPointer(Split.second.begin()), 1250 "invalid variant '" + Split.second + "'"); 1251 } 1252 } 1253 1254 MCSymbol *Sym = getContext().getInlineAsmLabel(SymbolName); 1255 if (!Sym) 1256 Sym = getContext().getOrCreateSymbol( 1257 MAI.shouldEmitLabelsInUpperCase() ? SymbolName.upper() : SymbolName); 1258 1259 // If this is an absolute variable reference, substitute it now to preserve 1260 // semantics in the face of reassignment. 1261 if (Sym->isVariable()) { 1262 auto V = Sym->getVariableValue(/*SetUsed*/ false); 1263 bool DoInline = isa<MCConstantExpr>(V) && !Variant; 1264 if (auto TV = dyn_cast<MCTargetExpr>(V)) 1265 DoInline = TV->inlineAssignedExpr(); 1266 if (DoInline) { 1267 if (Variant) 1268 return Error(EndLoc, "unexpected modifier on variable reference"); 1269 Res = Sym->getVariableValue(/*SetUsed*/ false); 1270 return false; 1271 } 1272 } 1273 1274 // Otherwise create a symbol ref. 1275 Res = MCSymbolRefExpr::create(Sym, Variant, getContext(), FirstTokenLoc); 1276 return false; 1277 } 1278 case AsmToken::BigNum: 1279 return TokError("literal value out of range for directive"); 1280 case AsmToken::Integer: { 1281 SMLoc Loc = getTok().getLoc(); 1282 int64_t IntVal = getTok().getIntVal(); 1283 Res = MCConstantExpr::create(IntVal, getContext()); 1284 EndLoc = Lexer.getTok().getEndLoc(); 1285 Lex(); // Eat token. 1286 // Look for 'b' or 'f' following an Integer as a directional label 1287 if (Lexer.getKind() == AsmToken::Identifier) { 1288 StringRef IDVal = getTok().getString(); 1289 // Lookup the symbol variant if used. 1290 std::pair<StringRef, StringRef> Split = IDVal.split('@'); 1291 MCSymbolRefExpr::VariantKind Variant = MCSymbolRefExpr::VK_None; 1292 if (Split.first.size() != IDVal.size()) { 1293 Variant = MCSymbolRefExpr::getVariantKindForName(Split.second); 1294 if (Variant == MCSymbolRefExpr::VK_Invalid) 1295 return TokError("invalid variant '" + Split.second + "'"); 1296 IDVal = Split.first; 1297 } 1298 if (IDVal == "f" || IDVal == "b") { 1299 MCSymbol *Sym = 1300 Ctx.getDirectionalLocalSymbol(IntVal, IDVal == "b"); 1301 Res = MCSymbolRefExpr::create(Sym, Variant, getContext()); 1302 if (IDVal == "b" && Sym->isUndefined()) 1303 return Error(Loc, "directional label undefined"); 1304 DirLabels.push_back(std::make_tuple(Loc, CppHashInfo, Sym)); 1305 EndLoc = Lexer.getTok().getEndLoc(); 1306 Lex(); // Eat identifier. 1307 } 1308 } 1309 return false; 1310 } 1311 case AsmToken::Real: { 1312 APFloat RealVal(APFloat::IEEEdouble(), getTok().getString()); 1313 uint64_t IntVal = RealVal.bitcastToAPInt().getZExtValue(); 1314 Res = MCConstantExpr::create(IntVal, getContext()); 1315 EndLoc = Lexer.getTok().getEndLoc(); 1316 Lex(); // Eat token. 1317 return false; 1318 } 1319 case AsmToken::Dot: { 1320 if (!MAI.getDotIsPC()) 1321 return TokError("cannot use . as current PC"); 1322 1323 // This is a '.' reference, which references the current PC. Emit a 1324 // temporary label to the streamer and refer to it. 1325 MCSymbol *Sym = Ctx.createTempSymbol(); 1326 Out.emitLabel(Sym); 1327 Res = MCSymbolRefExpr::create(Sym, MCSymbolRefExpr::VK_None, getContext()); 1328 EndLoc = Lexer.getTok().getEndLoc(); 1329 Lex(); // Eat identifier. 1330 return false; 1331 } 1332 case AsmToken::LParen: 1333 Lex(); // Eat the '('. 1334 return parseParenExpr(Res, EndLoc); 1335 case AsmToken::LBrac: 1336 if (!PlatformParser->HasBracketExpressions()) 1337 return TokError("brackets expression not supported on this target"); 1338 Lex(); // Eat the '['. 1339 return parseBracketExpr(Res, EndLoc); 1340 case AsmToken::Minus: 1341 Lex(); // Eat the operator. 1342 if (parsePrimaryExpr(Res, EndLoc, TypeInfo)) 1343 return true; 1344 Res = MCUnaryExpr::createMinus(Res, getContext(), FirstTokenLoc); 1345 return false; 1346 case AsmToken::Plus: 1347 Lex(); // Eat the operator. 1348 if (parsePrimaryExpr(Res, EndLoc, TypeInfo)) 1349 return true; 1350 Res = MCUnaryExpr::createPlus(Res, getContext(), FirstTokenLoc); 1351 return false; 1352 case AsmToken::Tilde: 1353 Lex(); // Eat the operator. 1354 if (parsePrimaryExpr(Res, EndLoc, TypeInfo)) 1355 return true; 1356 Res = MCUnaryExpr::createNot(Res, getContext(), FirstTokenLoc); 1357 return false; 1358 // MIPS unary expression operators. The lexer won't generate these tokens if 1359 // MCAsmInfo::HasMipsExpressions is false for the target. 1360 case AsmToken::PercentCall16: 1361 case AsmToken::PercentCall_Hi: 1362 case AsmToken::PercentCall_Lo: 1363 case AsmToken::PercentDtprel_Hi: 1364 case AsmToken::PercentDtprel_Lo: 1365 case AsmToken::PercentGot: 1366 case AsmToken::PercentGot_Disp: 1367 case AsmToken::PercentGot_Hi: 1368 case AsmToken::PercentGot_Lo: 1369 case AsmToken::PercentGot_Ofst: 1370 case AsmToken::PercentGot_Page: 1371 case AsmToken::PercentGottprel: 1372 case AsmToken::PercentGp_Rel: 1373 case AsmToken::PercentHi: 1374 case AsmToken::PercentHigher: 1375 case AsmToken::PercentHighest: 1376 case AsmToken::PercentLo: 1377 case AsmToken::PercentNeg: 1378 case AsmToken::PercentPcrel_Hi: 1379 case AsmToken::PercentPcrel_Lo: 1380 case AsmToken::PercentTlsgd: 1381 case AsmToken::PercentTlsldm: 1382 case AsmToken::PercentTprel_Hi: 1383 case AsmToken::PercentTprel_Lo: 1384 Lex(); // Eat the operator. 1385 if (Lexer.isNot(AsmToken::LParen)) 1386 return TokError("expected '(' after operator"); 1387 Lex(); // Eat the operator. 1388 if (parseExpression(Res, EndLoc)) 1389 return true; 1390 if (parseRParen()) 1391 return true; 1392 Res = getTargetParser().createTargetUnaryExpr(Res, FirstTokenKind, Ctx); 1393 return !Res; 1394 } 1395 } 1396 1397 bool AsmParser::parseExpression(const MCExpr *&Res) { 1398 SMLoc EndLoc; 1399 return parseExpression(Res, EndLoc); 1400 } 1401 1402 const MCExpr * 1403 AsmParser::applyModifierToExpr(const MCExpr *E, 1404 MCSymbolRefExpr::VariantKind Variant) { 1405 // Ask the target implementation about this expression first. 1406 const MCExpr *NewE = getTargetParser().applyModifierToExpr(E, Variant, Ctx); 1407 if (NewE) 1408 return NewE; 1409 // Recurse over the given expression, rebuilding it to apply the given variant 1410 // if there is exactly one symbol. 1411 switch (E->getKind()) { 1412 case MCExpr::Target: 1413 case MCExpr::Constant: 1414 return nullptr; 1415 1416 case MCExpr::SymbolRef: { 1417 const MCSymbolRefExpr *SRE = cast<MCSymbolRefExpr>(E); 1418 1419 if (SRE->getKind() != MCSymbolRefExpr::VK_None) { 1420 TokError("invalid variant on expression '" + getTok().getIdentifier() + 1421 "' (already modified)"); 1422 return E; 1423 } 1424 1425 return MCSymbolRefExpr::create(&SRE->getSymbol(), Variant, getContext()); 1426 } 1427 1428 case MCExpr::Unary: { 1429 const MCUnaryExpr *UE = cast<MCUnaryExpr>(E); 1430 const MCExpr *Sub = applyModifierToExpr(UE->getSubExpr(), Variant); 1431 if (!Sub) 1432 return nullptr; 1433 return MCUnaryExpr::create(UE->getOpcode(), Sub, getContext()); 1434 } 1435 1436 case MCExpr::Binary: { 1437 const MCBinaryExpr *BE = cast<MCBinaryExpr>(E); 1438 const MCExpr *LHS = applyModifierToExpr(BE->getLHS(), Variant); 1439 const MCExpr *RHS = applyModifierToExpr(BE->getRHS(), Variant); 1440 1441 if (!LHS && !RHS) 1442 return nullptr; 1443 1444 if (!LHS) 1445 LHS = BE->getLHS(); 1446 if (!RHS) 1447 RHS = BE->getRHS(); 1448 1449 return MCBinaryExpr::create(BE->getOpcode(), LHS, RHS, getContext()); 1450 } 1451 } 1452 1453 llvm_unreachable("Invalid expression kind!"); 1454 } 1455 1456 /// This function checks if the next token is <string> type or arithmetic. 1457 /// string that begin with character '<' must end with character '>'. 1458 /// otherwise it is arithmetics. 1459 /// If the function returns a 'true' value, 1460 /// the End argument will be filled with the last location pointed to the '>' 1461 /// character. 1462 1463 /// There is a gap between the AltMacro's documentation and the single quote 1464 /// implementation. GCC does not fully support this feature and so we will not 1465 /// support it. 1466 /// TODO: Adding single quote as a string. 1467 static bool isAngleBracketString(SMLoc &StrLoc, SMLoc &EndLoc) { 1468 assert((StrLoc.getPointer() != nullptr) && 1469 "Argument to the function cannot be a NULL value"); 1470 const char *CharPtr = StrLoc.getPointer(); 1471 while ((*CharPtr != '>') && (*CharPtr != '\n') && (*CharPtr != '\r') && 1472 (*CharPtr != '\0')) { 1473 if (*CharPtr == '!') 1474 CharPtr++; 1475 CharPtr++; 1476 } 1477 if (*CharPtr == '>') { 1478 EndLoc = StrLoc.getFromPointer(CharPtr + 1); 1479 return true; 1480 } 1481 return false; 1482 } 1483 1484 /// creating a string without the escape characters '!'. 1485 static std::string angleBracketString(StringRef AltMacroStr) { 1486 std::string Res; 1487 for (size_t Pos = 0; Pos < AltMacroStr.size(); Pos++) { 1488 if (AltMacroStr[Pos] == '!') 1489 Pos++; 1490 Res += AltMacroStr[Pos]; 1491 } 1492 return Res; 1493 } 1494 1495 /// Parse an expression and return it. 1496 /// 1497 /// expr ::= expr &&,|| expr -> lowest. 1498 /// expr ::= expr |,^,&,! expr 1499 /// expr ::= expr ==,!=,<>,<,<=,>,>= expr 1500 /// expr ::= expr <<,>> expr 1501 /// expr ::= expr +,- expr 1502 /// expr ::= expr *,/,% expr -> highest. 1503 /// expr ::= primaryexpr 1504 /// 1505 bool AsmParser::parseExpression(const MCExpr *&Res, SMLoc &EndLoc) { 1506 // Parse the expression. 1507 Res = nullptr; 1508 if (getTargetParser().parsePrimaryExpr(Res, EndLoc) || 1509 parseBinOpRHS(1, Res, EndLoc)) 1510 return true; 1511 1512 // As a special case, we support 'a op b @ modifier' by rewriting the 1513 // expression to include the modifier. This is inefficient, but in general we 1514 // expect users to use 'a@modifier op b'. 1515 if (Lexer.getKind() == AsmToken::At) { 1516 Lex(); 1517 1518 if (Lexer.isNot(AsmToken::Identifier)) 1519 return TokError("unexpected symbol modifier following '@'"); 1520 1521 MCSymbolRefExpr::VariantKind Variant = 1522 MCSymbolRefExpr::getVariantKindForName(getTok().getIdentifier()); 1523 if (Variant == MCSymbolRefExpr::VK_Invalid) 1524 return TokError("invalid variant '" + getTok().getIdentifier() + "'"); 1525 1526 const MCExpr *ModifiedRes = applyModifierToExpr(Res, Variant); 1527 if (!ModifiedRes) { 1528 return TokError("invalid modifier '" + getTok().getIdentifier() + 1529 "' (no symbols present)"); 1530 } 1531 1532 Res = ModifiedRes; 1533 Lex(); 1534 } 1535 1536 // Try to constant fold it up front, if possible. Do not exploit 1537 // assembler here. 1538 int64_t Value; 1539 if (Res->evaluateAsAbsolute(Value)) 1540 Res = MCConstantExpr::create(Value, getContext()); 1541 1542 return false; 1543 } 1544 1545 bool AsmParser::parseParenExpression(const MCExpr *&Res, SMLoc &EndLoc) { 1546 Res = nullptr; 1547 return parseParenExpr(Res, EndLoc) || parseBinOpRHS(1, Res, EndLoc); 1548 } 1549 1550 bool AsmParser::parseParenExprOfDepth(unsigned ParenDepth, const MCExpr *&Res, 1551 SMLoc &EndLoc) { 1552 if (parseParenExpr(Res, EndLoc)) 1553 return true; 1554 1555 for (; ParenDepth > 0; --ParenDepth) { 1556 if (parseBinOpRHS(1, Res, EndLoc)) 1557 return true; 1558 1559 // We don't Lex() the last RParen. 1560 // This is the same behavior as parseParenExpression(). 1561 if (ParenDepth - 1 > 0) { 1562 EndLoc = getTok().getEndLoc(); 1563 if (parseRParen()) 1564 return true; 1565 } 1566 } 1567 return false; 1568 } 1569 1570 bool AsmParser::parseAbsoluteExpression(int64_t &Res) { 1571 const MCExpr *Expr; 1572 1573 SMLoc StartLoc = Lexer.getLoc(); 1574 if (parseExpression(Expr)) 1575 return true; 1576 1577 if (!Expr->evaluateAsAbsolute(Res, getStreamer().getAssemblerPtr())) 1578 return Error(StartLoc, "expected absolute expression"); 1579 1580 return false; 1581 } 1582 1583 static unsigned getDarwinBinOpPrecedence(AsmToken::TokenKind K, 1584 MCBinaryExpr::Opcode &Kind, 1585 bool ShouldUseLogicalShr) { 1586 switch (K) { 1587 default: 1588 return 0; // not a binop. 1589 1590 // Lowest Precedence: &&, || 1591 case AsmToken::AmpAmp: 1592 Kind = MCBinaryExpr::LAnd; 1593 return 1; 1594 case AsmToken::PipePipe: 1595 Kind = MCBinaryExpr::LOr; 1596 return 1; 1597 1598 // Low Precedence: |, &, ^ 1599 case AsmToken::Pipe: 1600 Kind = MCBinaryExpr::Or; 1601 return 2; 1602 case AsmToken::Caret: 1603 Kind = MCBinaryExpr::Xor; 1604 return 2; 1605 case AsmToken::Amp: 1606 Kind = MCBinaryExpr::And; 1607 return 2; 1608 1609 // Low Intermediate Precedence: ==, !=, <>, <, <=, >, >= 1610 case AsmToken::EqualEqual: 1611 Kind = MCBinaryExpr::EQ; 1612 return 3; 1613 case AsmToken::ExclaimEqual: 1614 case AsmToken::LessGreater: 1615 Kind = MCBinaryExpr::NE; 1616 return 3; 1617 case AsmToken::Less: 1618 Kind = MCBinaryExpr::LT; 1619 return 3; 1620 case AsmToken::LessEqual: 1621 Kind = MCBinaryExpr::LTE; 1622 return 3; 1623 case AsmToken::Greater: 1624 Kind = MCBinaryExpr::GT; 1625 return 3; 1626 case AsmToken::GreaterEqual: 1627 Kind = MCBinaryExpr::GTE; 1628 return 3; 1629 1630 // Intermediate Precedence: <<, >> 1631 case AsmToken::LessLess: 1632 Kind = MCBinaryExpr::Shl; 1633 return 4; 1634 case AsmToken::GreaterGreater: 1635 Kind = ShouldUseLogicalShr ? MCBinaryExpr::LShr : MCBinaryExpr::AShr; 1636 return 4; 1637 1638 // High Intermediate Precedence: +, - 1639 case AsmToken::Plus: 1640 Kind = MCBinaryExpr::Add; 1641 return 5; 1642 case AsmToken::Minus: 1643 Kind = MCBinaryExpr::Sub; 1644 return 5; 1645 1646 // Highest Precedence: *, /, % 1647 case AsmToken::Star: 1648 Kind = MCBinaryExpr::Mul; 1649 return 6; 1650 case AsmToken::Slash: 1651 Kind = MCBinaryExpr::Div; 1652 return 6; 1653 case AsmToken::Percent: 1654 Kind = MCBinaryExpr::Mod; 1655 return 6; 1656 } 1657 } 1658 1659 static unsigned getGNUBinOpPrecedence(const MCAsmInfo &MAI, 1660 AsmToken::TokenKind K, 1661 MCBinaryExpr::Opcode &Kind, 1662 bool ShouldUseLogicalShr) { 1663 switch (K) { 1664 default: 1665 return 0; // not a binop. 1666 1667 // Lowest Precedence: &&, || 1668 case AsmToken::AmpAmp: 1669 Kind = MCBinaryExpr::LAnd; 1670 return 2; 1671 case AsmToken::PipePipe: 1672 Kind = MCBinaryExpr::LOr; 1673 return 1; 1674 1675 // Low Precedence: ==, !=, <>, <, <=, >, >= 1676 case AsmToken::EqualEqual: 1677 Kind = MCBinaryExpr::EQ; 1678 return 3; 1679 case AsmToken::ExclaimEqual: 1680 case AsmToken::LessGreater: 1681 Kind = MCBinaryExpr::NE; 1682 return 3; 1683 case AsmToken::Less: 1684 Kind = MCBinaryExpr::LT; 1685 return 3; 1686 case AsmToken::LessEqual: 1687 Kind = MCBinaryExpr::LTE; 1688 return 3; 1689 case AsmToken::Greater: 1690 Kind = MCBinaryExpr::GT; 1691 return 3; 1692 case AsmToken::GreaterEqual: 1693 Kind = MCBinaryExpr::GTE; 1694 return 3; 1695 1696 // Low Intermediate Precedence: +, - 1697 case AsmToken::Plus: 1698 Kind = MCBinaryExpr::Add; 1699 return 4; 1700 case AsmToken::Minus: 1701 Kind = MCBinaryExpr::Sub; 1702 return 4; 1703 1704 // High Intermediate Precedence: |, !, &, ^ 1705 // 1706 case AsmToken::Pipe: 1707 Kind = MCBinaryExpr::Or; 1708 return 5; 1709 case AsmToken::Exclaim: 1710 // Hack to support ARM compatible aliases (implied 'sp' operand in 'srs*' 1711 // instructions like 'srsda #31!') and not parse ! as an infix operator. 1712 if (MAI.getCommentString() == "@") 1713 return 0; 1714 Kind = MCBinaryExpr::OrNot; 1715 return 5; 1716 case AsmToken::Caret: 1717 Kind = MCBinaryExpr::Xor; 1718 return 5; 1719 case AsmToken::Amp: 1720 Kind = MCBinaryExpr::And; 1721 return 5; 1722 1723 // Highest Precedence: *, /, %, <<, >> 1724 case AsmToken::Star: 1725 Kind = MCBinaryExpr::Mul; 1726 return 6; 1727 case AsmToken::Slash: 1728 Kind = MCBinaryExpr::Div; 1729 return 6; 1730 case AsmToken::Percent: 1731 Kind = MCBinaryExpr::Mod; 1732 return 6; 1733 case AsmToken::LessLess: 1734 Kind = MCBinaryExpr::Shl; 1735 return 6; 1736 case AsmToken::GreaterGreater: 1737 Kind = ShouldUseLogicalShr ? MCBinaryExpr::LShr : MCBinaryExpr::AShr; 1738 return 6; 1739 } 1740 } 1741 1742 unsigned AsmParser::getBinOpPrecedence(AsmToken::TokenKind K, 1743 MCBinaryExpr::Opcode &Kind) { 1744 bool ShouldUseLogicalShr = MAI.shouldUseLogicalShr(); 1745 return IsDarwin ? getDarwinBinOpPrecedence(K, Kind, ShouldUseLogicalShr) 1746 : getGNUBinOpPrecedence(MAI, K, Kind, ShouldUseLogicalShr); 1747 } 1748 1749 /// Parse all binary operators with precedence >= 'Precedence'. 1750 /// Res contains the LHS of the expression on input. 1751 bool AsmParser::parseBinOpRHS(unsigned Precedence, const MCExpr *&Res, 1752 SMLoc &EndLoc) { 1753 SMLoc StartLoc = Lexer.getLoc(); 1754 while (true) { 1755 MCBinaryExpr::Opcode Kind = MCBinaryExpr::Add; 1756 unsigned TokPrec = getBinOpPrecedence(Lexer.getKind(), Kind); 1757 1758 // If the next token is lower precedence than we are allowed to eat, return 1759 // successfully with what we ate already. 1760 if (TokPrec < Precedence) 1761 return false; 1762 1763 Lex(); 1764 1765 // Eat the next primary expression. 1766 const MCExpr *RHS; 1767 if (getTargetParser().parsePrimaryExpr(RHS, EndLoc)) 1768 return true; 1769 1770 // If BinOp binds less tightly with RHS than the operator after RHS, let 1771 // the pending operator take RHS as its LHS. 1772 MCBinaryExpr::Opcode Dummy; 1773 unsigned NextTokPrec = getBinOpPrecedence(Lexer.getKind(), Dummy); 1774 if (TokPrec < NextTokPrec && parseBinOpRHS(TokPrec + 1, RHS, EndLoc)) 1775 return true; 1776 1777 // Merge LHS and RHS according to operator. 1778 Res = MCBinaryExpr::create(Kind, Res, RHS, getContext(), StartLoc); 1779 } 1780 } 1781 1782 /// ParseStatement: 1783 /// ::= EndOfStatement 1784 /// ::= Label* Directive ...Operands... EndOfStatement 1785 /// ::= Label* Identifier OperandList* EndOfStatement 1786 bool AsmParser::parseStatement(ParseStatementInfo &Info, 1787 MCAsmParserSemaCallback *SI) { 1788 assert(!hasPendingError() && "parseStatement started with pending error"); 1789 // Eat initial spaces and comments 1790 while (Lexer.is(AsmToken::Space)) 1791 Lex(); 1792 if (Lexer.is(AsmToken::EndOfStatement)) { 1793 // if this is a line comment we can drop it safely 1794 if (getTok().getString().empty() || getTok().getString().front() == '\r' || 1795 getTok().getString().front() == '\n') 1796 Out.addBlankLine(); 1797 Lex(); 1798 return false; 1799 } 1800 // Statements always start with an identifier. 1801 AsmToken ID = getTok(); 1802 SMLoc IDLoc = ID.getLoc(); 1803 StringRef IDVal; 1804 int64_t LocalLabelVal = -1; 1805 StartTokLoc = ID.getLoc(); 1806 if (Lexer.is(AsmToken::HashDirective)) 1807 return parseCppHashLineFilenameComment(IDLoc, 1808 !isInsideMacroInstantiation()); 1809 1810 // Allow an integer followed by a ':' as a directional local label. 1811 if (Lexer.is(AsmToken::Integer)) { 1812 LocalLabelVal = getTok().getIntVal(); 1813 if (LocalLabelVal < 0) { 1814 if (!TheCondState.Ignore) { 1815 Lex(); // always eat a token 1816 return Error(IDLoc, "unexpected token at start of statement"); 1817 } 1818 IDVal = ""; 1819 } else { 1820 IDVal = getTok().getString(); 1821 Lex(); // Consume the integer token to be used as an identifier token. 1822 if (Lexer.getKind() != AsmToken::Colon) { 1823 if (!TheCondState.Ignore) { 1824 Lex(); // always eat a token 1825 return Error(IDLoc, "unexpected token at start of statement"); 1826 } 1827 } 1828 } 1829 } else if (Lexer.is(AsmToken::Dot)) { 1830 // Treat '.' as a valid identifier in this context. 1831 Lex(); 1832 IDVal = "."; 1833 } else if (Lexer.is(AsmToken::LCurly)) { 1834 // Treat '{' as a valid identifier in this context. 1835 Lex(); 1836 IDVal = "{"; 1837 1838 } else if (Lexer.is(AsmToken::RCurly)) { 1839 // Treat '}' as a valid identifier in this context. 1840 Lex(); 1841 IDVal = "}"; 1842 } else if (Lexer.is(AsmToken::Star) && 1843 getTargetParser().starIsStartOfStatement()) { 1844 // Accept '*' as a valid start of statement. 1845 Lex(); 1846 IDVal = "*"; 1847 } else if (parseIdentifier(IDVal)) { 1848 if (!TheCondState.Ignore) { 1849 Lex(); // always eat a token 1850 return Error(IDLoc, "unexpected token at start of statement"); 1851 } 1852 IDVal = ""; 1853 } 1854 1855 // Handle conditional assembly here before checking for skipping. We 1856 // have to do this so that .endif isn't skipped in a ".if 0" block for 1857 // example. 1858 StringMap<DirectiveKind>::const_iterator DirKindIt = 1859 DirectiveKindMap.find(IDVal.lower()); 1860 DirectiveKind DirKind = (DirKindIt == DirectiveKindMap.end()) 1861 ? DK_NO_DIRECTIVE 1862 : DirKindIt->getValue(); 1863 switch (DirKind) { 1864 default: 1865 break; 1866 case DK_IF: 1867 case DK_IFEQ: 1868 case DK_IFGE: 1869 case DK_IFGT: 1870 case DK_IFLE: 1871 case DK_IFLT: 1872 case DK_IFNE: 1873 return parseDirectiveIf(IDLoc, DirKind); 1874 case DK_IFB: 1875 return parseDirectiveIfb(IDLoc, true); 1876 case DK_IFNB: 1877 return parseDirectiveIfb(IDLoc, false); 1878 case DK_IFC: 1879 return parseDirectiveIfc(IDLoc, true); 1880 case DK_IFEQS: 1881 return parseDirectiveIfeqs(IDLoc, true); 1882 case DK_IFNC: 1883 return parseDirectiveIfc(IDLoc, false); 1884 case DK_IFNES: 1885 return parseDirectiveIfeqs(IDLoc, false); 1886 case DK_IFDEF: 1887 return parseDirectiveIfdef(IDLoc, true); 1888 case DK_IFNDEF: 1889 case DK_IFNOTDEF: 1890 return parseDirectiveIfdef(IDLoc, false); 1891 case DK_ELSEIF: 1892 return parseDirectiveElseIf(IDLoc); 1893 case DK_ELSE: 1894 return parseDirectiveElse(IDLoc); 1895 case DK_ENDIF: 1896 return parseDirectiveEndIf(IDLoc); 1897 } 1898 1899 // Ignore the statement if in the middle of inactive conditional 1900 // (e.g. ".if 0"). 1901 if (TheCondState.Ignore) { 1902 eatToEndOfStatement(); 1903 return false; 1904 } 1905 1906 // FIXME: Recurse on local labels? 1907 1908 // Check for a label. 1909 // ::= identifier ':' 1910 // ::= number ':' 1911 if (Lexer.is(AsmToken::Colon) && getTargetParser().isLabel(ID)) { 1912 if (checkForValidSection()) 1913 return true; 1914 1915 Lex(); // Consume the ':'. 1916 1917 // Diagnose attempt to use '.' as a label. 1918 if (IDVal == ".") 1919 return Error(IDLoc, "invalid use of pseudo-symbol '.' as a label"); 1920 1921 // Diagnose attempt to use a variable as a label. 1922 // 1923 // FIXME: Diagnostics. Note the location of the definition as a label. 1924 // FIXME: This doesn't diagnose assignment to a symbol which has been 1925 // implicitly marked as external. 1926 MCSymbol *Sym; 1927 if (LocalLabelVal == -1) { 1928 if (ParsingMSInlineAsm && SI) { 1929 StringRef RewrittenLabel = 1930 SI->LookupInlineAsmLabel(IDVal, getSourceManager(), IDLoc, true); 1931 assert(!RewrittenLabel.empty() && 1932 "We should have an internal name here."); 1933 Info.AsmRewrites->emplace_back(AOK_Label, IDLoc, IDVal.size(), 1934 RewrittenLabel); 1935 IDVal = RewrittenLabel; 1936 } 1937 Sym = getContext().getOrCreateSymbol(IDVal); 1938 } else 1939 Sym = Ctx.createDirectionalLocalSymbol(LocalLabelVal); 1940 // End of Labels should be treated as end of line for lexing 1941 // purposes but that information is not available to the Lexer who 1942 // does not understand Labels. This may cause us to see a Hash 1943 // here instead of a preprocessor line comment. 1944 if (getTok().is(AsmToken::Hash)) { 1945 StringRef CommentStr = parseStringToEndOfStatement(); 1946 Lexer.Lex(); 1947 Lexer.UnLex(AsmToken(AsmToken::EndOfStatement, CommentStr)); 1948 } 1949 1950 // Consume any end of statement token, if present, to avoid spurious 1951 // addBlankLine calls(). 1952 if (getTok().is(AsmToken::EndOfStatement)) { 1953 Lex(); 1954 } 1955 1956 if (discardLTOSymbol(IDVal)) 1957 return false; 1958 1959 getTargetParser().doBeforeLabelEmit(Sym, IDLoc); 1960 1961 // Emit the label. 1962 if (!getTargetParser().isParsingMSInlineAsm()) 1963 Out.emitLabel(Sym, IDLoc); 1964 1965 // If we are generating dwarf for assembly source files then gather the 1966 // info to make a dwarf label entry for this label if needed. 1967 if (enabledGenDwarfForAssembly()) 1968 MCGenDwarfLabelEntry::Make(Sym, &getStreamer(), getSourceManager(), 1969 IDLoc); 1970 1971 getTargetParser().onLabelParsed(Sym); 1972 1973 return false; 1974 } 1975 1976 // Check for an assignment statement. 1977 // ::= identifier '=' 1978 if (Lexer.is(AsmToken::Equal) && getTargetParser().equalIsAsmAssignment()) { 1979 Lex(); 1980 return parseAssignment(IDVal, AssignmentKind::Equal); 1981 } 1982 1983 // If macros are enabled, check to see if this is a macro instantiation. 1984 if (areMacrosEnabled()) 1985 if (const MCAsmMacro *M = getContext().lookupMacro(IDVal)) { 1986 return handleMacroEntry(M, IDLoc); 1987 } 1988 1989 // Otherwise, we have a normal instruction or directive. 1990 1991 // Directives start with "." 1992 if (IDVal.startswith(".") && IDVal != ".") { 1993 // There are several entities interested in parsing directives: 1994 // 1995 // 1. The target-specific assembly parser. Some directives are target 1996 // specific or may potentially behave differently on certain targets. 1997 // 2. Asm parser extensions. For example, platform-specific parsers 1998 // (like the ELF parser) register themselves as extensions. 1999 // 3. The generic directive parser implemented by this class. These are 2000 // all the directives that behave in a target and platform independent 2001 // manner, or at least have a default behavior that's shared between 2002 // all targets and platforms. 2003 2004 getTargetParser().flushPendingInstructions(getStreamer()); 2005 2006 SMLoc StartTokLoc = getTok().getLoc(); 2007 bool TPDirectiveReturn = getTargetParser().ParseDirective(ID); 2008 2009 if (hasPendingError()) 2010 return true; 2011 // Currently the return value should be true if we are 2012 // uninterested but as this is at odds with the standard parsing 2013 // convention (return true = error) we have instances of a parsed 2014 // directive that fails returning true as an error. Catch these 2015 // cases as best as possible errors here. 2016 if (TPDirectiveReturn && StartTokLoc != getTok().getLoc()) 2017 return true; 2018 // Return if we did some parsing or believe we succeeded. 2019 if (!TPDirectiveReturn || StartTokLoc != getTok().getLoc()) 2020 return false; 2021 2022 // Next, check the extension directive map to see if any extension has 2023 // registered itself to parse this directive. 2024 std::pair<MCAsmParserExtension *, DirectiveHandler> Handler = 2025 ExtensionDirectiveMap.lookup(IDVal); 2026 if (Handler.first) 2027 return (*Handler.second)(Handler.first, IDVal, IDLoc); 2028 2029 // Finally, if no one else is interested in this directive, it must be 2030 // generic and familiar to this class. 2031 switch (DirKind) { 2032 default: 2033 break; 2034 case DK_SET: 2035 case DK_EQU: 2036 return parseDirectiveSet(IDVal, AssignmentKind::Set); 2037 case DK_EQUIV: 2038 return parseDirectiveSet(IDVal, AssignmentKind::Equiv); 2039 case DK_LTO_SET_CONDITIONAL: 2040 return parseDirectiveSet(IDVal, AssignmentKind::LTOSetConditional); 2041 case DK_ASCII: 2042 return parseDirectiveAscii(IDVal, false); 2043 case DK_ASCIZ: 2044 case DK_STRING: 2045 return parseDirectiveAscii(IDVal, true); 2046 case DK_BYTE: 2047 case DK_DC_B: 2048 return parseDirectiveValue(IDVal, 1); 2049 case DK_DC: 2050 case DK_DC_W: 2051 case DK_SHORT: 2052 case DK_VALUE: 2053 case DK_2BYTE: 2054 return parseDirectiveValue(IDVal, 2); 2055 case DK_LONG: 2056 case DK_INT: 2057 case DK_4BYTE: 2058 case DK_DC_L: 2059 return parseDirectiveValue(IDVal, 4); 2060 case DK_QUAD: 2061 case DK_8BYTE: 2062 return parseDirectiveValue(IDVal, 8); 2063 case DK_DC_A: 2064 return parseDirectiveValue( 2065 IDVal, getContext().getAsmInfo()->getCodePointerSize()); 2066 case DK_OCTA: 2067 return parseDirectiveOctaValue(IDVal); 2068 case DK_SINGLE: 2069 case DK_FLOAT: 2070 case DK_DC_S: 2071 return parseDirectiveRealValue(IDVal, APFloat::IEEEsingle()); 2072 case DK_DOUBLE: 2073 case DK_DC_D: 2074 return parseDirectiveRealValue(IDVal, APFloat::IEEEdouble()); 2075 case DK_ALIGN: { 2076 bool IsPow2 = !getContext().getAsmInfo()->getAlignmentIsInBytes(); 2077 return parseDirectiveAlign(IsPow2, /*ExprSize=*/1); 2078 } 2079 case DK_ALIGN32: { 2080 bool IsPow2 = !getContext().getAsmInfo()->getAlignmentIsInBytes(); 2081 return parseDirectiveAlign(IsPow2, /*ExprSize=*/4); 2082 } 2083 case DK_BALIGN: 2084 return parseDirectiveAlign(/*IsPow2=*/false, /*ExprSize=*/1); 2085 case DK_BALIGNW: 2086 return parseDirectiveAlign(/*IsPow2=*/false, /*ExprSize=*/2); 2087 case DK_BALIGNL: 2088 return parseDirectiveAlign(/*IsPow2=*/false, /*ExprSize=*/4); 2089 case DK_P2ALIGN: 2090 return parseDirectiveAlign(/*IsPow2=*/true, /*ExprSize=*/1); 2091 case DK_P2ALIGNW: 2092 return parseDirectiveAlign(/*IsPow2=*/true, /*ExprSize=*/2); 2093 case DK_P2ALIGNL: 2094 return parseDirectiveAlign(/*IsPow2=*/true, /*ExprSize=*/4); 2095 case DK_ORG: 2096 return parseDirectiveOrg(); 2097 case DK_FILL: 2098 return parseDirectiveFill(); 2099 case DK_ZERO: 2100 return parseDirectiveZero(); 2101 case DK_EXTERN: 2102 eatToEndOfStatement(); // .extern is the default, ignore it. 2103 return false; 2104 case DK_GLOBL: 2105 case DK_GLOBAL: 2106 return parseDirectiveSymbolAttribute(MCSA_Global); 2107 case DK_LAZY_REFERENCE: 2108 return parseDirectiveSymbolAttribute(MCSA_LazyReference); 2109 case DK_NO_DEAD_STRIP: 2110 return parseDirectiveSymbolAttribute(MCSA_NoDeadStrip); 2111 case DK_SYMBOL_RESOLVER: 2112 return parseDirectiveSymbolAttribute(MCSA_SymbolResolver); 2113 case DK_PRIVATE_EXTERN: 2114 return parseDirectiveSymbolAttribute(MCSA_PrivateExtern); 2115 case DK_REFERENCE: 2116 return parseDirectiveSymbolAttribute(MCSA_Reference); 2117 case DK_WEAK_DEFINITION: 2118 return parseDirectiveSymbolAttribute(MCSA_WeakDefinition); 2119 case DK_WEAK_REFERENCE: 2120 return parseDirectiveSymbolAttribute(MCSA_WeakReference); 2121 case DK_WEAK_DEF_CAN_BE_HIDDEN: 2122 return parseDirectiveSymbolAttribute(MCSA_WeakDefAutoPrivate); 2123 case DK_COLD: 2124 return parseDirectiveSymbolAttribute(MCSA_Cold); 2125 case DK_COMM: 2126 case DK_COMMON: 2127 return parseDirectiveComm(/*IsLocal=*/false); 2128 case DK_LCOMM: 2129 return parseDirectiveComm(/*IsLocal=*/true); 2130 case DK_ABORT: 2131 return parseDirectiveAbort(); 2132 case DK_INCLUDE: 2133 return parseDirectiveInclude(); 2134 case DK_INCBIN: 2135 return parseDirectiveIncbin(); 2136 case DK_CODE16: 2137 case DK_CODE16GCC: 2138 return TokError(Twine(IDVal) + 2139 " not currently supported for this target"); 2140 case DK_REPT: 2141 return parseDirectiveRept(IDLoc, IDVal); 2142 case DK_IRP: 2143 return parseDirectiveIrp(IDLoc); 2144 case DK_IRPC: 2145 return parseDirectiveIrpc(IDLoc); 2146 case DK_ENDR: 2147 return parseDirectiveEndr(IDLoc); 2148 case DK_BUNDLE_ALIGN_MODE: 2149 return parseDirectiveBundleAlignMode(); 2150 case DK_BUNDLE_LOCK: 2151 return parseDirectiveBundleLock(); 2152 case DK_BUNDLE_UNLOCK: 2153 return parseDirectiveBundleUnlock(); 2154 case DK_SLEB128: 2155 return parseDirectiveLEB128(true); 2156 case DK_ULEB128: 2157 return parseDirectiveLEB128(false); 2158 case DK_SPACE: 2159 case DK_SKIP: 2160 return parseDirectiveSpace(IDVal); 2161 case DK_FILE: 2162 return parseDirectiveFile(IDLoc); 2163 case DK_LINE: 2164 return parseDirectiveLine(); 2165 case DK_LOC: 2166 return parseDirectiveLoc(); 2167 case DK_STABS: 2168 return parseDirectiveStabs(); 2169 case DK_CV_FILE: 2170 return parseDirectiveCVFile(); 2171 case DK_CV_FUNC_ID: 2172 return parseDirectiveCVFuncId(); 2173 case DK_CV_INLINE_SITE_ID: 2174 return parseDirectiveCVInlineSiteId(); 2175 case DK_CV_LOC: 2176 return parseDirectiveCVLoc(); 2177 case DK_CV_LINETABLE: 2178 return parseDirectiveCVLinetable(); 2179 case DK_CV_INLINE_LINETABLE: 2180 return parseDirectiveCVInlineLinetable(); 2181 case DK_CV_DEF_RANGE: 2182 return parseDirectiveCVDefRange(); 2183 case DK_CV_STRING: 2184 return parseDirectiveCVString(); 2185 case DK_CV_STRINGTABLE: 2186 return parseDirectiveCVStringTable(); 2187 case DK_CV_FILECHECKSUMS: 2188 return parseDirectiveCVFileChecksums(); 2189 case DK_CV_FILECHECKSUM_OFFSET: 2190 return parseDirectiveCVFileChecksumOffset(); 2191 case DK_CV_FPO_DATA: 2192 return parseDirectiveCVFPOData(); 2193 case DK_CFI_SECTIONS: 2194 return parseDirectiveCFISections(); 2195 case DK_CFI_STARTPROC: 2196 return parseDirectiveCFIStartProc(); 2197 case DK_CFI_ENDPROC: 2198 return parseDirectiveCFIEndProc(); 2199 case DK_CFI_DEF_CFA: 2200 return parseDirectiveCFIDefCfa(IDLoc); 2201 case DK_CFI_DEF_CFA_OFFSET: 2202 return parseDirectiveCFIDefCfaOffset(); 2203 case DK_CFI_ADJUST_CFA_OFFSET: 2204 return parseDirectiveCFIAdjustCfaOffset(); 2205 case DK_CFI_DEF_CFA_REGISTER: 2206 return parseDirectiveCFIDefCfaRegister(IDLoc); 2207 case DK_CFI_LLVM_DEF_ASPACE_CFA: 2208 return parseDirectiveCFILLVMDefAspaceCfa(IDLoc); 2209 case DK_CFI_OFFSET: 2210 return parseDirectiveCFIOffset(IDLoc); 2211 case DK_CFI_REL_OFFSET: 2212 return parseDirectiveCFIRelOffset(IDLoc); 2213 case DK_CFI_PERSONALITY: 2214 return parseDirectiveCFIPersonalityOrLsda(true); 2215 case DK_CFI_LSDA: 2216 return parseDirectiveCFIPersonalityOrLsda(false); 2217 case DK_CFI_REMEMBER_STATE: 2218 return parseDirectiveCFIRememberState(); 2219 case DK_CFI_RESTORE_STATE: 2220 return parseDirectiveCFIRestoreState(); 2221 case DK_CFI_SAME_VALUE: 2222 return parseDirectiveCFISameValue(IDLoc); 2223 case DK_CFI_RESTORE: 2224 return parseDirectiveCFIRestore(IDLoc); 2225 case DK_CFI_ESCAPE: 2226 return parseDirectiveCFIEscape(); 2227 case DK_CFI_RETURN_COLUMN: 2228 return parseDirectiveCFIReturnColumn(IDLoc); 2229 case DK_CFI_SIGNAL_FRAME: 2230 return parseDirectiveCFISignalFrame(); 2231 case DK_CFI_UNDEFINED: 2232 return parseDirectiveCFIUndefined(IDLoc); 2233 case DK_CFI_REGISTER: 2234 return parseDirectiveCFIRegister(IDLoc); 2235 case DK_CFI_WINDOW_SAVE: 2236 return parseDirectiveCFIWindowSave(); 2237 case DK_MACROS_ON: 2238 case DK_MACROS_OFF: 2239 return parseDirectiveMacrosOnOff(IDVal); 2240 case DK_MACRO: 2241 return parseDirectiveMacro(IDLoc); 2242 case DK_ALTMACRO: 2243 case DK_NOALTMACRO: 2244 return parseDirectiveAltmacro(IDVal); 2245 case DK_EXITM: 2246 return parseDirectiveExitMacro(IDVal); 2247 case DK_ENDM: 2248 case DK_ENDMACRO: 2249 return parseDirectiveEndMacro(IDVal); 2250 case DK_PURGEM: 2251 return parseDirectivePurgeMacro(IDLoc); 2252 case DK_END: 2253 return parseDirectiveEnd(IDLoc); 2254 case DK_ERR: 2255 return parseDirectiveError(IDLoc, false); 2256 case DK_ERROR: 2257 return parseDirectiveError(IDLoc, true); 2258 case DK_WARNING: 2259 return parseDirectiveWarning(IDLoc); 2260 case DK_RELOC: 2261 return parseDirectiveReloc(IDLoc); 2262 case DK_DCB: 2263 case DK_DCB_W: 2264 return parseDirectiveDCB(IDVal, 2); 2265 case DK_DCB_B: 2266 return parseDirectiveDCB(IDVal, 1); 2267 case DK_DCB_D: 2268 return parseDirectiveRealDCB(IDVal, APFloat::IEEEdouble()); 2269 case DK_DCB_L: 2270 return parseDirectiveDCB(IDVal, 4); 2271 case DK_DCB_S: 2272 return parseDirectiveRealDCB(IDVal, APFloat::IEEEsingle()); 2273 case DK_DC_X: 2274 case DK_DCB_X: 2275 return TokError(Twine(IDVal) + 2276 " not currently supported for this target"); 2277 case DK_DS: 2278 case DK_DS_W: 2279 return parseDirectiveDS(IDVal, 2); 2280 case DK_DS_B: 2281 return parseDirectiveDS(IDVal, 1); 2282 case DK_DS_D: 2283 return parseDirectiveDS(IDVal, 8); 2284 case DK_DS_L: 2285 case DK_DS_S: 2286 return parseDirectiveDS(IDVal, 4); 2287 case DK_DS_P: 2288 case DK_DS_X: 2289 return parseDirectiveDS(IDVal, 12); 2290 case DK_PRINT: 2291 return parseDirectivePrint(IDLoc); 2292 case DK_ADDRSIG: 2293 return parseDirectiveAddrsig(); 2294 case DK_ADDRSIG_SYM: 2295 return parseDirectiveAddrsigSym(); 2296 case DK_PSEUDO_PROBE: 2297 return parseDirectivePseudoProbe(); 2298 case DK_LTO_DISCARD: 2299 return parseDirectiveLTODiscard(); 2300 case DK_MEMTAG: 2301 return parseDirectiveSymbolAttribute(MCSA_Memtag); 2302 } 2303 2304 return Error(IDLoc, "unknown directive"); 2305 } 2306 2307 // __asm _emit or __asm __emit 2308 if (ParsingMSInlineAsm && (IDVal == "_emit" || IDVal == "__emit" || 2309 IDVal == "_EMIT" || IDVal == "__EMIT")) 2310 return parseDirectiveMSEmit(IDLoc, Info, IDVal.size()); 2311 2312 // __asm align 2313 if (ParsingMSInlineAsm && (IDVal == "align" || IDVal == "ALIGN")) 2314 return parseDirectiveMSAlign(IDLoc, Info); 2315 2316 if (ParsingMSInlineAsm && (IDVal == "even" || IDVal == "EVEN")) 2317 Info.AsmRewrites->emplace_back(AOK_EVEN, IDLoc, 4); 2318 if (checkForValidSection()) 2319 return true; 2320 2321 return parseAndMatchAndEmitTargetInstruction(Info, IDVal, ID, IDLoc); 2322 } 2323 2324 bool AsmParser::parseAndMatchAndEmitTargetInstruction(ParseStatementInfo &Info, 2325 StringRef IDVal, 2326 AsmToken ID, 2327 SMLoc IDLoc) { 2328 // Canonicalize the opcode to lower case. 2329 std::string OpcodeStr = IDVal.lower(); 2330 ParseInstructionInfo IInfo(Info.AsmRewrites); 2331 bool ParseHadError = getTargetParser().ParseInstruction(IInfo, OpcodeStr, ID, 2332 Info.ParsedOperands); 2333 Info.ParseError = ParseHadError; 2334 2335 // Dump the parsed representation, if requested. 2336 if (getShowParsedOperands()) { 2337 SmallString<256> Str; 2338 raw_svector_ostream OS(Str); 2339 OS << "parsed instruction: ["; 2340 for (unsigned i = 0; i != Info.ParsedOperands.size(); ++i) { 2341 if (i != 0) 2342 OS << ", "; 2343 Info.ParsedOperands[i]->print(OS); 2344 } 2345 OS << "]"; 2346 2347 printMessage(IDLoc, SourceMgr::DK_Note, OS.str()); 2348 } 2349 2350 // Fail even if ParseInstruction erroneously returns false. 2351 if (hasPendingError() || ParseHadError) 2352 return true; 2353 2354 // If we are generating dwarf for the current section then generate a .loc 2355 // directive for the instruction. 2356 if (!ParseHadError && enabledGenDwarfForAssembly() && 2357 getContext().getGenDwarfSectionSyms().count( 2358 getStreamer().getCurrentSectionOnly())) { 2359 unsigned Line; 2360 if (ActiveMacros.empty()) 2361 Line = SrcMgr.FindLineNumber(IDLoc, CurBuffer); 2362 else 2363 Line = SrcMgr.FindLineNumber(ActiveMacros.front()->InstantiationLoc, 2364 ActiveMacros.front()->ExitBuffer); 2365 2366 // If we previously parsed a cpp hash file line comment then make sure the 2367 // current Dwarf File is for the CppHashFilename if not then emit the 2368 // Dwarf File table for it and adjust the line number for the .loc. 2369 if (!CppHashInfo.Filename.empty()) { 2370 unsigned FileNumber = getStreamer().emitDwarfFileDirective( 2371 0, StringRef(), CppHashInfo.Filename); 2372 getContext().setGenDwarfFileNumber(FileNumber); 2373 2374 unsigned CppHashLocLineNo = 2375 SrcMgr.FindLineNumber(CppHashInfo.Loc, CppHashInfo.Buf); 2376 Line = CppHashInfo.LineNumber - 1 + (Line - CppHashLocLineNo); 2377 } 2378 2379 getStreamer().emitDwarfLocDirective( 2380 getContext().getGenDwarfFileNumber(), Line, 0, 2381 DWARF2_LINE_DEFAULT_IS_STMT ? DWARF2_FLAG_IS_STMT : 0, 0, 0, 2382 StringRef()); 2383 } 2384 2385 // If parsing succeeded, match the instruction. 2386 if (!ParseHadError) { 2387 uint64_t ErrorInfo; 2388 if (getTargetParser().MatchAndEmitInstruction( 2389 IDLoc, Info.Opcode, Info.ParsedOperands, Out, ErrorInfo, 2390 getTargetParser().isParsingMSInlineAsm())) 2391 return true; 2392 } 2393 return false; 2394 } 2395 2396 // Parse and erase curly braces marking block start/end 2397 bool 2398 AsmParser::parseCurlyBlockScope(SmallVectorImpl<AsmRewrite> &AsmStrRewrites) { 2399 // Identify curly brace marking block start/end 2400 if (Lexer.isNot(AsmToken::LCurly) && Lexer.isNot(AsmToken::RCurly)) 2401 return false; 2402 2403 SMLoc StartLoc = Lexer.getLoc(); 2404 Lex(); // Eat the brace 2405 if (Lexer.is(AsmToken::EndOfStatement)) 2406 Lex(); // Eat EndOfStatement following the brace 2407 2408 // Erase the block start/end brace from the output asm string 2409 AsmStrRewrites.emplace_back(AOK_Skip, StartLoc, Lexer.getLoc().getPointer() - 2410 StartLoc.getPointer()); 2411 return true; 2412 } 2413 2414 /// parseCppHashLineFilenameComment as this: 2415 /// ::= # number "filename" 2416 bool AsmParser::parseCppHashLineFilenameComment(SMLoc L, bool SaveLocInfo) { 2417 Lex(); // Eat the hash token. 2418 // Lexer only ever emits HashDirective if it fully formed if it's 2419 // done the checking already so this is an internal error. 2420 assert(getTok().is(AsmToken::Integer) && 2421 "Lexing Cpp line comment: Expected Integer"); 2422 int64_t LineNumber = getTok().getIntVal(); 2423 Lex(); 2424 assert(getTok().is(AsmToken::String) && 2425 "Lexing Cpp line comment: Expected String"); 2426 StringRef Filename = getTok().getString(); 2427 Lex(); 2428 2429 if (!SaveLocInfo) 2430 return false; 2431 2432 // Get rid of the enclosing quotes. 2433 Filename = Filename.substr(1, Filename.size() - 2); 2434 2435 // Save the SMLoc, Filename and LineNumber for later use by diagnostics 2436 // and possibly DWARF file info. 2437 CppHashInfo.Loc = L; 2438 CppHashInfo.Filename = Filename; 2439 CppHashInfo.LineNumber = LineNumber; 2440 CppHashInfo.Buf = CurBuffer; 2441 if (FirstCppHashFilename.empty()) 2442 FirstCppHashFilename = Filename; 2443 return false; 2444 } 2445 2446 /// will use the last parsed cpp hash line filename comment 2447 /// for the Filename and LineNo if any in the diagnostic. 2448 void AsmParser::DiagHandler(const SMDiagnostic &Diag, void *Context) { 2449 auto *Parser = static_cast<AsmParser *>(Context); 2450 raw_ostream &OS = errs(); 2451 2452 const SourceMgr &DiagSrcMgr = *Diag.getSourceMgr(); 2453 SMLoc DiagLoc = Diag.getLoc(); 2454 unsigned DiagBuf = DiagSrcMgr.FindBufferContainingLoc(DiagLoc); 2455 unsigned CppHashBuf = 2456 Parser->SrcMgr.FindBufferContainingLoc(Parser->CppHashInfo.Loc); 2457 2458 // Like SourceMgr::printMessage() we need to print the include stack if any 2459 // before printing the message. 2460 unsigned DiagCurBuffer = DiagSrcMgr.FindBufferContainingLoc(DiagLoc); 2461 if (!Parser->SavedDiagHandler && DiagCurBuffer && 2462 DiagCurBuffer != DiagSrcMgr.getMainFileID()) { 2463 SMLoc ParentIncludeLoc = DiagSrcMgr.getParentIncludeLoc(DiagCurBuffer); 2464 DiagSrcMgr.PrintIncludeStack(ParentIncludeLoc, OS); 2465 } 2466 2467 // If we have not parsed a cpp hash line filename comment or the source 2468 // manager changed or buffer changed (like in a nested include) then just 2469 // print the normal diagnostic using its Filename and LineNo. 2470 if (!Parser->CppHashInfo.LineNumber || DiagBuf != CppHashBuf) { 2471 if (Parser->SavedDiagHandler) 2472 Parser->SavedDiagHandler(Diag, Parser->SavedDiagContext); 2473 else 2474 Parser->getContext().diagnose(Diag); 2475 return; 2476 } 2477 2478 // Use the CppHashFilename and calculate a line number based on the 2479 // CppHashInfo.Loc and CppHashInfo.LineNumber relative to this Diag's SMLoc 2480 // for the diagnostic. 2481 const std::string &Filename = std::string(Parser->CppHashInfo.Filename); 2482 2483 int DiagLocLineNo = DiagSrcMgr.FindLineNumber(DiagLoc, DiagBuf); 2484 int CppHashLocLineNo = 2485 Parser->SrcMgr.FindLineNumber(Parser->CppHashInfo.Loc, CppHashBuf); 2486 int LineNo = 2487 Parser->CppHashInfo.LineNumber - 1 + (DiagLocLineNo - CppHashLocLineNo); 2488 2489 SMDiagnostic NewDiag(*Diag.getSourceMgr(), Diag.getLoc(), Filename, LineNo, 2490 Diag.getColumnNo(), Diag.getKind(), Diag.getMessage(), 2491 Diag.getLineContents(), Diag.getRanges()); 2492 2493 if (Parser->SavedDiagHandler) 2494 Parser->SavedDiagHandler(Diag, Parser->SavedDiagContext); 2495 else 2496 Parser->getContext().diagnose(NewDiag); 2497 } 2498 2499 // FIXME: This is mostly duplicated from the function in AsmLexer.cpp. The 2500 // difference being that that function accepts '@' as part of identifiers and 2501 // we can't do that. AsmLexer.cpp should probably be changed to handle 2502 // '@' as a special case when needed. 2503 static bool isIdentifierChar(char c) { 2504 return isalnum(static_cast<unsigned char>(c)) || c == '_' || c == '$' || 2505 c == '.'; 2506 } 2507 2508 bool AsmParser::expandMacro(raw_svector_ostream &OS, StringRef Body, 2509 ArrayRef<MCAsmMacroParameter> Parameters, 2510 ArrayRef<MCAsmMacroArgument> A, 2511 bool EnableAtPseudoVariable, SMLoc L) { 2512 unsigned NParameters = Parameters.size(); 2513 bool HasVararg = NParameters ? Parameters.back().Vararg : false; 2514 if ((!IsDarwin || NParameters != 0) && NParameters != A.size()) 2515 return Error(L, "Wrong number of arguments"); 2516 2517 // A macro without parameters is handled differently on Darwin: 2518 // gas accepts no arguments and does no substitutions 2519 while (!Body.empty()) { 2520 // Scan for the next substitution. 2521 std::size_t End = Body.size(), Pos = 0; 2522 for (; Pos != End; ++Pos) { 2523 // Check for a substitution or escape. 2524 if (IsDarwin && !NParameters) { 2525 // This macro has no parameters, look for $0, $1, etc. 2526 if (Body[Pos] != '$' || Pos + 1 == End) 2527 continue; 2528 2529 char Next = Body[Pos + 1]; 2530 if (Next == '$' || Next == 'n' || 2531 isdigit(static_cast<unsigned char>(Next))) 2532 break; 2533 } else { 2534 // This macro has parameters, look for \foo, \bar, etc. 2535 if (Body[Pos] == '\\' && Pos + 1 != End) 2536 break; 2537 } 2538 } 2539 2540 // Add the prefix. 2541 OS << Body.slice(0, Pos); 2542 2543 // Check if we reached the end. 2544 if (Pos == End) 2545 break; 2546 2547 if (IsDarwin && !NParameters) { 2548 switch (Body[Pos + 1]) { 2549 // $$ => $ 2550 case '$': 2551 OS << '$'; 2552 break; 2553 2554 // $n => number of arguments 2555 case 'n': 2556 OS << A.size(); 2557 break; 2558 2559 // $[0-9] => argument 2560 default: { 2561 // Missing arguments are ignored. 2562 unsigned Index = Body[Pos + 1] - '0'; 2563 if (Index >= A.size()) 2564 break; 2565 2566 // Otherwise substitute with the token values, with spaces eliminated. 2567 for (const AsmToken &Token : A[Index]) 2568 OS << Token.getString(); 2569 break; 2570 } 2571 } 2572 Pos += 2; 2573 } else { 2574 unsigned I = Pos + 1; 2575 2576 // Check for the \@ pseudo-variable. 2577 if (EnableAtPseudoVariable && Body[I] == '@' && I + 1 != End) 2578 ++I; 2579 else 2580 while (isIdentifierChar(Body[I]) && I + 1 != End) 2581 ++I; 2582 2583 const char *Begin = Body.data() + Pos + 1; 2584 StringRef Argument(Begin, I - (Pos + 1)); 2585 unsigned Index = 0; 2586 2587 if (Argument == "@") { 2588 OS << NumOfMacroInstantiations; 2589 Pos += 2; 2590 } else { 2591 for (; Index < NParameters; ++Index) 2592 if (Parameters[Index].Name == Argument) 2593 break; 2594 2595 if (Index == NParameters) { 2596 if (Body[Pos + 1] == '(' && Body[Pos + 2] == ')') 2597 Pos += 3; 2598 else { 2599 OS << '\\' << Argument; 2600 Pos = I; 2601 } 2602 } else { 2603 bool VarargParameter = HasVararg && Index == (NParameters - 1); 2604 for (const AsmToken &Token : A[Index]) 2605 // For altmacro mode, you can write '%expr'. 2606 // The prefix '%' evaluates the expression 'expr' 2607 // and uses the result as a string (e.g. replace %(1+2) with the 2608 // string "3"). 2609 // Here, we identify the integer token which is the result of the 2610 // absolute expression evaluation and replace it with its string 2611 // representation. 2612 if (AltMacroMode && Token.getString().front() == '%' && 2613 Token.is(AsmToken::Integer)) 2614 // Emit an integer value to the buffer. 2615 OS << Token.getIntVal(); 2616 // Only Token that was validated as a string and begins with '<' 2617 // is considered altMacroString!!! 2618 else if (AltMacroMode && Token.getString().front() == '<' && 2619 Token.is(AsmToken::String)) { 2620 OS << angleBracketString(Token.getStringContents()); 2621 } 2622 // We expect no quotes around the string's contents when 2623 // parsing for varargs. 2624 else if (Token.isNot(AsmToken::String) || VarargParameter) 2625 OS << Token.getString(); 2626 else 2627 OS << Token.getStringContents(); 2628 2629 Pos += 1 + Argument.size(); 2630 } 2631 } 2632 } 2633 // Update the scan point. 2634 Body = Body.substr(Pos); 2635 } 2636 2637 return false; 2638 } 2639 2640 static bool isOperator(AsmToken::TokenKind kind) { 2641 switch (kind) { 2642 default: 2643 return false; 2644 case AsmToken::Plus: 2645 case AsmToken::Minus: 2646 case AsmToken::Tilde: 2647 case AsmToken::Slash: 2648 case AsmToken::Star: 2649 case AsmToken::Dot: 2650 case AsmToken::Equal: 2651 case AsmToken::EqualEqual: 2652 case AsmToken::Pipe: 2653 case AsmToken::PipePipe: 2654 case AsmToken::Caret: 2655 case AsmToken::Amp: 2656 case AsmToken::AmpAmp: 2657 case AsmToken::Exclaim: 2658 case AsmToken::ExclaimEqual: 2659 case AsmToken::Less: 2660 case AsmToken::LessEqual: 2661 case AsmToken::LessLess: 2662 case AsmToken::LessGreater: 2663 case AsmToken::Greater: 2664 case AsmToken::GreaterEqual: 2665 case AsmToken::GreaterGreater: 2666 return true; 2667 } 2668 } 2669 2670 namespace { 2671 2672 class AsmLexerSkipSpaceRAII { 2673 public: 2674 AsmLexerSkipSpaceRAII(AsmLexer &Lexer, bool SkipSpace) : Lexer(Lexer) { 2675 Lexer.setSkipSpace(SkipSpace); 2676 } 2677 2678 ~AsmLexerSkipSpaceRAII() { 2679 Lexer.setSkipSpace(true); 2680 } 2681 2682 private: 2683 AsmLexer &Lexer; 2684 }; 2685 2686 } // end anonymous namespace 2687 2688 bool AsmParser::parseMacroArgument(MCAsmMacroArgument &MA, bool Vararg) { 2689 2690 if (Vararg) { 2691 if (Lexer.isNot(AsmToken::EndOfStatement)) { 2692 StringRef Str = parseStringToEndOfStatement(); 2693 MA.emplace_back(AsmToken::String, Str); 2694 } 2695 return false; 2696 } 2697 2698 unsigned ParenLevel = 0; 2699 2700 // Darwin doesn't use spaces to delmit arguments. 2701 AsmLexerSkipSpaceRAII ScopedSkipSpace(Lexer, IsDarwin); 2702 2703 bool SpaceEaten; 2704 2705 while (true) { 2706 SpaceEaten = false; 2707 if (Lexer.is(AsmToken::Eof) || Lexer.is(AsmToken::Equal)) 2708 return TokError("unexpected token in macro instantiation"); 2709 2710 if (ParenLevel == 0) { 2711 2712 if (Lexer.is(AsmToken::Comma)) 2713 break; 2714 2715 if (Lexer.is(AsmToken::Space)) { 2716 SpaceEaten = true; 2717 Lexer.Lex(); // Eat spaces 2718 } 2719 2720 // Spaces can delimit parameters, but could also be part an expression. 2721 // If the token after a space is an operator, add the token and the next 2722 // one into this argument 2723 if (!IsDarwin) { 2724 if (isOperator(Lexer.getKind())) { 2725 MA.push_back(getTok()); 2726 Lexer.Lex(); 2727 2728 // Whitespace after an operator can be ignored. 2729 if (Lexer.is(AsmToken::Space)) 2730 Lexer.Lex(); 2731 2732 continue; 2733 } 2734 } 2735 if (SpaceEaten) 2736 break; 2737 } 2738 2739 // handleMacroEntry relies on not advancing the lexer here 2740 // to be able to fill in the remaining default parameter values 2741 if (Lexer.is(AsmToken::EndOfStatement)) 2742 break; 2743 2744 // Adjust the current parentheses level. 2745 if (Lexer.is(AsmToken::LParen)) 2746 ++ParenLevel; 2747 else if (Lexer.is(AsmToken::RParen) && ParenLevel) 2748 --ParenLevel; 2749 2750 // Append the token to the current argument list. 2751 MA.push_back(getTok()); 2752 Lexer.Lex(); 2753 } 2754 2755 if (ParenLevel != 0) 2756 return TokError("unbalanced parentheses in macro argument"); 2757 return false; 2758 } 2759 2760 // Parse the macro instantiation arguments. 2761 bool AsmParser::parseMacroArguments(const MCAsmMacro *M, 2762 MCAsmMacroArguments &A) { 2763 const unsigned NParameters = M ? M->Parameters.size() : 0; 2764 bool NamedParametersFound = false; 2765 SmallVector<SMLoc, 4> FALocs; 2766 2767 A.resize(NParameters); 2768 FALocs.resize(NParameters); 2769 2770 // Parse two kinds of macro invocations: 2771 // - macros defined without any parameters accept an arbitrary number of them 2772 // - macros defined with parameters accept at most that many of them 2773 bool HasVararg = NParameters ? M->Parameters.back().Vararg : false; 2774 for (unsigned Parameter = 0; !NParameters || Parameter < NParameters; 2775 ++Parameter) { 2776 SMLoc IDLoc = Lexer.getLoc(); 2777 MCAsmMacroParameter FA; 2778 2779 if (Lexer.is(AsmToken::Identifier) && Lexer.peekTok().is(AsmToken::Equal)) { 2780 if (parseIdentifier(FA.Name)) 2781 return Error(IDLoc, "invalid argument identifier for formal argument"); 2782 2783 if (Lexer.isNot(AsmToken::Equal)) 2784 return TokError("expected '=' after formal parameter identifier"); 2785 2786 Lex(); 2787 2788 NamedParametersFound = true; 2789 } 2790 bool Vararg = HasVararg && Parameter == (NParameters - 1); 2791 2792 if (NamedParametersFound && FA.Name.empty()) 2793 return Error(IDLoc, "cannot mix positional and keyword arguments"); 2794 2795 SMLoc StrLoc = Lexer.getLoc(); 2796 SMLoc EndLoc; 2797 if (AltMacroMode && Lexer.is(AsmToken::Percent)) { 2798 const MCExpr *AbsoluteExp; 2799 int64_t Value; 2800 /// Eat '%' 2801 Lex(); 2802 if (parseExpression(AbsoluteExp, EndLoc)) 2803 return false; 2804 if (!AbsoluteExp->evaluateAsAbsolute(Value, 2805 getStreamer().getAssemblerPtr())) 2806 return Error(StrLoc, "expected absolute expression"); 2807 const char *StrChar = StrLoc.getPointer(); 2808 const char *EndChar = EndLoc.getPointer(); 2809 AsmToken newToken(AsmToken::Integer, 2810 StringRef(StrChar, EndChar - StrChar), Value); 2811 FA.Value.push_back(newToken); 2812 } else if (AltMacroMode && Lexer.is(AsmToken::Less) && 2813 isAngleBracketString(StrLoc, EndLoc)) { 2814 const char *StrChar = StrLoc.getPointer(); 2815 const char *EndChar = EndLoc.getPointer(); 2816 jumpToLoc(EndLoc, CurBuffer); 2817 /// Eat from '<' to '>' 2818 Lex(); 2819 AsmToken newToken(AsmToken::String, 2820 StringRef(StrChar, EndChar - StrChar)); 2821 FA.Value.push_back(newToken); 2822 } else if(parseMacroArgument(FA.Value, Vararg)) 2823 return true; 2824 2825 unsigned PI = Parameter; 2826 if (!FA.Name.empty()) { 2827 unsigned FAI = 0; 2828 for (FAI = 0; FAI < NParameters; ++FAI) 2829 if (M->Parameters[FAI].Name == FA.Name) 2830 break; 2831 2832 if (FAI >= NParameters) { 2833 assert(M && "expected macro to be defined"); 2834 return Error(IDLoc, "parameter named '" + FA.Name + 2835 "' does not exist for macro '" + M->Name + "'"); 2836 } 2837 PI = FAI; 2838 } 2839 2840 if (!FA.Value.empty()) { 2841 if (A.size() <= PI) 2842 A.resize(PI + 1); 2843 A[PI] = FA.Value; 2844 2845 if (FALocs.size() <= PI) 2846 FALocs.resize(PI + 1); 2847 2848 FALocs[PI] = Lexer.getLoc(); 2849 } 2850 2851 // At the end of the statement, fill in remaining arguments that have 2852 // default values. If there aren't any, then the next argument is 2853 // required but missing 2854 if (Lexer.is(AsmToken::EndOfStatement)) { 2855 bool Failure = false; 2856 for (unsigned FAI = 0; FAI < NParameters; ++FAI) { 2857 if (A[FAI].empty()) { 2858 if (M->Parameters[FAI].Required) { 2859 Error(FALocs[FAI].isValid() ? FALocs[FAI] : Lexer.getLoc(), 2860 "missing value for required parameter " 2861 "'" + M->Parameters[FAI].Name + "' in macro '" + M->Name + "'"); 2862 Failure = true; 2863 } 2864 2865 if (!M->Parameters[FAI].Value.empty()) 2866 A[FAI] = M->Parameters[FAI].Value; 2867 } 2868 } 2869 return Failure; 2870 } 2871 2872 if (Lexer.is(AsmToken::Comma)) 2873 Lex(); 2874 } 2875 2876 return TokError("too many positional arguments"); 2877 } 2878 2879 bool AsmParser::handleMacroEntry(const MCAsmMacro *M, SMLoc NameLoc) { 2880 // Arbitrarily limit macro nesting depth (default matches 'as'). We can 2881 // eliminate this, although we should protect against infinite loops. 2882 unsigned MaxNestingDepth = AsmMacroMaxNestingDepth; 2883 if (ActiveMacros.size() == MaxNestingDepth) { 2884 std::ostringstream MaxNestingDepthError; 2885 MaxNestingDepthError << "macros cannot be nested more than " 2886 << MaxNestingDepth << " levels deep." 2887 << " Use -asm-macro-max-nesting-depth to increase " 2888 "this limit."; 2889 return TokError(MaxNestingDepthError.str()); 2890 } 2891 2892 MCAsmMacroArguments A; 2893 if (parseMacroArguments(M, A)) 2894 return true; 2895 2896 // Macro instantiation is lexical, unfortunately. We construct a new buffer 2897 // to hold the macro body with substitutions. 2898 SmallString<256> Buf; 2899 StringRef Body = M->Body; 2900 raw_svector_ostream OS(Buf); 2901 2902 if (expandMacro(OS, Body, M->Parameters, A, true, getTok().getLoc())) 2903 return true; 2904 2905 // We include the .endmacro in the buffer as our cue to exit the macro 2906 // instantiation. 2907 OS << ".endmacro\n"; 2908 2909 std::unique_ptr<MemoryBuffer> Instantiation = 2910 MemoryBuffer::getMemBufferCopy(OS.str(), "<instantiation>"); 2911 2912 // Create the macro instantiation object and add to the current macro 2913 // instantiation stack. 2914 MacroInstantiation *MI = new MacroInstantiation{ 2915 NameLoc, CurBuffer, getTok().getLoc(), TheCondStack.size()}; 2916 ActiveMacros.push_back(MI); 2917 2918 ++NumOfMacroInstantiations; 2919 2920 // Jump to the macro instantiation and prime the lexer. 2921 CurBuffer = SrcMgr.AddNewSourceBuffer(std::move(Instantiation), SMLoc()); 2922 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer()); 2923 Lex(); 2924 2925 return false; 2926 } 2927 2928 void AsmParser::handleMacroExit() { 2929 // Jump to the EndOfStatement we should return to, and consume it. 2930 jumpToLoc(ActiveMacros.back()->ExitLoc, ActiveMacros.back()->ExitBuffer); 2931 Lex(); 2932 2933 // Pop the instantiation entry. 2934 delete ActiveMacros.back(); 2935 ActiveMacros.pop_back(); 2936 } 2937 2938 bool AsmParser::parseAssignment(StringRef Name, AssignmentKind Kind) { 2939 MCSymbol *Sym; 2940 const MCExpr *Value; 2941 SMLoc ExprLoc = getTok().getLoc(); 2942 bool AllowRedef = 2943 Kind == AssignmentKind::Set || Kind == AssignmentKind::Equal; 2944 if (MCParserUtils::parseAssignmentExpression(Name, AllowRedef, *this, Sym, 2945 Value)) 2946 return true; 2947 2948 if (!Sym) { 2949 // In the case where we parse an expression starting with a '.', we will 2950 // not generate an error, nor will we create a symbol. In this case we 2951 // should just return out. 2952 return false; 2953 } 2954 2955 if (discardLTOSymbol(Name)) 2956 return false; 2957 2958 // Do the assignment. 2959 switch (Kind) { 2960 case AssignmentKind::Equal: 2961 Out.emitAssignment(Sym, Value); 2962 break; 2963 case AssignmentKind::Set: 2964 case AssignmentKind::Equiv: 2965 Out.emitAssignment(Sym, Value); 2966 Out.emitSymbolAttribute(Sym, MCSA_NoDeadStrip); 2967 break; 2968 case AssignmentKind::LTOSetConditional: 2969 if (Value->getKind() != MCExpr::SymbolRef) 2970 return Error(ExprLoc, "expected identifier"); 2971 2972 Out.emitConditionalAssignment(Sym, Value); 2973 break; 2974 } 2975 2976 return false; 2977 } 2978 2979 /// parseIdentifier: 2980 /// ::= identifier 2981 /// ::= string 2982 bool AsmParser::parseIdentifier(StringRef &Res) { 2983 // The assembler has relaxed rules for accepting identifiers, in particular we 2984 // allow things like '.globl $foo' and '.def @feat.00', which would normally be 2985 // separate tokens. At this level, we have already lexed so we cannot (currently) 2986 // handle this as a context dependent token, instead we detect adjacent tokens 2987 // and return the combined identifier. 2988 if (Lexer.is(AsmToken::Dollar) || Lexer.is(AsmToken::At)) { 2989 SMLoc PrefixLoc = getLexer().getLoc(); 2990 2991 // Consume the prefix character, and check for a following identifier. 2992 2993 AsmToken Buf[1]; 2994 Lexer.peekTokens(Buf, false); 2995 2996 if (Buf[0].isNot(AsmToken::Identifier) && Buf[0].isNot(AsmToken::Integer)) 2997 return true; 2998 2999 // We have a '$' or '@' followed by an identifier or integer token, make 3000 // sure they are adjacent. 3001 if (PrefixLoc.getPointer() + 1 != Buf[0].getLoc().getPointer()) 3002 return true; 3003 3004 // eat $ or @ 3005 Lexer.Lex(); // Lexer's Lex guarantees consecutive token. 3006 // Construct the joined identifier and consume the token. 3007 Res = StringRef(PrefixLoc.getPointer(), getTok().getString().size() + 1); 3008 Lex(); // Parser Lex to maintain invariants. 3009 return false; 3010 } 3011 3012 if (Lexer.isNot(AsmToken::Identifier) && Lexer.isNot(AsmToken::String)) 3013 return true; 3014 3015 Res = getTok().getIdentifier(); 3016 3017 Lex(); // Consume the identifier token. 3018 3019 return false; 3020 } 3021 3022 /// parseDirectiveSet: 3023 /// ::= .equ identifier ',' expression 3024 /// ::= .equiv identifier ',' expression 3025 /// ::= .set identifier ',' expression 3026 /// ::= .lto_set_conditional identifier ',' expression 3027 bool AsmParser::parseDirectiveSet(StringRef IDVal, AssignmentKind Kind) { 3028 StringRef Name; 3029 if (check(parseIdentifier(Name), "expected identifier") || parseComma() || 3030 parseAssignment(Name, Kind)) 3031 return true; 3032 return false; 3033 } 3034 3035 bool AsmParser::parseEscapedString(std::string &Data) { 3036 if (check(getTok().isNot(AsmToken::String), "expected string")) 3037 return true; 3038 3039 Data = ""; 3040 StringRef Str = getTok().getStringContents(); 3041 for (unsigned i = 0, e = Str.size(); i != e; ++i) { 3042 if (Str[i] != '\\') { 3043 Data += Str[i]; 3044 continue; 3045 } 3046 3047 // Recognize escaped characters. Note that this escape semantics currently 3048 // loosely follows Darwin 'as'. 3049 ++i; 3050 if (i == e) 3051 return TokError("unexpected backslash at end of string"); 3052 3053 // Recognize hex sequences similarly to GNU 'as'. 3054 if (Str[i] == 'x' || Str[i] == 'X') { 3055 size_t length = Str.size(); 3056 if (i + 1 >= length || !isHexDigit(Str[i + 1])) 3057 return TokError("invalid hexadecimal escape sequence"); 3058 3059 // Consume hex characters. GNU 'as' reads all hexadecimal characters and 3060 // then truncates to the lower 16 bits. Seems reasonable. 3061 unsigned Value = 0; 3062 while (i + 1 < length && isHexDigit(Str[i + 1])) 3063 Value = Value * 16 + hexDigitValue(Str[++i]); 3064 3065 Data += (unsigned char)(Value & 0xFF); 3066 continue; 3067 } 3068 3069 // Recognize octal sequences. 3070 if ((unsigned)(Str[i] - '0') <= 7) { 3071 // Consume up to three octal characters. 3072 unsigned Value = Str[i] - '0'; 3073 3074 if (i + 1 != e && ((unsigned)(Str[i + 1] - '0')) <= 7) { 3075 ++i; 3076 Value = Value * 8 + (Str[i] - '0'); 3077 3078 if (i + 1 != e && ((unsigned)(Str[i + 1] - '0')) <= 7) { 3079 ++i; 3080 Value = Value * 8 + (Str[i] - '0'); 3081 } 3082 } 3083 3084 if (Value > 255) 3085 return TokError("invalid octal escape sequence (out of range)"); 3086 3087 Data += (unsigned char)Value; 3088 continue; 3089 } 3090 3091 // Otherwise recognize individual escapes. 3092 switch (Str[i]) { 3093 default: 3094 // Just reject invalid escape sequences for now. 3095 return TokError("invalid escape sequence (unrecognized character)"); 3096 3097 case 'b': Data += '\b'; break; 3098 case 'f': Data += '\f'; break; 3099 case 'n': Data += '\n'; break; 3100 case 'r': Data += '\r'; break; 3101 case 't': Data += '\t'; break; 3102 case '"': Data += '"'; break; 3103 case '\\': Data += '\\'; break; 3104 } 3105 } 3106 3107 Lex(); 3108 return false; 3109 } 3110 3111 bool AsmParser::parseAngleBracketString(std::string &Data) { 3112 SMLoc EndLoc, StartLoc = getTok().getLoc(); 3113 if (isAngleBracketString(StartLoc, EndLoc)) { 3114 const char *StartChar = StartLoc.getPointer() + 1; 3115 const char *EndChar = EndLoc.getPointer() - 1; 3116 jumpToLoc(EndLoc, CurBuffer); 3117 /// Eat from '<' to '>' 3118 Lex(); 3119 3120 Data = angleBracketString(StringRef(StartChar, EndChar - StartChar)); 3121 return false; 3122 } 3123 return true; 3124 } 3125 3126 /// parseDirectiveAscii: 3127 // ::= .ascii [ "string"+ ( , "string"+ )* ] 3128 /// ::= ( .asciz | .string ) [ "string" ( , "string" )* ] 3129 bool AsmParser::parseDirectiveAscii(StringRef IDVal, bool ZeroTerminated) { 3130 auto parseOp = [&]() -> bool { 3131 std::string Data; 3132 if (checkForValidSection()) 3133 return true; 3134 // Only support spaces as separators for .ascii directive for now. See the 3135 // discusssion at https://reviews.llvm.org/D91460 for more details. 3136 do { 3137 if (parseEscapedString(Data)) 3138 return true; 3139 getStreamer().emitBytes(Data); 3140 } while (!ZeroTerminated && getTok().is(AsmToken::String)); 3141 if (ZeroTerminated) 3142 getStreamer().emitBytes(StringRef("\0", 1)); 3143 return false; 3144 }; 3145 3146 return parseMany(parseOp); 3147 } 3148 3149 /// parseDirectiveReloc 3150 /// ::= .reloc expression , identifier [ , expression ] 3151 bool AsmParser::parseDirectiveReloc(SMLoc DirectiveLoc) { 3152 const MCExpr *Offset; 3153 const MCExpr *Expr = nullptr; 3154 SMLoc OffsetLoc = Lexer.getTok().getLoc(); 3155 3156 if (parseExpression(Offset)) 3157 return true; 3158 if (parseComma() || 3159 check(getTok().isNot(AsmToken::Identifier), "expected relocation name")) 3160 return true; 3161 3162 SMLoc NameLoc = Lexer.getTok().getLoc(); 3163 StringRef Name = Lexer.getTok().getIdentifier(); 3164 Lex(); 3165 3166 if (Lexer.is(AsmToken::Comma)) { 3167 Lex(); 3168 SMLoc ExprLoc = Lexer.getLoc(); 3169 if (parseExpression(Expr)) 3170 return true; 3171 3172 MCValue Value; 3173 if (!Expr->evaluateAsRelocatable(Value, nullptr, nullptr)) 3174 return Error(ExprLoc, "expression must be relocatable"); 3175 } 3176 3177 if (parseEOL()) 3178 return true; 3179 3180 const MCTargetAsmParser &MCT = getTargetParser(); 3181 const MCSubtargetInfo &STI = MCT.getSTI(); 3182 if (std::optional<std::pair<bool, std::string>> Err = 3183 getStreamer().emitRelocDirective(*Offset, Name, Expr, DirectiveLoc, 3184 STI)) 3185 return Error(Err->first ? NameLoc : OffsetLoc, Err->second); 3186 3187 return false; 3188 } 3189 3190 /// parseDirectiveValue 3191 /// ::= (.byte | .short | ... ) [ expression (, expression)* ] 3192 bool AsmParser::parseDirectiveValue(StringRef IDVal, unsigned Size) { 3193 auto parseOp = [&]() -> bool { 3194 const MCExpr *Value; 3195 SMLoc ExprLoc = getLexer().getLoc(); 3196 if (checkForValidSection() || parseExpression(Value)) 3197 return true; 3198 // Special case constant expressions to match code generator. 3199 if (const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value)) { 3200 assert(Size <= 8 && "Invalid size"); 3201 uint64_t IntValue = MCE->getValue(); 3202 if (!isUIntN(8 * Size, IntValue) && !isIntN(8 * Size, IntValue)) 3203 return Error(ExprLoc, "out of range literal value"); 3204 getStreamer().emitIntValue(IntValue, Size); 3205 } else 3206 getStreamer().emitValue(Value, Size, ExprLoc); 3207 return false; 3208 }; 3209 3210 return parseMany(parseOp); 3211 } 3212 3213 static bool parseHexOcta(AsmParser &Asm, uint64_t &hi, uint64_t &lo) { 3214 if (Asm.getTok().isNot(AsmToken::Integer) && 3215 Asm.getTok().isNot(AsmToken::BigNum)) 3216 return Asm.TokError("unknown token in expression"); 3217 SMLoc ExprLoc = Asm.getTok().getLoc(); 3218 APInt IntValue = Asm.getTok().getAPIntVal(); 3219 Asm.Lex(); 3220 if (!IntValue.isIntN(128)) 3221 return Asm.Error(ExprLoc, "out of range literal value"); 3222 if (!IntValue.isIntN(64)) { 3223 hi = IntValue.getHiBits(IntValue.getBitWidth() - 64).getZExtValue(); 3224 lo = IntValue.getLoBits(64).getZExtValue(); 3225 } else { 3226 hi = 0; 3227 lo = IntValue.getZExtValue(); 3228 } 3229 return false; 3230 } 3231 3232 /// ParseDirectiveOctaValue 3233 /// ::= .octa [ hexconstant (, hexconstant)* ] 3234 3235 bool AsmParser::parseDirectiveOctaValue(StringRef IDVal) { 3236 auto parseOp = [&]() -> bool { 3237 if (checkForValidSection()) 3238 return true; 3239 uint64_t hi, lo; 3240 if (parseHexOcta(*this, hi, lo)) 3241 return true; 3242 if (MAI.isLittleEndian()) { 3243 getStreamer().emitInt64(lo); 3244 getStreamer().emitInt64(hi); 3245 } else { 3246 getStreamer().emitInt64(hi); 3247 getStreamer().emitInt64(lo); 3248 } 3249 return false; 3250 }; 3251 3252 return parseMany(parseOp); 3253 } 3254 3255 bool AsmParser::parseRealValue(const fltSemantics &Semantics, APInt &Res) { 3256 // We don't truly support arithmetic on floating point expressions, so we 3257 // have to manually parse unary prefixes. 3258 bool IsNeg = false; 3259 if (getLexer().is(AsmToken::Minus)) { 3260 Lexer.Lex(); 3261 IsNeg = true; 3262 } else if (getLexer().is(AsmToken::Plus)) 3263 Lexer.Lex(); 3264 3265 if (Lexer.is(AsmToken::Error)) 3266 return TokError(Lexer.getErr()); 3267 if (Lexer.isNot(AsmToken::Integer) && Lexer.isNot(AsmToken::Real) && 3268 Lexer.isNot(AsmToken::Identifier)) 3269 return TokError("unexpected token in directive"); 3270 3271 // Convert to an APFloat. 3272 APFloat Value(Semantics); 3273 StringRef IDVal = getTok().getString(); 3274 if (getLexer().is(AsmToken::Identifier)) { 3275 if (!IDVal.compare_insensitive("infinity") || 3276 !IDVal.compare_insensitive("inf")) 3277 Value = APFloat::getInf(Semantics); 3278 else if (!IDVal.compare_insensitive("nan")) 3279 Value = APFloat::getNaN(Semantics, false, ~0); 3280 else 3281 return TokError("invalid floating point literal"); 3282 } else if (errorToBool( 3283 Value.convertFromString(IDVal, APFloat::rmNearestTiesToEven) 3284 .takeError())) 3285 return TokError("invalid floating point literal"); 3286 if (IsNeg) 3287 Value.changeSign(); 3288 3289 // Consume the numeric token. 3290 Lex(); 3291 3292 Res = Value.bitcastToAPInt(); 3293 3294 return false; 3295 } 3296 3297 /// parseDirectiveRealValue 3298 /// ::= (.single | .double) [ expression (, expression)* ] 3299 bool AsmParser::parseDirectiveRealValue(StringRef IDVal, 3300 const fltSemantics &Semantics) { 3301 auto parseOp = [&]() -> bool { 3302 APInt AsInt; 3303 if (checkForValidSection() || parseRealValue(Semantics, AsInt)) 3304 return true; 3305 getStreamer().emitIntValue(AsInt.getLimitedValue(), 3306 AsInt.getBitWidth() / 8); 3307 return false; 3308 }; 3309 3310 return parseMany(parseOp); 3311 } 3312 3313 /// parseDirectiveZero 3314 /// ::= .zero expression 3315 bool AsmParser::parseDirectiveZero() { 3316 SMLoc NumBytesLoc = Lexer.getLoc(); 3317 const MCExpr *NumBytes; 3318 if (checkForValidSection() || parseExpression(NumBytes)) 3319 return true; 3320 3321 int64_t Val = 0; 3322 if (getLexer().is(AsmToken::Comma)) { 3323 Lex(); 3324 if (parseAbsoluteExpression(Val)) 3325 return true; 3326 } 3327 3328 if (parseEOL()) 3329 return true; 3330 getStreamer().emitFill(*NumBytes, Val, NumBytesLoc); 3331 3332 return false; 3333 } 3334 3335 /// parseDirectiveFill 3336 /// ::= .fill expression [ , expression [ , expression ] ] 3337 bool AsmParser::parseDirectiveFill() { 3338 SMLoc NumValuesLoc = Lexer.getLoc(); 3339 const MCExpr *NumValues; 3340 if (checkForValidSection() || parseExpression(NumValues)) 3341 return true; 3342 3343 int64_t FillSize = 1; 3344 int64_t FillExpr = 0; 3345 3346 SMLoc SizeLoc, ExprLoc; 3347 3348 if (parseOptionalToken(AsmToken::Comma)) { 3349 SizeLoc = getTok().getLoc(); 3350 if (parseAbsoluteExpression(FillSize)) 3351 return true; 3352 if (parseOptionalToken(AsmToken::Comma)) { 3353 ExprLoc = getTok().getLoc(); 3354 if (parseAbsoluteExpression(FillExpr)) 3355 return true; 3356 } 3357 } 3358 if (parseEOL()) 3359 return true; 3360 3361 if (FillSize < 0) { 3362 Warning(SizeLoc, "'.fill' directive with negative size has no effect"); 3363 return false; 3364 } 3365 if (FillSize > 8) { 3366 Warning(SizeLoc, "'.fill' directive with size greater than 8 has been truncated to 8"); 3367 FillSize = 8; 3368 } 3369 3370 if (!isUInt<32>(FillExpr) && FillSize > 4) 3371 Warning(ExprLoc, "'.fill' directive pattern has been truncated to 32-bits"); 3372 3373 getStreamer().emitFill(*NumValues, FillSize, FillExpr, NumValuesLoc); 3374 3375 return false; 3376 } 3377 3378 /// parseDirectiveOrg 3379 /// ::= .org expression [ , expression ] 3380 bool AsmParser::parseDirectiveOrg() { 3381 const MCExpr *Offset; 3382 SMLoc OffsetLoc = Lexer.getLoc(); 3383 if (checkForValidSection() || parseExpression(Offset)) 3384 return true; 3385 3386 // Parse optional fill expression. 3387 int64_t FillExpr = 0; 3388 if (parseOptionalToken(AsmToken::Comma)) 3389 if (parseAbsoluteExpression(FillExpr)) 3390 return true; 3391 if (parseEOL()) 3392 return true; 3393 3394 getStreamer().emitValueToOffset(Offset, FillExpr, OffsetLoc); 3395 return false; 3396 } 3397 3398 /// parseDirectiveAlign 3399 /// ::= {.align, ...} expression [ , expression [ , expression ]] 3400 bool AsmParser::parseDirectiveAlign(bool IsPow2, unsigned ValueSize) { 3401 SMLoc AlignmentLoc = getLexer().getLoc(); 3402 int64_t Alignment; 3403 SMLoc MaxBytesLoc; 3404 bool HasFillExpr = false; 3405 int64_t FillExpr = 0; 3406 int64_t MaxBytesToFill = 0; 3407 3408 auto parseAlign = [&]() -> bool { 3409 if (parseAbsoluteExpression(Alignment)) 3410 return true; 3411 if (parseOptionalToken(AsmToken::Comma)) { 3412 // The fill expression can be omitted while specifying a maximum number of 3413 // alignment bytes, e.g: 3414 // .align 3,,4 3415 if (getTok().isNot(AsmToken::Comma)) { 3416 HasFillExpr = true; 3417 if (parseAbsoluteExpression(FillExpr)) 3418 return true; 3419 } 3420 if (parseOptionalToken(AsmToken::Comma)) 3421 if (parseTokenLoc(MaxBytesLoc) || 3422 parseAbsoluteExpression(MaxBytesToFill)) 3423 return true; 3424 } 3425 return parseEOL(); 3426 }; 3427 3428 if (checkForValidSection()) 3429 return true; 3430 // Ignore empty '.p2align' directives for GNU-as compatibility 3431 if (IsPow2 && (ValueSize == 1) && getTok().is(AsmToken::EndOfStatement)) { 3432 Warning(AlignmentLoc, "p2align directive with no operand(s) is ignored"); 3433 return parseEOL(); 3434 } 3435 if (parseAlign()) 3436 return true; 3437 3438 // Always emit an alignment here even if we thrown an error. 3439 bool ReturnVal = false; 3440 3441 // Compute alignment in bytes. 3442 if (IsPow2) { 3443 // FIXME: Diagnose overflow. 3444 if (Alignment >= 32) { 3445 ReturnVal |= Error(AlignmentLoc, "invalid alignment value"); 3446 Alignment = 31; 3447 } 3448 3449 Alignment = 1ULL << Alignment; 3450 } else { 3451 // Reject alignments that aren't either a power of two or zero, 3452 // for gas compatibility. Alignment of zero is silently rounded 3453 // up to one. 3454 if (Alignment == 0) 3455 Alignment = 1; 3456 else if (!isPowerOf2_64(Alignment)) { 3457 ReturnVal |= Error(AlignmentLoc, "alignment must be a power of 2"); 3458 Alignment = PowerOf2Floor(Alignment); 3459 } 3460 if (!isUInt<32>(Alignment)) { 3461 ReturnVal |= Error(AlignmentLoc, "alignment must be smaller than 2**32"); 3462 Alignment = 1u << 31; 3463 } 3464 } 3465 3466 // Diagnose non-sensical max bytes to align. 3467 if (MaxBytesLoc.isValid()) { 3468 if (MaxBytesToFill < 1) { 3469 ReturnVal |= Error(MaxBytesLoc, 3470 "alignment directive can never be satisfied in this " 3471 "many bytes, ignoring maximum bytes expression"); 3472 MaxBytesToFill = 0; 3473 } 3474 3475 if (MaxBytesToFill >= Alignment) { 3476 Warning(MaxBytesLoc, "maximum bytes expression exceeds alignment and " 3477 "has no effect"); 3478 MaxBytesToFill = 0; 3479 } 3480 } 3481 3482 // Check whether we should use optimal code alignment for this .align 3483 // directive. 3484 const MCSection *Section = getStreamer().getCurrentSectionOnly(); 3485 assert(Section && "must have section to emit alignment"); 3486 bool useCodeAlign = Section->useCodeAlign(); 3487 if ((!HasFillExpr || Lexer.getMAI().getTextAlignFillValue() == FillExpr) && 3488 ValueSize == 1 && useCodeAlign) { 3489 getStreamer().emitCodeAlignment( 3490 Align(Alignment), &getTargetParser().getSTI(), MaxBytesToFill); 3491 } else { 3492 // FIXME: Target specific behavior about how the "extra" bytes are filled. 3493 getStreamer().emitValueToAlignment(Align(Alignment), FillExpr, ValueSize, 3494 MaxBytesToFill); 3495 } 3496 3497 return ReturnVal; 3498 } 3499 3500 /// parseDirectiveFile 3501 /// ::= .file filename 3502 /// ::= .file number [directory] filename [md5 checksum] [source source-text] 3503 bool AsmParser::parseDirectiveFile(SMLoc DirectiveLoc) { 3504 // FIXME: I'm not sure what this is. 3505 int64_t FileNumber = -1; 3506 if (getLexer().is(AsmToken::Integer)) { 3507 FileNumber = getTok().getIntVal(); 3508 Lex(); 3509 3510 if (FileNumber < 0) 3511 return TokError("negative file number"); 3512 } 3513 3514 std::string Path; 3515 3516 // Usually the directory and filename together, otherwise just the directory. 3517 // Allow the strings to have escaped octal character sequence. 3518 if (parseEscapedString(Path)) 3519 return true; 3520 3521 StringRef Directory; 3522 StringRef Filename; 3523 std::string FilenameData; 3524 if (getLexer().is(AsmToken::String)) { 3525 if (check(FileNumber == -1, 3526 "explicit path specified, but no file number") || 3527 parseEscapedString(FilenameData)) 3528 return true; 3529 Filename = FilenameData; 3530 Directory = Path; 3531 } else { 3532 Filename = Path; 3533 } 3534 3535 uint64_t MD5Hi, MD5Lo; 3536 bool HasMD5 = false; 3537 3538 std::optional<StringRef> Source; 3539 bool HasSource = false; 3540 std::string SourceString; 3541 3542 while (!parseOptionalToken(AsmToken::EndOfStatement)) { 3543 StringRef Keyword; 3544 if (check(getTok().isNot(AsmToken::Identifier), 3545 "unexpected token in '.file' directive") || 3546 parseIdentifier(Keyword)) 3547 return true; 3548 if (Keyword == "md5") { 3549 HasMD5 = true; 3550 if (check(FileNumber == -1, 3551 "MD5 checksum specified, but no file number") || 3552 parseHexOcta(*this, MD5Hi, MD5Lo)) 3553 return true; 3554 } else if (Keyword == "source") { 3555 HasSource = true; 3556 if (check(FileNumber == -1, 3557 "source specified, but no file number") || 3558 check(getTok().isNot(AsmToken::String), 3559 "unexpected token in '.file' directive") || 3560 parseEscapedString(SourceString)) 3561 return true; 3562 } else { 3563 return TokError("unexpected token in '.file' directive"); 3564 } 3565 } 3566 3567 if (FileNumber == -1) { 3568 // Ignore the directive if there is no number and the target doesn't support 3569 // numberless .file directives. This allows some portability of assembler 3570 // between different object file formats. 3571 if (getContext().getAsmInfo()->hasSingleParameterDotFile()) 3572 getStreamer().emitFileDirective(Filename); 3573 } else { 3574 // In case there is a -g option as well as debug info from directive .file, 3575 // we turn off the -g option, directly use the existing debug info instead. 3576 // Throw away any implicit file table for the assembler source. 3577 if (Ctx.getGenDwarfForAssembly()) { 3578 Ctx.getMCDwarfLineTable(0).resetFileTable(); 3579 Ctx.setGenDwarfForAssembly(false); 3580 } 3581 3582 std::optional<MD5::MD5Result> CKMem; 3583 if (HasMD5) { 3584 MD5::MD5Result Sum; 3585 for (unsigned i = 0; i != 8; ++i) { 3586 Sum[i] = uint8_t(MD5Hi >> ((7 - i) * 8)); 3587 Sum[i + 8] = uint8_t(MD5Lo >> ((7 - i) * 8)); 3588 } 3589 CKMem = Sum; 3590 } 3591 if (HasSource) { 3592 char *SourceBuf = static_cast<char *>(Ctx.allocate(SourceString.size())); 3593 memcpy(SourceBuf, SourceString.data(), SourceString.size()); 3594 Source = StringRef(SourceBuf, SourceString.size()); 3595 } 3596 if (FileNumber == 0) { 3597 // Upgrade to Version 5 for assembly actions like clang -c a.s. 3598 if (Ctx.getDwarfVersion() < 5) 3599 Ctx.setDwarfVersion(5); 3600 getStreamer().emitDwarfFile0Directive(Directory, Filename, CKMem, Source); 3601 } else { 3602 Expected<unsigned> FileNumOrErr = getStreamer().tryEmitDwarfFileDirective( 3603 FileNumber, Directory, Filename, CKMem, Source); 3604 if (!FileNumOrErr) 3605 return Error(DirectiveLoc, toString(FileNumOrErr.takeError())); 3606 } 3607 // Alert the user if there are some .file directives with MD5 and some not. 3608 // But only do that once. 3609 if (!ReportedInconsistentMD5 && !Ctx.isDwarfMD5UsageConsistent(0)) { 3610 ReportedInconsistentMD5 = true; 3611 return Warning(DirectiveLoc, "inconsistent use of MD5 checksums"); 3612 } 3613 } 3614 3615 return false; 3616 } 3617 3618 /// parseDirectiveLine 3619 /// ::= .line [number] 3620 bool AsmParser::parseDirectiveLine() { 3621 int64_t LineNumber; 3622 if (getLexer().is(AsmToken::Integer)) { 3623 if (parseIntToken(LineNumber, "unexpected token in '.line' directive")) 3624 return true; 3625 (void)LineNumber; 3626 // FIXME: Do something with the .line. 3627 } 3628 return parseEOL(); 3629 } 3630 3631 /// parseDirectiveLoc 3632 /// ::= .loc FileNumber [LineNumber] [ColumnPos] [basic_block] [prologue_end] 3633 /// [epilogue_begin] [is_stmt VALUE] [isa VALUE] 3634 /// The first number is a file number, must have been previously assigned with 3635 /// a .file directive, the second number is the line number and optionally the 3636 /// third number is a column position (zero if not specified). The remaining 3637 /// optional items are .loc sub-directives. 3638 bool AsmParser::parseDirectiveLoc() { 3639 int64_t FileNumber = 0, LineNumber = 0; 3640 SMLoc Loc = getTok().getLoc(); 3641 if (parseIntToken(FileNumber, "unexpected token in '.loc' directive") || 3642 check(FileNumber < 1 && Ctx.getDwarfVersion() < 5, Loc, 3643 "file number less than one in '.loc' directive") || 3644 check(!getContext().isValidDwarfFileNumber(FileNumber), Loc, 3645 "unassigned file number in '.loc' directive")) 3646 return true; 3647 3648 // optional 3649 if (getLexer().is(AsmToken::Integer)) { 3650 LineNumber = getTok().getIntVal(); 3651 if (LineNumber < 0) 3652 return TokError("line number less than zero in '.loc' directive"); 3653 Lex(); 3654 } 3655 3656 int64_t ColumnPos = 0; 3657 if (getLexer().is(AsmToken::Integer)) { 3658 ColumnPos = getTok().getIntVal(); 3659 if (ColumnPos < 0) 3660 return TokError("column position less than zero in '.loc' directive"); 3661 Lex(); 3662 } 3663 3664 auto PrevFlags = getContext().getCurrentDwarfLoc().getFlags(); 3665 unsigned Flags = PrevFlags & DWARF2_FLAG_IS_STMT; 3666 unsigned Isa = 0; 3667 int64_t Discriminator = 0; 3668 3669 auto parseLocOp = [&]() -> bool { 3670 StringRef Name; 3671 SMLoc Loc = getTok().getLoc(); 3672 if (parseIdentifier(Name)) 3673 return TokError("unexpected token in '.loc' directive"); 3674 3675 if (Name == "basic_block") 3676 Flags |= DWARF2_FLAG_BASIC_BLOCK; 3677 else if (Name == "prologue_end") 3678 Flags |= DWARF2_FLAG_PROLOGUE_END; 3679 else if (Name == "epilogue_begin") 3680 Flags |= DWARF2_FLAG_EPILOGUE_BEGIN; 3681 else if (Name == "is_stmt") { 3682 Loc = getTok().getLoc(); 3683 const MCExpr *Value; 3684 if (parseExpression(Value)) 3685 return true; 3686 // The expression must be the constant 0 or 1. 3687 if (const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value)) { 3688 int Value = MCE->getValue(); 3689 if (Value == 0) 3690 Flags &= ~DWARF2_FLAG_IS_STMT; 3691 else if (Value == 1) 3692 Flags |= DWARF2_FLAG_IS_STMT; 3693 else 3694 return Error(Loc, "is_stmt value not 0 or 1"); 3695 } else { 3696 return Error(Loc, "is_stmt value not the constant value of 0 or 1"); 3697 } 3698 } else if (Name == "isa") { 3699 Loc = getTok().getLoc(); 3700 const MCExpr *Value; 3701 if (parseExpression(Value)) 3702 return true; 3703 // The expression must be a constant greater or equal to 0. 3704 if (const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value)) { 3705 int Value = MCE->getValue(); 3706 if (Value < 0) 3707 return Error(Loc, "isa number less than zero"); 3708 Isa = Value; 3709 } else { 3710 return Error(Loc, "isa number not a constant value"); 3711 } 3712 } else if (Name == "discriminator") { 3713 if (parseAbsoluteExpression(Discriminator)) 3714 return true; 3715 } else { 3716 return Error(Loc, "unknown sub-directive in '.loc' directive"); 3717 } 3718 return false; 3719 }; 3720 3721 if (parseMany(parseLocOp, false /*hasComma*/)) 3722 return true; 3723 3724 getStreamer().emitDwarfLocDirective(FileNumber, LineNumber, ColumnPos, Flags, 3725 Isa, Discriminator, StringRef()); 3726 3727 return false; 3728 } 3729 3730 /// parseDirectiveStabs 3731 /// ::= .stabs string, number, number, number 3732 bool AsmParser::parseDirectiveStabs() { 3733 return TokError("unsupported directive '.stabs'"); 3734 } 3735 3736 /// parseDirectiveCVFile 3737 /// ::= .cv_file number filename [checksum] [checksumkind] 3738 bool AsmParser::parseDirectiveCVFile() { 3739 SMLoc FileNumberLoc = getTok().getLoc(); 3740 int64_t FileNumber; 3741 std::string Filename; 3742 std::string Checksum; 3743 int64_t ChecksumKind = 0; 3744 3745 if (parseIntToken(FileNumber, 3746 "expected file number in '.cv_file' directive") || 3747 check(FileNumber < 1, FileNumberLoc, "file number less than one") || 3748 check(getTok().isNot(AsmToken::String), 3749 "unexpected token in '.cv_file' directive") || 3750 parseEscapedString(Filename)) 3751 return true; 3752 if (!parseOptionalToken(AsmToken::EndOfStatement)) { 3753 if (check(getTok().isNot(AsmToken::String), 3754 "unexpected token in '.cv_file' directive") || 3755 parseEscapedString(Checksum) || 3756 parseIntToken(ChecksumKind, 3757 "expected checksum kind in '.cv_file' directive") || 3758 parseEOL()) 3759 return true; 3760 } 3761 3762 Checksum = fromHex(Checksum); 3763 void *CKMem = Ctx.allocate(Checksum.size(), 1); 3764 memcpy(CKMem, Checksum.data(), Checksum.size()); 3765 ArrayRef<uint8_t> ChecksumAsBytes(reinterpret_cast<const uint8_t *>(CKMem), 3766 Checksum.size()); 3767 3768 if (!getStreamer().emitCVFileDirective(FileNumber, Filename, ChecksumAsBytes, 3769 static_cast<uint8_t>(ChecksumKind))) 3770 return Error(FileNumberLoc, "file number already allocated"); 3771 3772 return false; 3773 } 3774 3775 bool AsmParser::parseCVFunctionId(int64_t &FunctionId, 3776 StringRef DirectiveName) { 3777 SMLoc Loc; 3778 return parseTokenLoc(Loc) || 3779 parseIntToken(FunctionId, "expected function id in '" + DirectiveName + 3780 "' directive") || 3781 check(FunctionId < 0 || FunctionId >= UINT_MAX, Loc, 3782 "expected function id within range [0, UINT_MAX)"); 3783 } 3784 3785 bool AsmParser::parseCVFileId(int64_t &FileNumber, StringRef DirectiveName) { 3786 SMLoc Loc; 3787 return parseTokenLoc(Loc) || 3788 parseIntToken(FileNumber, "expected integer in '" + DirectiveName + 3789 "' directive") || 3790 check(FileNumber < 1, Loc, "file number less than one in '" + 3791 DirectiveName + "' directive") || 3792 check(!getCVContext().isValidFileNumber(FileNumber), Loc, 3793 "unassigned file number in '" + DirectiveName + "' directive"); 3794 } 3795 3796 /// parseDirectiveCVFuncId 3797 /// ::= .cv_func_id FunctionId 3798 /// 3799 /// Introduces a function ID that can be used with .cv_loc. 3800 bool AsmParser::parseDirectiveCVFuncId() { 3801 SMLoc FunctionIdLoc = getTok().getLoc(); 3802 int64_t FunctionId; 3803 3804 if (parseCVFunctionId(FunctionId, ".cv_func_id") || parseEOL()) 3805 return true; 3806 3807 if (!getStreamer().emitCVFuncIdDirective(FunctionId)) 3808 return Error(FunctionIdLoc, "function id already allocated"); 3809 3810 return false; 3811 } 3812 3813 /// parseDirectiveCVInlineSiteId 3814 /// ::= .cv_inline_site_id FunctionId 3815 /// "within" IAFunc 3816 /// "inlined_at" IAFile IALine [IACol] 3817 /// 3818 /// Introduces a function ID that can be used with .cv_loc. Includes "inlined 3819 /// at" source location information for use in the line table of the caller, 3820 /// whether the caller is a real function or another inlined call site. 3821 bool AsmParser::parseDirectiveCVInlineSiteId() { 3822 SMLoc FunctionIdLoc = getTok().getLoc(); 3823 int64_t FunctionId; 3824 int64_t IAFunc; 3825 int64_t IAFile; 3826 int64_t IALine; 3827 int64_t IACol = 0; 3828 3829 // FunctionId 3830 if (parseCVFunctionId(FunctionId, ".cv_inline_site_id")) 3831 return true; 3832 3833 // "within" 3834 if (check((getLexer().isNot(AsmToken::Identifier) || 3835 getTok().getIdentifier() != "within"), 3836 "expected 'within' identifier in '.cv_inline_site_id' directive")) 3837 return true; 3838 Lex(); 3839 3840 // IAFunc 3841 if (parseCVFunctionId(IAFunc, ".cv_inline_site_id")) 3842 return true; 3843 3844 // "inlined_at" 3845 if (check((getLexer().isNot(AsmToken::Identifier) || 3846 getTok().getIdentifier() != "inlined_at"), 3847 "expected 'inlined_at' identifier in '.cv_inline_site_id' " 3848 "directive") ) 3849 return true; 3850 Lex(); 3851 3852 // IAFile IALine 3853 if (parseCVFileId(IAFile, ".cv_inline_site_id") || 3854 parseIntToken(IALine, "expected line number after 'inlined_at'")) 3855 return true; 3856 3857 // [IACol] 3858 if (getLexer().is(AsmToken::Integer)) { 3859 IACol = getTok().getIntVal(); 3860 Lex(); 3861 } 3862 3863 if (parseEOL()) 3864 return true; 3865 3866 if (!getStreamer().emitCVInlineSiteIdDirective(FunctionId, IAFunc, IAFile, 3867 IALine, IACol, FunctionIdLoc)) 3868 return Error(FunctionIdLoc, "function id already allocated"); 3869 3870 return false; 3871 } 3872 3873 /// parseDirectiveCVLoc 3874 /// ::= .cv_loc FunctionId FileNumber [LineNumber] [ColumnPos] [prologue_end] 3875 /// [is_stmt VALUE] 3876 /// The first number is a file number, must have been previously assigned with 3877 /// a .file directive, the second number is the line number and optionally the 3878 /// third number is a column position (zero if not specified). The remaining 3879 /// optional items are .loc sub-directives. 3880 bool AsmParser::parseDirectiveCVLoc() { 3881 SMLoc DirectiveLoc = getTok().getLoc(); 3882 int64_t FunctionId, FileNumber; 3883 if (parseCVFunctionId(FunctionId, ".cv_loc") || 3884 parseCVFileId(FileNumber, ".cv_loc")) 3885 return true; 3886 3887 int64_t LineNumber = 0; 3888 if (getLexer().is(AsmToken::Integer)) { 3889 LineNumber = getTok().getIntVal(); 3890 if (LineNumber < 0) 3891 return TokError("line number less than zero in '.cv_loc' directive"); 3892 Lex(); 3893 } 3894 3895 int64_t ColumnPos = 0; 3896 if (getLexer().is(AsmToken::Integer)) { 3897 ColumnPos = getTok().getIntVal(); 3898 if (ColumnPos < 0) 3899 return TokError("column position less than zero in '.cv_loc' directive"); 3900 Lex(); 3901 } 3902 3903 bool PrologueEnd = false; 3904 uint64_t IsStmt = 0; 3905 3906 auto parseOp = [&]() -> bool { 3907 StringRef Name; 3908 SMLoc Loc = getTok().getLoc(); 3909 if (parseIdentifier(Name)) 3910 return TokError("unexpected token in '.cv_loc' directive"); 3911 if (Name == "prologue_end") 3912 PrologueEnd = true; 3913 else if (Name == "is_stmt") { 3914 Loc = getTok().getLoc(); 3915 const MCExpr *Value; 3916 if (parseExpression(Value)) 3917 return true; 3918 // The expression must be the constant 0 or 1. 3919 IsStmt = ~0ULL; 3920 if (const auto *MCE = dyn_cast<MCConstantExpr>(Value)) 3921 IsStmt = MCE->getValue(); 3922 3923 if (IsStmt > 1) 3924 return Error(Loc, "is_stmt value not 0 or 1"); 3925 } else { 3926 return Error(Loc, "unknown sub-directive in '.cv_loc' directive"); 3927 } 3928 return false; 3929 }; 3930 3931 if (parseMany(parseOp, false /*hasComma*/)) 3932 return true; 3933 3934 getStreamer().emitCVLocDirective(FunctionId, FileNumber, LineNumber, 3935 ColumnPos, PrologueEnd, IsStmt, StringRef(), 3936 DirectiveLoc); 3937 return false; 3938 } 3939 3940 /// parseDirectiveCVLinetable 3941 /// ::= .cv_linetable FunctionId, FnStart, FnEnd 3942 bool AsmParser::parseDirectiveCVLinetable() { 3943 int64_t FunctionId; 3944 StringRef FnStartName, FnEndName; 3945 SMLoc Loc = getTok().getLoc(); 3946 if (parseCVFunctionId(FunctionId, ".cv_linetable") || parseComma() || 3947 parseTokenLoc(Loc) || 3948 check(parseIdentifier(FnStartName), Loc, 3949 "expected identifier in directive") || 3950 parseComma() || parseTokenLoc(Loc) || 3951 check(parseIdentifier(FnEndName), Loc, 3952 "expected identifier in directive")) 3953 return true; 3954 3955 MCSymbol *FnStartSym = getContext().getOrCreateSymbol(FnStartName); 3956 MCSymbol *FnEndSym = getContext().getOrCreateSymbol(FnEndName); 3957 3958 getStreamer().emitCVLinetableDirective(FunctionId, FnStartSym, FnEndSym); 3959 return false; 3960 } 3961 3962 /// parseDirectiveCVInlineLinetable 3963 /// ::= .cv_inline_linetable PrimaryFunctionId FileId LineNum FnStart FnEnd 3964 bool AsmParser::parseDirectiveCVInlineLinetable() { 3965 int64_t PrimaryFunctionId, SourceFileId, SourceLineNum; 3966 StringRef FnStartName, FnEndName; 3967 SMLoc Loc = getTok().getLoc(); 3968 if (parseCVFunctionId(PrimaryFunctionId, ".cv_inline_linetable") || 3969 parseTokenLoc(Loc) || 3970 parseIntToken( 3971 SourceFileId, 3972 "expected SourceField in '.cv_inline_linetable' directive") || 3973 check(SourceFileId <= 0, Loc, 3974 "File id less than zero in '.cv_inline_linetable' directive") || 3975 parseTokenLoc(Loc) || 3976 parseIntToken( 3977 SourceLineNum, 3978 "expected SourceLineNum in '.cv_inline_linetable' directive") || 3979 check(SourceLineNum < 0, Loc, 3980 "Line number less than zero in '.cv_inline_linetable' directive") || 3981 parseTokenLoc(Loc) || check(parseIdentifier(FnStartName), Loc, 3982 "expected identifier in directive") || 3983 parseTokenLoc(Loc) || check(parseIdentifier(FnEndName), Loc, 3984 "expected identifier in directive")) 3985 return true; 3986 3987 if (parseEOL()) 3988 return true; 3989 3990 MCSymbol *FnStartSym = getContext().getOrCreateSymbol(FnStartName); 3991 MCSymbol *FnEndSym = getContext().getOrCreateSymbol(FnEndName); 3992 getStreamer().emitCVInlineLinetableDirective(PrimaryFunctionId, SourceFileId, 3993 SourceLineNum, FnStartSym, 3994 FnEndSym); 3995 return false; 3996 } 3997 3998 void AsmParser::initializeCVDefRangeTypeMap() { 3999 CVDefRangeTypeMap["reg"] = CVDR_DEFRANGE_REGISTER; 4000 CVDefRangeTypeMap["frame_ptr_rel"] = CVDR_DEFRANGE_FRAMEPOINTER_REL; 4001 CVDefRangeTypeMap["subfield_reg"] = CVDR_DEFRANGE_SUBFIELD_REGISTER; 4002 CVDefRangeTypeMap["reg_rel"] = CVDR_DEFRANGE_REGISTER_REL; 4003 } 4004 4005 /// parseDirectiveCVDefRange 4006 /// ::= .cv_def_range RangeStart RangeEnd (GapStart GapEnd)*, bytes* 4007 bool AsmParser::parseDirectiveCVDefRange() { 4008 SMLoc Loc; 4009 std::vector<std::pair<const MCSymbol *, const MCSymbol *>> Ranges; 4010 while (getLexer().is(AsmToken::Identifier)) { 4011 Loc = getLexer().getLoc(); 4012 StringRef GapStartName; 4013 if (parseIdentifier(GapStartName)) 4014 return Error(Loc, "expected identifier in directive"); 4015 MCSymbol *GapStartSym = getContext().getOrCreateSymbol(GapStartName); 4016 4017 Loc = getLexer().getLoc(); 4018 StringRef GapEndName; 4019 if (parseIdentifier(GapEndName)) 4020 return Error(Loc, "expected identifier in directive"); 4021 MCSymbol *GapEndSym = getContext().getOrCreateSymbol(GapEndName); 4022 4023 Ranges.push_back({GapStartSym, GapEndSym}); 4024 } 4025 4026 StringRef CVDefRangeTypeStr; 4027 if (parseToken( 4028 AsmToken::Comma, 4029 "expected comma before def_range type in .cv_def_range directive") || 4030 parseIdentifier(CVDefRangeTypeStr)) 4031 return Error(Loc, "expected def_range type in directive"); 4032 4033 StringMap<CVDefRangeType>::const_iterator CVTypeIt = 4034 CVDefRangeTypeMap.find(CVDefRangeTypeStr); 4035 CVDefRangeType CVDRType = (CVTypeIt == CVDefRangeTypeMap.end()) 4036 ? CVDR_DEFRANGE 4037 : CVTypeIt->getValue(); 4038 switch (CVDRType) { 4039 case CVDR_DEFRANGE_REGISTER: { 4040 int64_t DRRegister; 4041 if (parseToken(AsmToken::Comma, "expected comma before register number in " 4042 ".cv_def_range directive") || 4043 parseAbsoluteExpression(DRRegister)) 4044 return Error(Loc, "expected register number"); 4045 4046 codeview::DefRangeRegisterHeader DRHdr; 4047 DRHdr.Register = DRRegister; 4048 DRHdr.MayHaveNoName = 0; 4049 getStreamer().emitCVDefRangeDirective(Ranges, DRHdr); 4050 break; 4051 } 4052 case CVDR_DEFRANGE_FRAMEPOINTER_REL: { 4053 int64_t DROffset; 4054 if (parseToken(AsmToken::Comma, 4055 "expected comma before offset in .cv_def_range directive") || 4056 parseAbsoluteExpression(DROffset)) 4057 return Error(Loc, "expected offset value"); 4058 4059 codeview::DefRangeFramePointerRelHeader DRHdr; 4060 DRHdr.Offset = DROffset; 4061 getStreamer().emitCVDefRangeDirective(Ranges, DRHdr); 4062 break; 4063 } 4064 case CVDR_DEFRANGE_SUBFIELD_REGISTER: { 4065 int64_t DRRegister; 4066 int64_t DROffsetInParent; 4067 if (parseToken(AsmToken::Comma, "expected comma before register number in " 4068 ".cv_def_range directive") || 4069 parseAbsoluteExpression(DRRegister)) 4070 return Error(Loc, "expected register number"); 4071 if (parseToken(AsmToken::Comma, 4072 "expected comma before offset in .cv_def_range directive") || 4073 parseAbsoluteExpression(DROffsetInParent)) 4074 return Error(Loc, "expected offset value"); 4075 4076 codeview::DefRangeSubfieldRegisterHeader DRHdr; 4077 DRHdr.Register = DRRegister; 4078 DRHdr.MayHaveNoName = 0; 4079 DRHdr.OffsetInParent = DROffsetInParent; 4080 getStreamer().emitCVDefRangeDirective(Ranges, DRHdr); 4081 break; 4082 } 4083 case CVDR_DEFRANGE_REGISTER_REL: { 4084 int64_t DRRegister; 4085 int64_t DRFlags; 4086 int64_t DRBasePointerOffset; 4087 if (parseToken(AsmToken::Comma, "expected comma before register number in " 4088 ".cv_def_range directive") || 4089 parseAbsoluteExpression(DRRegister)) 4090 return Error(Loc, "expected register value"); 4091 if (parseToken( 4092 AsmToken::Comma, 4093 "expected comma before flag value in .cv_def_range directive") || 4094 parseAbsoluteExpression(DRFlags)) 4095 return Error(Loc, "expected flag value"); 4096 if (parseToken(AsmToken::Comma, "expected comma before base pointer offset " 4097 "in .cv_def_range directive") || 4098 parseAbsoluteExpression(DRBasePointerOffset)) 4099 return Error(Loc, "expected base pointer offset value"); 4100 4101 codeview::DefRangeRegisterRelHeader DRHdr; 4102 DRHdr.Register = DRRegister; 4103 DRHdr.Flags = DRFlags; 4104 DRHdr.BasePointerOffset = DRBasePointerOffset; 4105 getStreamer().emitCVDefRangeDirective(Ranges, DRHdr); 4106 break; 4107 } 4108 default: 4109 return Error(Loc, "unexpected def_range type in .cv_def_range directive"); 4110 } 4111 return true; 4112 } 4113 4114 /// parseDirectiveCVString 4115 /// ::= .cv_stringtable "string" 4116 bool AsmParser::parseDirectiveCVString() { 4117 std::string Data; 4118 if (checkForValidSection() || parseEscapedString(Data)) 4119 return true; 4120 4121 // Put the string in the table and emit the offset. 4122 std::pair<StringRef, unsigned> Insertion = 4123 getCVContext().addToStringTable(Data); 4124 getStreamer().emitInt32(Insertion.second); 4125 return false; 4126 } 4127 4128 /// parseDirectiveCVStringTable 4129 /// ::= .cv_stringtable 4130 bool AsmParser::parseDirectiveCVStringTable() { 4131 getStreamer().emitCVStringTableDirective(); 4132 return false; 4133 } 4134 4135 /// parseDirectiveCVFileChecksums 4136 /// ::= .cv_filechecksums 4137 bool AsmParser::parseDirectiveCVFileChecksums() { 4138 getStreamer().emitCVFileChecksumsDirective(); 4139 return false; 4140 } 4141 4142 /// parseDirectiveCVFileChecksumOffset 4143 /// ::= .cv_filechecksumoffset fileno 4144 bool AsmParser::parseDirectiveCVFileChecksumOffset() { 4145 int64_t FileNo; 4146 if (parseIntToken(FileNo, "expected identifier in directive")) 4147 return true; 4148 if (parseEOL()) 4149 return true; 4150 getStreamer().emitCVFileChecksumOffsetDirective(FileNo); 4151 return false; 4152 } 4153 4154 /// parseDirectiveCVFPOData 4155 /// ::= .cv_fpo_data procsym 4156 bool AsmParser::parseDirectiveCVFPOData() { 4157 SMLoc DirLoc = getLexer().getLoc(); 4158 StringRef ProcName; 4159 if (parseIdentifier(ProcName)) 4160 return TokError("expected symbol name"); 4161 if (parseEOL()) 4162 return true; 4163 MCSymbol *ProcSym = getContext().getOrCreateSymbol(ProcName); 4164 getStreamer().emitCVFPOData(ProcSym, DirLoc); 4165 return false; 4166 } 4167 4168 /// parseDirectiveCFISections 4169 /// ::= .cfi_sections section [, section] 4170 bool AsmParser::parseDirectiveCFISections() { 4171 StringRef Name; 4172 bool EH = false; 4173 bool Debug = false; 4174 4175 if (!parseOptionalToken(AsmToken::EndOfStatement)) { 4176 for (;;) { 4177 if (parseIdentifier(Name)) 4178 return TokError("expected .eh_frame or .debug_frame"); 4179 if (Name == ".eh_frame") 4180 EH = true; 4181 else if (Name == ".debug_frame") 4182 Debug = true; 4183 if (parseOptionalToken(AsmToken::EndOfStatement)) 4184 break; 4185 if (parseComma()) 4186 return true; 4187 } 4188 } 4189 getStreamer().emitCFISections(EH, Debug); 4190 return false; 4191 } 4192 4193 /// parseDirectiveCFIStartProc 4194 /// ::= .cfi_startproc [simple] 4195 bool AsmParser::parseDirectiveCFIStartProc() { 4196 StringRef Simple; 4197 if (!parseOptionalToken(AsmToken::EndOfStatement)) { 4198 if (check(parseIdentifier(Simple) || Simple != "simple", 4199 "unexpected token") || 4200 parseEOL()) 4201 return true; 4202 } 4203 4204 // TODO(kristina): Deal with a corner case of incorrect diagnostic context 4205 // being produced if this directive is emitted as part of preprocessor macro 4206 // expansion which can *ONLY* happen if Clang's cc1as is the API consumer. 4207 // Tools like llvm-mc on the other hand are not affected by it, and report 4208 // correct context information. 4209 getStreamer().emitCFIStartProc(!Simple.empty(), Lexer.getLoc()); 4210 return false; 4211 } 4212 4213 /// parseDirectiveCFIEndProc 4214 /// ::= .cfi_endproc 4215 bool AsmParser::parseDirectiveCFIEndProc() { 4216 if (parseEOL()) 4217 return true; 4218 getStreamer().emitCFIEndProc(); 4219 return false; 4220 } 4221 4222 /// parse register name or number. 4223 bool AsmParser::parseRegisterOrRegisterNumber(int64_t &Register, 4224 SMLoc DirectiveLoc) { 4225 MCRegister RegNo; 4226 4227 if (getLexer().isNot(AsmToken::Integer)) { 4228 if (getTargetParser().parseRegister(RegNo, DirectiveLoc, DirectiveLoc)) 4229 return true; 4230 Register = getContext().getRegisterInfo()->getDwarfRegNum(RegNo, true); 4231 } else 4232 return parseAbsoluteExpression(Register); 4233 4234 return false; 4235 } 4236 4237 /// parseDirectiveCFIDefCfa 4238 /// ::= .cfi_def_cfa register, offset 4239 bool AsmParser::parseDirectiveCFIDefCfa(SMLoc DirectiveLoc) { 4240 int64_t Register = 0, Offset = 0; 4241 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc) || parseComma() || 4242 parseAbsoluteExpression(Offset) || parseEOL()) 4243 return true; 4244 4245 getStreamer().emitCFIDefCfa(Register, Offset); 4246 return false; 4247 } 4248 4249 /// parseDirectiveCFIDefCfaOffset 4250 /// ::= .cfi_def_cfa_offset offset 4251 bool AsmParser::parseDirectiveCFIDefCfaOffset() { 4252 int64_t Offset = 0; 4253 if (parseAbsoluteExpression(Offset) || parseEOL()) 4254 return true; 4255 4256 getStreamer().emitCFIDefCfaOffset(Offset); 4257 return false; 4258 } 4259 4260 /// parseDirectiveCFIRegister 4261 /// ::= .cfi_register register, register 4262 bool AsmParser::parseDirectiveCFIRegister(SMLoc DirectiveLoc) { 4263 int64_t Register1 = 0, Register2 = 0; 4264 if (parseRegisterOrRegisterNumber(Register1, DirectiveLoc) || parseComma() || 4265 parseRegisterOrRegisterNumber(Register2, DirectiveLoc) || parseEOL()) 4266 return true; 4267 4268 getStreamer().emitCFIRegister(Register1, Register2); 4269 return false; 4270 } 4271 4272 /// parseDirectiveCFIWindowSave 4273 /// ::= .cfi_window_save 4274 bool AsmParser::parseDirectiveCFIWindowSave() { 4275 if (parseEOL()) 4276 return true; 4277 getStreamer().emitCFIWindowSave(); 4278 return false; 4279 } 4280 4281 /// parseDirectiveCFIAdjustCfaOffset 4282 /// ::= .cfi_adjust_cfa_offset adjustment 4283 bool AsmParser::parseDirectiveCFIAdjustCfaOffset() { 4284 int64_t Adjustment = 0; 4285 if (parseAbsoluteExpression(Adjustment) || parseEOL()) 4286 return true; 4287 4288 getStreamer().emitCFIAdjustCfaOffset(Adjustment); 4289 return false; 4290 } 4291 4292 /// parseDirectiveCFIDefCfaRegister 4293 /// ::= .cfi_def_cfa_register register 4294 bool AsmParser::parseDirectiveCFIDefCfaRegister(SMLoc DirectiveLoc) { 4295 int64_t Register = 0; 4296 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc) || parseEOL()) 4297 return true; 4298 4299 getStreamer().emitCFIDefCfaRegister(Register); 4300 return false; 4301 } 4302 4303 /// parseDirectiveCFILLVMDefAspaceCfa 4304 /// ::= .cfi_llvm_def_aspace_cfa register, offset, address_space 4305 bool AsmParser::parseDirectiveCFILLVMDefAspaceCfa(SMLoc DirectiveLoc) { 4306 int64_t Register = 0, Offset = 0, AddressSpace = 0; 4307 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc) || parseComma() || 4308 parseAbsoluteExpression(Offset) || parseComma() || 4309 parseAbsoluteExpression(AddressSpace) || parseEOL()) 4310 return true; 4311 4312 getStreamer().emitCFILLVMDefAspaceCfa(Register, Offset, AddressSpace); 4313 return false; 4314 } 4315 4316 /// parseDirectiveCFIOffset 4317 /// ::= .cfi_offset register, offset 4318 bool AsmParser::parseDirectiveCFIOffset(SMLoc DirectiveLoc) { 4319 int64_t Register = 0; 4320 int64_t Offset = 0; 4321 4322 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc) || parseComma() || 4323 parseAbsoluteExpression(Offset) || parseEOL()) 4324 return true; 4325 4326 getStreamer().emitCFIOffset(Register, Offset); 4327 return false; 4328 } 4329 4330 /// parseDirectiveCFIRelOffset 4331 /// ::= .cfi_rel_offset register, offset 4332 bool AsmParser::parseDirectiveCFIRelOffset(SMLoc DirectiveLoc) { 4333 int64_t Register = 0, Offset = 0; 4334 4335 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc) || parseComma() || 4336 parseAbsoluteExpression(Offset) || parseEOL()) 4337 return true; 4338 4339 getStreamer().emitCFIRelOffset(Register, Offset); 4340 return false; 4341 } 4342 4343 static bool isValidEncoding(int64_t Encoding) { 4344 if (Encoding & ~0xff) 4345 return false; 4346 4347 if (Encoding == dwarf::DW_EH_PE_omit) 4348 return true; 4349 4350 const unsigned Format = Encoding & 0xf; 4351 if (Format != dwarf::DW_EH_PE_absptr && Format != dwarf::DW_EH_PE_udata2 && 4352 Format != dwarf::DW_EH_PE_udata4 && Format != dwarf::DW_EH_PE_udata8 && 4353 Format != dwarf::DW_EH_PE_sdata2 && Format != dwarf::DW_EH_PE_sdata4 && 4354 Format != dwarf::DW_EH_PE_sdata8 && Format != dwarf::DW_EH_PE_signed) 4355 return false; 4356 4357 const unsigned Application = Encoding & 0x70; 4358 if (Application != dwarf::DW_EH_PE_absptr && 4359 Application != dwarf::DW_EH_PE_pcrel) 4360 return false; 4361 4362 return true; 4363 } 4364 4365 /// parseDirectiveCFIPersonalityOrLsda 4366 /// IsPersonality true for cfi_personality, false for cfi_lsda 4367 /// ::= .cfi_personality encoding, [symbol_name] 4368 /// ::= .cfi_lsda encoding, [symbol_name] 4369 bool AsmParser::parseDirectiveCFIPersonalityOrLsda(bool IsPersonality) { 4370 int64_t Encoding = 0; 4371 if (parseAbsoluteExpression(Encoding)) 4372 return true; 4373 if (Encoding == dwarf::DW_EH_PE_omit) 4374 return false; 4375 4376 StringRef Name; 4377 if (check(!isValidEncoding(Encoding), "unsupported encoding.") || 4378 parseComma() || 4379 check(parseIdentifier(Name), "expected identifier in directive") || 4380 parseEOL()) 4381 return true; 4382 4383 MCSymbol *Sym = getContext().getOrCreateSymbol(Name); 4384 4385 if (IsPersonality) 4386 getStreamer().emitCFIPersonality(Sym, Encoding); 4387 else 4388 getStreamer().emitCFILsda(Sym, Encoding); 4389 return false; 4390 } 4391 4392 /// parseDirectiveCFIRememberState 4393 /// ::= .cfi_remember_state 4394 bool AsmParser::parseDirectiveCFIRememberState() { 4395 if (parseEOL()) 4396 return true; 4397 getStreamer().emitCFIRememberState(); 4398 return false; 4399 } 4400 4401 /// parseDirectiveCFIRestoreState 4402 /// ::= .cfi_remember_state 4403 bool AsmParser::parseDirectiveCFIRestoreState() { 4404 if (parseEOL()) 4405 return true; 4406 getStreamer().emitCFIRestoreState(); 4407 return false; 4408 } 4409 4410 /// parseDirectiveCFISameValue 4411 /// ::= .cfi_same_value register 4412 bool AsmParser::parseDirectiveCFISameValue(SMLoc DirectiveLoc) { 4413 int64_t Register = 0; 4414 4415 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc) || parseEOL()) 4416 return true; 4417 4418 getStreamer().emitCFISameValue(Register); 4419 return false; 4420 } 4421 4422 /// parseDirectiveCFIRestore 4423 /// ::= .cfi_restore register 4424 bool AsmParser::parseDirectiveCFIRestore(SMLoc DirectiveLoc) { 4425 int64_t Register = 0; 4426 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc) || parseEOL()) 4427 return true; 4428 4429 getStreamer().emitCFIRestore(Register); 4430 return false; 4431 } 4432 4433 /// parseDirectiveCFIEscape 4434 /// ::= .cfi_escape expression[,...] 4435 bool AsmParser::parseDirectiveCFIEscape() { 4436 std::string Values; 4437 int64_t CurrValue; 4438 if (parseAbsoluteExpression(CurrValue)) 4439 return true; 4440 4441 Values.push_back((uint8_t)CurrValue); 4442 4443 while (getLexer().is(AsmToken::Comma)) { 4444 Lex(); 4445 4446 if (parseAbsoluteExpression(CurrValue)) 4447 return true; 4448 4449 Values.push_back((uint8_t)CurrValue); 4450 } 4451 4452 getStreamer().emitCFIEscape(Values); 4453 return false; 4454 } 4455 4456 /// parseDirectiveCFIReturnColumn 4457 /// ::= .cfi_return_column register 4458 bool AsmParser::parseDirectiveCFIReturnColumn(SMLoc DirectiveLoc) { 4459 int64_t Register = 0; 4460 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc) || parseEOL()) 4461 return true; 4462 getStreamer().emitCFIReturnColumn(Register); 4463 return false; 4464 } 4465 4466 /// parseDirectiveCFISignalFrame 4467 /// ::= .cfi_signal_frame 4468 bool AsmParser::parseDirectiveCFISignalFrame() { 4469 if (parseEOL()) 4470 return true; 4471 4472 getStreamer().emitCFISignalFrame(); 4473 return false; 4474 } 4475 4476 /// parseDirectiveCFIUndefined 4477 /// ::= .cfi_undefined register 4478 bool AsmParser::parseDirectiveCFIUndefined(SMLoc DirectiveLoc) { 4479 int64_t Register = 0; 4480 4481 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc) || parseEOL()) 4482 return true; 4483 4484 getStreamer().emitCFIUndefined(Register); 4485 return false; 4486 } 4487 4488 /// parseDirectiveAltmacro 4489 /// ::= .altmacro 4490 /// ::= .noaltmacro 4491 bool AsmParser::parseDirectiveAltmacro(StringRef Directive) { 4492 if (parseEOL()) 4493 return true; 4494 AltMacroMode = (Directive == ".altmacro"); 4495 return false; 4496 } 4497 4498 /// parseDirectiveMacrosOnOff 4499 /// ::= .macros_on 4500 /// ::= .macros_off 4501 bool AsmParser::parseDirectiveMacrosOnOff(StringRef Directive) { 4502 if (parseEOL()) 4503 return true; 4504 setMacrosEnabled(Directive == ".macros_on"); 4505 return false; 4506 } 4507 4508 /// parseDirectiveMacro 4509 /// ::= .macro name[,] [parameters] 4510 bool AsmParser::parseDirectiveMacro(SMLoc DirectiveLoc) { 4511 StringRef Name; 4512 if (parseIdentifier(Name)) 4513 return TokError("expected identifier in '.macro' directive"); 4514 4515 if (getLexer().is(AsmToken::Comma)) 4516 Lex(); 4517 4518 MCAsmMacroParameters Parameters; 4519 while (getLexer().isNot(AsmToken::EndOfStatement)) { 4520 4521 if (!Parameters.empty() && Parameters.back().Vararg) 4522 return Error(Lexer.getLoc(), "vararg parameter '" + 4523 Parameters.back().Name + 4524 "' should be the last parameter"); 4525 4526 MCAsmMacroParameter Parameter; 4527 if (parseIdentifier(Parameter.Name)) 4528 return TokError("expected identifier in '.macro' directive"); 4529 4530 // Emit an error if two (or more) named parameters share the same name 4531 for (const MCAsmMacroParameter& CurrParam : Parameters) 4532 if (CurrParam.Name.equals(Parameter.Name)) 4533 return TokError("macro '" + Name + "' has multiple parameters" 4534 " named '" + Parameter.Name + "'"); 4535 4536 if (Lexer.is(AsmToken::Colon)) { 4537 Lex(); // consume ':' 4538 4539 SMLoc QualLoc; 4540 StringRef Qualifier; 4541 4542 QualLoc = Lexer.getLoc(); 4543 if (parseIdentifier(Qualifier)) 4544 return Error(QualLoc, "missing parameter qualifier for " 4545 "'" + Parameter.Name + "' in macro '" + Name + "'"); 4546 4547 if (Qualifier == "req") 4548 Parameter.Required = true; 4549 else if (Qualifier == "vararg") 4550 Parameter.Vararg = true; 4551 else 4552 return Error(QualLoc, Qualifier + " is not a valid parameter qualifier " 4553 "for '" + Parameter.Name + "' in macro '" + Name + "'"); 4554 } 4555 4556 if (getLexer().is(AsmToken::Equal)) { 4557 Lex(); 4558 4559 SMLoc ParamLoc; 4560 4561 ParamLoc = Lexer.getLoc(); 4562 if (parseMacroArgument(Parameter.Value, /*Vararg=*/false )) 4563 return true; 4564 4565 if (Parameter.Required) 4566 Warning(ParamLoc, "pointless default value for required parameter " 4567 "'" + Parameter.Name + "' in macro '" + Name + "'"); 4568 } 4569 4570 Parameters.push_back(std::move(Parameter)); 4571 4572 if (getLexer().is(AsmToken::Comma)) 4573 Lex(); 4574 } 4575 4576 // Eat just the end of statement. 4577 Lexer.Lex(); 4578 4579 // Consuming deferred text, so use Lexer.Lex to ignore Lexing Errors 4580 AsmToken EndToken, StartToken = getTok(); 4581 unsigned MacroDepth = 0; 4582 // Lex the macro definition. 4583 while (true) { 4584 // Ignore Lexing errors in macros. 4585 while (Lexer.is(AsmToken::Error)) { 4586 Lexer.Lex(); 4587 } 4588 4589 // Check whether we have reached the end of the file. 4590 if (getLexer().is(AsmToken::Eof)) 4591 return Error(DirectiveLoc, "no matching '.endmacro' in definition"); 4592 4593 // Otherwise, check whether we have reach the .endmacro or the start of a 4594 // preprocessor line marker. 4595 if (getLexer().is(AsmToken::Identifier)) { 4596 if (getTok().getIdentifier() == ".endm" || 4597 getTok().getIdentifier() == ".endmacro") { 4598 if (MacroDepth == 0) { // Outermost macro. 4599 EndToken = getTok(); 4600 Lexer.Lex(); 4601 if (getLexer().isNot(AsmToken::EndOfStatement)) 4602 return TokError("unexpected token in '" + EndToken.getIdentifier() + 4603 "' directive"); 4604 break; 4605 } else { 4606 // Otherwise we just found the end of an inner macro. 4607 --MacroDepth; 4608 } 4609 } else if (getTok().getIdentifier() == ".macro") { 4610 // We allow nested macros. Those aren't instantiated until the outermost 4611 // macro is expanded so just ignore them for now. 4612 ++MacroDepth; 4613 } 4614 } else if (Lexer.is(AsmToken::HashDirective)) { 4615 (void)parseCppHashLineFilenameComment(getLexer().getLoc()); 4616 } 4617 4618 // Otherwise, scan til the end of the statement. 4619 eatToEndOfStatement(); 4620 } 4621 4622 if (getContext().lookupMacro(Name)) { 4623 return Error(DirectiveLoc, "macro '" + Name + "' is already defined"); 4624 } 4625 4626 const char *BodyStart = StartToken.getLoc().getPointer(); 4627 const char *BodyEnd = EndToken.getLoc().getPointer(); 4628 StringRef Body = StringRef(BodyStart, BodyEnd - BodyStart); 4629 checkForBadMacro(DirectiveLoc, Name, Body, Parameters); 4630 MCAsmMacro Macro(Name, Body, std::move(Parameters)); 4631 DEBUG_WITH_TYPE("asm-macros", dbgs() << "Defining new macro:\n"; 4632 Macro.dump()); 4633 getContext().defineMacro(Name, std::move(Macro)); 4634 return false; 4635 } 4636 4637 /// checkForBadMacro 4638 /// 4639 /// With the support added for named parameters there may be code out there that 4640 /// is transitioning from positional parameters. In versions of gas that did 4641 /// not support named parameters they would be ignored on the macro definition. 4642 /// But to support both styles of parameters this is not possible so if a macro 4643 /// definition has named parameters but does not use them and has what appears 4644 /// to be positional parameters, strings like $1, $2, ... and $n, then issue a 4645 /// warning that the positional parameter found in body which have no effect. 4646 /// Hoping the developer will either remove the named parameters from the macro 4647 /// definition so the positional parameters get used if that was what was 4648 /// intended or change the macro to use the named parameters. It is possible 4649 /// this warning will trigger when the none of the named parameters are used 4650 /// and the strings like $1 are infact to simply to be passed trough unchanged. 4651 void AsmParser::checkForBadMacro(SMLoc DirectiveLoc, StringRef Name, 4652 StringRef Body, 4653 ArrayRef<MCAsmMacroParameter> Parameters) { 4654 // If this macro is not defined with named parameters the warning we are 4655 // checking for here doesn't apply. 4656 unsigned NParameters = Parameters.size(); 4657 if (NParameters == 0) 4658 return; 4659 4660 bool NamedParametersFound = false; 4661 bool PositionalParametersFound = false; 4662 4663 // Look at the body of the macro for use of both the named parameters and what 4664 // are likely to be positional parameters. This is what expandMacro() is 4665 // doing when it finds the parameters in the body. 4666 while (!Body.empty()) { 4667 // Scan for the next possible parameter. 4668 std::size_t End = Body.size(), Pos = 0; 4669 for (; Pos != End; ++Pos) { 4670 // Check for a substitution or escape. 4671 // This macro is defined with parameters, look for \foo, \bar, etc. 4672 if (Body[Pos] == '\\' && Pos + 1 != End) 4673 break; 4674 4675 // This macro should have parameters, but look for $0, $1, ..., $n too. 4676 if (Body[Pos] != '$' || Pos + 1 == End) 4677 continue; 4678 char Next = Body[Pos + 1]; 4679 if (Next == '$' || Next == 'n' || 4680 isdigit(static_cast<unsigned char>(Next))) 4681 break; 4682 } 4683 4684 // Check if we reached the end. 4685 if (Pos == End) 4686 break; 4687 4688 if (Body[Pos] == '$') { 4689 switch (Body[Pos + 1]) { 4690 // $$ => $ 4691 case '$': 4692 break; 4693 4694 // $n => number of arguments 4695 case 'n': 4696 PositionalParametersFound = true; 4697 break; 4698 4699 // $[0-9] => argument 4700 default: { 4701 PositionalParametersFound = true; 4702 break; 4703 } 4704 } 4705 Pos += 2; 4706 } else { 4707 unsigned I = Pos + 1; 4708 while (isIdentifierChar(Body[I]) && I + 1 != End) 4709 ++I; 4710 4711 const char *Begin = Body.data() + Pos + 1; 4712 StringRef Argument(Begin, I - (Pos + 1)); 4713 unsigned Index = 0; 4714 for (; Index < NParameters; ++Index) 4715 if (Parameters[Index].Name == Argument) 4716 break; 4717 4718 if (Index == NParameters) { 4719 if (Body[Pos + 1] == '(' && Body[Pos + 2] == ')') 4720 Pos += 3; 4721 else { 4722 Pos = I; 4723 } 4724 } else { 4725 NamedParametersFound = true; 4726 Pos += 1 + Argument.size(); 4727 } 4728 } 4729 // Update the scan point. 4730 Body = Body.substr(Pos); 4731 } 4732 4733 if (!NamedParametersFound && PositionalParametersFound) 4734 Warning(DirectiveLoc, "macro defined with named parameters which are not " 4735 "used in macro body, possible positional parameter " 4736 "found in body which will have no effect"); 4737 } 4738 4739 /// parseDirectiveExitMacro 4740 /// ::= .exitm 4741 bool AsmParser::parseDirectiveExitMacro(StringRef Directive) { 4742 if (parseEOL()) 4743 return true; 4744 4745 if (!isInsideMacroInstantiation()) 4746 return TokError("unexpected '" + Directive + "' in file, " 4747 "no current macro definition"); 4748 4749 // Exit all conditionals that are active in the current macro. 4750 while (TheCondStack.size() != ActiveMacros.back()->CondStackDepth) { 4751 TheCondState = TheCondStack.back(); 4752 TheCondStack.pop_back(); 4753 } 4754 4755 handleMacroExit(); 4756 return false; 4757 } 4758 4759 /// parseDirectiveEndMacro 4760 /// ::= .endm 4761 /// ::= .endmacro 4762 bool AsmParser::parseDirectiveEndMacro(StringRef Directive) { 4763 if (getLexer().isNot(AsmToken::EndOfStatement)) 4764 return TokError("unexpected token in '" + Directive + "' directive"); 4765 4766 // If we are inside a macro instantiation, terminate the current 4767 // instantiation. 4768 if (isInsideMacroInstantiation()) { 4769 handleMacroExit(); 4770 return false; 4771 } 4772 4773 // Otherwise, this .endmacro is a stray entry in the file; well formed 4774 // .endmacro directives are handled during the macro definition parsing. 4775 return TokError("unexpected '" + Directive + "' in file, " 4776 "no current macro definition"); 4777 } 4778 4779 /// parseDirectivePurgeMacro 4780 /// ::= .purgem name 4781 bool AsmParser::parseDirectivePurgeMacro(SMLoc DirectiveLoc) { 4782 StringRef Name; 4783 SMLoc Loc; 4784 if (parseTokenLoc(Loc) || 4785 check(parseIdentifier(Name), Loc, 4786 "expected identifier in '.purgem' directive") || 4787 parseEOL()) 4788 return true; 4789 4790 if (!getContext().lookupMacro(Name)) 4791 return Error(DirectiveLoc, "macro '" + Name + "' is not defined"); 4792 4793 getContext().undefineMacro(Name); 4794 DEBUG_WITH_TYPE("asm-macros", dbgs() 4795 << "Un-defining macro: " << Name << "\n"); 4796 return false; 4797 } 4798 4799 /// parseDirectiveBundleAlignMode 4800 /// ::= {.bundle_align_mode} expression 4801 bool AsmParser::parseDirectiveBundleAlignMode() { 4802 // Expect a single argument: an expression that evaluates to a constant 4803 // in the inclusive range 0-30. 4804 SMLoc ExprLoc = getLexer().getLoc(); 4805 int64_t AlignSizePow2; 4806 if (checkForValidSection() || parseAbsoluteExpression(AlignSizePow2) || 4807 parseEOL() || 4808 check(AlignSizePow2 < 0 || AlignSizePow2 > 30, ExprLoc, 4809 "invalid bundle alignment size (expected between 0 and 30)")) 4810 return true; 4811 4812 getStreamer().emitBundleAlignMode(Align(1ULL << AlignSizePow2)); 4813 return false; 4814 } 4815 4816 /// parseDirectiveBundleLock 4817 /// ::= {.bundle_lock} [align_to_end] 4818 bool AsmParser::parseDirectiveBundleLock() { 4819 if (checkForValidSection()) 4820 return true; 4821 bool AlignToEnd = false; 4822 4823 StringRef Option; 4824 SMLoc Loc = getTok().getLoc(); 4825 const char *kInvalidOptionError = 4826 "invalid option for '.bundle_lock' directive"; 4827 4828 if (!parseOptionalToken(AsmToken::EndOfStatement)) { 4829 if (check(parseIdentifier(Option), Loc, kInvalidOptionError) || 4830 check(Option != "align_to_end", Loc, kInvalidOptionError) || parseEOL()) 4831 return true; 4832 AlignToEnd = true; 4833 } 4834 4835 getStreamer().emitBundleLock(AlignToEnd); 4836 return false; 4837 } 4838 4839 /// parseDirectiveBundleLock 4840 /// ::= {.bundle_lock} 4841 bool AsmParser::parseDirectiveBundleUnlock() { 4842 if (checkForValidSection() || parseEOL()) 4843 return true; 4844 4845 getStreamer().emitBundleUnlock(); 4846 return false; 4847 } 4848 4849 /// parseDirectiveSpace 4850 /// ::= (.skip | .space) expression [ , expression ] 4851 bool AsmParser::parseDirectiveSpace(StringRef IDVal) { 4852 SMLoc NumBytesLoc = Lexer.getLoc(); 4853 const MCExpr *NumBytes; 4854 if (checkForValidSection() || parseExpression(NumBytes)) 4855 return true; 4856 4857 int64_t FillExpr = 0; 4858 if (parseOptionalToken(AsmToken::Comma)) 4859 if (parseAbsoluteExpression(FillExpr)) 4860 return true; 4861 if (parseEOL()) 4862 return true; 4863 4864 // FIXME: Sometimes the fill expr is 'nop' if it isn't supplied, instead of 0. 4865 getStreamer().emitFill(*NumBytes, FillExpr, NumBytesLoc); 4866 4867 return false; 4868 } 4869 4870 /// parseDirectiveDCB 4871 /// ::= .dcb.{b, l, w} expression, expression 4872 bool AsmParser::parseDirectiveDCB(StringRef IDVal, unsigned Size) { 4873 SMLoc NumValuesLoc = Lexer.getLoc(); 4874 int64_t NumValues; 4875 if (checkForValidSection() || parseAbsoluteExpression(NumValues)) 4876 return true; 4877 4878 if (NumValues < 0) { 4879 Warning(NumValuesLoc, "'" + Twine(IDVal) + "' directive with negative repeat count has no effect"); 4880 return false; 4881 } 4882 4883 if (parseComma()) 4884 return true; 4885 4886 const MCExpr *Value; 4887 SMLoc ExprLoc = getLexer().getLoc(); 4888 if (parseExpression(Value)) 4889 return true; 4890 4891 // Special case constant expressions to match code generator. 4892 if (const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value)) { 4893 assert(Size <= 8 && "Invalid size"); 4894 uint64_t IntValue = MCE->getValue(); 4895 if (!isUIntN(8 * Size, IntValue) && !isIntN(8 * Size, IntValue)) 4896 return Error(ExprLoc, "literal value out of range for directive"); 4897 for (uint64_t i = 0, e = NumValues; i != e; ++i) 4898 getStreamer().emitIntValue(IntValue, Size); 4899 } else { 4900 for (uint64_t i = 0, e = NumValues; i != e; ++i) 4901 getStreamer().emitValue(Value, Size, ExprLoc); 4902 } 4903 4904 return parseEOL(); 4905 } 4906 4907 /// parseDirectiveRealDCB 4908 /// ::= .dcb.{d, s} expression, expression 4909 bool AsmParser::parseDirectiveRealDCB(StringRef IDVal, const fltSemantics &Semantics) { 4910 SMLoc NumValuesLoc = Lexer.getLoc(); 4911 int64_t NumValues; 4912 if (checkForValidSection() || parseAbsoluteExpression(NumValues)) 4913 return true; 4914 4915 if (NumValues < 0) { 4916 Warning(NumValuesLoc, "'" + Twine(IDVal) + "' directive with negative repeat count has no effect"); 4917 return false; 4918 } 4919 4920 if (parseComma()) 4921 return true; 4922 4923 APInt AsInt; 4924 if (parseRealValue(Semantics, AsInt) || parseEOL()) 4925 return true; 4926 4927 for (uint64_t i = 0, e = NumValues; i != e; ++i) 4928 getStreamer().emitIntValue(AsInt.getLimitedValue(), 4929 AsInt.getBitWidth() / 8); 4930 4931 return false; 4932 } 4933 4934 /// parseDirectiveDS 4935 /// ::= .ds.{b, d, l, p, s, w, x} expression 4936 bool AsmParser::parseDirectiveDS(StringRef IDVal, unsigned Size) { 4937 SMLoc NumValuesLoc = Lexer.getLoc(); 4938 int64_t NumValues; 4939 if (checkForValidSection() || parseAbsoluteExpression(NumValues) || 4940 parseEOL()) 4941 return true; 4942 4943 if (NumValues < 0) { 4944 Warning(NumValuesLoc, "'" + Twine(IDVal) + "' directive with negative repeat count has no effect"); 4945 return false; 4946 } 4947 4948 for (uint64_t i = 0, e = NumValues; i != e; ++i) 4949 getStreamer().emitFill(Size, 0); 4950 4951 return false; 4952 } 4953 4954 /// parseDirectiveLEB128 4955 /// ::= (.sleb128 | .uleb128) [ expression (, expression)* ] 4956 bool AsmParser::parseDirectiveLEB128(bool Signed) { 4957 if (checkForValidSection()) 4958 return true; 4959 4960 auto parseOp = [&]() -> bool { 4961 const MCExpr *Value; 4962 if (parseExpression(Value)) 4963 return true; 4964 if (Signed) 4965 getStreamer().emitSLEB128Value(Value); 4966 else 4967 getStreamer().emitULEB128Value(Value); 4968 return false; 4969 }; 4970 4971 return parseMany(parseOp); 4972 } 4973 4974 /// parseDirectiveSymbolAttribute 4975 /// ::= { ".globl", ".weak", ... } [ identifier ( , identifier )* ] 4976 bool AsmParser::parseDirectiveSymbolAttribute(MCSymbolAttr Attr) { 4977 auto parseOp = [&]() -> bool { 4978 StringRef Name; 4979 SMLoc Loc = getTok().getLoc(); 4980 if (parseIdentifier(Name)) 4981 return Error(Loc, "expected identifier"); 4982 4983 if (discardLTOSymbol(Name)) 4984 return false; 4985 4986 MCSymbol *Sym = getContext().getOrCreateSymbol(Name); 4987 4988 // Assembler local symbols don't make any sense here, except for directives 4989 // that the symbol should be tagged. 4990 if (Sym->isTemporary() && Attr != MCSA_Memtag) 4991 return Error(Loc, "non-local symbol required"); 4992 4993 if (!getStreamer().emitSymbolAttribute(Sym, Attr)) 4994 return Error(Loc, "unable to emit symbol attribute"); 4995 return false; 4996 }; 4997 4998 return parseMany(parseOp); 4999 } 5000 5001 /// parseDirectiveComm 5002 /// ::= ( .comm | .lcomm ) identifier , size_expression [ , align_expression ] 5003 bool AsmParser::parseDirectiveComm(bool IsLocal) { 5004 if (checkForValidSection()) 5005 return true; 5006 5007 SMLoc IDLoc = getLexer().getLoc(); 5008 StringRef Name; 5009 if (parseIdentifier(Name)) 5010 return TokError("expected identifier in directive"); 5011 5012 // Handle the identifier as the key symbol. 5013 MCSymbol *Sym = getContext().getOrCreateSymbol(Name); 5014 5015 if (parseComma()) 5016 return true; 5017 5018 int64_t Size; 5019 SMLoc SizeLoc = getLexer().getLoc(); 5020 if (parseAbsoluteExpression(Size)) 5021 return true; 5022 5023 int64_t Pow2Alignment = 0; 5024 SMLoc Pow2AlignmentLoc; 5025 if (getLexer().is(AsmToken::Comma)) { 5026 Lex(); 5027 Pow2AlignmentLoc = getLexer().getLoc(); 5028 if (parseAbsoluteExpression(Pow2Alignment)) 5029 return true; 5030 5031 LCOMM::LCOMMType LCOMM = Lexer.getMAI().getLCOMMDirectiveAlignmentType(); 5032 if (IsLocal && LCOMM == LCOMM::NoAlignment) 5033 return Error(Pow2AlignmentLoc, "alignment not supported on this target"); 5034 5035 // If this target takes alignments in bytes (not log) validate and convert. 5036 if ((!IsLocal && Lexer.getMAI().getCOMMDirectiveAlignmentIsInBytes()) || 5037 (IsLocal && LCOMM == LCOMM::ByteAlignment)) { 5038 if (!isPowerOf2_64(Pow2Alignment)) 5039 return Error(Pow2AlignmentLoc, "alignment must be a power of 2"); 5040 Pow2Alignment = Log2_64(Pow2Alignment); 5041 } 5042 } 5043 5044 if (parseEOL()) 5045 return true; 5046 5047 // NOTE: a size of zero for a .comm should create a undefined symbol 5048 // but a size of .lcomm creates a bss symbol of size zero. 5049 if (Size < 0) 5050 return Error(SizeLoc, "size must be non-negative"); 5051 5052 Sym->redefineIfPossible(); 5053 if (!Sym->isUndefined()) 5054 return Error(IDLoc, "invalid symbol redefinition"); 5055 5056 // Create the Symbol as a common or local common with Size and Pow2Alignment 5057 if (IsLocal) { 5058 getStreamer().emitLocalCommonSymbol(Sym, Size, 5059 Align(1ULL << Pow2Alignment)); 5060 return false; 5061 } 5062 5063 getStreamer().emitCommonSymbol(Sym, Size, Align(1ULL << Pow2Alignment)); 5064 return false; 5065 } 5066 5067 /// parseDirectiveAbort 5068 /// ::= .abort [... message ...] 5069 bool AsmParser::parseDirectiveAbort() { 5070 // FIXME: Use loc from directive. 5071 SMLoc Loc = getLexer().getLoc(); 5072 5073 StringRef Str = parseStringToEndOfStatement(); 5074 if (parseEOL()) 5075 return true; 5076 5077 if (Str.empty()) 5078 return Error(Loc, ".abort detected. Assembly stopping."); 5079 else 5080 return Error(Loc, ".abort '" + Str + "' detected. Assembly stopping."); 5081 // FIXME: Actually abort assembly here. 5082 5083 return false; 5084 } 5085 5086 /// parseDirectiveInclude 5087 /// ::= .include "filename" 5088 bool AsmParser::parseDirectiveInclude() { 5089 // Allow the strings to have escaped octal character sequence. 5090 std::string Filename; 5091 SMLoc IncludeLoc = getTok().getLoc(); 5092 5093 if (check(getTok().isNot(AsmToken::String), 5094 "expected string in '.include' directive") || 5095 parseEscapedString(Filename) || 5096 check(getTok().isNot(AsmToken::EndOfStatement), 5097 "unexpected token in '.include' directive") || 5098 // Attempt to switch the lexer to the included file before consuming the 5099 // end of statement to avoid losing it when we switch. 5100 check(enterIncludeFile(Filename), IncludeLoc, 5101 "Could not find include file '" + Filename + "'")) 5102 return true; 5103 5104 return false; 5105 } 5106 5107 /// parseDirectiveIncbin 5108 /// ::= .incbin "filename" [ , skip [ , count ] ] 5109 bool AsmParser::parseDirectiveIncbin() { 5110 // Allow the strings to have escaped octal character sequence. 5111 std::string Filename; 5112 SMLoc IncbinLoc = getTok().getLoc(); 5113 if (check(getTok().isNot(AsmToken::String), 5114 "expected string in '.incbin' directive") || 5115 parseEscapedString(Filename)) 5116 return true; 5117 5118 int64_t Skip = 0; 5119 const MCExpr *Count = nullptr; 5120 SMLoc SkipLoc, CountLoc; 5121 if (parseOptionalToken(AsmToken::Comma)) { 5122 // The skip expression can be omitted while specifying the count, e.g: 5123 // .incbin "filename",,4 5124 if (getTok().isNot(AsmToken::Comma)) { 5125 if (parseTokenLoc(SkipLoc) || parseAbsoluteExpression(Skip)) 5126 return true; 5127 } 5128 if (parseOptionalToken(AsmToken::Comma)) { 5129 CountLoc = getTok().getLoc(); 5130 if (parseExpression(Count)) 5131 return true; 5132 } 5133 } 5134 5135 if (parseEOL()) 5136 return true; 5137 5138 if (check(Skip < 0, SkipLoc, "skip is negative")) 5139 return true; 5140 5141 // Attempt to process the included file. 5142 if (processIncbinFile(Filename, Skip, Count, CountLoc)) 5143 return Error(IncbinLoc, "Could not find incbin file '" + Filename + "'"); 5144 return false; 5145 } 5146 5147 /// parseDirectiveIf 5148 /// ::= .if{,eq,ge,gt,le,lt,ne} expression 5149 bool AsmParser::parseDirectiveIf(SMLoc DirectiveLoc, DirectiveKind DirKind) { 5150 TheCondStack.push_back(TheCondState); 5151 TheCondState.TheCond = AsmCond::IfCond; 5152 if (TheCondState.Ignore) { 5153 eatToEndOfStatement(); 5154 } else { 5155 int64_t ExprValue; 5156 if (parseAbsoluteExpression(ExprValue) || parseEOL()) 5157 return true; 5158 5159 switch (DirKind) { 5160 default: 5161 llvm_unreachable("unsupported directive"); 5162 case DK_IF: 5163 case DK_IFNE: 5164 break; 5165 case DK_IFEQ: 5166 ExprValue = ExprValue == 0; 5167 break; 5168 case DK_IFGE: 5169 ExprValue = ExprValue >= 0; 5170 break; 5171 case DK_IFGT: 5172 ExprValue = ExprValue > 0; 5173 break; 5174 case DK_IFLE: 5175 ExprValue = ExprValue <= 0; 5176 break; 5177 case DK_IFLT: 5178 ExprValue = ExprValue < 0; 5179 break; 5180 } 5181 5182 TheCondState.CondMet = ExprValue; 5183 TheCondState.Ignore = !TheCondState.CondMet; 5184 } 5185 5186 return false; 5187 } 5188 5189 /// parseDirectiveIfb 5190 /// ::= .ifb string 5191 bool AsmParser::parseDirectiveIfb(SMLoc DirectiveLoc, bool ExpectBlank) { 5192 TheCondStack.push_back(TheCondState); 5193 TheCondState.TheCond = AsmCond::IfCond; 5194 5195 if (TheCondState.Ignore) { 5196 eatToEndOfStatement(); 5197 } else { 5198 StringRef Str = parseStringToEndOfStatement(); 5199 5200 if (parseEOL()) 5201 return true; 5202 5203 TheCondState.CondMet = ExpectBlank == Str.empty(); 5204 TheCondState.Ignore = !TheCondState.CondMet; 5205 } 5206 5207 return false; 5208 } 5209 5210 /// parseDirectiveIfc 5211 /// ::= .ifc string1, string2 5212 /// ::= .ifnc string1, string2 5213 bool AsmParser::parseDirectiveIfc(SMLoc DirectiveLoc, bool ExpectEqual) { 5214 TheCondStack.push_back(TheCondState); 5215 TheCondState.TheCond = AsmCond::IfCond; 5216 5217 if (TheCondState.Ignore) { 5218 eatToEndOfStatement(); 5219 } else { 5220 StringRef Str1 = parseStringToComma(); 5221 5222 if (parseComma()) 5223 return true; 5224 5225 StringRef Str2 = parseStringToEndOfStatement(); 5226 5227 if (parseEOL()) 5228 return true; 5229 5230 TheCondState.CondMet = ExpectEqual == (Str1.trim() == Str2.trim()); 5231 TheCondState.Ignore = !TheCondState.CondMet; 5232 } 5233 5234 return false; 5235 } 5236 5237 /// parseDirectiveIfeqs 5238 /// ::= .ifeqs string1, string2 5239 bool AsmParser::parseDirectiveIfeqs(SMLoc DirectiveLoc, bool ExpectEqual) { 5240 if (Lexer.isNot(AsmToken::String)) { 5241 if (ExpectEqual) 5242 return TokError("expected string parameter for '.ifeqs' directive"); 5243 return TokError("expected string parameter for '.ifnes' directive"); 5244 } 5245 5246 StringRef String1 = getTok().getStringContents(); 5247 Lex(); 5248 5249 if (Lexer.isNot(AsmToken::Comma)) { 5250 if (ExpectEqual) 5251 return TokError( 5252 "expected comma after first string for '.ifeqs' directive"); 5253 return TokError("expected comma after first string for '.ifnes' directive"); 5254 } 5255 5256 Lex(); 5257 5258 if (Lexer.isNot(AsmToken::String)) { 5259 if (ExpectEqual) 5260 return TokError("expected string parameter for '.ifeqs' directive"); 5261 return TokError("expected string parameter for '.ifnes' directive"); 5262 } 5263 5264 StringRef String2 = getTok().getStringContents(); 5265 Lex(); 5266 5267 TheCondStack.push_back(TheCondState); 5268 TheCondState.TheCond = AsmCond::IfCond; 5269 TheCondState.CondMet = ExpectEqual == (String1 == String2); 5270 TheCondState.Ignore = !TheCondState.CondMet; 5271 5272 return false; 5273 } 5274 5275 /// parseDirectiveIfdef 5276 /// ::= .ifdef symbol 5277 bool AsmParser::parseDirectiveIfdef(SMLoc DirectiveLoc, bool expect_defined) { 5278 StringRef Name; 5279 TheCondStack.push_back(TheCondState); 5280 TheCondState.TheCond = AsmCond::IfCond; 5281 5282 if (TheCondState.Ignore) { 5283 eatToEndOfStatement(); 5284 } else { 5285 if (check(parseIdentifier(Name), "expected identifier after '.ifdef'") || 5286 parseEOL()) 5287 return true; 5288 5289 MCSymbol *Sym = getContext().lookupSymbol(Name); 5290 5291 if (expect_defined) 5292 TheCondState.CondMet = (Sym && !Sym->isUndefined(false)); 5293 else 5294 TheCondState.CondMet = (!Sym || Sym->isUndefined(false)); 5295 TheCondState.Ignore = !TheCondState.CondMet; 5296 } 5297 5298 return false; 5299 } 5300 5301 /// parseDirectiveElseIf 5302 /// ::= .elseif expression 5303 bool AsmParser::parseDirectiveElseIf(SMLoc DirectiveLoc) { 5304 if (TheCondState.TheCond != AsmCond::IfCond && 5305 TheCondState.TheCond != AsmCond::ElseIfCond) 5306 return Error(DirectiveLoc, "Encountered a .elseif that doesn't follow an" 5307 " .if or an .elseif"); 5308 TheCondState.TheCond = AsmCond::ElseIfCond; 5309 5310 bool LastIgnoreState = false; 5311 if (!TheCondStack.empty()) 5312 LastIgnoreState = TheCondStack.back().Ignore; 5313 if (LastIgnoreState || TheCondState.CondMet) { 5314 TheCondState.Ignore = true; 5315 eatToEndOfStatement(); 5316 } else { 5317 int64_t ExprValue; 5318 if (parseAbsoluteExpression(ExprValue)) 5319 return true; 5320 5321 if (parseEOL()) 5322 return true; 5323 5324 TheCondState.CondMet = ExprValue; 5325 TheCondState.Ignore = !TheCondState.CondMet; 5326 } 5327 5328 return false; 5329 } 5330 5331 /// parseDirectiveElse 5332 /// ::= .else 5333 bool AsmParser::parseDirectiveElse(SMLoc DirectiveLoc) { 5334 if (parseEOL()) 5335 return true; 5336 5337 if (TheCondState.TheCond != AsmCond::IfCond && 5338 TheCondState.TheCond != AsmCond::ElseIfCond) 5339 return Error(DirectiveLoc, "Encountered a .else that doesn't follow " 5340 " an .if or an .elseif"); 5341 TheCondState.TheCond = AsmCond::ElseCond; 5342 bool LastIgnoreState = false; 5343 if (!TheCondStack.empty()) 5344 LastIgnoreState = TheCondStack.back().Ignore; 5345 if (LastIgnoreState || TheCondState.CondMet) 5346 TheCondState.Ignore = true; 5347 else 5348 TheCondState.Ignore = false; 5349 5350 return false; 5351 } 5352 5353 /// parseDirectiveEnd 5354 /// ::= .end 5355 bool AsmParser::parseDirectiveEnd(SMLoc DirectiveLoc) { 5356 if (parseEOL()) 5357 return true; 5358 5359 while (Lexer.isNot(AsmToken::Eof)) 5360 Lexer.Lex(); 5361 5362 return false; 5363 } 5364 5365 /// parseDirectiveError 5366 /// ::= .err 5367 /// ::= .error [string] 5368 bool AsmParser::parseDirectiveError(SMLoc L, bool WithMessage) { 5369 if (!TheCondStack.empty()) { 5370 if (TheCondStack.back().Ignore) { 5371 eatToEndOfStatement(); 5372 return false; 5373 } 5374 } 5375 5376 if (!WithMessage) 5377 return Error(L, ".err encountered"); 5378 5379 StringRef Message = ".error directive invoked in source file"; 5380 if (Lexer.isNot(AsmToken::EndOfStatement)) { 5381 if (Lexer.isNot(AsmToken::String)) 5382 return TokError(".error argument must be a string"); 5383 5384 Message = getTok().getStringContents(); 5385 Lex(); 5386 } 5387 5388 return Error(L, Message); 5389 } 5390 5391 /// parseDirectiveWarning 5392 /// ::= .warning [string] 5393 bool AsmParser::parseDirectiveWarning(SMLoc L) { 5394 if (!TheCondStack.empty()) { 5395 if (TheCondStack.back().Ignore) { 5396 eatToEndOfStatement(); 5397 return false; 5398 } 5399 } 5400 5401 StringRef Message = ".warning directive invoked in source file"; 5402 5403 if (!parseOptionalToken(AsmToken::EndOfStatement)) { 5404 if (Lexer.isNot(AsmToken::String)) 5405 return TokError(".warning argument must be a string"); 5406 5407 Message = getTok().getStringContents(); 5408 Lex(); 5409 if (parseEOL()) 5410 return true; 5411 } 5412 5413 return Warning(L, Message); 5414 } 5415 5416 /// parseDirectiveEndIf 5417 /// ::= .endif 5418 bool AsmParser::parseDirectiveEndIf(SMLoc DirectiveLoc) { 5419 if (parseEOL()) 5420 return true; 5421 5422 if ((TheCondState.TheCond == AsmCond::NoCond) || TheCondStack.empty()) 5423 return Error(DirectiveLoc, "Encountered a .endif that doesn't follow " 5424 "an .if or .else"); 5425 if (!TheCondStack.empty()) { 5426 TheCondState = TheCondStack.back(); 5427 TheCondStack.pop_back(); 5428 } 5429 5430 return false; 5431 } 5432 5433 void AsmParser::initializeDirectiveKindMap() { 5434 /* Lookup will be done with the directive 5435 * converted to lower case, so all these 5436 * keys should be lower case. 5437 * (target specific directives are handled 5438 * elsewhere) 5439 */ 5440 DirectiveKindMap[".set"] = DK_SET; 5441 DirectiveKindMap[".equ"] = DK_EQU; 5442 DirectiveKindMap[".equiv"] = DK_EQUIV; 5443 DirectiveKindMap[".ascii"] = DK_ASCII; 5444 DirectiveKindMap[".asciz"] = DK_ASCIZ; 5445 DirectiveKindMap[".string"] = DK_STRING; 5446 DirectiveKindMap[".byte"] = DK_BYTE; 5447 DirectiveKindMap[".short"] = DK_SHORT; 5448 DirectiveKindMap[".value"] = DK_VALUE; 5449 DirectiveKindMap[".2byte"] = DK_2BYTE; 5450 DirectiveKindMap[".long"] = DK_LONG; 5451 DirectiveKindMap[".int"] = DK_INT; 5452 DirectiveKindMap[".4byte"] = DK_4BYTE; 5453 DirectiveKindMap[".quad"] = DK_QUAD; 5454 DirectiveKindMap[".8byte"] = DK_8BYTE; 5455 DirectiveKindMap[".octa"] = DK_OCTA; 5456 DirectiveKindMap[".single"] = DK_SINGLE; 5457 DirectiveKindMap[".float"] = DK_FLOAT; 5458 DirectiveKindMap[".double"] = DK_DOUBLE; 5459 DirectiveKindMap[".align"] = DK_ALIGN; 5460 DirectiveKindMap[".align32"] = DK_ALIGN32; 5461 DirectiveKindMap[".balign"] = DK_BALIGN; 5462 DirectiveKindMap[".balignw"] = DK_BALIGNW; 5463 DirectiveKindMap[".balignl"] = DK_BALIGNL; 5464 DirectiveKindMap[".p2align"] = DK_P2ALIGN; 5465 DirectiveKindMap[".p2alignw"] = DK_P2ALIGNW; 5466 DirectiveKindMap[".p2alignl"] = DK_P2ALIGNL; 5467 DirectiveKindMap[".org"] = DK_ORG; 5468 DirectiveKindMap[".fill"] = DK_FILL; 5469 DirectiveKindMap[".zero"] = DK_ZERO; 5470 DirectiveKindMap[".extern"] = DK_EXTERN; 5471 DirectiveKindMap[".globl"] = DK_GLOBL; 5472 DirectiveKindMap[".global"] = DK_GLOBAL; 5473 DirectiveKindMap[".lazy_reference"] = DK_LAZY_REFERENCE; 5474 DirectiveKindMap[".no_dead_strip"] = DK_NO_DEAD_STRIP; 5475 DirectiveKindMap[".symbol_resolver"] = DK_SYMBOL_RESOLVER; 5476 DirectiveKindMap[".private_extern"] = DK_PRIVATE_EXTERN; 5477 DirectiveKindMap[".reference"] = DK_REFERENCE; 5478 DirectiveKindMap[".weak_definition"] = DK_WEAK_DEFINITION; 5479 DirectiveKindMap[".weak_reference"] = DK_WEAK_REFERENCE; 5480 DirectiveKindMap[".weak_def_can_be_hidden"] = DK_WEAK_DEF_CAN_BE_HIDDEN; 5481 DirectiveKindMap[".cold"] = DK_COLD; 5482 DirectiveKindMap[".comm"] = DK_COMM; 5483 DirectiveKindMap[".common"] = DK_COMMON; 5484 DirectiveKindMap[".lcomm"] = DK_LCOMM; 5485 DirectiveKindMap[".abort"] = DK_ABORT; 5486 DirectiveKindMap[".include"] = DK_INCLUDE; 5487 DirectiveKindMap[".incbin"] = DK_INCBIN; 5488 DirectiveKindMap[".code16"] = DK_CODE16; 5489 DirectiveKindMap[".code16gcc"] = DK_CODE16GCC; 5490 DirectiveKindMap[".rept"] = DK_REPT; 5491 DirectiveKindMap[".rep"] = DK_REPT; 5492 DirectiveKindMap[".irp"] = DK_IRP; 5493 DirectiveKindMap[".irpc"] = DK_IRPC; 5494 DirectiveKindMap[".endr"] = DK_ENDR; 5495 DirectiveKindMap[".bundle_align_mode"] = DK_BUNDLE_ALIGN_MODE; 5496 DirectiveKindMap[".bundle_lock"] = DK_BUNDLE_LOCK; 5497 DirectiveKindMap[".bundle_unlock"] = DK_BUNDLE_UNLOCK; 5498 DirectiveKindMap[".if"] = DK_IF; 5499 DirectiveKindMap[".ifeq"] = DK_IFEQ; 5500 DirectiveKindMap[".ifge"] = DK_IFGE; 5501 DirectiveKindMap[".ifgt"] = DK_IFGT; 5502 DirectiveKindMap[".ifle"] = DK_IFLE; 5503 DirectiveKindMap[".iflt"] = DK_IFLT; 5504 DirectiveKindMap[".ifne"] = DK_IFNE; 5505 DirectiveKindMap[".ifb"] = DK_IFB; 5506 DirectiveKindMap[".ifnb"] = DK_IFNB; 5507 DirectiveKindMap[".ifc"] = DK_IFC; 5508 DirectiveKindMap[".ifeqs"] = DK_IFEQS; 5509 DirectiveKindMap[".ifnc"] = DK_IFNC; 5510 DirectiveKindMap[".ifnes"] = DK_IFNES; 5511 DirectiveKindMap[".ifdef"] = DK_IFDEF; 5512 DirectiveKindMap[".ifndef"] = DK_IFNDEF; 5513 DirectiveKindMap[".ifnotdef"] = DK_IFNOTDEF; 5514 DirectiveKindMap[".elseif"] = DK_ELSEIF; 5515 DirectiveKindMap[".else"] = DK_ELSE; 5516 DirectiveKindMap[".end"] = DK_END; 5517 DirectiveKindMap[".endif"] = DK_ENDIF; 5518 DirectiveKindMap[".skip"] = DK_SKIP; 5519 DirectiveKindMap[".space"] = DK_SPACE; 5520 DirectiveKindMap[".file"] = DK_FILE; 5521 DirectiveKindMap[".line"] = DK_LINE; 5522 DirectiveKindMap[".loc"] = DK_LOC; 5523 DirectiveKindMap[".stabs"] = DK_STABS; 5524 DirectiveKindMap[".cv_file"] = DK_CV_FILE; 5525 DirectiveKindMap[".cv_func_id"] = DK_CV_FUNC_ID; 5526 DirectiveKindMap[".cv_loc"] = DK_CV_LOC; 5527 DirectiveKindMap[".cv_linetable"] = DK_CV_LINETABLE; 5528 DirectiveKindMap[".cv_inline_linetable"] = DK_CV_INLINE_LINETABLE; 5529 DirectiveKindMap[".cv_inline_site_id"] = DK_CV_INLINE_SITE_ID; 5530 DirectiveKindMap[".cv_def_range"] = DK_CV_DEF_RANGE; 5531 DirectiveKindMap[".cv_string"] = DK_CV_STRING; 5532 DirectiveKindMap[".cv_stringtable"] = DK_CV_STRINGTABLE; 5533 DirectiveKindMap[".cv_filechecksums"] = DK_CV_FILECHECKSUMS; 5534 DirectiveKindMap[".cv_filechecksumoffset"] = DK_CV_FILECHECKSUM_OFFSET; 5535 DirectiveKindMap[".cv_fpo_data"] = DK_CV_FPO_DATA; 5536 DirectiveKindMap[".sleb128"] = DK_SLEB128; 5537 DirectiveKindMap[".uleb128"] = DK_ULEB128; 5538 DirectiveKindMap[".cfi_sections"] = DK_CFI_SECTIONS; 5539 DirectiveKindMap[".cfi_startproc"] = DK_CFI_STARTPROC; 5540 DirectiveKindMap[".cfi_endproc"] = DK_CFI_ENDPROC; 5541 DirectiveKindMap[".cfi_def_cfa"] = DK_CFI_DEF_CFA; 5542 DirectiveKindMap[".cfi_def_cfa_offset"] = DK_CFI_DEF_CFA_OFFSET; 5543 DirectiveKindMap[".cfi_adjust_cfa_offset"] = DK_CFI_ADJUST_CFA_OFFSET; 5544 DirectiveKindMap[".cfi_def_cfa_register"] = DK_CFI_DEF_CFA_REGISTER; 5545 DirectiveKindMap[".cfi_llvm_def_aspace_cfa"] = DK_CFI_LLVM_DEF_ASPACE_CFA; 5546 DirectiveKindMap[".cfi_offset"] = DK_CFI_OFFSET; 5547 DirectiveKindMap[".cfi_rel_offset"] = DK_CFI_REL_OFFSET; 5548 DirectiveKindMap[".cfi_personality"] = DK_CFI_PERSONALITY; 5549 DirectiveKindMap[".cfi_lsda"] = DK_CFI_LSDA; 5550 DirectiveKindMap[".cfi_remember_state"] = DK_CFI_REMEMBER_STATE; 5551 DirectiveKindMap[".cfi_restore_state"] = DK_CFI_RESTORE_STATE; 5552 DirectiveKindMap[".cfi_same_value"] = DK_CFI_SAME_VALUE; 5553 DirectiveKindMap[".cfi_restore"] = DK_CFI_RESTORE; 5554 DirectiveKindMap[".cfi_escape"] = DK_CFI_ESCAPE; 5555 DirectiveKindMap[".cfi_return_column"] = DK_CFI_RETURN_COLUMN; 5556 DirectiveKindMap[".cfi_signal_frame"] = DK_CFI_SIGNAL_FRAME; 5557 DirectiveKindMap[".cfi_undefined"] = DK_CFI_UNDEFINED; 5558 DirectiveKindMap[".cfi_register"] = DK_CFI_REGISTER; 5559 DirectiveKindMap[".cfi_window_save"] = DK_CFI_WINDOW_SAVE; 5560 DirectiveKindMap[".cfi_b_key_frame"] = DK_CFI_B_KEY_FRAME; 5561 DirectiveKindMap[".cfi_mte_tagged_frame"] = DK_CFI_MTE_TAGGED_FRAME; 5562 DirectiveKindMap[".macros_on"] = DK_MACROS_ON; 5563 DirectiveKindMap[".macros_off"] = DK_MACROS_OFF; 5564 DirectiveKindMap[".macro"] = DK_MACRO; 5565 DirectiveKindMap[".exitm"] = DK_EXITM; 5566 DirectiveKindMap[".endm"] = DK_ENDM; 5567 DirectiveKindMap[".endmacro"] = DK_ENDMACRO; 5568 DirectiveKindMap[".purgem"] = DK_PURGEM; 5569 DirectiveKindMap[".err"] = DK_ERR; 5570 DirectiveKindMap[".error"] = DK_ERROR; 5571 DirectiveKindMap[".warning"] = DK_WARNING; 5572 DirectiveKindMap[".altmacro"] = DK_ALTMACRO; 5573 DirectiveKindMap[".noaltmacro"] = DK_NOALTMACRO; 5574 DirectiveKindMap[".reloc"] = DK_RELOC; 5575 DirectiveKindMap[".dc"] = DK_DC; 5576 DirectiveKindMap[".dc.a"] = DK_DC_A; 5577 DirectiveKindMap[".dc.b"] = DK_DC_B; 5578 DirectiveKindMap[".dc.d"] = DK_DC_D; 5579 DirectiveKindMap[".dc.l"] = DK_DC_L; 5580 DirectiveKindMap[".dc.s"] = DK_DC_S; 5581 DirectiveKindMap[".dc.w"] = DK_DC_W; 5582 DirectiveKindMap[".dc.x"] = DK_DC_X; 5583 DirectiveKindMap[".dcb"] = DK_DCB; 5584 DirectiveKindMap[".dcb.b"] = DK_DCB_B; 5585 DirectiveKindMap[".dcb.d"] = DK_DCB_D; 5586 DirectiveKindMap[".dcb.l"] = DK_DCB_L; 5587 DirectiveKindMap[".dcb.s"] = DK_DCB_S; 5588 DirectiveKindMap[".dcb.w"] = DK_DCB_W; 5589 DirectiveKindMap[".dcb.x"] = DK_DCB_X; 5590 DirectiveKindMap[".ds"] = DK_DS; 5591 DirectiveKindMap[".ds.b"] = DK_DS_B; 5592 DirectiveKindMap[".ds.d"] = DK_DS_D; 5593 DirectiveKindMap[".ds.l"] = DK_DS_L; 5594 DirectiveKindMap[".ds.p"] = DK_DS_P; 5595 DirectiveKindMap[".ds.s"] = DK_DS_S; 5596 DirectiveKindMap[".ds.w"] = DK_DS_W; 5597 DirectiveKindMap[".ds.x"] = DK_DS_X; 5598 DirectiveKindMap[".print"] = DK_PRINT; 5599 DirectiveKindMap[".addrsig"] = DK_ADDRSIG; 5600 DirectiveKindMap[".addrsig_sym"] = DK_ADDRSIG_SYM; 5601 DirectiveKindMap[".pseudoprobe"] = DK_PSEUDO_PROBE; 5602 DirectiveKindMap[".lto_discard"] = DK_LTO_DISCARD; 5603 DirectiveKindMap[".lto_set_conditional"] = DK_LTO_SET_CONDITIONAL; 5604 DirectiveKindMap[".memtag"] = DK_MEMTAG; 5605 } 5606 5607 MCAsmMacro *AsmParser::parseMacroLikeBody(SMLoc DirectiveLoc) { 5608 AsmToken EndToken, StartToken = getTok(); 5609 5610 unsigned NestLevel = 0; 5611 while (true) { 5612 // Check whether we have reached the end of the file. 5613 if (getLexer().is(AsmToken::Eof)) { 5614 printError(DirectiveLoc, "no matching '.endr' in definition"); 5615 return nullptr; 5616 } 5617 5618 if (Lexer.is(AsmToken::Identifier) && 5619 (getTok().getIdentifier() == ".rep" || 5620 getTok().getIdentifier() == ".rept" || 5621 getTok().getIdentifier() == ".irp" || 5622 getTok().getIdentifier() == ".irpc")) { 5623 ++NestLevel; 5624 } 5625 5626 // Otherwise, check whether we have reached the .endr. 5627 if (Lexer.is(AsmToken::Identifier) && getTok().getIdentifier() == ".endr") { 5628 if (NestLevel == 0) { 5629 EndToken = getTok(); 5630 Lex(); 5631 if (Lexer.isNot(AsmToken::EndOfStatement)) { 5632 printError(getTok().getLoc(), 5633 "unexpected token in '.endr' directive"); 5634 return nullptr; 5635 } 5636 break; 5637 } 5638 --NestLevel; 5639 } 5640 5641 // Otherwise, scan till the end of the statement. 5642 eatToEndOfStatement(); 5643 } 5644 5645 const char *BodyStart = StartToken.getLoc().getPointer(); 5646 const char *BodyEnd = EndToken.getLoc().getPointer(); 5647 StringRef Body = StringRef(BodyStart, BodyEnd - BodyStart); 5648 5649 // We Are Anonymous. 5650 MacroLikeBodies.emplace_back(StringRef(), Body, MCAsmMacroParameters()); 5651 return &MacroLikeBodies.back(); 5652 } 5653 5654 void AsmParser::instantiateMacroLikeBody(MCAsmMacro *M, SMLoc DirectiveLoc, 5655 raw_svector_ostream &OS) { 5656 OS << ".endr\n"; 5657 5658 std::unique_ptr<MemoryBuffer> Instantiation = 5659 MemoryBuffer::getMemBufferCopy(OS.str(), "<instantiation>"); 5660 5661 // Create the macro instantiation object and add to the current macro 5662 // instantiation stack. 5663 MacroInstantiation *MI = new MacroInstantiation{ 5664 DirectiveLoc, CurBuffer, getTok().getLoc(), TheCondStack.size()}; 5665 ActiveMacros.push_back(MI); 5666 5667 // Jump to the macro instantiation and prime the lexer. 5668 CurBuffer = SrcMgr.AddNewSourceBuffer(std::move(Instantiation), SMLoc()); 5669 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer()); 5670 Lex(); 5671 } 5672 5673 /// parseDirectiveRept 5674 /// ::= .rep | .rept count 5675 bool AsmParser::parseDirectiveRept(SMLoc DirectiveLoc, StringRef Dir) { 5676 const MCExpr *CountExpr; 5677 SMLoc CountLoc = getTok().getLoc(); 5678 if (parseExpression(CountExpr)) 5679 return true; 5680 5681 int64_t Count; 5682 if (!CountExpr->evaluateAsAbsolute(Count, getStreamer().getAssemblerPtr())) { 5683 return Error(CountLoc, "unexpected token in '" + Dir + "' directive"); 5684 } 5685 5686 if (check(Count < 0, CountLoc, "Count is negative") || parseEOL()) 5687 return true; 5688 5689 // Lex the rept definition. 5690 MCAsmMacro *M = parseMacroLikeBody(DirectiveLoc); 5691 if (!M) 5692 return true; 5693 5694 // Macro instantiation is lexical, unfortunately. We construct a new buffer 5695 // to hold the macro body with substitutions. 5696 SmallString<256> Buf; 5697 raw_svector_ostream OS(Buf); 5698 while (Count--) { 5699 // Note that the AtPseudoVariable is disabled for instantiations of .rep(t). 5700 if (expandMacro(OS, M->Body, std::nullopt, std::nullopt, false, 5701 getTok().getLoc())) 5702 return true; 5703 } 5704 instantiateMacroLikeBody(M, DirectiveLoc, OS); 5705 5706 return false; 5707 } 5708 5709 /// parseDirectiveIrp 5710 /// ::= .irp symbol,values 5711 bool AsmParser::parseDirectiveIrp(SMLoc DirectiveLoc) { 5712 MCAsmMacroParameter Parameter; 5713 MCAsmMacroArguments A; 5714 if (check(parseIdentifier(Parameter.Name), 5715 "expected identifier in '.irp' directive") || 5716 parseComma() || parseMacroArguments(nullptr, A) || parseEOL()) 5717 return true; 5718 5719 // Lex the irp definition. 5720 MCAsmMacro *M = parseMacroLikeBody(DirectiveLoc); 5721 if (!M) 5722 return true; 5723 5724 // Macro instantiation is lexical, unfortunately. We construct a new buffer 5725 // to hold the macro body with substitutions. 5726 SmallString<256> Buf; 5727 raw_svector_ostream OS(Buf); 5728 5729 for (const MCAsmMacroArgument &Arg : A) { 5730 // Note that the AtPseudoVariable is enabled for instantiations of .irp. 5731 // This is undocumented, but GAS seems to support it. 5732 if (expandMacro(OS, M->Body, Parameter, Arg, true, getTok().getLoc())) 5733 return true; 5734 } 5735 5736 instantiateMacroLikeBody(M, DirectiveLoc, OS); 5737 5738 return false; 5739 } 5740 5741 /// parseDirectiveIrpc 5742 /// ::= .irpc symbol,values 5743 bool AsmParser::parseDirectiveIrpc(SMLoc DirectiveLoc) { 5744 MCAsmMacroParameter Parameter; 5745 MCAsmMacroArguments A; 5746 5747 if (check(parseIdentifier(Parameter.Name), 5748 "expected identifier in '.irpc' directive") || 5749 parseComma() || parseMacroArguments(nullptr, A)) 5750 return true; 5751 5752 if (A.size() != 1 || A.front().size() != 1) 5753 return TokError("unexpected token in '.irpc' directive"); 5754 if (parseEOL()) 5755 return true; 5756 5757 // Lex the irpc definition. 5758 MCAsmMacro *M = parseMacroLikeBody(DirectiveLoc); 5759 if (!M) 5760 return true; 5761 5762 // Macro instantiation is lexical, unfortunately. We construct a new buffer 5763 // to hold the macro body with substitutions. 5764 SmallString<256> Buf; 5765 raw_svector_ostream OS(Buf); 5766 5767 StringRef Values = A.front().front().getString(); 5768 for (std::size_t I = 0, End = Values.size(); I != End; ++I) { 5769 MCAsmMacroArgument Arg; 5770 Arg.emplace_back(AsmToken::Identifier, Values.slice(I, I + 1)); 5771 5772 // Note that the AtPseudoVariable is enabled for instantiations of .irpc. 5773 // This is undocumented, but GAS seems to support it. 5774 if (expandMacro(OS, M->Body, Parameter, Arg, true, getTok().getLoc())) 5775 return true; 5776 } 5777 5778 instantiateMacroLikeBody(M, DirectiveLoc, OS); 5779 5780 return false; 5781 } 5782 5783 bool AsmParser::parseDirectiveEndr(SMLoc DirectiveLoc) { 5784 if (ActiveMacros.empty()) 5785 return TokError("unmatched '.endr' directive"); 5786 5787 // The only .repl that should get here are the ones created by 5788 // instantiateMacroLikeBody. 5789 assert(getLexer().is(AsmToken::EndOfStatement)); 5790 5791 handleMacroExit(); 5792 return false; 5793 } 5794 5795 bool AsmParser::parseDirectiveMSEmit(SMLoc IDLoc, ParseStatementInfo &Info, 5796 size_t Len) { 5797 const MCExpr *Value; 5798 SMLoc ExprLoc = getLexer().getLoc(); 5799 if (parseExpression(Value)) 5800 return true; 5801 const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value); 5802 if (!MCE) 5803 return Error(ExprLoc, "unexpected expression in _emit"); 5804 uint64_t IntValue = MCE->getValue(); 5805 if (!isUInt<8>(IntValue) && !isInt<8>(IntValue)) 5806 return Error(ExprLoc, "literal value out of range for directive"); 5807 5808 Info.AsmRewrites->emplace_back(AOK_Emit, IDLoc, Len); 5809 return false; 5810 } 5811 5812 bool AsmParser::parseDirectiveMSAlign(SMLoc IDLoc, ParseStatementInfo &Info) { 5813 const MCExpr *Value; 5814 SMLoc ExprLoc = getLexer().getLoc(); 5815 if (parseExpression(Value)) 5816 return true; 5817 const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value); 5818 if (!MCE) 5819 return Error(ExprLoc, "unexpected expression in align"); 5820 uint64_t IntValue = MCE->getValue(); 5821 if (!isPowerOf2_64(IntValue)) 5822 return Error(ExprLoc, "literal value not a power of two greater then zero"); 5823 5824 Info.AsmRewrites->emplace_back(AOK_Align, IDLoc, 5, Log2_64(IntValue)); 5825 return false; 5826 } 5827 5828 bool AsmParser::parseDirectivePrint(SMLoc DirectiveLoc) { 5829 const AsmToken StrTok = getTok(); 5830 Lex(); 5831 if (StrTok.isNot(AsmToken::String) || StrTok.getString().front() != '"') 5832 return Error(DirectiveLoc, "expected double quoted string after .print"); 5833 if (parseEOL()) 5834 return true; 5835 llvm::outs() << StrTok.getStringContents() << '\n'; 5836 return false; 5837 } 5838 5839 bool AsmParser::parseDirectiveAddrsig() { 5840 if (parseEOL()) 5841 return true; 5842 getStreamer().emitAddrsig(); 5843 return false; 5844 } 5845 5846 bool AsmParser::parseDirectiveAddrsigSym() { 5847 StringRef Name; 5848 if (check(parseIdentifier(Name), "expected identifier") || parseEOL()) 5849 return true; 5850 MCSymbol *Sym = getContext().getOrCreateSymbol(Name); 5851 getStreamer().emitAddrsigSym(Sym); 5852 return false; 5853 } 5854 5855 bool AsmParser::parseDirectivePseudoProbe() { 5856 int64_t Guid; 5857 int64_t Index; 5858 int64_t Type; 5859 int64_t Attr; 5860 5861 if (getLexer().is(AsmToken::Integer)) { 5862 if (parseIntToken(Guid, "unexpected token in '.pseudoprobe' directive")) 5863 return true; 5864 } 5865 5866 if (getLexer().is(AsmToken::Integer)) { 5867 if (parseIntToken(Index, "unexpected token in '.pseudoprobe' directive")) 5868 return true; 5869 } 5870 5871 if (getLexer().is(AsmToken::Integer)) { 5872 if (parseIntToken(Type, "unexpected token in '.pseudoprobe' directive")) 5873 return true; 5874 } 5875 5876 if (getLexer().is(AsmToken::Integer)) { 5877 if (parseIntToken(Attr, "unexpected token in '.pseudoprobe' directive")) 5878 return true; 5879 } 5880 5881 // Parse inline stack like @ GUID:11:12 @ GUID:1:11 @ GUID:3:21 5882 MCPseudoProbeInlineStack InlineStack; 5883 5884 while (getLexer().is(AsmToken::At)) { 5885 // eat @ 5886 Lex(); 5887 5888 int64_t CallerGuid = 0; 5889 if (getLexer().is(AsmToken::Integer)) { 5890 if (parseIntToken(CallerGuid, 5891 "unexpected token in '.pseudoprobe' directive")) 5892 return true; 5893 } 5894 5895 // eat colon 5896 if (getLexer().is(AsmToken::Colon)) 5897 Lex(); 5898 5899 int64_t CallerProbeId = 0; 5900 if (getLexer().is(AsmToken::Integer)) { 5901 if (parseIntToken(CallerProbeId, 5902 "unexpected token in '.pseudoprobe' directive")) 5903 return true; 5904 } 5905 5906 InlineSite Site(CallerGuid, CallerProbeId); 5907 InlineStack.push_back(Site); 5908 } 5909 5910 // Parse function entry name 5911 StringRef FnName; 5912 if (parseIdentifier(FnName)) 5913 return Error(getLexer().getLoc(), "unexpected token in '.pseudoprobe' directive"); 5914 MCSymbol *FnSym = getContext().lookupSymbol(FnName); 5915 5916 if (parseEOL()) 5917 return true; 5918 5919 getStreamer().emitPseudoProbe(Guid, Index, Type, Attr, InlineStack, FnSym); 5920 return false; 5921 } 5922 5923 /// parseDirectiveLTODiscard 5924 /// ::= ".lto_discard" [ identifier ( , identifier )* ] 5925 /// The LTO library emits this directive to discard non-prevailing symbols. 5926 /// We ignore symbol assignments and attribute changes for the specified 5927 /// symbols. 5928 bool AsmParser::parseDirectiveLTODiscard() { 5929 auto ParseOp = [&]() -> bool { 5930 StringRef Name; 5931 SMLoc Loc = getTok().getLoc(); 5932 if (parseIdentifier(Name)) 5933 return Error(Loc, "expected identifier"); 5934 LTODiscardSymbols.insert(Name); 5935 return false; 5936 }; 5937 5938 LTODiscardSymbols.clear(); 5939 return parseMany(ParseOp); 5940 } 5941 5942 // We are comparing pointers, but the pointers are relative to a single string. 5943 // Thus, this should always be deterministic. 5944 static int rewritesSort(const AsmRewrite *AsmRewriteA, 5945 const AsmRewrite *AsmRewriteB) { 5946 if (AsmRewriteA->Loc.getPointer() < AsmRewriteB->Loc.getPointer()) 5947 return -1; 5948 if (AsmRewriteB->Loc.getPointer() < AsmRewriteA->Loc.getPointer()) 5949 return 1; 5950 5951 // It's possible to have a SizeDirective, Imm/ImmPrefix and an Input/Output 5952 // rewrite to the same location. Make sure the SizeDirective rewrite is 5953 // performed first, then the Imm/ImmPrefix and finally the Input/Output. This 5954 // ensures the sort algorithm is stable. 5955 if (AsmRewritePrecedence[AsmRewriteA->Kind] > 5956 AsmRewritePrecedence[AsmRewriteB->Kind]) 5957 return -1; 5958 5959 if (AsmRewritePrecedence[AsmRewriteA->Kind] < 5960 AsmRewritePrecedence[AsmRewriteB->Kind]) 5961 return 1; 5962 llvm_unreachable("Unstable rewrite sort."); 5963 } 5964 5965 bool AsmParser::parseMSInlineAsm( 5966 std::string &AsmString, unsigned &NumOutputs, unsigned &NumInputs, 5967 SmallVectorImpl<std::pair<void *, bool>> &OpDecls, 5968 SmallVectorImpl<std::string> &Constraints, 5969 SmallVectorImpl<std::string> &Clobbers, const MCInstrInfo *MII, 5970 const MCInstPrinter *IP, MCAsmParserSemaCallback &SI) { 5971 SmallVector<void *, 4> InputDecls; 5972 SmallVector<void *, 4> OutputDecls; 5973 SmallVector<bool, 4> InputDeclsAddressOf; 5974 SmallVector<bool, 4> OutputDeclsAddressOf; 5975 SmallVector<std::string, 4> InputConstraints; 5976 SmallVector<std::string, 4> OutputConstraints; 5977 SmallVector<unsigned, 4> ClobberRegs; 5978 5979 SmallVector<AsmRewrite, 4> AsmStrRewrites; 5980 5981 // Prime the lexer. 5982 Lex(); 5983 5984 // While we have input, parse each statement. 5985 unsigned InputIdx = 0; 5986 unsigned OutputIdx = 0; 5987 while (getLexer().isNot(AsmToken::Eof)) { 5988 // Parse curly braces marking block start/end 5989 if (parseCurlyBlockScope(AsmStrRewrites)) 5990 continue; 5991 5992 ParseStatementInfo Info(&AsmStrRewrites); 5993 bool StatementErr = parseStatement(Info, &SI); 5994 5995 if (StatementErr || Info.ParseError) { 5996 // Emit pending errors if any exist. 5997 printPendingErrors(); 5998 return true; 5999 } 6000 6001 // No pending error should exist here. 6002 assert(!hasPendingError() && "unexpected error from parseStatement"); 6003 6004 if (Info.Opcode == ~0U) 6005 continue; 6006 6007 const MCInstrDesc &Desc = MII->get(Info.Opcode); 6008 6009 // Build the list of clobbers, outputs and inputs. 6010 for (unsigned i = 1, e = Info.ParsedOperands.size(); i != e; ++i) { 6011 MCParsedAsmOperand &Operand = *Info.ParsedOperands[i]; 6012 6013 // Register operand. 6014 if (Operand.isReg() && !Operand.needAddressOf() && 6015 !getTargetParser().OmitRegisterFromClobberLists(Operand.getReg())) { 6016 unsigned NumDefs = Desc.getNumDefs(); 6017 // Clobber. 6018 if (NumDefs && Operand.getMCOperandNum() < NumDefs) 6019 ClobberRegs.push_back(Operand.getReg()); 6020 continue; 6021 } 6022 6023 // Expr/Input or Output. 6024 StringRef SymName = Operand.getSymName(); 6025 if (SymName.empty()) 6026 continue; 6027 6028 void *OpDecl = Operand.getOpDecl(); 6029 if (!OpDecl) 6030 continue; 6031 6032 StringRef Constraint = Operand.getConstraint(); 6033 if (Operand.isImm()) { 6034 // Offset as immediate 6035 if (Operand.isOffsetOfLocal()) 6036 Constraint = "r"; 6037 else 6038 Constraint = "i"; 6039 } 6040 6041 bool isOutput = (i == 1) && Desc.mayStore(); 6042 bool Restricted = Operand.isMemUseUpRegs(); 6043 SMLoc Start = SMLoc::getFromPointer(SymName.data()); 6044 if (isOutput) { 6045 ++InputIdx; 6046 OutputDecls.push_back(OpDecl); 6047 OutputDeclsAddressOf.push_back(Operand.needAddressOf()); 6048 OutputConstraints.push_back(("=" + Constraint).str()); 6049 AsmStrRewrites.emplace_back(AOK_Output, Start, SymName.size(), 0, 6050 Restricted); 6051 } else { 6052 InputDecls.push_back(OpDecl); 6053 InputDeclsAddressOf.push_back(Operand.needAddressOf()); 6054 InputConstraints.push_back(Constraint.str()); 6055 if (Desc.operands()[i - 1].isBranchTarget()) 6056 AsmStrRewrites.emplace_back(AOK_CallInput, Start, SymName.size(), 0, 6057 Restricted); 6058 else 6059 AsmStrRewrites.emplace_back(AOK_Input, Start, SymName.size(), 0, 6060 Restricted); 6061 } 6062 } 6063 6064 // Consider implicit defs to be clobbers. Think of cpuid and push. 6065 llvm::append_range(ClobberRegs, Desc.implicit_defs()); 6066 } 6067 6068 // Set the number of Outputs and Inputs. 6069 NumOutputs = OutputDecls.size(); 6070 NumInputs = InputDecls.size(); 6071 6072 // Set the unique clobbers. 6073 array_pod_sort(ClobberRegs.begin(), ClobberRegs.end()); 6074 ClobberRegs.erase(std::unique(ClobberRegs.begin(), ClobberRegs.end()), 6075 ClobberRegs.end()); 6076 Clobbers.assign(ClobberRegs.size(), std::string()); 6077 for (unsigned I = 0, E = ClobberRegs.size(); I != E; ++I) { 6078 raw_string_ostream OS(Clobbers[I]); 6079 IP->printRegName(OS, ClobberRegs[I]); 6080 } 6081 6082 // Merge the various outputs and inputs. Output are expected first. 6083 if (NumOutputs || NumInputs) { 6084 unsigned NumExprs = NumOutputs + NumInputs; 6085 OpDecls.resize(NumExprs); 6086 Constraints.resize(NumExprs); 6087 for (unsigned i = 0; i < NumOutputs; ++i) { 6088 OpDecls[i] = std::make_pair(OutputDecls[i], OutputDeclsAddressOf[i]); 6089 Constraints[i] = OutputConstraints[i]; 6090 } 6091 for (unsigned i = 0, j = NumOutputs; i < NumInputs; ++i, ++j) { 6092 OpDecls[j] = std::make_pair(InputDecls[i], InputDeclsAddressOf[i]); 6093 Constraints[j] = InputConstraints[i]; 6094 } 6095 } 6096 6097 // Build the IR assembly string. 6098 std::string AsmStringIR; 6099 raw_string_ostream OS(AsmStringIR); 6100 StringRef ASMString = 6101 SrcMgr.getMemoryBuffer(SrcMgr.getMainFileID())->getBuffer(); 6102 const char *AsmStart = ASMString.begin(); 6103 const char *AsmEnd = ASMString.end(); 6104 array_pod_sort(AsmStrRewrites.begin(), AsmStrRewrites.end(), rewritesSort); 6105 for (auto it = AsmStrRewrites.begin(); it != AsmStrRewrites.end(); ++it) { 6106 const AsmRewrite &AR = *it; 6107 // Check if this has already been covered by another rewrite... 6108 if (AR.Done) 6109 continue; 6110 AsmRewriteKind Kind = AR.Kind; 6111 6112 const char *Loc = AR.Loc.getPointer(); 6113 assert(Loc >= AsmStart && "Expected Loc to be at or after Start!"); 6114 6115 // Emit everything up to the immediate/expression. 6116 if (unsigned Len = Loc - AsmStart) 6117 OS << StringRef(AsmStart, Len); 6118 6119 // Skip the original expression. 6120 if (Kind == AOK_Skip) { 6121 AsmStart = Loc + AR.Len; 6122 continue; 6123 } 6124 6125 unsigned AdditionalSkip = 0; 6126 // Rewrite expressions in $N notation. 6127 switch (Kind) { 6128 default: 6129 break; 6130 case AOK_IntelExpr: 6131 assert(AR.IntelExp.isValid() && "cannot write invalid intel expression"); 6132 if (AR.IntelExp.NeedBracs) 6133 OS << "["; 6134 if (AR.IntelExp.hasBaseReg()) 6135 OS << AR.IntelExp.BaseReg; 6136 if (AR.IntelExp.hasIndexReg()) 6137 OS << (AR.IntelExp.hasBaseReg() ? " + " : "") 6138 << AR.IntelExp.IndexReg; 6139 if (AR.IntelExp.Scale > 1) 6140 OS << " * $$" << AR.IntelExp.Scale; 6141 if (AR.IntelExp.hasOffset()) { 6142 if (AR.IntelExp.hasRegs()) 6143 OS << " + "; 6144 // Fuse this rewrite with a rewrite of the offset name, if present. 6145 StringRef OffsetName = AR.IntelExp.OffsetName; 6146 SMLoc OffsetLoc = SMLoc::getFromPointer(AR.IntelExp.OffsetName.data()); 6147 size_t OffsetLen = OffsetName.size(); 6148 auto rewrite_it = std::find_if( 6149 it, AsmStrRewrites.end(), [&](const AsmRewrite &FusingAR) { 6150 return FusingAR.Loc == OffsetLoc && FusingAR.Len == OffsetLen && 6151 (FusingAR.Kind == AOK_Input || 6152 FusingAR.Kind == AOK_CallInput); 6153 }); 6154 if (rewrite_it == AsmStrRewrites.end()) { 6155 OS << "offset " << OffsetName; 6156 } else if (rewrite_it->Kind == AOK_CallInput) { 6157 OS << "${" << InputIdx++ << ":P}"; 6158 rewrite_it->Done = true; 6159 } else { 6160 OS << '$' << InputIdx++; 6161 rewrite_it->Done = true; 6162 } 6163 } 6164 if (AR.IntelExp.Imm || AR.IntelExp.emitImm()) 6165 OS << (AR.IntelExp.emitImm() ? "$$" : " + $$") << AR.IntelExp.Imm; 6166 if (AR.IntelExp.NeedBracs) 6167 OS << "]"; 6168 break; 6169 case AOK_Label: 6170 OS << Ctx.getAsmInfo()->getPrivateLabelPrefix() << AR.Label; 6171 break; 6172 case AOK_Input: 6173 if (AR.IntelExpRestricted) 6174 OS << "${" << InputIdx++ << ":P}"; 6175 else 6176 OS << '$' << InputIdx++; 6177 break; 6178 case AOK_CallInput: 6179 OS << "${" << InputIdx++ << ":P}"; 6180 break; 6181 case AOK_Output: 6182 if (AR.IntelExpRestricted) 6183 OS << "${" << OutputIdx++ << ":P}"; 6184 else 6185 OS << '$' << OutputIdx++; 6186 break; 6187 case AOK_SizeDirective: 6188 switch (AR.Val) { 6189 default: break; 6190 case 8: OS << "byte ptr "; break; 6191 case 16: OS << "word ptr "; break; 6192 case 32: OS << "dword ptr "; break; 6193 case 64: OS << "qword ptr "; break; 6194 case 80: OS << "xword ptr "; break; 6195 case 128: OS << "xmmword ptr "; break; 6196 case 256: OS << "ymmword ptr "; break; 6197 } 6198 break; 6199 case AOK_Emit: 6200 OS << ".byte"; 6201 break; 6202 case AOK_Align: { 6203 // MS alignment directives are measured in bytes. If the native assembler 6204 // measures alignment in bytes, we can pass it straight through. 6205 OS << ".align"; 6206 if (getContext().getAsmInfo()->getAlignmentIsInBytes()) 6207 break; 6208 6209 // Alignment is in log2 form, so print that instead and skip the original 6210 // immediate. 6211 unsigned Val = AR.Val; 6212 OS << ' ' << Val; 6213 assert(Val < 10 && "Expected alignment less then 2^10."); 6214 AdditionalSkip = (Val < 4) ? 2 : Val < 7 ? 3 : 4; 6215 break; 6216 } 6217 case AOK_EVEN: 6218 OS << ".even"; 6219 break; 6220 case AOK_EndOfStatement: 6221 OS << "\n\t"; 6222 break; 6223 } 6224 6225 // Skip the original expression. 6226 AsmStart = Loc + AR.Len + AdditionalSkip; 6227 } 6228 6229 // Emit the remainder of the asm string. 6230 if (AsmStart != AsmEnd) 6231 OS << StringRef(AsmStart, AsmEnd - AsmStart); 6232 6233 AsmString = OS.str(); 6234 return false; 6235 } 6236 6237 bool HLASMAsmParser::parseAsHLASMLabel(ParseStatementInfo &Info, 6238 MCAsmParserSemaCallback *SI) { 6239 AsmToken LabelTok = getTok(); 6240 SMLoc LabelLoc = LabelTok.getLoc(); 6241 StringRef LabelVal; 6242 6243 if (parseIdentifier(LabelVal)) 6244 return Error(LabelLoc, "The HLASM Label has to be an Identifier"); 6245 6246 // We have validated whether the token is an Identifier. 6247 // Now we have to validate whether the token is a 6248 // valid HLASM Label. 6249 if (!getTargetParser().isLabel(LabelTok) || checkForValidSection()) 6250 return true; 6251 6252 // Lex leading spaces to get to the next operand. 6253 lexLeadingSpaces(); 6254 6255 // We shouldn't emit the label if there is nothing else after the label. 6256 // i.e asm("<token>\n") 6257 if (getTok().is(AsmToken::EndOfStatement)) 6258 return Error(LabelLoc, 6259 "Cannot have just a label for an HLASM inline asm statement"); 6260 6261 MCSymbol *Sym = getContext().getOrCreateSymbol( 6262 getContext().getAsmInfo()->shouldEmitLabelsInUpperCase() 6263 ? LabelVal.upper() 6264 : LabelVal); 6265 6266 getTargetParser().doBeforeLabelEmit(Sym, LabelLoc); 6267 6268 // Emit the label. 6269 Out.emitLabel(Sym, LabelLoc); 6270 6271 // If we are generating dwarf for assembly source files then gather the 6272 // info to make a dwarf label entry for this label if needed. 6273 if (enabledGenDwarfForAssembly()) 6274 MCGenDwarfLabelEntry::Make(Sym, &getStreamer(), getSourceManager(), 6275 LabelLoc); 6276 6277 getTargetParser().onLabelParsed(Sym); 6278 6279 return false; 6280 } 6281 6282 bool HLASMAsmParser::parseAsMachineInstruction(ParseStatementInfo &Info, 6283 MCAsmParserSemaCallback *SI) { 6284 AsmToken OperationEntryTok = Lexer.getTok(); 6285 SMLoc OperationEntryLoc = OperationEntryTok.getLoc(); 6286 StringRef OperationEntryVal; 6287 6288 // Attempt to parse the first token as an Identifier 6289 if (parseIdentifier(OperationEntryVal)) 6290 return Error(OperationEntryLoc, "unexpected token at start of statement"); 6291 6292 // Once we've parsed the operation entry successfully, lex 6293 // any spaces to get to the OperandEntries. 6294 lexLeadingSpaces(); 6295 6296 return parseAndMatchAndEmitTargetInstruction( 6297 Info, OperationEntryVal, OperationEntryTok, OperationEntryLoc); 6298 } 6299 6300 bool HLASMAsmParser::parseStatement(ParseStatementInfo &Info, 6301 MCAsmParserSemaCallback *SI) { 6302 assert(!hasPendingError() && "parseStatement started with pending error"); 6303 6304 // Should the first token be interpreted as a HLASM Label. 6305 bool ShouldParseAsHLASMLabel = false; 6306 6307 // If a Name Entry exists, it should occur at the very 6308 // start of the string. In this case, we should parse the 6309 // first non-space token as a Label. 6310 // If the Name entry is missing (i.e. there's some other 6311 // token), then we attempt to parse the first non-space 6312 // token as a Machine Instruction. 6313 if (getTok().isNot(AsmToken::Space)) 6314 ShouldParseAsHLASMLabel = true; 6315 6316 // If we have an EndOfStatement (which includes the target's comment 6317 // string) we can appropriately lex it early on) 6318 if (Lexer.is(AsmToken::EndOfStatement)) { 6319 // if this is a line comment we can drop it safely 6320 if (getTok().getString().empty() || getTok().getString().front() == '\r' || 6321 getTok().getString().front() == '\n') 6322 Out.addBlankLine(); 6323 Lex(); 6324 return false; 6325 } 6326 6327 // We have established how to parse the inline asm statement. 6328 // Now we can safely lex any leading spaces to get to the 6329 // first token. 6330 lexLeadingSpaces(); 6331 6332 // If we see a new line or carriage return as the first operand, 6333 // after lexing leading spaces, emit the new line and lex the 6334 // EndOfStatement token. 6335 if (Lexer.is(AsmToken::EndOfStatement)) { 6336 if (getTok().getString().front() == '\n' || 6337 getTok().getString().front() == '\r') { 6338 Out.addBlankLine(); 6339 Lex(); 6340 return false; 6341 } 6342 } 6343 6344 // Handle the label first if we have to before processing the rest 6345 // of the tokens as a machine instruction. 6346 if (ShouldParseAsHLASMLabel) { 6347 // If there were any errors while handling and emitting the label, 6348 // early return. 6349 if (parseAsHLASMLabel(Info, SI)) { 6350 // If we know we've failed in parsing, simply eat until end of the 6351 // statement. This ensures that we don't process any other statements. 6352 eatToEndOfStatement(); 6353 return true; 6354 } 6355 } 6356 6357 return parseAsMachineInstruction(Info, SI); 6358 } 6359 6360 namespace llvm { 6361 namespace MCParserUtils { 6362 6363 /// Returns whether the given symbol is used anywhere in the given expression, 6364 /// or subexpressions. 6365 static bool isSymbolUsedInExpression(const MCSymbol *Sym, const MCExpr *Value) { 6366 switch (Value->getKind()) { 6367 case MCExpr::Binary: { 6368 const MCBinaryExpr *BE = static_cast<const MCBinaryExpr *>(Value); 6369 return isSymbolUsedInExpression(Sym, BE->getLHS()) || 6370 isSymbolUsedInExpression(Sym, BE->getRHS()); 6371 } 6372 case MCExpr::Target: 6373 case MCExpr::Constant: 6374 return false; 6375 case MCExpr::SymbolRef: { 6376 const MCSymbol &S = 6377 static_cast<const MCSymbolRefExpr *>(Value)->getSymbol(); 6378 if (S.isVariable()) 6379 return isSymbolUsedInExpression(Sym, S.getVariableValue()); 6380 return &S == Sym; 6381 } 6382 case MCExpr::Unary: 6383 return isSymbolUsedInExpression( 6384 Sym, static_cast<const MCUnaryExpr *>(Value)->getSubExpr()); 6385 } 6386 6387 llvm_unreachable("Unknown expr kind!"); 6388 } 6389 6390 bool parseAssignmentExpression(StringRef Name, bool allow_redef, 6391 MCAsmParser &Parser, MCSymbol *&Sym, 6392 const MCExpr *&Value) { 6393 6394 // FIXME: Use better location, we should use proper tokens. 6395 SMLoc EqualLoc = Parser.getTok().getLoc(); 6396 if (Parser.parseExpression(Value)) 6397 return Parser.TokError("missing expression"); 6398 6399 // Note: we don't count b as used in "a = b". This is to allow 6400 // a = b 6401 // b = c 6402 6403 if (Parser.parseEOL()) 6404 return true; 6405 6406 // Validate that the LHS is allowed to be a variable (either it has not been 6407 // used as a symbol, or it is an absolute symbol). 6408 Sym = Parser.getContext().lookupSymbol(Name); 6409 if (Sym) { 6410 // Diagnose assignment to a label. 6411 // 6412 // FIXME: Diagnostics. Note the location of the definition as a label. 6413 // FIXME: Diagnose assignment to protected identifier (e.g., register name). 6414 if (isSymbolUsedInExpression(Sym, Value)) 6415 return Parser.Error(EqualLoc, "Recursive use of '" + Name + "'"); 6416 else if (Sym->isUndefined(/*SetUsed*/ false) && !Sym->isUsed() && 6417 !Sym->isVariable()) 6418 ; // Allow redefinitions of undefined symbols only used in directives. 6419 else if (Sym->isVariable() && !Sym->isUsed() && allow_redef) 6420 ; // Allow redefinitions of variables that haven't yet been used. 6421 else if (!Sym->isUndefined() && (!Sym->isVariable() || !allow_redef)) 6422 return Parser.Error(EqualLoc, "redefinition of '" + Name + "'"); 6423 else if (!Sym->isVariable()) 6424 return Parser.Error(EqualLoc, "invalid assignment to '" + Name + "'"); 6425 else if (!isa<MCConstantExpr>(Sym->getVariableValue())) 6426 return Parser.Error(EqualLoc, 6427 "invalid reassignment of non-absolute variable '" + 6428 Name + "'"); 6429 } else if (Name == ".") { 6430 Parser.getStreamer().emitValueToOffset(Value, 0, EqualLoc); 6431 return false; 6432 } else 6433 Sym = Parser.getContext().getOrCreateSymbol(Name); 6434 6435 Sym->setRedefinable(allow_redef); 6436 6437 return false; 6438 } 6439 6440 } // end namespace MCParserUtils 6441 } // end namespace llvm 6442 6443 /// Create an MCAsmParser instance. 6444 MCAsmParser *llvm::createMCAsmParser(SourceMgr &SM, MCContext &C, 6445 MCStreamer &Out, const MCAsmInfo &MAI, 6446 unsigned CB) { 6447 if (C.getTargetTriple().isSystemZ() && C.getTargetTriple().isOSzOS()) 6448 return new HLASMAsmParser(SM, C, Out, MAI, CB); 6449 6450 return new AsmParser(SM, C, Out, MAI, CB); 6451 } 6452