Revert "[AMDGPU][CodeGen] Do not backtrace invalid -regalloc param (#119687)"
[llvm-project.git] / llvm / lib / MC / MCParser / AsmParser.cpp
blob153c1070a68c82a6d2bbcdfc3663888481511280
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.
11 //===----------------------------------------------------------------------===//
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/MCSymbolMachO.h"
48 #include "llvm/MC/MCTargetOptions.h"
49 #include "llvm/MC/MCValue.h"
50 #include "llvm/Support/Casting.h"
51 #include "llvm/Support/CommandLine.h"
52 #include "llvm/Support/ErrorHandling.h"
53 #include "llvm/Support/MD5.h"
54 #include "llvm/Support/MathExtras.h"
55 #include "llvm/Support/MemoryBuffer.h"
56 #include "llvm/Support/SMLoc.h"
57 #include "llvm/Support/SourceMgr.h"
58 #include "llvm/Support/raw_ostream.h"
59 #include <algorithm>
60 #include <cassert>
61 #include <cctype>
62 #include <climits>
63 #include <cstddef>
64 #include <cstdint>
65 #include <deque>
66 #include <memory>
67 #include <optional>
68 #include <sstream>
69 #include <string>
70 #include <tuple>
71 #include <utility>
72 #include <vector>
74 using namespace llvm;
76 MCAsmParserSemaCallback::~MCAsmParserSemaCallback() = default;
78 namespace {
80 /// Helper types for tracking macro definitions.
81 typedef std::vector<AsmToken> MCAsmMacroArgument;
82 typedef std::vector<MCAsmMacroArgument> MCAsmMacroArguments;
84 /// Helper class for storing information about an active macro
85 /// instantiation.
86 struct MacroInstantiation {
87 /// The location of the instantiation.
88 SMLoc InstantiationLoc;
90 /// The buffer where parsing should resume upon instantiation completion.
91 unsigned ExitBuffer;
93 /// The location where parsing should resume upon instantiation completion.
94 SMLoc ExitLoc;
96 /// The depth of TheCondStack at the start of the instantiation.
97 size_t CondStackDepth;
100 struct ParseStatementInfo {
101 /// The parsed operands from the last parsed statement.
102 SmallVector<std::unique_ptr<MCParsedAsmOperand>, 8> ParsedOperands;
104 /// The opcode from the last parsed instruction.
105 unsigned Opcode = ~0U;
107 /// Was there an error parsing the inline assembly?
108 bool ParseError = false;
110 SmallVectorImpl<AsmRewrite> *AsmRewrites = nullptr;
112 ParseStatementInfo() = delete;
113 ParseStatementInfo(SmallVectorImpl<AsmRewrite> *rewrites)
114 : AsmRewrites(rewrites) {}
117 /// The concrete assembly parser instance.
118 class AsmParser : public MCAsmParser {
119 private:
120 AsmLexer Lexer;
121 MCContext &Ctx;
122 MCStreamer &Out;
123 const MCAsmInfo &MAI;
124 SourceMgr &SrcMgr;
125 SourceMgr::DiagHandlerTy SavedDiagHandler;
126 void *SavedDiagContext;
127 std::unique_ptr<MCAsmParserExtension> PlatformParser;
128 SMLoc StartTokLoc;
129 std::optional<SMLoc> CFIStartProcLoc;
131 /// This is the current buffer index we're lexing from as managed by the
132 /// SourceMgr object.
133 unsigned CurBuffer;
135 AsmCond TheCondState;
136 std::vector<AsmCond> TheCondStack;
138 /// maps directive names to handler methods in parser
139 /// extensions. Extensions register themselves in this map by calling
140 /// addDirectiveHandler.
141 StringMap<ExtensionDirectiveHandler> ExtensionDirectiveMap;
143 /// Stack of active macro instantiations.
144 std::vector<MacroInstantiation*> ActiveMacros;
146 /// List of bodies of anonymous macros.
147 std::deque<MCAsmMacro> MacroLikeBodies;
149 /// Boolean tracking whether macro substitution is enabled.
150 unsigned MacrosEnabledFlag : 1;
152 /// Keeps track of how many .macro's have been instantiated.
153 unsigned NumOfMacroInstantiations;
155 /// The values from the last parsed cpp hash file line comment if any.
156 struct CppHashInfoTy {
157 StringRef Filename;
158 int64_t LineNumber;
159 SMLoc Loc;
160 unsigned Buf;
161 CppHashInfoTy() : LineNumber(0), Buf(0) {}
163 CppHashInfoTy CppHashInfo;
165 /// Have we seen any file line comment.
166 bool HadCppHashFilename = false;
168 /// List of forward directional labels for diagnosis at the end.
169 SmallVector<std::tuple<SMLoc, CppHashInfoTy, MCSymbol *>, 4> DirLabels;
171 SmallSet<StringRef, 2> LTODiscardSymbols;
173 /// AssemblerDialect. ~OU means unset value and use value provided by MAI.
174 unsigned AssemblerDialect = ~0U;
176 /// is Darwin compatibility enabled?
177 bool IsDarwin = false;
179 /// Are we parsing ms-style inline assembly?
180 bool ParsingMSInlineAsm = false;
182 /// Did we already inform the user about inconsistent MD5 usage?
183 bool ReportedInconsistentMD5 = false;
185 // Is alt macro mode enabled.
186 bool AltMacroMode = false;
188 protected:
189 virtual bool parseStatement(ParseStatementInfo &Info,
190 MCAsmParserSemaCallback *SI);
192 /// This routine uses the target specific ParseInstruction function to
193 /// parse an instruction into Operands, and then call the target specific
194 /// MatchAndEmit function to match and emit the instruction.
195 bool parseAndMatchAndEmitTargetInstruction(ParseStatementInfo &Info,
196 StringRef IDVal, AsmToken ID,
197 SMLoc IDLoc);
199 /// Should we emit DWARF describing this assembler source? (Returns false if
200 /// the source has .file directives, which means we don't want to generate
201 /// info describing the assembler source itself.)
202 bool enabledGenDwarfForAssembly();
204 public:
205 AsmParser(SourceMgr &SM, MCContext &Ctx, MCStreamer &Out,
206 const MCAsmInfo &MAI, unsigned CB);
207 AsmParser(const AsmParser &) = delete;
208 AsmParser &operator=(const AsmParser &) = delete;
209 ~AsmParser() override;
211 bool Run(bool NoInitialTextSection, bool NoFinalize = false) override;
213 void addDirectiveHandler(StringRef Directive,
214 ExtensionDirectiveHandler Handler) override {
215 ExtensionDirectiveMap[Directive] = Handler;
218 void addAliasForDirective(StringRef Directive, StringRef Alias) override {
219 DirectiveKindMap[Directive.lower()] = DirectiveKindMap[Alias.lower()];
222 /// @name MCAsmParser Interface
223 /// {
225 SourceMgr &getSourceManager() override { return SrcMgr; }
226 MCAsmLexer &getLexer() override { return Lexer; }
227 MCContext &getContext() override { return Ctx; }
228 MCStreamer &getStreamer() override { return Out; }
230 CodeViewContext &getCVContext() { return Ctx.getCVContext(); }
232 unsigned getAssemblerDialect() override {
233 if (AssemblerDialect == ~0U)
234 return MAI.getAssemblerDialect();
235 else
236 return AssemblerDialect;
238 void setAssemblerDialect(unsigned i) override {
239 AssemblerDialect = i;
242 void Note(SMLoc L, const Twine &Msg, SMRange Range = std::nullopt) override;
243 bool Warning(SMLoc L, const Twine &Msg,
244 SMRange Range = std::nullopt) override;
245 bool printError(SMLoc L, const Twine &Msg,
246 SMRange Range = std::nullopt) override;
248 const AsmToken &Lex() override;
250 void setParsingMSInlineAsm(bool V) override {
251 ParsingMSInlineAsm = V;
252 // When parsing MS inline asm, we must lex 0b1101 and 0ABCH as binary and
253 // hex integer literals.
254 Lexer.setLexMasmIntegers(V);
256 bool isParsingMSInlineAsm() override { return ParsingMSInlineAsm; }
258 bool discardLTOSymbol(StringRef Name) const override {
259 return LTODiscardSymbols.contains(Name);
262 bool parseMSInlineAsm(std::string &AsmString, unsigned &NumOutputs,
263 unsigned &NumInputs,
264 SmallVectorImpl<std::pair<void *, bool>> &OpDecls,
265 SmallVectorImpl<std::string> &Constraints,
266 SmallVectorImpl<std::string> &Clobbers,
267 const MCInstrInfo *MII, MCInstPrinter *IP,
268 MCAsmParserSemaCallback &SI) override;
270 bool parseExpression(const MCExpr *&Res);
271 bool parseExpression(const MCExpr *&Res, SMLoc &EndLoc) override;
272 bool parsePrimaryExpr(const MCExpr *&Res, SMLoc &EndLoc,
273 AsmTypeInfo *TypeInfo) override;
274 bool parseParenExpression(const MCExpr *&Res, SMLoc &EndLoc) override;
275 bool parseParenExprOfDepth(unsigned ParenDepth, const MCExpr *&Res,
276 SMLoc &EndLoc) override;
277 bool parseAbsoluteExpression(int64_t &Res) override;
279 /// Parse a floating point expression using the float \p Semantics
280 /// and set \p Res to the value.
281 bool parseRealValue(const fltSemantics &Semantics, APInt &Res);
283 /// Parse an identifier or string (as a quoted identifier)
284 /// and set \p Res to the identifier contents.
285 bool parseIdentifier(StringRef &Res) override;
286 void eatToEndOfStatement() override;
288 bool checkForValidSection() override;
290 /// }
292 private:
293 bool parseCurlyBlockScope(SmallVectorImpl<AsmRewrite>& AsmStrRewrites);
294 bool parseCppHashLineFilenameComment(SMLoc L, bool SaveLocInfo = true);
296 void checkForBadMacro(SMLoc DirectiveLoc, StringRef Name, StringRef Body,
297 ArrayRef<MCAsmMacroParameter> Parameters);
298 bool expandMacro(raw_svector_ostream &OS, MCAsmMacro &Macro,
299 ArrayRef<MCAsmMacroParameter> Parameters,
300 ArrayRef<MCAsmMacroArgument> A, bool EnableAtPseudoVariable);
302 /// Are macros enabled in the parser?
303 bool areMacrosEnabled() {return MacrosEnabledFlag;}
305 /// Control a flag in the parser that enables or disables macros.
306 void setMacrosEnabled(bool Flag) {MacrosEnabledFlag = Flag;}
308 /// Are we inside a macro instantiation?
309 bool isInsideMacroInstantiation() {return !ActiveMacros.empty();}
311 /// Handle entry to macro instantiation.
313 /// \param M The macro.
314 /// \param NameLoc Instantiation location.
315 bool handleMacroEntry(MCAsmMacro *M, SMLoc NameLoc);
317 /// Handle exit from macro instantiation.
318 void handleMacroExit();
320 /// Extract AsmTokens for a macro argument.
321 bool parseMacroArgument(MCAsmMacroArgument &MA, bool Vararg);
323 /// Parse all macro arguments for a given macro.
324 bool parseMacroArguments(const MCAsmMacro *M, MCAsmMacroArguments &A);
326 void printMacroInstantiations();
327 void printMessage(SMLoc Loc, SourceMgr::DiagKind Kind, const Twine &Msg,
328 SMRange Range = std::nullopt) const {
329 ArrayRef<SMRange> Ranges(Range);
330 SrcMgr.PrintMessage(Loc, Kind, Msg, Ranges);
332 static void DiagHandler(const SMDiagnostic &Diag, void *Context);
334 /// Enter the specified file. This returns true on failure.
335 bool enterIncludeFile(const std::string &Filename);
337 /// Process the specified file for the .incbin directive.
338 /// This returns true on failure.
339 bool processIncbinFile(const std::string &Filename, int64_t Skip = 0,
340 const MCExpr *Count = nullptr, SMLoc Loc = SMLoc());
342 /// Reset the current lexer position to that given by \p Loc. The
343 /// current token is not set; clients should ensure Lex() is called
344 /// subsequently.
346 /// \param InBuffer If not 0, should be the known buffer id that contains the
347 /// location.
348 void jumpToLoc(SMLoc Loc, unsigned InBuffer = 0);
350 /// Parse up to the end of statement and a return the contents from the
351 /// current token until the end of the statement; the current token on exit
352 /// will be either the EndOfStatement or EOF.
353 StringRef parseStringToEndOfStatement() override;
355 /// Parse until the end of a statement or a comma is encountered,
356 /// return the contents from the current token up to the end or comma.
357 StringRef parseStringToComma();
359 enum class AssignmentKind {
360 Set,
361 Equiv,
362 Equal,
363 LTOSetConditional,
366 bool parseAssignment(StringRef Name, AssignmentKind Kind);
368 unsigned getBinOpPrecedence(AsmToken::TokenKind K,
369 MCBinaryExpr::Opcode &Kind);
371 bool parseBinOpRHS(unsigned Precedence, const MCExpr *&Res, SMLoc &EndLoc);
372 bool parseParenExpr(const MCExpr *&Res, SMLoc &EndLoc);
373 bool parseBracketExpr(const MCExpr *&Res, SMLoc &EndLoc);
375 bool parseRegisterOrRegisterNumber(int64_t &Register, SMLoc DirectiveLoc);
377 bool parseCVFunctionId(int64_t &FunctionId, StringRef DirectiveName);
378 bool parseCVFileId(int64_t &FileId, StringRef DirectiveName);
380 // Generic (target and platform independent) directive parsing.
381 enum DirectiveKind {
382 DK_NO_DIRECTIVE, // Placeholder
383 DK_SET,
384 DK_EQU,
385 DK_EQUIV,
386 DK_ASCII,
387 DK_ASCIZ,
388 DK_STRING,
389 DK_BYTE,
390 DK_SHORT,
391 DK_RELOC,
392 DK_VALUE,
393 DK_2BYTE,
394 DK_LONG,
395 DK_INT,
396 DK_4BYTE,
397 DK_QUAD,
398 DK_8BYTE,
399 DK_OCTA,
400 DK_DC,
401 DK_DC_A,
402 DK_DC_B,
403 DK_DC_D,
404 DK_DC_L,
405 DK_DC_S,
406 DK_DC_W,
407 DK_DC_X,
408 DK_DCB,
409 DK_DCB_B,
410 DK_DCB_D,
411 DK_DCB_L,
412 DK_DCB_S,
413 DK_DCB_W,
414 DK_DCB_X,
415 DK_DS,
416 DK_DS_B,
417 DK_DS_D,
418 DK_DS_L,
419 DK_DS_P,
420 DK_DS_S,
421 DK_DS_W,
422 DK_DS_X,
423 DK_SINGLE,
424 DK_FLOAT,
425 DK_DOUBLE,
426 DK_ALIGN,
427 DK_ALIGN32,
428 DK_BALIGN,
429 DK_BALIGNW,
430 DK_BALIGNL,
431 DK_P2ALIGN,
432 DK_P2ALIGNW,
433 DK_P2ALIGNL,
434 DK_ORG,
435 DK_FILL,
436 DK_ENDR,
437 DK_BUNDLE_ALIGN_MODE,
438 DK_BUNDLE_LOCK,
439 DK_BUNDLE_UNLOCK,
440 DK_ZERO,
441 DK_EXTERN,
442 DK_GLOBL,
443 DK_GLOBAL,
444 DK_LAZY_REFERENCE,
445 DK_NO_DEAD_STRIP,
446 DK_SYMBOL_RESOLVER,
447 DK_PRIVATE_EXTERN,
448 DK_REFERENCE,
449 DK_WEAK_DEFINITION,
450 DK_WEAK_REFERENCE,
451 DK_WEAK_DEF_CAN_BE_HIDDEN,
452 DK_COLD,
453 DK_COMM,
454 DK_COMMON,
455 DK_LCOMM,
456 DK_ABORT,
457 DK_INCLUDE,
458 DK_INCBIN,
459 DK_CODE16,
460 DK_CODE16GCC,
461 DK_REPT,
462 DK_IRP,
463 DK_IRPC,
464 DK_IF,
465 DK_IFEQ,
466 DK_IFGE,
467 DK_IFGT,
468 DK_IFLE,
469 DK_IFLT,
470 DK_IFNE,
471 DK_IFB,
472 DK_IFNB,
473 DK_IFC,
474 DK_IFEQS,
475 DK_IFNC,
476 DK_IFNES,
477 DK_IFDEF,
478 DK_IFNDEF,
479 DK_IFNOTDEF,
480 DK_ELSEIF,
481 DK_ELSE,
482 DK_ENDIF,
483 DK_SPACE,
484 DK_SKIP,
485 DK_FILE,
486 DK_LINE,
487 DK_LOC,
488 DK_LOC_LABEL,
489 DK_STABS,
490 DK_CV_FILE,
491 DK_CV_FUNC_ID,
492 DK_CV_INLINE_SITE_ID,
493 DK_CV_LOC,
494 DK_CV_LINETABLE,
495 DK_CV_INLINE_LINETABLE,
496 DK_CV_DEF_RANGE,
497 DK_CV_STRINGTABLE,
498 DK_CV_STRING,
499 DK_CV_FILECHECKSUMS,
500 DK_CV_FILECHECKSUM_OFFSET,
501 DK_CV_FPO_DATA,
502 DK_CFI_SECTIONS,
503 DK_CFI_STARTPROC,
504 DK_CFI_ENDPROC,
505 DK_CFI_DEF_CFA,
506 DK_CFI_DEF_CFA_OFFSET,
507 DK_CFI_ADJUST_CFA_OFFSET,
508 DK_CFI_DEF_CFA_REGISTER,
509 DK_CFI_LLVM_DEF_ASPACE_CFA,
510 DK_CFI_OFFSET,
511 DK_CFI_REL_OFFSET,
512 DK_CFI_PERSONALITY,
513 DK_CFI_LSDA,
514 DK_CFI_REMEMBER_STATE,
515 DK_CFI_RESTORE_STATE,
516 DK_CFI_SAME_VALUE,
517 DK_CFI_RESTORE,
518 DK_CFI_ESCAPE,
519 DK_CFI_RETURN_COLUMN,
520 DK_CFI_SIGNAL_FRAME,
521 DK_CFI_UNDEFINED,
522 DK_CFI_REGISTER,
523 DK_CFI_WINDOW_SAVE,
524 DK_CFI_LABEL,
525 DK_CFI_B_KEY_FRAME,
526 DK_CFI_VAL_OFFSET,
527 DK_MACROS_ON,
528 DK_MACROS_OFF,
529 DK_ALTMACRO,
530 DK_NOALTMACRO,
531 DK_MACRO,
532 DK_EXITM,
533 DK_ENDM,
534 DK_ENDMACRO,
535 DK_PURGEM,
536 DK_SLEB128,
537 DK_ULEB128,
538 DK_ERR,
539 DK_ERROR,
540 DK_WARNING,
541 DK_PRINT,
542 DK_ADDRSIG,
543 DK_ADDRSIG_SYM,
544 DK_PSEUDO_PROBE,
545 DK_LTO_DISCARD,
546 DK_LTO_SET_CONDITIONAL,
547 DK_CFI_MTE_TAGGED_FRAME,
548 DK_MEMTAG,
549 DK_END
552 /// Maps directive name --> DirectiveKind enum, for
553 /// directives parsed by this class.
554 StringMap<DirectiveKind> DirectiveKindMap;
556 // Codeview def_range type parsing.
557 enum CVDefRangeType {
558 CVDR_DEFRANGE = 0, // Placeholder
559 CVDR_DEFRANGE_REGISTER,
560 CVDR_DEFRANGE_FRAMEPOINTER_REL,
561 CVDR_DEFRANGE_SUBFIELD_REGISTER,
562 CVDR_DEFRANGE_REGISTER_REL
565 /// Maps Codeview def_range types --> CVDefRangeType enum, for
566 /// Codeview def_range types parsed by this class.
567 StringMap<CVDefRangeType> CVDefRangeTypeMap;
569 // ".ascii", ".asciz", ".string"
570 bool parseDirectiveAscii(StringRef IDVal, bool ZeroTerminated);
571 bool parseDirectiveReloc(SMLoc DirectiveLoc); // ".reloc"
572 bool parseDirectiveValue(StringRef IDVal,
573 unsigned Size); // ".byte", ".long", ...
574 bool parseDirectiveOctaValue(StringRef IDVal); // ".octa", ...
575 bool parseDirectiveRealValue(StringRef IDVal,
576 const fltSemantics &); // ".single", ...
577 bool parseDirectiveFill(); // ".fill"
578 bool parseDirectiveZero(); // ".zero"
579 // ".set", ".equ", ".equiv", ".lto_set_conditional"
580 bool parseDirectiveSet(StringRef IDVal, AssignmentKind Kind);
581 bool parseDirectiveOrg(); // ".org"
582 // ".align{,32}", ".p2align{,w,l}"
583 bool parseDirectiveAlign(bool IsPow2, unsigned ValueSize);
585 // ".file", ".line", ".loc", ".loc_label", ".stabs"
586 bool parseDirectiveFile(SMLoc DirectiveLoc);
587 bool parseDirectiveLine();
588 bool parseDirectiveLoc();
589 bool parseDirectiveLocLabel(SMLoc DirectiveLoc);
590 bool parseDirectiveStabs();
592 // ".cv_file", ".cv_func_id", ".cv_inline_site_id", ".cv_loc", ".cv_linetable",
593 // ".cv_inline_linetable", ".cv_def_range", ".cv_string"
594 bool parseDirectiveCVFile();
595 bool parseDirectiveCVFuncId();
596 bool parseDirectiveCVInlineSiteId();
597 bool parseDirectiveCVLoc();
598 bool parseDirectiveCVLinetable();
599 bool parseDirectiveCVInlineLinetable();
600 bool parseDirectiveCVDefRange();
601 bool parseDirectiveCVString();
602 bool parseDirectiveCVStringTable();
603 bool parseDirectiveCVFileChecksums();
604 bool parseDirectiveCVFileChecksumOffset();
605 bool parseDirectiveCVFPOData();
607 // .cfi directives
608 bool parseDirectiveCFIRegister(SMLoc DirectiveLoc);
609 bool parseDirectiveCFIWindowSave(SMLoc DirectiveLoc);
610 bool parseDirectiveCFISections();
611 bool parseDirectiveCFIStartProc();
612 bool parseDirectiveCFIEndProc();
613 bool parseDirectiveCFIDefCfaOffset(SMLoc DirectiveLoc);
614 bool parseDirectiveCFIDefCfa(SMLoc DirectiveLoc);
615 bool parseDirectiveCFIAdjustCfaOffset(SMLoc DirectiveLoc);
616 bool parseDirectiveCFIDefCfaRegister(SMLoc DirectiveLoc);
617 bool parseDirectiveCFILLVMDefAspaceCfa(SMLoc DirectiveLoc);
618 bool parseDirectiveCFIOffset(SMLoc DirectiveLoc);
619 bool parseDirectiveCFIRelOffset(SMLoc DirectiveLoc);
620 bool parseDirectiveCFIPersonalityOrLsda(bool IsPersonality);
621 bool parseDirectiveCFIRememberState(SMLoc DirectiveLoc);
622 bool parseDirectiveCFIRestoreState(SMLoc DirectiveLoc);
623 bool parseDirectiveCFISameValue(SMLoc DirectiveLoc);
624 bool parseDirectiveCFIRestore(SMLoc DirectiveLoc);
625 bool parseDirectiveCFIEscape(SMLoc DirectiveLoc);
626 bool parseDirectiveCFIReturnColumn(SMLoc DirectiveLoc);
627 bool parseDirectiveCFISignalFrame(SMLoc DirectiveLoc);
628 bool parseDirectiveCFIUndefined(SMLoc DirectiveLoc);
629 bool parseDirectiveCFILabel(SMLoc DirectiveLoc);
630 bool parseDirectiveCFIValOffset(SMLoc DirectiveLoc);
632 // macro directives
633 bool parseDirectivePurgeMacro(SMLoc DirectiveLoc);
634 bool parseDirectiveExitMacro(StringRef Directive);
635 bool parseDirectiveEndMacro(StringRef Directive);
636 bool parseDirectiveMacro(SMLoc DirectiveLoc);
637 bool parseDirectiveMacrosOnOff(StringRef Directive);
638 // alternate macro mode directives
639 bool parseDirectiveAltmacro(StringRef Directive);
640 // ".bundle_align_mode"
641 bool parseDirectiveBundleAlignMode();
642 // ".bundle_lock"
643 bool parseDirectiveBundleLock();
644 // ".bundle_unlock"
645 bool parseDirectiveBundleUnlock();
647 // ".space", ".skip"
648 bool parseDirectiveSpace(StringRef IDVal);
650 // ".dcb"
651 bool parseDirectiveDCB(StringRef IDVal, unsigned Size);
652 bool parseDirectiveRealDCB(StringRef IDVal, const fltSemantics &);
653 // ".ds"
654 bool parseDirectiveDS(StringRef IDVal, unsigned Size);
656 // .sleb128 (Signed=true) and .uleb128 (Signed=false)
657 bool parseDirectiveLEB128(bool Signed);
659 /// Parse a directive like ".globl" which
660 /// accepts a single symbol (which should be a label or an external).
661 bool parseDirectiveSymbolAttribute(MCSymbolAttr Attr);
663 bool parseDirectiveComm(bool IsLocal); // ".comm" and ".lcomm"
665 bool parseDirectiveAbort(SMLoc DirectiveLoc); // ".abort"
666 bool parseDirectiveInclude(); // ".include"
667 bool parseDirectiveIncbin(); // ".incbin"
669 // ".if", ".ifeq", ".ifge", ".ifgt" , ".ifle", ".iflt" or ".ifne"
670 bool parseDirectiveIf(SMLoc DirectiveLoc, DirectiveKind DirKind);
671 // ".ifb" or ".ifnb", depending on ExpectBlank.
672 bool parseDirectiveIfb(SMLoc DirectiveLoc, bool ExpectBlank);
673 // ".ifc" or ".ifnc", depending on ExpectEqual.
674 bool parseDirectiveIfc(SMLoc DirectiveLoc, bool ExpectEqual);
675 // ".ifeqs" or ".ifnes", depending on ExpectEqual.
676 bool parseDirectiveIfeqs(SMLoc DirectiveLoc, bool ExpectEqual);
677 // ".ifdef" or ".ifndef", depending on expect_defined
678 bool parseDirectiveIfdef(SMLoc DirectiveLoc, bool expect_defined);
679 bool parseDirectiveElseIf(SMLoc DirectiveLoc); // ".elseif"
680 bool parseDirectiveElse(SMLoc DirectiveLoc); // ".else"
681 bool parseDirectiveEndIf(SMLoc DirectiveLoc); // .endif
682 bool parseEscapedString(std::string &Data) override;
683 bool parseAngleBracketString(std::string &Data) override;
685 const MCExpr *applyModifierToExpr(const MCExpr *E,
686 MCSymbolRefExpr::VariantKind Variant);
688 // Macro-like directives
689 MCAsmMacro *parseMacroLikeBody(SMLoc DirectiveLoc);
690 void instantiateMacroLikeBody(MCAsmMacro *M, SMLoc DirectiveLoc,
691 raw_svector_ostream &OS);
692 bool parseDirectiveRept(SMLoc DirectiveLoc, StringRef Directive);
693 bool parseDirectiveIrp(SMLoc DirectiveLoc); // ".irp"
694 bool parseDirectiveIrpc(SMLoc DirectiveLoc); // ".irpc"
695 bool parseDirectiveEndr(SMLoc DirectiveLoc); // ".endr"
697 // "_emit" or "__emit"
698 bool parseDirectiveMSEmit(SMLoc DirectiveLoc, ParseStatementInfo &Info,
699 size_t Len);
701 // "align"
702 bool parseDirectiveMSAlign(SMLoc DirectiveLoc, ParseStatementInfo &Info);
704 // "end"
705 bool parseDirectiveEnd(SMLoc DirectiveLoc);
707 // ".err" or ".error"
708 bool parseDirectiveError(SMLoc DirectiveLoc, bool WithMessage);
710 // ".warning"
711 bool parseDirectiveWarning(SMLoc DirectiveLoc);
713 // .print <double-quotes-string>
714 bool parseDirectivePrint(SMLoc DirectiveLoc);
716 // .pseudoprobe
717 bool parseDirectivePseudoProbe();
719 // ".lto_discard"
720 bool parseDirectiveLTODiscard();
722 // Directives to support address-significance tables.
723 bool parseDirectiveAddrsig();
724 bool parseDirectiveAddrsigSym();
726 void initializeDirectiveKindMap();
727 void initializeCVDefRangeTypeMap();
730 class HLASMAsmParser final : public AsmParser {
731 private:
732 MCAsmLexer &Lexer;
733 MCStreamer &Out;
735 void lexLeadingSpaces() {
736 while (Lexer.is(AsmToken::Space))
737 Lexer.Lex();
740 bool parseAsHLASMLabel(ParseStatementInfo &Info, MCAsmParserSemaCallback *SI);
741 bool parseAsMachineInstruction(ParseStatementInfo &Info,
742 MCAsmParserSemaCallback *SI);
744 public:
745 HLASMAsmParser(SourceMgr &SM, MCContext &Ctx, MCStreamer &Out,
746 const MCAsmInfo &MAI, unsigned CB = 0)
747 : AsmParser(SM, Ctx, Out, MAI, CB), Lexer(getLexer()), Out(Out) {
748 Lexer.setSkipSpace(false);
749 Lexer.setAllowHashInIdentifier(true);
750 Lexer.setLexHLASMIntegers(true);
751 Lexer.setLexHLASMStrings(true);
754 ~HLASMAsmParser() { Lexer.setSkipSpace(true); }
756 bool parseStatement(ParseStatementInfo &Info,
757 MCAsmParserSemaCallback *SI) override;
760 } // end anonymous namespace
762 namespace llvm {
764 extern cl::opt<unsigned> AsmMacroMaxNestingDepth;
766 extern MCAsmParserExtension *createDarwinAsmParser();
767 extern MCAsmParserExtension *createELFAsmParser();
768 extern MCAsmParserExtension *createCOFFAsmParser();
769 extern MCAsmParserExtension *createGOFFAsmParser();
770 extern MCAsmParserExtension *createXCOFFAsmParser();
771 extern MCAsmParserExtension *createWasmAsmParser();
773 } // end namespace llvm
775 enum { DEFAULT_ADDRSPACE = 0 };
777 AsmParser::AsmParser(SourceMgr &SM, MCContext &Ctx, MCStreamer &Out,
778 const MCAsmInfo &MAI, unsigned CB = 0)
779 : Lexer(MAI), Ctx(Ctx), Out(Out), MAI(MAI), SrcMgr(SM),
780 CurBuffer(CB ? CB : SM.getMainFileID()), MacrosEnabledFlag(true) {
781 HadError = false;
782 // Save the old handler.
783 SavedDiagHandler = SrcMgr.getDiagHandler();
784 SavedDiagContext = SrcMgr.getDiagContext();
785 // Set our own handler which calls the saved handler.
786 SrcMgr.setDiagHandler(DiagHandler, this);
787 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer());
788 // Make MCStreamer aware of the StartTokLoc for locations in diagnostics.
789 Out.setStartTokLocPtr(&StartTokLoc);
791 // Initialize the platform / file format parser.
792 switch (Ctx.getObjectFileType()) {
793 case MCContext::IsCOFF:
794 PlatformParser.reset(createCOFFAsmParser());
795 break;
796 case MCContext::IsMachO:
797 PlatformParser.reset(createDarwinAsmParser());
798 IsDarwin = true;
799 break;
800 case MCContext::IsELF:
801 PlatformParser.reset(createELFAsmParser());
802 break;
803 case MCContext::IsGOFF:
804 PlatformParser.reset(createGOFFAsmParser());
805 break;
806 case MCContext::IsSPIRV:
807 report_fatal_error(
808 "Need to implement createSPIRVAsmParser for SPIRV format.");
809 break;
810 case MCContext::IsWasm:
811 PlatformParser.reset(createWasmAsmParser());
812 break;
813 case MCContext::IsXCOFF:
814 PlatformParser.reset(createXCOFFAsmParser());
815 break;
816 case MCContext::IsDXContainer:
817 report_fatal_error("DXContainer is not supported yet");
818 break;
821 PlatformParser->Initialize(*this);
822 initializeDirectiveKindMap();
823 initializeCVDefRangeTypeMap();
825 NumOfMacroInstantiations = 0;
828 AsmParser::~AsmParser() {
829 assert((HadError || ActiveMacros.empty()) &&
830 "Unexpected active macro instantiation!");
832 // Remove MCStreamer's reference to the parser SMLoc.
833 Out.setStartTokLocPtr(nullptr);
834 // Restore the saved diagnostics handler and context for use during
835 // finalization.
836 SrcMgr.setDiagHandler(SavedDiagHandler, SavedDiagContext);
839 void AsmParser::printMacroInstantiations() {
840 // Print the active macro instantiation stack.
841 for (MacroInstantiation *M : reverse(ActiveMacros))
842 printMessage(M->InstantiationLoc, SourceMgr::DK_Note,
843 "while in macro instantiation");
846 void AsmParser::Note(SMLoc L, const Twine &Msg, SMRange Range) {
847 printPendingErrors();
848 printMessage(L, SourceMgr::DK_Note, Msg, Range);
849 printMacroInstantiations();
852 bool AsmParser::Warning(SMLoc L, const Twine &Msg, SMRange Range) {
853 if(getTargetParser().getTargetOptions().MCNoWarn)
854 return false;
855 if (getTargetParser().getTargetOptions().MCFatalWarnings)
856 return Error(L, Msg, Range);
857 printMessage(L, SourceMgr::DK_Warning, Msg, Range);
858 printMacroInstantiations();
859 return false;
862 bool AsmParser::printError(SMLoc L, const Twine &Msg, SMRange Range) {
863 HadError = true;
864 printMessage(L, SourceMgr::DK_Error, Msg, Range);
865 printMacroInstantiations();
866 return true;
869 bool AsmParser::enterIncludeFile(const std::string &Filename) {
870 std::string IncludedFile;
871 unsigned NewBuf =
872 SrcMgr.AddIncludeFile(Filename, Lexer.getLoc(), IncludedFile);
873 if (!NewBuf)
874 return true;
876 CurBuffer = NewBuf;
877 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer());
878 return false;
881 /// Process the specified .incbin file by searching for it in the include paths
882 /// then just emitting the byte contents of the file to the streamer. This
883 /// returns true on failure.
884 bool AsmParser::processIncbinFile(const std::string &Filename, int64_t Skip,
885 const MCExpr *Count, SMLoc Loc) {
886 std::string IncludedFile;
887 unsigned NewBuf =
888 SrcMgr.AddIncludeFile(Filename, Lexer.getLoc(), IncludedFile);
889 if (!NewBuf)
890 return true;
892 // Pick up the bytes from the file and emit them.
893 StringRef Bytes = SrcMgr.getMemoryBuffer(NewBuf)->getBuffer();
894 Bytes = Bytes.drop_front(Skip);
895 if (Count) {
896 int64_t Res;
897 if (!Count->evaluateAsAbsolute(Res, getStreamer().getAssemblerPtr()))
898 return Error(Loc, "expected absolute expression");
899 if (Res < 0)
900 return Warning(Loc, "negative count has no effect");
901 Bytes = Bytes.take_front(Res);
903 getStreamer().emitBytes(Bytes);
904 return false;
907 void AsmParser::jumpToLoc(SMLoc Loc, unsigned InBuffer) {
908 CurBuffer = InBuffer ? InBuffer : SrcMgr.FindBufferContainingLoc(Loc);
909 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer(),
910 Loc.getPointer());
913 const AsmToken &AsmParser::Lex() {
914 if (Lexer.getTok().is(AsmToken::Error))
915 Error(Lexer.getErrLoc(), Lexer.getErr());
917 // if it's a end of statement with a comment in it
918 if (getTok().is(AsmToken::EndOfStatement)) {
919 // if this is a line comment output it.
920 if (!getTok().getString().empty() && getTok().getString().front() != '\n' &&
921 getTok().getString().front() != '\r' && MAI.preserveAsmComments())
922 Out.addExplicitComment(Twine(getTok().getString()));
925 const AsmToken *tok = &Lexer.Lex();
927 // Parse comments here to be deferred until end of next statement.
928 while (tok->is(AsmToken::Comment)) {
929 if (MAI.preserveAsmComments())
930 Out.addExplicitComment(Twine(tok->getString()));
931 tok = &Lexer.Lex();
934 if (tok->is(AsmToken::Eof)) {
935 // If this is the end of an included file, pop the parent file off the
936 // include stack.
937 SMLoc ParentIncludeLoc = SrcMgr.getParentIncludeLoc(CurBuffer);
938 if (ParentIncludeLoc != SMLoc()) {
939 jumpToLoc(ParentIncludeLoc);
940 return Lex();
944 return *tok;
947 bool AsmParser::enabledGenDwarfForAssembly() {
948 // Check whether the user specified -g.
949 if (!getContext().getGenDwarfForAssembly())
950 return false;
951 // If we haven't encountered any .file directives (which would imply that
952 // the assembler source was produced with debug info already) then emit one
953 // describing the assembler source file itself.
954 if (getContext().getGenDwarfFileNumber() == 0) {
955 const MCDwarfFile &RootFile =
956 getContext().getMCDwarfLineTable(/*CUID=*/0).getRootFile();
957 getContext().setGenDwarfFileNumber(getStreamer().emitDwarfFileDirective(
958 /*CUID=*/0, getContext().getCompilationDir(), RootFile.Name,
959 RootFile.Checksum, RootFile.Source));
961 return true;
964 bool AsmParser::Run(bool NoInitialTextSection, bool NoFinalize) {
965 LTODiscardSymbols.clear();
967 // Create the initial section, if requested.
968 if (!NoInitialTextSection)
969 Out.initSections(false, getTargetParser().getSTI());
971 // Prime the lexer.
972 Lex();
974 HadError = false;
975 AsmCond StartingCondState = TheCondState;
976 SmallVector<AsmRewrite, 4> AsmStrRewrites;
978 // If we are generating dwarf for assembly source files save the initial text
979 // section. (Don't use enabledGenDwarfForAssembly() here, as we aren't
980 // emitting any actual debug info yet and haven't had a chance to parse any
981 // embedded .file directives.)
982 if (getContext().getGenDwarfForAssembly()) {
983 MCSection *Sec = getStreamer().getCurrentSectionOnly();
984 if (!Sec->getBeginSymbol()) {
985 MCSymbol *SectionStartSym = getContext().createTempSymbol();
986 getStreamer().emitLabel(SectionStartSym);
987 Sec->setBeginSymbol(SectionStartSym);
989 bool InsertResult = getContext().addGenDwarfSection(Sec);
990 assert(InsertResult && ".text section should not have debug info yet");
991 (void)InsertResult;
994 getTargetParser().onBeginOfFile();
996 // While we have input, parse each statement.
997 while (Lexer.isNot(AsmToken::Eof)) {
998 ParseStatementInfo Info(&AsmStrRewrites);
999 bool Parsed = parseStatement(Info, nullptr);
1001 // If we have a Lexer Error we are on an Error Token. Load in Lexer Error
1002 // for printing ErrMsg via Lex() only if no (presumably better) parser error
1003 // exists.
1004 if (Parsed && !hasPendingError() && Lexer.getTok().is(AsmToken::Error)) {
1005 Lex();
1008 // parseStatement returned true so may need to emit an error.
1009 printPendingErrors();
1011 // Skipping to the next line if needed.
1012 if (Parsed && !getLexer().isAtStartOfStatement())
1013 eatToEndOfStatement();
1016 getTargetParser().onEndOfFile();
1017 printPendingErrors();
1019 // All errors should have been emitted.
1020 assert(!hasPendingError() && "unexpected error from parseStatement");
1022 getTargetParser().flushPendingInstructions(getStreamer());
1024 if (TheCondState.TheCond != StartingCondState.TheCond ||
1025 TheCondState.Ignore != StartingCondState.Ignore)
1026 printError(getTok().getLoc(), "unmatched .ifs or .elses");
1027 // Check to see there are no empty DwarfFile slots.
1028 const auto &LineTables = getContext().getMCDwarfLineTables();
1029 if (!LineTables.empty()) {
1030 unsigned Index = 0;
1031 for (const auto &File : LineTables.begin()->second.getMCDwarfFiles()) {
1032 if (File.Name.empty() && Index != 0)
1033 printError(getTok().getLoc(), "unassigned file number: " +
1034 Twine(Index) +
1035 " for .file directives");
1036 ++Index;
1040 // Check to see that all assembler local symbols were actually defined.
1041 // Targets that don't do subsections via symbols may not want this, though,
1042 // so conservatively exclude them. Only do this if we're finalizing, though,
1043 // as otherwise we won't necessarily have seen everything yet.
1044 if (!NoFinalize) {
1045 if (MAI.hasSubsectionsViaSymbols()) {
1046 for (const auto &TableEntry : getContext().getSymbols()) {
1047 MCSymbol *Sym = TableEntry.getValue().Symbol;
1048 // Variable symbols may not be marked as defined, so check those
1049 // explicitly. If we know it's a variable, we have a definition for
1050 // the purposes of this check.
1051 if (Sym && Sym->isTemporary() && !Sym->isVariable() &&
1052 !Sym->isDefined())
1053 // FIXME: We would really like to refer back to where the symbol was
1054 // first referenced for a source location. We need to add something
1055 // to track that. Currently, we just point to the end of the file.
1056 printError(getTok().getLoc(), "assembler local symbol '" +
1057 Sym->getName() + "' not defined");
1061 // Temporary symbols like the ones for directional jumps don't go in the
1062 // symbol table. They also need to be diagnosed in all (final) cases.
1063 for (std::tuple<SMLoc, CppHashInfoTy, MCSymbol *> &LocSym : DirLabels) {
1064 if (std::get<2>(LocSym)->isUndefined()) {
1065 // Reset the state of any "# line file" directives we've seen to the
1066 // context as it was at the diagnostic site.
1067 CppHashInfo = std::get<1>(LocSym);
1068 printError(std::get<0>(LocSym), "directional label undefined");
1072 // Finalize the output stream if there are no errors and if the client wants
1073 // us to.
1074 if (!HadError && !NoFinalize) {
1075 if (auto *TS = Out.getTargetStreamer())
1076 TS->emitConstantPools();
1078 Out.finish(Lexer.getLoc());
1081 return HadError || getContext().hadError();
1084 bool AsmParser::checkForValidSection() {
1085 if (!ParsingMSInlineAsm && !getStreamer().getCurrentFragment()) {
1086 Out.initSections(false, getTargetParser().getSTI());
1087 return Error(getTok().getLoc(),
1088 "expected section directive before assembly directive");
1090 return false;
1093 /// Throw away the rest of the line for testing purposes.
1094 void AsmParser::eatToEndOfStatement() {
1095 while (Lexer.isNot(AsmToken::EndOfStatement) && Lexer.isNot(AsmToken::Eof))
1096 Lexer.Lex();
1098 // Eat EOL.
1099 if (Lexer.is(AsmToken::EndOfStatement))
1100 Lexer.Lex();
1103 StringRef AsmParser::parseStringToEndOfStatement() {
1104 const char *Start = getTok().getLoc().getPointer();
1106 while (Lexer.isNot(AsmToken::EndOfStatement) && Lexer.isNot(AsmToken::Eof))
1107 Lexer.Lex();
1109 const char *End = getTok().getLoc().getPointer();
1110 return StringRef(Start, End - Start);
1113 StringRef AsmParser::parseStringToComma() {
1114 const char *Start = getTok().getLoc().getPointer();
1116 while (Lexer.isNot(AsmToken::EndOfStatement) &&
1117 Lexer.isNot(AsmToken::Comma) && Lexer.isNot(AsmToken::Eof))
1118 Lexer.Lex();
1120 const char *End = getTok().getLoc().getPointer();
1121 return StringRef(Start, End - Start);
1124 /// Parse a paren expression and return it.
1125 /// NOTE: This assumes the leading '(' has already been consumed.
1127 /// parenexpr ::= expr)
1129 bool AsmParser::parseParenExpr(const MCExpr *&Res, SMLoc &EndLoc) {
1130 if (parseExpression(Res))
1131 return true;
1132 EndLoc = Lexer.getTok().getEndLoc();
1133 return parseRParen();
1136 /// Parse a bracket expression and return it.
1137 /// NOTE: This assumes the leading '[' has already been consumed.
1139 /// bracketexpr ::= expr]
1141 bool AsmParser::parseBracketExpr(const MCExpr *&Res, SMLoc &EndLoc) {
1142 if (parseExpression(Res))
1143 return true;
1144 EndLoc = getTok().getEndLoc();
1145 if (parseToken(AsmToken::RBrac, "expected ']' in brackets expression"))
1146 return true;
1147 return false;
1150 /// Parse a primary expression and return it.
1151 /// primaryexpr ::= (parenexpr
1152 /// primaryexpr ::= symbol
1153 /// primaryexpr ::= number
1154 /// primaryexpr ::= '.'
1155 /// primaryexpr ::= ~,+,- primaryexpr
1156 bool AsmParser::parsePrimaryExpr(const MCExpr *&Res, SMLoc &EndLoc,
1157 AsmTypeInfo *TypeInfo) {
1158 SMLoc FirstTokenLoc = getLexer().getLoc();
1159 AsmToken::TokenKind FirstTokenKind = Lexer.getKind();
1160 switch (FirstTokenKind) {
1161 default:
1162 return TokError("unknown token in expression");
1163 // If we have an error assume that we've already handled it.
1164 case AsmToken::Error:
1165 return true;
1166 case AsmToken::Exclaim:
1167 Lex(); // Eat the operator.
1168 if (parsePrimaryExpr(Res, EndLoc, TypeInfo))
1169 return true;
1170 Res = MCUnaryExpr::createLNot(Res, getContext(), FirstTokenLoc);
1171 return false;
1172 case AsmToken::Dollar:
1173 case AsmToken::Star:
1174 case AsmToken::At:
1175 case AsmToken::String:
1176 case AsmToken::Identifier: {
1177 StringRef Identifier;
1178 if (parseIdentifier(Identifier)) {
1179 // We may have failed but '$'|'*' may be a valid token in context of
1180 // the current PC.
1181 if (getTok().is(AsmToken::Dollar) || getTok().is(AsmToken::Star)) {
1182 bool ShouldGenerateTempSymbol = false;
1183 if ((getTok().is(AsmToken::Dollar) && MAI.getDollarIsPC()) ||
1184 (getTok().is(AsmToken::Star) && MAI.getStarIsPC()))
1185 ShouldGenerateTempSymbol = true;
1187 if (!ShouldGenerateTempSymbol)
1188 return Error(FirstTokenLoc, "invalid token in expression");
1190 // Eat the '$'|'*' token.
1191 Lex();
1192 // This is either a '$'|'*' reference, which references the current PC.
1193 // Emit a temporary label to the streamer and refer to it.
1194 MCSymbol *Sym = Ctx.createTempSymbol();
1195 Out.emitLabel(Sym);
1196 Res = MCSymbolRefExpr::create(Sym, MCSymbolRefExpr::VK_None,
1197 getContext());
1198 EndLoc = FirstTokenLoc;
1199 return false;
1202 // Parse symbol variant
1203 std::pair<StringRef, StringRef> Split;
1204 if (!MAI.useParensForSymbolVariant()) {
1205 if (FirstTokenKind == AsmToken::String) {
1206 if (Lexer.is(AsmToken::At)) {
1207 Lex(); // eat @
1208 SMLoc AtLoc = getLexer().getLoc();
1209 StringRef VName;
1210 if (parseIdentifier(VName))
1211 return Error(AtLoc, "expected symbol variant after '@'");
1213 Split = std::make_pair(Identifier, VName);
1215 } else {
1216 Split = Identifier.split('@');
1218 } else if (Lexer.is(AsmToken::LParen)) {
1219 Lex(); // eat '('.
1220 StringRef VName;
1221 parseIdentifier(VName);
1222 if (parseRParen())
1223 return true;
1224 Split = std::make_pair(Identifier, VName);
1227 EndLoc = SMLoc::getFromPointer(Identifier.end());
1229 // This is a symbol reference.
1230 StringRef SymbolName = Identifier;
1231 if (SymbolName.empty())
1232 return Error(getLexer().getLoc(), "expected a symbol reference");
1234 MCSymbolRefExpr::VariantKind Variant = MCSymbolRefExpr::VK_None;
1236 // Lookup the symbol variant if used.
1237 if (!Split.second.empty()) {
1238 Variant = getTargetParser().getVariantKindForName(Split.second);
1239 if (Variant != MCSymbolRefExpr::VK_Invalid) {
1240 SymbolName = Split.first;
1241 } else if (MAI.doesAllowAtInName() && !MAI.useParensForSymbolVariant()) {
1242 Variant = MCSymbolRefExpr::VK_None;
1243 } else {
1244 return Error(SMLoc::getFromPointer(Split.second.begin()),
1245 "invalid variant '" + Split.second + "'");
1249 MCSymbol *Sym = getContext().getInlineAsmLabel(SymbolName);
1250 if (!Sym)
1251 Sym = getContext().getOrCreateSymbol(
1252 MAI.shouldEmitLabelsInUpperCase() ? SymbolName.upper() : SymbolName);
1254 // If this is an absolute variable reference, substitute it now to preserve
1255 // semantics in the face of reassignment.
1256 if (Sym->isVariable()) {
1257 auto V = Sym->getVariableValue(/*SetUsed*/ false);
1258 bool DoInline = isa<MCConstantExpr>(V) && !Variant;
1259 if (auto TV = dyn_cast<MCTargetExpr>(V))
1260 DoInline = TV->inlineAssignedExpr();
1261 if (DoInline) {
1262 if (Variant)
1263 return Error(EndLoc, "unexpected modifier on variable reference");
1264 Res = Sym->getVariableValue(/*SetUsed*/ false);
1265 return false;
1269 // Otherwise create a symbol ref.
1270 Res = MCSymbolRefExpr::create(Sym, Variant, getContext(), FirstTokenLoc);
1271 return false;
1273 case AsmToken::BigNum:
1274 return TokError("literal value out of range for directive");
1275 case AsmToken::Integer: {
1276 SMLoc Loc = getTok().getLoc();
1277 int64_t IntVal = getTok().getIntVal();
1278 Res = MCConstantExpr::create(IntVal, getContext());
1279 EndLoc = Lexer.getTok().getEndLoc();
1280 Lex(); // Eat token.
1281 // Look for 'b' or 'f' following an Integer as a directional label
1282 if (Lexer.getKind() == AsmToken::Identifier) {
1283 StringRef IDVal = getTok().getString();
1284 // Lookup the symbol variant if used.
1285 std::pair<StringRef, StringRef> Split = IDVal.split('@');
1286 MCSymbolRefExpr::VariantKind Variant = MCSymbolRefExpr::VK_None;
1287 if (Split.first.size() != IDVal.size()) {
1288 Variant = MCSymbolRefExpr::getVariantKindForName(Split.second);
1289 if (Variant == MCSymbolRefExpr::VK_Invalid)
1290 return TokError("invalid variant '" + Split.second + "'");
1291 IDVal = Split.first;
1293 if (IDVal == "f" || IDVal == "b") {
1294 MCSymbol *Sym =
1295 Ctx.getDirectionalLocalSymbol(IntVal, IDVal == "b");
1296 Res = MCSymbolRefExpr::create(Sym, Variant, getContext());
1297 if (IDVal == "b" && Sym->isUndefined())
1298 return Error(Loc, "directional label undefined");
1299 DirLabels.push_back(std::make_tuple(Loc, CppHashInfo, Sym));
1300 EndLoc = Lexer.getTok().getEndLoc();
1301 Lex(); // Eat identifier.
1304 return false;
1306 case AsmToken::Real: {
1307 APFloat RealVal(APFloat::IEEEdouble(), getTok().getString());
1308 uint64_t IntVal = RealVal.bitcastToAPInt().getZExtValue();
1309 Res = MCConstantExpr::create(IntVal, getContext());
1310 EndLoc = Lexer.getTok().getEndLoc();
1311 Lex(); // Eat token.
1312 return false;
1314 case AsmToken::Dot: {
1315 if (!MAI.getDotIsPC())
1316 return TokError("cannot use . as current PC");
1318 // This is a '.' reference, which references the current PC. Emit a
1319 // temporary label to the streamer and refer to it.
1320 MCSymbol *Sym = Ctx.createTempSymbol();
1321 Out.emitLabel(Sym);
1322 Res = MCSymbolRefExpr::create(Sym, MCSymbolRefExpr::VK_None, getContext());
1323 EndLoc = Lexer.getTok().getEndLoc();
1324 Lex(); // Eat identifier.
1325 return false;
1327 case AsmToken::LParen:
1328 Lex(); // Eat the '('.
1329 return parseParenExpr(Res, EndLoc);
1330 case AsmToken::LBrac:
1331 if (!PlatformParser->HasBracketExpressions())
1332 return TokError("brackets expression not supported on this target");
1333 Lex(); // Eat the '['.
1334 return parseBracketExpr(Res, EndLoc);
1335 case AsmToken::Minus:
1336 Lex(); // Eat the operator.
1337 if (parsePrimaryExpr(Res, EndLoc, TypeInfo))
1338 return true;
1339 Res = MCUnaryExpr::createMinus(Res, getContext(), FirstTokenLoc);
1340 return false;
1341 case AsmToken::Plus:
1342 Lex(); // Eat the operator.
1343 if (parsePrimaryExpr(Res, EndLoc, TypeInfo))
1344 return true;
1345 Res = MCUnaryExpr::createPlus(Res, getContext(), FirstTokenLoc);
1346 return false;
1347 case AsmToken::Tilde:
1348 Lex(); // Eat the operator.
1349 if (parsePrimaryExpr(Res, EndLoc, TypeInfo))
1350 return true;
1351 Res = MCUnaryExpr::createNot(Res, getContext(), FirstTokenLoc);
1352 return false;
1353 // MIPS unary expression operators. The lexer won't generate these tokens if
1354 // MCAsmInfo::HasMipsExpressions is false for the target.
1355 case AsmToken::PercentCall16:
1356 case AsmToken::PercentCall_Hi:
1357 case AsmToken::PercentCall_Lo:
1358 case AsmToken::PercentDtprel_Hi:
1359 case AsmToken::PercentDtprel_Lo:
1360 case AsmToken::PercentGot:
1361 case AsmToken::PercentGot_Disp:
1362 case AsmToken::PercentGot_Hi:
1363 case AsmToken::PercentGot_Lo:
1364 case AsmToken::PercentGot_Ofst:
1365 case AsmToken::PercentGot_Page:
1366 case AsmToken::PercentGottprel:
1367 case AsmToken::PercentGp_Rel:
1368 case AsmToken::PercentHi:
1369 case AsmToken::PercentHigher:
1370 case AsmToken::PercentHighest:
1371 case AsmToken::PercentLo:
1372 case AsmToken::PercentNeg:
1373 case AsmToken::PercentPcrel_Hi:
1374 case AsmToken::PercentPcrel_Lo:
1375 case AsmToken::PercentTlsgd:
1376 case AsmToken::PercentTlsldm:
1377 case AsmToken::PercentTprel_Hi:
1378 case AsmToken::PercentTprel_Lo:
1379 Lex(); // Eat the operator.
1380 if (Lexer.isNot(AsmToken::LParen))
1381 return TokError("expected '(' after operator");
1382 Lex(); // Eat the operator.
1383 if (parseExpression(Res, EndLoc))
1384 return true;
1385 if (parseRParen())
1386 return true;
1387 Res = getTargetParser().createTargetUnaryExpr(Res, FirstTokenKind, Ctx);
1388 return !Res;
1392 bool AsmParser::parseExpression(const MCExpr *&Res) {
1393 SMLoc EndLoc;
1394 return parseExpression(Res, EndLoc);
1397 const MCExpr *
1398 AsmParser::applyModifierToExpr(const MCExpr *E,
1399 MCSymbolRefExpr::VariantKind Variant) {
1400 // Ask the target implementation about this expression first.
1401 const MCExpr *NewE = getTargetParser().applyModifierToExpr(E, Variant, Ctx);
1402 if (NewE)
1403 return NewE;
1404 // Recurse over the given expression, rebuilding it to apply the given variant
1405 // if there is exactly one symbol.
1406 switch (E->getKind()) {
1407 case MCExpr::Target:
1408 case MCExpr::Constant:
1409 return nullptr;
1411 case MCExpr::SymbolRef: {
1412 const MCSymbolRefExpr *SRE = cast<MCSymbolRefExpr>(E);
1414 if (SRE->getKind() != MCSymbolRefExpr::VK_None) {
1415 TokError("invalid variant on expression '" + getTok().getIdentifier() +
1416 "' (already modified)");
1417 return E;
1420 return MCSymbolRefExpr::create(&SRE->getSymbol(), Variant, getContext());
1423 case MCExpr::Unary: {
1424 const MCUnaryExpr *UE = cast<MCUnaryExpr>(E);
1425 const MCExpr *Sub = applyModifierToExpr(UE->getSubExpr(), Variant);
1426 if (!Sub)
1427 return nullptr;
1428 return MCUnaryExpr::create(UE->getOpcode(), Sub, getContext());
1431 case MCExpr::Binary: {
1432 const MCBinaryExpr *BE = cast<MCBinaryExpr>(E);
1433 const MCExpr *LHS = applyModifierToExpr(BE->getLHS(), Variant);
1434 const MCExpr *RHS = applyModifierToExpr(BE->getRHS(), Variant);
1436 if (!LHS && !RHS)
1437 return nullptr;
1439 if (!LHS)
1440 LHS = BE->getLHS();
1441 if (!RHS)
1442 RHS = BE->getRHS();
1444 return MCBinaryExpr::create(BE->getOpcode(), LHS, RHS, getContext());
1448 llvm_unreachable("Invalid expression kind!");
1451 /// This function checks if the next token is <string> type or arithmetic.
1452 /// string that begin with character '<' must end with character '>'.
1453 /// otherwise it is arithmetics.
1454 /// If the function returns a 'true' value,
1455 /// the End argument will be filled with the last location pointed to the '>'
1456 /// character.
1458 /// There is a gap between the AltMacro's documentation and the single quote
1459 /// implementation. GCC does not fully support this feature and so we will not
1460 /// support it.
1461 /// TODO: Adding single quote as a string.
1462 static bool isAngleBracketString(SMLoc &StrLoc, SMLoc &EndLoc) {
1463 assert((StrLoc.getPointer() != nullptr) &&
1464 "Argument to the function cannot be a NULL value");
1465 const char *CharPtr = StrLoc.getPointer();
1466 while ((*CharPtr != '>') && (*CharPtr != '\n') && (*CharPtr != '\r') &&
1467 (*CharPtr != '\0')) {
1468 if (*CharPtr == '!')
1469 CharPtr++;
1470 CharPtr++;
1472 if (*CharPtr == '>') {
1473 EndLoc = StrLoc.getFromPointer(CharPtr + 1);
1474 return true;
1476 return false;
1479 /// creating a string without the escape characters '!'.
1480 static std::string angleBracketString(StringRef AltMacroStr) {
1481 std::string Res;
1482 for (size_t Pos = 0; Pos < AltMacroStr.size(); Pos++) {
1483 if (AltMacroStr[Pos] == '!')
1484 Pos++;
1485 Res += AltMacroStr[Pos];
1487 return Res;
1490 /// Parse an expression and return it.
1492 /// expr ::= expr &&,|| expr -> lowest.
1493 /// expr ::= expr |,^,&,! expr
1494 /// expr ::= expr ==,!=,<>,<,<=,>,>= expr
1495 /// expr ::= expr <<,>> expr
1496 /// expr ::= expr +,- expr
1497 /// expr ::= expr *,/,% expr -> highest.
1498 /// expr ::= primaryexpr
1500 bool AsmParser::parseExpression(const MCExpr *&Res, SMLoc &EndLoc) {
1501 // Parse the expression.
1502 Res = nullptr;
1503 if (getTargetParser().parsePrimaryExpr(Res, EndLoc) ||
1504 parseBinOpRHS(1, Res, EndLoc))
1505 return true;
1507 // As a special case, we support 'a op b @ modifier' by rewriting the
1508 // expression to include the modifier. This is inefficient, but in general we
1509 // expect users to use 'a@modifier op b'.
1510 if (parseOptionalToken(AsmToken::At)) {
1511 if (Lexer.isNot(AsmToken::Identifier))
1512 return TokError("unexpected symbol modifier following '@'");
1514 MCSymbolRefExpr::VariantKind Variant =
1515 MCSymbolRefExpr::getVariantKindForName(getTok().getIdentifier());
1516 if (Variant == MCSymbolRefExpr::VK_Invalid)
1517 return TokError("invalid variant '" + getTok().getIdentifier() + "'");
1519 const MCExpr *ModifiedRes = applyModifierToExpr(Res, Variant);
1520 if (!ModifiedRes) {
1521 return TokError("invalid modifier '" + getTok().getIdentifier() +
1522 "' (no symbols present)");
1525 Res = ModifiedRes;
1526 Lex();
1529 // Try to constant fold it up front, if possible. Do not exploit
1530 // assembler here.
1531 int64_t Value;
1532 if (Res->evaluateAsAbsolute(Value))
1533 Res = MCConstantExpr::create(Value, getContext());
1535 return false;
1538 bool AsmParser::parseParenExpression(const MCExpr *&Res, SMLoc &EndLoc) {
1539 Res = nullptr;
1540 return parseParenExpr(Res, EndLoc) || parseBinOpRHS(1, Res, EndLoc);
1543 bool AsmParser::parseParenExprOfDepth(unsigned ParenDepth, const MCExpr *&Res,
1544 SMLoc &EndLoc) {
1545 if (parseParenExpr(Res, EndLoc))
1546 return true;
1548 for (; ParenDepth > 0; --ParenDepth) {
1549 if (parseBinOpRHS(1, Res, EndLoc))
1550 return true;
1552 // We don't Lex() the last RParen.
1553 // This is the same behavior as parseParenExpression().
1554 if (ParenDepth - 1 > 0) {
1555 EndLoc = getTok().getEndLoc();
1556 if (parseRParen())
1557 return true;
1560 return false;
1563 bool AsmParser::parseAbsoluteExpression(int64_t &Res) {
1564 const MCExpr *Expr;
1566 SMLoc StartLoc = Lexer.getLoc();
1567 if (parseExpression(Expr))
1568 return true;
1570 if (!Expr->evaluateAsAbsolute(Res, getStreamer().getAssemblerPtr()))
1571 return Error(StartLoc, "expected absolute expression");
1573 return false;
1576 static unsigned getDarwinBinOpPrecedence(AsmToken::TokenKind K,
1577 MCBinaryExpr::Opcode &Kind,
1578 bool ShouldUseLogicalShr) {
1579 switch (K) {
1580 default:
1581 return 0; // not a binop.
1583 // Lowest Precedence: &&, ||
1584 case AsmToken::AmpAmp:
1585 Kind = MCBinaryExpr::LAnd;
1586 return 1;
1587 case AsmToken::PipePipe:
1588 Kind = MCBinaryExpr::LOr;
1589 return 1;
1591 // Low Precedence: |, &, ^
1592 case AsmToken::Pipe:
1593 Kind = MCBinaryExpr::Or;
1594 return 2;
1595 case AsmToken::Caret:
1596 Kind = MCBinaryExpr::Xor;
1597 return 2;
1598 case AsmToken::Amp:
1599 Kind = MCBinaryExpr::And;
1600 return 2;
1602 // Low Intermediate Precedence: ==, !=, <>, <, <=, >, >=
1603 case AsmToken::EqualEqual:
1604 Kind = MCBinaryExpr::EQ;
1605 return 3;
1606 case AsmToken::ExclaimEqual:
1607 case AsmToken::LessGreater:
1608 Kind = MCBinaryExpr::NE;
1609 return 3;
1610 case AsmToken::Less:
1611 Kind = MCBinaryExpr::LT;
1612 return 3;
1613 case AsmToken::LessEqual:
1614 Kind = MCBinaryExpr::LTE;
1615 return 3;
1616 case AsmToken::Greater:
1617 Kind = MCBinaryExpr::GT;
1618 return 3;
1619 case AsmToken::GreaterEqual:
1620 Kind = MCBinaryExpr::GTE;
1621 return 3;
1623 // Intermediate Precedence: <<, >>
1624 case AsmToken::LessLess:
1625 Kind = MCBinaryExpr::Shl;
1626 return 4;
1627 case AsmToken::GreaterGreater:
1628 Kind = ShouldUseLogicalShr ? MCBinaryExpr::LShr : MCBinaryExpr::AShr;
1629 return 4;
1631 // High Intermediate Precedence: +, -
1632 case AsmToken::Plus:
1633 Kind = MCBinaryExpr::Add;
1634 return 5;
1635 case AsmToken::Minus:
1636 Kind = MCBinaryExpr::Sub;
1637 return 5;
1639 // Highest Precedence: *, /, %
1640 case AsmToken::Star:
1641 Kind = MCBinaryExpr::Mul;
1642 return 6;
1643 case AsmToken::Slash:
1644 Kind = MCBinaryExpr::Div;
1645 return 6;
1646 case AsmToken::Percent:
1647 Kind = MCBinaryExpr::Mod;
1648 return 6;
1652 static unsigned getGNUBinOpPrecedence(const MCAsmInfo &MAI,
1653 AsmToken::TokenKind K,
1654 MCBinaryExpr::Opcode &Kind,
1655 bool ShouldUseLogicalShr) {
1656 switch (K) {
1657 default:
1658 return 0; // not a binop.
1660 // Lowest Precedence: &&, ||
1661 case AsmToken::AmpAmp:
1662 Kind = MCBinaryExpr::LAnd;
1663 return 2;
1664 case AsmToken::PipePipe:
1665 Kind = MCBinaryExpr::LOr;
1666 return 1;
1668 // Low Precedence: ==, !=, <>, <, <=, >, >=
1669 case AsmToken::EqualEqual:
1670 Kind = MCBinaryExpr::EQ;
1671 return 3;
1672 case AsmToken::ExclaimEqual:
1673 case AsmToken::LessGreater:
1674 Kind = MCBinaryExpr::NE;
1675 return 3;
1676 case AsmToken::Less:
1677 Kind = MCBinaryExpr::LT;
1678 return 3;
1679 case AsmToken::LessEqual:
1680 Kind = MCBinaryExpr::LTE;
1681 return 3;
1682 case AsmToken::Greater:
1683 Kind = MCBinaryExpr::GT;
1684 return 3;
1685 case AsmToken::GreaterEqual:
1686 Kind = MCBinaryExpr::GTE;
1687 return 3;
1689 // Low Intermediate Precedence: +, -
1690 case AsmToken::Plus:
1691 Kind = MCBinaryExpr::Add;
1692 return 4;
1693 case AsmToken::Minus:
1694 Kind = MCBinaryExpr::Sub;
1695 return 4;
1697 // High Intermediate Precedence: |, !, &, ^
1699 case AsmToken::Pipe:
1700 Kind = MCBinaryExpr::Or;
1701 return 5;
1702 case AsmToken::Exclaim:
1703 // Hack to support ARM compatible aliases (implied 'sp' operand in 'srs*'
1704 // instructions like 'srsda #31!') and not parse ! as an infix operator.
1705 if (MAI.getCommentString() == "@")
1706 return 0;
1707 Kind = MCBinaryExpr::OrNot;
1708 return 5;
1709 case AsmToken::Caret:
1710 Kind = MCBinaryExpr::Xor;
1711 return 5;
1712 case AsmToken::Amp:
1713 Kind = MCBinaryExpr::And;
1714 return 5;
1716 // Highest Precedence: *, /, %, <<, >>
1717 case AsmToken::Star:
1718 Kind = MCBinaryExpr::Mul;
1719 return 6;
1720 case AsmToken::Slash:
1721 Kind = MCBinaryExpr::Div;
1722 return 6;
1723 case AsmToken::Percent:
1724 Kind = MCBinaryExpr::Mod;
1725 return 6;
1726 case AsmToken::LessLess:
1727 Kind = MCBinaryExpr::Shl;
1728 return 6;
1729 case AsmToken::GreaterGreater:
1730 Kind = ShouldUseLogicalShr ? MCBinaryExpr::LShr : MCBinaryExpr::AShr;
1731 return 6;
1735 unsigned AsmParser::getBinOpPrecedence(AsmToken::TokenKind K,
1736 MCBinaryExpr::Opcode &Kind) {
1737 bool ShouldUseLogicalShr = MAI.shouldUseLogicalShr();
1738 return IsDarwin ? getDarwinBinOpPrecedence(K, Kind, ShouldUseLogicalShr)
1739 : getGNUBinOpPrecedence(MAI, K, Kind, ShouldUseLogicalShr);
1742 /// Parse all binary operators with precedence >= 'Precedence'.
1743 /// Res contains the LHS of the expression on input.
1744 bool AsmParser::parseBinOpRHS(unsigned Precedence, const MCExpr *&Res,
1745 SMLoc &EndLoc) {
1746 SMLoc StartLoc = Lexer.getLoc();
1747 while (true) {
1748 MCBinaryExpr::Opcode Kind = MCBinaryExpr::Add;
1749 unsigned TokPrec = getBinOpPrecedence(Lexer.getKind(), Kind);
1751 // If the next token is lower precedence than we are allowed to eat, return
1752 // successfully with what we ate already.
1753 if (TokPrec < Precedence)
1754 return false;
1756 Lex();
1758 // Eat the next primary expression.
1759 const MCExpr *RHS;
1760 if (getTargetParser().parsePrimaryExpr(RHS, EndLoc))
1761 return true;
1763 // If BinOp binds less tightly with RHS than the operator after RHS, let
1764 // the pending operator take RHS as its LHS.
1765 MCBinaryExpr::Opcode Dummy;
1766 unsigned NextTokPrec = getBinOpPrecedence(Lexer.getKind(), Dummy);
1767 if (TokPrec < NextTokPrec && parseBinOpRHS(TokPrec + 1, RHS, EndLoc))
1768 return true;
1770 // Merge LHS and RHS according to operator.
1771 Res = MCBinaryExpr::create(Kind, Res, RHS, getContext(), StartLoc);
1775 /// ParseStatement:
1776 /// ::= EndOfStatement
1777 /// ::= Label* Directive ...Operands... EndOfStatement
1778 /// ::= Label* Identifier OperandList* EndOfStatement
1779 bool AsmParser::parseStatement(ParseStatementInfo &Info,
1780 MCAsmParserSemaCallback *SI) {
1781 assert(!hasPendingError() && "parseStatement started with pending error");
1782 // Eat initial spaces and comments
1783 while (Lexer.is(AsmToken::Space))
1784 Lex();
1785 if (Lexer.is(AsmToken::EndOfStatement)) {
1786 // if this is a line comment we can drop it safely
1787 if (getTok().getString().empty() || getTok().getString().front() == '\r' ||
1788 getTok().getString().front() == '\n')
1789 Out.addBlankLine();
1790 Lex();
1791 return false;
1793 // Statements always start with an identifier.
1794 AsmToken ID = getTok();
1795 SMLoc IDLoc = ID.getLoc();
1796 StringRef IDVal;
1797 int64_t LocalLabelVal = -1;
1798 StartTokLoc = ID.getLoc();
1799 if (Lexer.is(AsmToken::HashDirective))
1800 return parseCppHashLineFilenameComment(IDLoc,
1801 !isInsideMacroInstantiation());
1803 // Allow an integer followed by a ':' as a directional local label.
1804 if (Lexer.is(AsmToken::Integer)) {
1805 LocalLabelVal = getTok().getIntVal();
1806 if (LocalLabelVal < 0) {
1807 if (!TheCondState.Ignore) {
1808 Lex(); // always eat a token
1809 return Error(IDLoc, "unexpected token at start of statement");
1811 IDVal = "";
1812 } else {
1813 IDVal = getTok().getString();
1814 Lex(); // Consume the integer token to be used as an identifier token.
1815 if (Lexer.getKind() != AsmToken::Colon) {
1816 if (!TheCondState.Ignore) {
1817 Lex(); // always eat a token
1818 return Error(IDLoc, "unexpected token at start of statement");
1822 } else if (Lexer.is(AsmToken::Dot)) {
1823 // Treat '.' as a valid identifier in this context.
1824 Lex();
1825 IDVal = ".";
1826 } else if (Lexer.is(AsmToken::LCurly)) {
1827 // Treat '{' as a valid identifier in this context.
1828 Lex();
1829 IDVal = "{";
1831 } else if (Lexer.is(AsmToken::RCurly)) {
1832 // Treat '}' as a valid identifier in this context.
1833 Lex();
1834 IDVal = "}";
1835 } else if (Lexer.is(AsmToken::Star) &&
1836 getTargetParser().starIsStartOfStatement()) {
1837 // Accept '*' as a valid start of statement.
1838 Lex();
1839 IDVal = "*";
1840 } else if (parseIdentifier(IDVal)) {
1841 if (!TheCondState.Ignore) {
1842 Lex(); // always eat a token
1843 return Error(IDLoc, "unexpected token at start of statement");
1845 IDVal = "";
1848 // Handle conditional assembly here before checking for skipping. We
1849 // have to do this so that .endif isn't skipped in a ".if 0" block for
1850 // example.
1851 StringMap<DirectiveKind>::const_iterator DirKindIt =
1852 DirectiveKindMap.find(IDVal.lower());
1853 DirectiveKind DirKind = (DirKindIt == DirectiveKindMap.end())
1854 ? DK_NO_DIRECTIVE
1855 : DirKindIt->getValue();
1856 switch (DirKind) {
1857 default:
1858 break;
1859 case DK_IF:
1860 case DK_IFEQ:
1861 case DK_IFGE:
1862 case DK_IFGT:
1863 case DK_IFLE:
1864 case DK_IFLT:
1865 case DK_IFNE:
1866 return parseDirectiveIf(IDLoc, DirKind);
1867 case DK_IFB:
1868 return parseDirectiveIfb(IDLoc, true);
1869 case DK_IFNB:
1870 return parseDirectiveIfb(IDLoc, false);
1871 case DK_IFC:
1872 return parseDirectiveIfc(IDLoc, true);
1873 case DK_IFEQS:
1874 return parseDirectiveIfeqs(IDLoc, true);
1875 case DK_IFNC:
1876 return parseDirectiveIfc(IDLoc, false);
1877 case DK_IFNES:
1878 return parseDirectiveIfeqs(IDLoc, false);
1879 case DK_IFDEF:
1880 return parseDirectiveIfdef(IDLoc, true);
1881 case DK_IFNDEF:
1882 case DK_IFNOTDEF:
1883 return parseDirectiveIfdef(IDLoc, false);
1884 case DK_ELSEIF:
1885 return parseDirectiveElseIf(IDLoc);
1886 case DK_ELSE:
1887 return parseDirectiveElse(IDLoc);
1888 case DK_ENDIF:
1889 return parseDirectiveEndIf(IDLoc);
1892 // Ignore the statement if in the middle of inactive conditional
1893 // (e.g. ".if 0").
1894 if (TheCondState.Ignore) {
1895 eatToEndOfStatement();
1896 return false;
1899 // FIXME: Recurse on local labels?
1901 // Check for a label.
1902 // ::= identifier ':'
1903 // ::= number ':'
1904 if (Lexer.is(AsmToken::Colon) && getTargetParser().isLabel(ID)) {
1905 if (checkForValidSection())
1906 return true;
1908 Lex(); // Consume the ':'.
1910 // Diagnose attempt to use '.' as a label.
1911 if (IDVal == ".")
1912 return Error(IDLoc, "invalid use of pseudo-symbol '.' as a label");
1914 // Diagnose attempt to use a variable as a label.
1916 // FIXME: Diagnostics. Note the location of the definition as a label.
1917 // FIXME: This doesn't diagnose assignment to a symbol which has been
1918 // implicitly marked as external.
1919 MCSymbol *Sym;
1920 if (LocalLabelVal == -1) {
1921 if (ParsingMSInlineAsm && SI) {
1922 StringRef RewrittenLabel =
1923 SI->LookupInlineAsmLabel(IDVal, getSourceManager(), IDLoc, true);
1924 assert(!RewrittenLabel.empty() &&
1925 "We should have an internal name here.");
1926 Info.AsmRewrites->emplace_back(AOK_Label, IDLoc, IDVal.size(),
1927 RewrittenLabel);
1928 IDVal = RewrittenLabel;
1930 Sym = getContext().getOrCreateSymbol(IDVal);
1931 } else
1932 Sym = Ctx.createDirectionalLocalSymbol(LocalLabelVal);
1933 // End of Labels should be treated as end of line for lexing
1934 // purposes but that information is not available to the Lexer who
1935 // does not understand Labels. This may cause us to see a Hash
1936 // here instead of a preprocessor line comment.
1937 if (getTok().is(AsmToken::Hash)) {
1938 StringRef CommentStr = parseStringToEndOfStatement();
1939 Lexer.Lex();
1940 Lexer.UnLex(AsmToken(AsmToken::EndOfStatement, CommentStr));
1943 // Consume any end of statement token, if present, to avoid spurious
1944 // addBlankLine calls().
1945 if (getTok().is(AsmToken::EndOfStatement)) {
1946 Lex();
1949 if (MAI.hasSubsectionsViaSymbols() && CFIStartProcLoc &&
1950 Sym->isExternal() && !cast<MCSymbolMachO>(Sym)->isAltEntry())
1951 return Error(StartTokLoc, "non-private labels cannot appear between "
1952 ".cfi_startproc / .cfi_endproc pairs") &&
1953 Error(*CFIStartProcLoc, "previous .cfi_startproc was here");
1955 if (discardLTOSymbol(IDVal))
1956 return false;
1958 getTargetParser().doBeforeLabelEmit(Sym, IDLoc);
1960 // Emit the label.
1961 if (!getTargetParser().isParsingMSInlineAsm())
1962 Out.emitLabel(Sym, IDLoc);
1964 // If we are generating dwarf for assembly source files then gather the
1965 // info to make a dwarf label entry for this label if needed.
1966 if (enabledGenDwarfForAssembly())
1967 MCGenDwarfLabelEntry::Make(Sym, &getStreamer(), getSourceManager(),
1968 IDLoc);
1970 getTargetParser().onLabelParsed(Sym);
1972 return false;
1975 // Check for an assignment statement.
1976 // ::= identifier '='
1977 if (Lexer.is(AsmToken::Equal) && getTargetParser().equalIsAsmAssignment()) {
1978 Lex();
1979 return parseAssignment(IDVal, AssignmentKind::Equal);
1982 // If macros are enabled, check to see if this is a macro instantiation.
1983 if (areMacrosEnabled())
1984 if (MCAsmMacro *M = getContext().lookupMacro(IDVal))
1985 return handleMacroEntry(M, IDLoc);
1987 // Otherwise, we have a normal instruction or directive.
1989 // Directives start with "."
1990 if (IDVal.starts_with(".") && IDVal != ".") {
1991 // There are several entities interested in parsing directives:
1993 // 1. The target-specific assembly parser. Some directives are target
1994 // specific or may potentially behave differently on certain targets.
1995 // 2. Asm parser extensions. For example, platform-specific parsers
1996 // (like the ELF parser) register themselves as extensions.
1997 // 3. The generic directive parser implemented by this class. These are
1998 // all the directives that behave in a target and platform independent
1999 // manner, or at least have a default behavior that's shared between
2000 // all targets and platforms.
2002 getTargetParser().flushPendingInstructions(getStreamer());
2004 ParseStatus TPDirectiveReturn = getTargetParser().parseDirective(ID);
2005 assert(TPDirectiveReturn.isFailure() == hasPendingError() &&
2006 "Should only return Failure iff there was an error");
2007 if (TPDirectiveReturn.isFailure())
2008 return true;
2009 if (TPDirectiveReturn.isSuccess())
2010 return false;
2012 // Next, check the extension directive map to see if any extension has
2013 // registered itself to parse this directive.
2014 std::pair<MCAsmParserExtension *, DirectiveHandler> Handler =
2015 ExtensionDirectiveMap.lookup(IDVal);
2016 if (Handler.first)
2017 return (*Handler.second)(Handler.first, IDVal, IDLoc);
2019 // Finally, if no one else is interested in this directive, it must be
2020 // generic and familiar to this class.
2021 switch (DirKind) {
2022 default:
2023 break;
2024 case DK_SET:
2025 case DK_EQU:
2026 return parseDirectiveSet(IDVal, AssignmentKind::Set);
2027 case DK_EQUIV:
2028 return parseDirectiveSet(IDVal, AssignmentKind::Equiv);
2029 case DK_LTO_SET_CONDITIONAL:
2030 return parseDirectiveSet(IDVal, AssignmentKind::LTOSetConditional);
2031 case DK_ASCII:
2032 return parseDirectiveAscii(IDVal, false);
2033 case DK_ASCIZ:
2034 case DK_STRING:
2035 return parseDirectiveAscii(IDVal, true);
2036 case DK_BYTE:
2037 case DK_DC_B:
2038 return parseDirectiveValue(IDVal, 1);
2039 case DK_DC:
2040 case DK_DC_W:
2041 case DK_SHORT:
2042 case DK_VALUE:
2043 case DK_2BYTE:
2044 return parseDirectiveValue(IDVal, 2);
2045 case DK_LONG:
2046 case DK_INT:
2047 case DK_4BYTE:
2048 case DK_DC_L:
2049 return parseDirectiveValue(IDVal, 4);
2050 case DK_QUAD:
2051 case DK_8BYTE:
2052 return parseDirectiveValue(IDVal, 8);
2053 case DK_DC_A:
2054 return parseDirectiveValue(
2055 IDVal, getContext().getAsmInfo()->getCodePointerSize());
2056 case DK_OCTA:
2057 return parseDirectiveOctaValue(IDVal);
2058 case DK_SINGLE:
2059 case DK_FLOAT:
2060 case DK_DC_S:
2061 return parseDirectiveRealValue(IDVal, APFloat::IEEEsingle());
2062 case DK_DOUBLE:
2063 case DK_DC_D:
2064 return parseDirectiveRealValue(IDVal, APFloat::IEEEdouble());
2065 case DK_ALIGN: {
2066 bool IsPow2 = !getContext().getAsmInfo()->getAlignmentIsInBytes();
2067 return parseDirectiveAlign(IsPow2, /*ExprSize=*/1);
2069 case DK_ALIGN32: {
2070 bool IsPow2 = !getContext().getAsmInfo()->getAlignmentIsInBytes();
2071 return parseDirectiveAlign(IsPow2, /*ExprSize=*/4);
2073 case DK_BALIGN:
2074 return parseDirectiveAlign(/*IsPow2=*/false, /*ExprSize=*/1);
2075 case DK_BALIGNW:
2076 return parseDirectiveAlign(/*IsPow2=*/false, /*ExprSize=*/2);
2077 case DK_BALIGNL:
2078 return parseDirectiveAlign(/*IsPow2=*/false, /*ExprSize=*/4);
2079 case DK_P2ALIGN:
2080 return parseDirectiveAlign(/*IsPow2=*/true, /*ExprSize=*/1);
2081 case DK_P2ALIGNW:
2082 return parseDirectiveAlign(/*IsPow2=*/true, /*ExprSize=*/2);
2083 case DK_P2ALIGNL:
2084 return parseDirectiveAlign(/*IsPow2=*/true, /*ExprSize=*/4);
2085 case DK_ORG:
2086 return parseDirectiveOrg();
2087 case DK_FILL:
2088 return parseDirectiveFill();
2089 case DK_ZERO:
2090 return parseDirectiveZero();
2091 case DK_EXTERN:
2092 eatToEndOfStatement(); // .extern is the default, ignore it.
2093 return false;
2094 case DK_GLOBL:
2095 case DK_GLOBAL:
2096 return parseDirectiveSymbolAttribute(MCSA_Global);
2097 case DK_LAZY_REFERENCE:
2098 return parseDirectiveSymbolAttribute(MCSA_LazyReference);
2099 case DK_NO_DEAD_STRIP:
2100 return parseDirectiveSymbolAttribute(MCSA_NoDeadStrip);
2101 case DK_SYMBOL_RESOLVER:
2102 return parseDirectiveSymbolAttribute(MCSA_SymbolResolver);
2103 case DK_PRIVATE_EXTERN:
2104 return parseDirectiveSymbolAttribute(MCSA_PrivateExtern);
2105 case DK_REFERENCE:
2106 return parseDirectiveSymbolAttribute(MCSA_Reference);
2107 case DK_WEAK_DEFINITION:
2108 return parseDirectiveSymbolAttribute(MCSA_WeakDefinition);
2109 case DK_WEAK_REFERENCE:
2110 return parseDirectiveSymbolAttribute(MCSA_WeakReference);
2111 case DK_WEAK_DEF_CAN_BE_HIDDEN:
2112 return parseDirectiveSymbolAttribute(MCSA_WeakDefAutoPrivate);
2113 case DK_COLD:
2114 return parseDirectiveSymbolAttribute(MCSA_Cold);
2115 case DK_COMM:
2116 case DK_COMMON:
2117 return parseDirectiveComm(/*IsLocal=*/false);
2118 case DK_LCOMM:
2119 return parseDirectiveComm(/*IsLocal=*/true);
2120 case DK_ABORT:
2121 return parseDirectiveAbort(IDLoc);
2122 case DK_INCLUDE:
2123 return parseDirectiveInclude();
2124 case DK_INCBIN:
2125 return parseDirectiveIncbin();
2126 case DK_CODE16:
2127 case DK_CODE16GCC:
2128 return TokError(Twine(IDVal) +
2129 " not currently supported for this target");
2130 case DK_REPT:
2131 return parseDirectiveRept(IDLoc, IDVal);
2132 case DK_IRP:
2133 return parseDirectiveIrp(IDLoc);
2134 case DK_IRPC:
2135 return parseDirectiveIrpc(IDLoc);
2136 case DK_ENDR:
2137 return parseDirectiveEndr(IDLoc);
2138 case DK_BUNDLE_ALIGN_MODE:
2139 return parseDirectiveBundleAlignMode();
2140 case DK_BUNDLE_LOCK:
2141 return parseDirectiveBundleLock();
2142 case DK_BUNDLE_UNLOCK:
2143 return parseDirectiveBundleUnlock();
2144 case DK_SLEB128:
2145 return parseDirectiveLEB128(true);
2146 case DK_ULEB128:
2147 return parseDirectiveLEB128(false);
2148 case DK_SPACE:
2149 case DK_SKIP:
2150 return parseDirectiveSpace(IDVal);
2151 case DK_FILE:
2152 return parseDirectiveFile(IDLoc);
2153 case DK_LINE:
2154 return parseDirectiveLine();
2155 case DK_LOC:
2156 return parseDirectiveLoc();
2157 case DK_LOC_LABEL:
2158 return parseDirectiveLocLabel(IDLoc);
2159 case DK_STABS:
2160 return parseDirectiveStabs();
2161 case DK_CV_FILE:
2162 return parseDirectiveCVFile();
2163 case DK_CV_FUNC_ID:
2164 return parseDirectiveCVFuncId();
2165 case DK_CV_INLINE_SITE_ID:
2166 return parseDirectiveCVInlineSiteId();
2167 case DK_CV_LOC:
2168 return parseDirectiveCVLoc();
2169 case DK_CV_LINETABLE:
2170 return parseDirectiveCVLinetable();
2171 case DK_CV_INLINE_LINETABLE:
2172 return parseDirectiveCVInlineLinetable();
2173 case DK_CV_DEF_RANGE:
2174 return parseDirectiveCVDefRange();
2175 case DK_CV_STRING:
2176 return parseDirectiveCVString();
2177 case DK_CV_STRINGTABLE:
2178 return parseDirectiveCVStringTable();
2179 case DK_CV_FILECHECKSUMS:
2180 return parseDirectiveCVFileChecksums();
2181 case DK_CV_FILECHECKSUM_OFFSET:
2182 return parseDirectiveCVFileChecksumOffset();
2183 case DK_CV_FPO_DATA:
2184 return parseDirectiveCVFPOData();
2185 case DK_CFI_SECTIONS:
2186 return parseDirectiveCFISections();
2187 case DK_CFI_STARTPROC:
2188 return parseDirectiveCFIStartProc();
2189 case DK_CFI_ENDPROC:
2190 return parseDirectiveCFIEndProc();
2191 case DK_CFI_DEF_CFA:
2192 return parseDirectiveCFIDefCfa(IDLoc);
2193 case DK_CFI_DEF_CFA_OFFSET:
2194 return parseDirectiveCFIDefCfaOffset(IDLoc);
2195 case DK_CFI_ADJUST_CFA_OFFSET:
2196 return parseDirectiveCFIAdjustCfaOffset(IDLoc);
2197 case DK_CFI_DEF_CFA_REGISTER:
2198 return parseDirectiveCFIDefCfaRegister(IDLoc);
2199 case DK_CFI_LLVM_DEF_ASPACE_CFA:
2200 return parseDirectiveCFILLVMDefAspaceCfa(IDLoc);
2201 case DK_CFI_OFFSET:
2202 return parseDirectiveCFIOffset(IDLoc);
2203 case DK_CFI_REL_OFFSET:
2204 return parseDirectiveCFIRelOffset(IDLoc);
2205 case DK_CFI_PERSONALITY:
2206 return parseDirectiveCFIPersonalityOrLsda(true);
2207 case DK_CFI_LSDA:
2208 return parseDirectiveCFIPersonalityOrLsda(false);
2209 case DK_CFI_REMEMBER_STATE:
2210 return parseDirectiveCFIRememberState(IDLoc);
2211 case DK_CFI_RESTORE_STATE:
2212 return parseDirectiveCFIRestoreState(IDLoc);
2213 case DK_CFI_SAME_VALUE:
2214 return parseDirectiveCFISameValue(IDLoc);
2215 case DK_CFI_RESTORE:
2216 return parseDirectiveCFIRestore(IDLoc);
2217 case DK_CFI_ESCAPE:
2218 return parseDirectiveCFIEscape(IDLoc);
2219 case DK_CFI_RETURN_COLUMN:
2220 return parseDirectiveCFIReturnColumn(IDLoc);
2221 case DK_CFI_SIGNAL_FRAME:
2222 return parseDirectiveCFISignalFrame(IDLoc);
2223 case DK_CFI_UNDEFINED:
2224 return parseDirectiveCFIUndefined(IDLoc);
2225 case DK_CFI_REGISTER:
2226 return parseDirectiveCFIRegister(IDLoc);
2227 case DK_CFI_WINDOW_SAVE:
2228 return parseDirectiveCFIWindowSave(IDLoc);
2229 case DK_CFI_LABEL:
2230 return parseDirectiveCFILabel(IDLoc);
2231 case DK_CFI_VAL_OFFSET:
2232 return parseDirectiveCFIValOffset(IDLoc);
2233 case DK_MACROS_ON:
2234 case DK_MACROS_OFF:
2235 return parseDirectiveMacrosOnOff(IDVal);
2236 case DK_MACRO:
2237 return parseDirectiveMacro(IDLoc);
2238 case DK_ALTMACRO:
2239 case DK_NOALTMACRO:
2240 return parseDirectiveAltmacro(IDVal);
2241 case DK_EXITM:
2242 return parseDirectiveExitMacro(IDVal);
2243 case DK_ENDM:
2244 case DK_ENDMACRO:
2245 return parseDirectiveEndMacro(IDVal);
2246 case DK_PURGEM:
2247 return parseDirectivePurgeMacro(IDLoc);
2248 case DK_END:
2249 return parseDirectiveEnd(IDLoc);
2250 case DK_ERR:
2251 return parseDirectiveError(IDLoc, false);
2252 case DK_ERROR:
2253 return parseDirectiveError(IDLoc, true);
2254 case DK_WARNING:
2255 return parseDirectiveWarning(IDLoc);
2256 case DK_RELOC:
2257 return parseDirectiveReloc(IDLoc);
2258 case DK_DCB:
2259 case DK_DCB_W:
2260 return parseDirectiveDCB(IDVal, 2);
2261 case DK_DCB_B:
2262 return parseDirectiveDCB(IDVal, 1);
2263 case DK_DCB_D:
2264 return parseDirectiveRealDCB(IDVal, APFloat::IEEEdouble());
2265 case DK_DCB_L:
2266 return parseDirectiveDCB(IDVal, 4);
2267 case DK_DCB_S:
2268 return parseDirectiveRealDCB(IDVal, APFloat::IEEEsingle());
2269 case DK_DC_X:
2270 case DK_DCB_X:
2271 return TokError(Twine(IDVal) +
2272 " not currently supported for this target");
2273 case DK_DS:
2274 case DK_DS_W:
2275 return parseDirectiveDS(IDVal, 2);
2276 case DK_DS_B:
2277 return parseDirectiveDS(IDVal, 1);
2278 case DK_DS_D:
2279 return parseDirectiveDS(IDVal, 8);
2280 case DK_DS_L:
2281 case DK_DS_S:
2282 return parseDirectiveDS(IDVal, 4);
2283 case DK_DS_P:
2284 case DK_DS_X:
2285 return parseDirectiveDS(IDVal, 12);
2286 case DK_PRINT:
2287 return parseDirectivePrint(IDLoc);
2288 case DK_ADDRSIG:
2289 return parseDirectiveAddrsig();
2290 case DK_ADDRSIG_SYM:
2291 return parseDirectiveAddrsigSym();
2292 case DK_PSEUDO_PROBE:
2293 return parseDirectivePseudoProbe();
2294 case DK_LTO_DISCARD:
2295 return parseDirectiveLTODiscard();
2296 case DK_MEMTAG:
2297 return parseDirectiveSymbolAttribute(MCSA_Memtag);
2300 return Error(IDLoc, "unknown directive");
2303 // __asm _emit or __asm __emit
2304 if (ParsingMSInlineAsm && (IDVal == "_emit" || IDVal == "__emit" ||
2305 IDVal == "_EMIT" || IDVal == "__EMIT"))
2306 return parseDirectiveMSEmit(IDLoc, Info, IDVal.size());
2308 // __asm align
2309 if (ParsingMSInlineAsm && (IDVal == "align" || IDVal == "ALIGN"))
2310 return parseDirectiveMSAlign(IDLoc, Info);
2312 if (ParsingMSInlineAsm && (IDVal == "even" || IDVal == "EVEN"))
2313 Info.AsmRewrites->emplace_back(AOK_EVEN, IDLoc, 4);
2314 if (checkForValidSection())
2315 return true;
2317 return parseAndMatchAndEmitTargetInstruction(Info, IDVal, ID, IDLoc);
2320 bool AsmParser::parseAndMatchAndEmitTargetInstruction(ParseStatementInfo &Info,
2321 StringRef IDVal,
2322 AsmToken ID,
2323 SMLoc IDLoc) {
2324 // Canonicalize the opcode to lower case.
2325 std::string OpcodeStr = IDVal.lower();
2326 ParseInstructionInfo IInfo(Info.AsmRewrites);
2327 bool ParseHadError = getTargetParser().parseInstruction(IInfo, OpcodeStr, ID,
2328 Info.ParsedOperands);
2329 Info.ParseError = ParseHadError;
2331 // Dump the parsed representation, if requested.
2332 if (getShowParsedOperands()) {
2333 SmallString<256> Str;
2334 raw_svector_ostream OS(Str);
2335 OS << "parsed instruction: [";
2336 for (unsigned i = 0; i != Info.ParsedOperands.size(); ++i) {
2337 if (i != 0)
2338 OS << ", ";
2339 Info.ParsedOperands[i]->print(OS);
2341 OS << "]";
2343 printMessage(IDLoc, SourceMgr::DK_Note, OS.str());
2346 // Fail even if ParseInstruction erroneously returns false.
2347 if (hasPendingError() || ParseHadError)
2348 return true;
2350 // If we are generating dwarf for the current section then generate a .loc
2351 // directive for the instruction.
2352 if (!ParseHadError && enabledGenDwarfForAssembly() &&
2353 getContext().getGenDwarfSectionSyms().count(
2354 getStreamer().getCurrentSectionOnly())) {
2355 unsigned Line;
2356 if (ActiveMacros.empty())
2357 Line = SrcMgr.FindLineNumber(IDLoc, CurBuffer);
2358 else
2359 Line = SrcMgr.FindLineNumber(ActiveMacros.front()->InstantiationLoc,
2360 ActiveMacros.front()->ExitBuffer);
2362 // If we previously parsed a cpp hash file line comment then make sure the
2363 // current Dwarf File is for the CppHashFilename if not then emit the
2364 // Dwarf File table for it and adjust the line number for the .loc.
2365 if (!CppHashInfo.Filename.empty()) {
2366 unsigned FileNumber = getStreamer().emitDwarfFileDirective(
2367 0, StringRef(), CppHashInfo.Filename);
2368 getContext().setGenDwarfFileNumber(FileNumber);
2370 unsigned CppHashLocLineNo =
2371 SrcMgr.FindLineNumber(CppHashInfo.Loc, CppHashInfo.Buf);
2372 Line = CppHashInfo.LineNumber - 1 + (Line - CppHashLocLineNo);
2375 getStreamer().emitDwarfLocDirective(
2376 getContext().getGenDwarfFileNumber(), Line, 0,
2377 DWARF2_LINE_DEFAULT_IS_STMT ? DWARF2_FLAG_IS_STMT : 0, 0, 0,
2378 StringRef());
2381 // If parsing succeeded, match the instruction.
2382 if (!ParseHadError) {
2383 uint64_t ErrorInfo;
2384 if (getTargetParser().matchAndEmitInstruction(
2385 IDLoc, Info.Opcode, Info.ParsedOperands, Out, ErrorInfo,
2386 getTargetParser().isParsingMSInlineAsm()))
2387 return true;
2389 return false;
2392 // Parse and erase curly braces marking block start/end
2393 bool
2394 AsmParser::parseCurlyBlockScope(SmallVectorImpl<AsmRewrite> &AsmStrRewrites) {
2395 // Identify curly brace marking block start/end
2396 if (Lexer.isNot(AsmToken::LCurly) && Lexer.isNot(AsmToken::RCurly))
2397 return false;
2399 SMLoc StartLoc = Lexer.getLoc();
2400 Lex(); // Eat the brace
2401 if (Lexer.is(AsmToken::EndOfStatement))
2402 Lex(); // Eat EndOfStatement following the brace
2404 // Erase the block start/end brace from the output asm string
2405 AsmStrRewrites.emplace_back(AOK_Skip, StartLoc, Lexer.getLoc().getPointer() -
2406 StartLoc.getPointer());
2407 return true;
2410 /// parseCppHashLineFilenameComment as this:
2411 /// ::= # number "filename"
2412 bool AsmParser::parseCppHashLineFilenameComment(SMLoc L, bool SaveLocInfo) {
2413 Lex(); // Eat the hash token.
2414 // Lexer only ever emits HashDirective if it fully formed if it's
2415 // done the checking already so this is an internal error.
2416 assert(getTok().is(AsmToken::Integer) &&
2417 "Lexing Cpp line comment: Expected Integer");
2418 int64_t LineNumber = getTok().getIntVal();
2419 Lex();
2420 assert(getTok().is(AsmToken::String) &&
2421 "Lexing Cpp line comment: Expected String");
2422 StringRef Filename = getTok().getString();
2423 Lex();
2425 if (!SaveLocInfo)
2426 return false;
2428 // Get rid of the enclosing quotes.
2429 Filename = Filename.substr(1, Filename.size() - 2);
2431 // Save the SMLoc, Filename and LineNumber for later use by diagnostics
2432 // and possibly DWARF file info.
2433 CppHashInfo.Loc = L;
2434 CppHashInfo.Filename = Filename;
2435 CppHashInfo.LineNumber = LineNumber;
2436 CppHashInfo.Buf = CurBuffer;
2437 if (!HadCppHashFilename) {
2438 HadCppHashFilename = true;
2439 // If we haven't encountered any .file directives, then the first #line
2440 // directive describes the "root" file and directory of the compilation
2441 // unit.
2442 if (getContext().getGenDwarfForAssembly() &&
2443 getContext().getGenDwarfFileNumber() == 0) {
2444 // It's preprocessed, so there is no checksum, and of course no source
2445 // directive.
2446 getContext().setMCLineTableRootFile(
2447 /*CUID=*/0, getContext().getCompilationDir(), Filename,
2448 /*Cksum=*/std::nullopt, /*Source=*/std::nullopt);
2451 return false;
2454 /// will use the last parsed cpp hash line filename comment
2455 /// for the Filename and LineNo if any in the diagnostic.
2456 void AsmParser::DiagHandler(const SMDiagnostic &Diag, void *Context) {
2457 auto *Parser = static_cast<AsmParser *>(Context);
2458 raw_ostream &OS = errs();
2460 const SourceMgr &DiagSrcMgr = *Diag.getSourceMgr();
2461 SMLoc DiagLoc = Diag.getLoc();
2462 unsigned DiagBuf = DiagSrcMgr.FindBufferContainingLoc(DiagLoc);
2463 unsigned CppHashBuf =
2464 Parser->SrcMgr.FindBufferContainingLoc(Parser->CppHashInfo.Loc);
2466 // Like SourceMgr::printMessage() we need to print the include stack if any
2467 // before printing the message.
2468 unsigned DiagCurBuffer = DiagSrcMgr.FindBufferContainingLoc(DiagLoc);
2469 if (!Parser->SavedDiagHandler && DiagCurBuffer &&
2470 DiagCurBuffer != DiagSrcMgr.getMainFileID()) {
2471 SMLoc ParentIncludeLoc = DiagSrcMgr.getParentIncludeLoc(DiagCurBuffer);
2472 DiagSrcMgr.PrintIncludeStack(ParentIncludeLoc, OS);
2475 // If we have not parsed a cpp hash line filename comment or the source
2476 // manager changed or buffer changed (like in a nested include) then just
2477 // print the normal diagnostic using its Filename and LineNo.
2478 if (!Parser->CppHashInfo.LineNumber || DiagBuf != CppHashBuf) {
2479 if (Parser->SavedDiagHandler)
2480 Parser->SavedDiagHandler(Diag, Parser->SavedDiagContext);
2481 else
2482 Parser->getContext().diagnose(Diag);
2483 return;
2486 // Use the CppHashFilename and calculate a line number based on the
2487 // CppHashInfo.Loc and CppHashInfo.LineNumber relative to this Diag's SMLoc
2488 // for the diagnostic.
2489 const std::string &Filename = std::string(Parser->CppHashInfo.Filename);
2491 int DiagLocLineNo = DiagSrcMgr.FindLineNumber(DiagLoc, DiagBuf);
2492 int CppHashLocLineNo =
2493 Parser->SrcMgr.FindLineNumber(Parser->CppHashInfo.Loc, CppHashBuf);
2494 int LineNo =
2495 Parser->CppHashInfo.LineNumber - 1 + (DiagLocLineNo - CppHashLocLineNo);
2497 SMDiagnostic NewDiag(*Diag.getSourceMgr(), Diag.getLoc(), Filename, LineNo,
2498 Diag.getColumnNo(), Diag.getKind(), Diag.getMessage(),
2499 Diag.getLineContents(), Diag.getRanges());
2501 if (Parser->SavedDiagHandler)
2502 Parser->SavedDiagHandler(Diag, Parser->SavedDiagContext);
2503 else
2504 Parser->getContext().diagnose(NewDiag);
2507 // FIXME: This is mostly duplicated from the function in AsmLexer.cpp. The
2508 // difference being that that function accepts '@' as part of identifiers and
2509 // we can't do that. AsmLexer.cpp should probably be changed to handle
2510 // '@' as a special case when needed.
2511 static bool isIdentifierChar(char c) {
2512 return isalnum(static_cast<unsigned char>(c)) || c == '_' || c == '$' ||
2513 c == '.';
2516 bool AsmParser::expandMacro(raw_svector_ostream &OS, MCAsmMacro &Macro,
2517 ArrayRef<MCAsmMacroParameter> Parameters,
2518 ArrayRef<MCAsmMacroArgument> A,
2519 bool EnableAtPseudoVariable) {
2520 unsigned NParameters = Parameters.size();
2521 auto expandArg = [&](unsigned Index) {
2522 bool HasVararg = NParameters ? Parameters.back().Vararg : false;
2523 bool VarargParameter = HasVararg && Index == (NParameters - 1);
2524 for (const AsmToken &Token : A[Index])
2525 // For altmacro mode, you can write '%expr'.
2526 // The prefix '%' evaluates the expression 'expr'
2527 // and uses the result as a string (e.g. replace %(1+2) with the
2528 // string "3").
2529 // Here, we identify the integer token which is the result of the
2530 // absolute expression evaluation and replace it with its string
2531 // representation.
2532 if (AltMacroMode && Token.getString().front() == '%' &&
2533 Token.is(AsmToken::Integer))
2534 // Emit an integer value to the buffer.
2535 OS << Token.getIntVal();
2536 // Only Token that was validated as a string and begins with '<'
2537 // is considered altMacroString!!!
2538 else if (AltMacroMode && Token.getString().front() == '<' &&
2539 Token.is(AsmToken::String)) {
2540 OS << angleBracketString(Token.getStringContents());
2542 // We expect no quotes around the string's contents when
2543 // parsing for varargs.
2544 else if (Token.isNot(AsmToken::String) || VarargParameter)
2545 OS << Token.getString();
2546 else
2547 OS << Token.getStringContents();
2550 // A macro without parameters is handled differently on Darwin:
2551 // gas accepts no arguments and does no substitutions
2552 StringRef Body = Macro.Body;
2553 size_t I = 0, End = Body.size();
2554 while (I != End) {
2555 if (Body[I] == '\\' && I + 1 != End) {
2556 // Check for \@ and \+ pseudo variables.
2557 if (EnableAtPseudoVariable && Body[I + 1] == '@') {
2558 OS << NumOfMacroInstantiations;
2559 I += 2;
2560 continue;
2562 if (Body[I + 1] == '+') {
2563 OS << Macro.Count;
2564 I += 2;
2565 continue;
2567 if (Body[I + 1] == '(' && Body[I + 2] == ')') {
2568 I += 3;
2569 continue;
2572 size_t Pos = ++I;
2573 while (I != End && isIdentifierChar(Body[I]))
2574 ++I;
2575 StringRef Argument(Body.data() + Pos, I - Pos);
2576 if (AltMacroMode && I != End && Body[I] == '&')
2577 ++I;
2578 unsigned Index = 0;
2579 for (; Index < NParameters; ++Index)
2580 if (Parameters[Index].Name == Argument)
2581 break;
2582 if (Index == NParameters)
2583 OS << '\\' << Argument;
2584 else
2585 expandArg(Index);
2586 continue;
2589 // In Darwin mode, $ is used for macro expansion, not considered an
2590 // identifier char.
2591 if (Body[I] == '$' && I + 1 != End && IsDarwin && !NParameters) {
2592 // This macro has no parameters, look for $0, $1, etc.
2593 switch (Body[I + 1]) {
2594 // $$ => $
2595 case '$':
2596 OS << '$';
2597 I += 2;
2598 continue;
2599 // $n => number of arguments
2600 case 'n':
2601 OS << A.size();
2602 I += 2;
2603 continue;
2604 default: {
2605 if (!isDigit(Body[I + 1]))
2606 break;
2607 // $[0-9] => argument
2608 // Missing arguments are ignored.
2609 unsigned Index = Body[I + 1] - '0';
2610 if (Index < A.size())
2611 for (const AsmToken &Token : A[Index])
2612 OS << Token.getString();
2613 I += 2;
2614 continue;
2619 if (!isIdentifierChar(Body[I]) || IsDarwin) {
2620 OS << Body[I++];
2621 continue;
2624 const size_t Start = I;
2625 while (++I && isIdentifierChar(Body[I])) {
2627 StringRef Token(Body.data() + Start, I - Start);
2628 if (AltMacroMode) {
2629 unsigned Index = 0;
2630 for (; Index != NParameters; ++Index)
2631 if (Parameters[Index].Name == Token)
2632 break;
2633 if (Index != NParameters) {
2634 expandArg(Index);
2635 if (I != End && Body[I] == '&')
2636 ++I;
2637 continue;
2640 OS << Token;
2643 ++Macro.Count;
2644 return false;
2647 static bool isOperator(AsmToken::TokenKind kind) {
2648 switch (kind) {
2649 default:
2650 return false;
2651 case AsmToken::Plus:
2652 case AsmToken::Minus:
2653 case AsmToken::Tilde:
2654 case AsmToken::Slash:
2655 case AsmToken::Star:
2656 case AsmToken::Dot:
2657 case AsmToken::Equal:
2658 case AsmToken::EqualEqual:
2659 case AsmToken::Pipe:
2660 case AsmToken::PipePipe:
2661 case AsmToken::Caret:
2662 case AsmToken::Amp:
2663 case AsmToken::AmpAmp:
2664 case AsmToken::Exclaim:
2665 case AsmToken::ExclaimEqual:
2666 case AsmToken::Less:
2667 case AsmToken::LessEqual:
2668 case AsmToken::LessLess:
2669 case AsmToken::LessGreater:
2670 case AsmToken::Greater:
2671 case AsmToken::GreaterEqual:
2672 case AsmToken::GreaterGreater:
2673 return true;
2677 namespace {
2679 class AsmLexerSkipSpaceRAII {
2680 public:
2681 AsmLexerSkipSpaceRAII(AsmLexer &Lexer, bool SkipSpace) : Lexer(Lexer) {
2682 Lexer.setSkipSpace(SkipSpace);
2685 ~AsmLexerSkipSpaceRAII() {
2686 Lexer.setSkipSpace(true);
2689 private:
2690 AsmLexer &Lexer;
2693 } // end anonymous namespace
2695 bool AsmParser::parseMacroArgument(MCAsmMacroArgument &MA, bool Vararg) {
2697 if (Vararg) {
2698 if (Lexer.isNot(AsmToken::EndOfStatement)) {
2699 StringRef Str = parseStringToEndOfStatement();
2700 MA.emplace_back(AsmToken::String, Str);
2702 return false;
2705 unsigned ParenLevel = 0;
2707 // Darwin doesn't use spaces to delmit arguments.
2708 AsmLexerSkipSpaceRAII ScopedSkipSpace(Lexer, IsDarwin);
2710 bool SpaceEaten;
2712 while (true) {
2713 SpaceEaten = false;
2714 if (Lexer.is(AsmToken::Eof) || Lexer.is(AsmToken::Equal))
2715 return TokError("unexpected token in macro instantiation");
2717 if (ParenLevel == 0) {
2719 if (Lexer.is(AsmToken::Comma))
2720 break;
2722 if (parseOptionalToken(AsmToken::Space))
2723 SpaceEaten = true;
2725 // Spaces can delimit parameters, but could also be part an expression.
2726 // If the token after a space is an operator, add the token and the next
2727 // one into this argument
2728 if (!IsDarwin) {
2729 if (isOperator(Lexer.getKind())) {
2730 MA.push_back(getTok());
2731 Lexer.Lex();
2733 // Whitespace after an operator can be ignored.
2734 parseOptionalToken(AsmToken::Space);
2735 continue;
2738 if (SpaceEaten)
2739 break;
2742 // handleMacroEntry relies on not advancing the lexer here
2743 // to be able to fill in the remaining default parameter values
2744 if (Lexer.is(AsmToken::EndOfStatement))
2745 break;
2747 // Adjust the current parentheses level.
2748 if (Lexer.is(AsmToken::LParen))
2749 ++ParenLevel;
2750 else if (Lexer.is(AsmToken::RParen) && ParenLevel)
2751 --ParenLevel;
2753 // Append the token to the current argument list.
2754 MA.push_back(getTok());
2755 Lexer.Lex();
2758 if (ParenLevel != 0)
2759 return TokError("unbalanced parentheses in macro argument");
2760 return false;
2763 // Parse the macro instantiation arguments.
2764 bool AsmParser::parseMacroArguments(const MCAsmMacro *M,
2765 MCAsmMacroArguments &A) {
2766 const unsigned NParameters = M ? M->Parameters.size() : 0;
2767 bool NamedParametersFound = false;
2768 SmallVector<SMLoc, 4> FALocs;
2770 A.resize(NParameters);
2771 FALocs.resize(NParameters);
2773 // Parse two kinds of macro invocations:
2774 // - macros defined without any parameters accept an arbitrary number of them
2775 // - macros defined with parameters accept at most that many of them
2776 bool HasVararg = NParameters ? M->Parameters.back().Vararg : false;
2777 for (unsigned Parameter = 0; !NParameters || Parameter < NParameters;
2778 ++Parameter) {
2779 SMLoc IDLoc = Lexer.getLoc();
2780 MCAsmMacroParameter FA;
2782 if (Lexer.is(AsmToken::Identifier) && Lexer.peekTok().is(AsmToken::Equal)) {
2783 if (parseIdentifier(FA.Name))
2784 return Error(IDLoc, "invalid argument identifier for formal argument");
2786 if (Lexer.isNot(AsmToken::Equal))
2787 return TokError("expected '=' after formal parameter identifier");
2789 Lex();
2791 NamedParametersFound = true;
2793 bool Vararg = HasVararg && Parameter == (NParameters - 1);
2795 if (NamedParametersFound && FA.Name.empty())
2796 return Error(IDLoc, "cannot mix positional and keyword arguments");
2798 SMLoc StrLoc = Lexer.getLoc();
2799 SMLoc EndLoc;
2800 if (AltMacroMode && Lexer.is(AsmToken::Percent)) {
2801 const MCExpr *AbsoluteExp;
2802 int64_t Value;
2803 /// Eat '%'
2804 Lex();
2805 if (parseExpression(AbsoluteExp, EndLoc))
2806 return false;
2807 if (!AbsoluteExp->evaluateAsAbsolute(Value,
2808 getStreamer().getAssemblerPtr()))
2809 return Error(StrLoc, "expected absolute expression");
2810 const char *StrChar = StrLoc.getPointer();
2811 const char *EndChar = EndLoc.getPointer();
2812 AsmToken newToken(AsmToken::Integer,
2813 StringRef(StrChar, EndChar - StrChar), Value);
2814 FA.Value.push_back(newToken);
2815 } else if (AltMacroMode && Lexer.is(AsmToken::Less) &&
2816 isAngleBracketString(StrLoc, EndLoc)) {
2817 const char *StrChar = StrLoc.getPointer();
2818 const char *EndChar = EndLoc.getPointer();
2819 jumpToLoc(EndLoc, CurBuffer);
2820 /// Eat from '<' to '>'
2821 Lex();
2822 AsmToken newToken(AsmToken::String,
2823 StringRef(StrChar, EndChar - StrChar));
2824 FA.Value.push_back(newToken);
2825 } else if(parseMacroArgument(FA.Value, Vararg))
2826 return true;
2828 unsigned PI = Parameter;
2829 if (!FA.Name.empty()) {
2830 unsigned FAI = 0;
2831 for (FAI = 0; FAI < NParameters; ++FAI)
2832 if (M->Parameters[FAI].Name == FA.Name)
2833 break;
2835 if (FAI >= NParameters) {
2836 assert(M && "expected macro to be defined");
2837 return Error(IDLoc, "parameter named '" + FA.Name +
2838 "' does not exist for macro '" + M->Name + "'");
2840 PI = FAI;
2843 if (!FA.Value.empty()) {
2844 if (A.size() <= PI)
2845 A.resize(PI + 1);
2846 A[PI] = FA.Value;
2848 if (FALocs.size() <= PI)
2849 FALocs.resize(PI + 1);
2851 FALocs[PI] = Lexer.getLoc();
2854 // At the end of the statement, fill in remaining arguments that have
2855 // default values. If there aren't any, then the next argument is
2856 // required but missing
2857 if (Lexer.is(AsmToken::EndOfStatement)) {
2858 bool Failure = false;
2859 for (unsigned FAI = 0; FAI < NParameters; ++FAI) {
2860 if (A[FAI].empty()) {
2861 if (M->Parameters[FAI].Required) {
2862 Error(FALocs[FAI].isValid() ? FALocs[FAI] : Lexer.getLoc(),
2863 "missing value for required parameter "
2864 "'" + M->Parameters[FAI].Name + "' in macro '" + M->Name + "'");
2865 Failure = true;
2868 if (!M->Parameters[FAI].Value.empty())
2869 A[FAI] = M->Parameters[FAI].Value;
2872 return Failure;
2875 parseOptionalToken(AsmToken::Comma);
2878 return TokError("too many positional arguments");
2881 bool AsmParser::handleMacroEntry(MCAsmMacro *M, SMLoc NameLoc) {
2882 // Arbitrarily limit macro nesting depth (default matches 'as'). We can
2883 // eliminate this, although we should protect against infinite loops.
2884 unsigned MaxNestingDepth = AsmMacroMaxNestingDepth;
2885 if (ActiveMacros.size() == MaxNestingDepth) {
2886 std::ostringstream MaxNestingDepthError;
2887 MaxNestingDepthError << "macros cannot be nested more than "
2888 << MaxNestingDepth << " levels deep."
2889 << " Use -asm-macro-max-nesting-depth to increase "
2890 "this limit.";
2891 return TokError(MaxNestingDepthError.str());
2894 MCAsmMacroArguments A;
2895 if (parseMacroArguments(M, A))
2896 return true;
2898 // Macro instantiation is lexical, unfortunately. We construct a new buffer
2899 // to hold the macro body with substitutions.
2900 SmallString<256> Buf;
2901 raw_svector_ostream OS(Buf);
2903 if ((!IsDarwin || M->Parameters.size()) && M->Parameters.size() != A.size())
2904 return Error(getTok().getLoc(), "Wrong number of arguments");
2905 if (expandMacro(OS, *M, M->Parameters, A, true))
2906 return true;
2908 // We include the .endmacro in the buffer as our cue to exit the macro
2909 // instantiation.
2910 OS << ".endmacro\n";
2912 std::unique_ptr<MemoryBuffer> Instantiation =
2913 MemoryBuffer::getMemBufferCopy(OS.str(), "<instantiation>");
2915 // Create the macro instantiation object and add to the current macro
2916 // instantiation stack.
2917 MacroInstantiation *MI = new MacroInstantiation{
2918 NameLoc, CurBuffer, getTok().getLoc(), TheCondStack.size()};
2919 ActiveMacros.push_back(MI);
2921 ++NumOfMacroInstantiations;
2923 // Jump to the macro instantiation and prime the lexer.
2924 CurBuffer = SrcMgr.AddNewSourceBuffer(std::move(Instantiation), SMLoc());
2925 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer());
2926 Lex();
2928 return false;
2931 void AsmParser::handleMacroExit() {
2932 // Jump to the EndOfStatement we should return to, and consume it.
2933 jumpToLoc(ActiveMacros.back()->ExitLoc, ActiveMacros.back()->ExitBuffer);
2934 Lex();
2935 // If .endm/.endr is followed by \n instead of a comment, consume it so that
2936 // we don't print an excess \n.
2937 if (getTok().is(AsmToken::EndOfStatement))
2938 Lex();
2940 // Pop the instantiation entry.
2941 delete ActiveMacros.back();
2942 ActiveMacros.pop_back();
2945 bool AsmParser::parseAssignment(StringRef Name, AssignmentKind Kind) {
2946 MCSymbol *Sym;
2947 const MCExpr *Value;
2948 SMLoc ExprLoc = getTok().getLoc();
2949 bool AllowRedef =
2950 Kind == AssignmentKind::Set || Kind == AssignmentKind::Equal;
2951 if (MCParserUtils::parseAssignmentExpression(Name, AllowRedef, *this, Sym,
2952 Value))
2953 return true;
2955 if (!Sym) {
2956 // In the case where we parse an expression starting with a '.', we will
2957 // not generate an error, nor will we create a symbol. In this case we
2958 // should just return out.
2959 return false;
2962 if (discardLTOSymbol(Name))
2963 return false;
2965 // Do the assignment.
2966 switch (Kind) {
2967 case AssignmentKind::Equal:
2968 Out.emitAssignment(Sym, Value);
2969 break;
2970 case AssignmentKind::Set:
2971 case AssignmentKind::Equiv:
2972 Out.emitAssignment(Sym, Value);
2973 Out.emitSymbolAttribute(Sym, MCSA_NoDeadStrip);
2974 break;
2975 case AssignmentKind::LTOSetConditional:
2976 if (Value->getKind() != MCExpr::SymbolRef)
2977 return Error(ExprLoc, "expected identifier");
2979 Out.emitConditionalAssignment(Sym, Value);
2980 break;
2983 return false;
2986 /// parseIdentifier:
2987 /// ::= identifier
2988 /// ::= string
2989 bool AsmParser::parseIdentifier(StringRef &Res) {
2990 // The assembler has relaxed rules for accepting identifiers, in particular we
2991 // allow things like '.globl $foo' and '.def @feat.00', which would normally be
2992 // separate tokens. At this level, we have already lexed so we cannot (currently)
2993 // handle this as a context dependent token, instead we detect adjacent tokens
2994 // and return the combined identifier.
2995 if (Lexer.is(AsmToken::Dollar) || Lexer.is(AsmToken::At)) {
2996 SMLoc PrefixLoc = getLexer().getLoc();
2998 // Consume the prefix character, and check for a following identifier.
3000 AsmToken Buf[1];
3001 Lexer.peekTokens(Buf, false);
3003 if (Buf[0].isNot(AsmToken::Identifier) && Buf[0].isNot(AsmToken::Integer))
3004 return true;
3006 // We have a '$' or '@' followed by an identifier or integer token, make
3007 // sure they are adjacent.
3008 if (PrefixLoc.getPointer() + 1 != Buf[0].getLoc().getPointer())
3009 return true;
3011 // eat $ or @
3012 Lexer.Lex(); // Lexer's Lex guarantees consecutive token.
3013 // Construct the joined identifier and consume the token.
3014 Res = StringRef(PrefixLoc.getPointer(), getTok().getString().size() + 1);
3015 Lex(); // Parser Lex to maintain invariants.
3016 return false;
3019 if (Lexer.isNot(AsmToken::Identifier) && Lexer.isNot(AsmToken::String))
3020 return true;
3022 Res = getTok().getIdentifier();
3024 Lex(); // Consume the identifier token.
3026 return false;
3029 /// parseDirectiveSet:
3030 /// ::= .equ identifier ',' expression
3031 /// ::= .equiv identifier ',' expression
3032 /// ::= .set identifier ',' expression
3033 /// ::= .lto_set_conditional identifier ',' expression
3034 bool AsmParser::parseDirectiveSet(StringRef IDVal, AssignmentKind Kind) {
3035 StringRef Name;
3036 if (check(parseIdentifier(Name), "expected identifier") || parseComma() ||
3037 parseAssignment(Name, Kind))
3038 return true;
3039 return false;
3042 bool AsmParser::parseEscapedString(std::string &Data) {
3043 if (check(getTok().isNot(AsmToken::String), "expected string"))
3044 return true;
3046 Data = "";
3047 StringRef Str = getTok().getStringContents();
3048 for (unsigned i = 0, e = Str.size(); i != e; ++i) {
3049 if (Str[i] != '\\') {
3050 if ((Str[i] == '\n') || (Str[i] == '\r')) {
3051 // Don't double-warn for Windows newlines.
3052 if ((Str[i] == '\n') && (i > 0) && (Str[i - 1] == '\r'))
3053 continue;
3055 SMLoc NewlineLoc = SMLoc::getFromPointer(Str.data() + i);
3056 if (Warning(NewlineLoc, "unterminated string; newline inserted"))
3057 return true;
3059 Data += Str[i];
3060 continue;
3063 // Recognize escaped characters. Note that this escape semantics currently
3064 // loosely follows Darwin 'as'.
3065 ++i;
3066 if (i == e)
3067 return TokError("unexpected backslash at end of string");
3069 // Recognize hex sequences similarly to GNU 'as'.
3070 if (Str[i] == 'x' || Str[i] == 'X') {
3071 size_t length = Str.size();
3072 if (i + 1 >= length || !isHexDigit(Str[i + 1]))
3073 return TokError("invalid hexadecimal escape sequence");
3075 // Consume hex characters. GNU 'as' reads all hexadecimal characters and
3076 // then truncates to the lower 16 bits. Seems reasonable.
3077 unsigned Value = 0;
3078 while (i + 1 < length && isHexDigit(Str[i + 1]))
3079 Value = Value * 16 + hexDigitValue(Str[++i]);
3081 Data += (unsigned char)(Value & 0xFF);
3082 continue;
3085 // Recognize octal sequences.
3086 if ((unsigned)(Str[i] - '0') <= 7) {
3087 // Consume up to three octal characters.
3088 unsigned Value = Str[i] - '0';
3090 if (i + 1 != e && ((unsigned)(Str[i + 1] - '0')) <= 7) {
3091 ++i;
3092 Value = Value * 8 + (Str[i] - '0');
3094 if (i + 1 != e && ((unsigned)(Str[i + 1] - '0')) <= 7) {
3095 ++i;
3096 Value = Value * 8 + (Str[i] - '0');
3100 if (Value > 255)
3101 return TokError("invalid octal escape sequence (out of range)");
3103 Data += (unsigned char)Value;
3104 continue;
3107 // Otherwise recognize individual escapes.
3108 switch (Str[i]) {
3109 default:
3110 // Just reject invalid escape sequences for now.
3111 return TokError("invalid escape sequence (unrecognized character)");
3113 case 'b': Data += '\b'; break;
3114 case 'f': Data += '\f'; break;
3115 case 'n': Data += '\n'; break;
3116 case 'r': Data += '\r'; break;
3117 case 't': Data += '\t'; break;
3118 case '"': Data += '"'; break;
3119 case '\\': Data += '\\'; break;
3123 Lex();
3124 return false;
3127 bool AsmParser::parseAngleBracketString(std::string &Data) {
3128 SMLoc EndLoc, StartLoc = getTok().getLoc();
3129 if (isAngleBracketString(StartLoc, EndLoc)) {
3130 const char *StartChar = StartLoc.getPointer() + 1;
3131 const char *EndChar = EndLoc.getPointer() - 1;
3132 jumpToLoc(EndLoc, CurBuffer);
3133 /// Eat from '<' to '>'
3134 Lex();
3136 Data = angleBracketString(StringRef(StartChar, EndChar - StartChar));
3137 return false;
3139 return true;
3142 /// parseDirectiveAscii:
3143 // ::= .ascii [ "string"+ ( , "string"+ )* ]
3144 /// ::= ( .asciz | .string ) [ "string" ( , "string" )* ]
3145 bool AsmParser::parseDirectiveAscii(StringRef IDVal, bool ZeroTerminated) {
3146 auto parseOp = [&]() -> bool {
3147 std::string Data;
3148 if (checkForValidSection())
3149 return true;
3150 // Only support spaces as separators for .ascii directive for now. See the
3151 // discusssion at https://reviews.llvm.org/D91460 for more details.
3152 do {
3153 if (parseEscapedString(Data))
3154 return true;
3155 getStreamer().emitBytes(Data);
3156 } while (!ZeroTerminated && getTok().is(AsmToken::String));
3157 if (ZeroTerminated)
3158 getStreamer().emitBytes(StringRef("\0", 1));
3159 return false;
3162 return parseMany(parseOp);
3165 /// parseDirectiveReloc
3166 /// ::= .reloc expression , identifier [ , expression ]
3167 bool AsmParser::parseDirectiveReloc(SMLoc DirectiveLoc) {
3168 const MCExpr *Offset;
3169 const MCExpr *Expr = nullptr;
3170 SMLoc OffsetLoc = Lexer.getTok().getLoc();
3172 if (parseExpression(Offset))
3173 return true;
3174 if (parseComma() ||
3175 check(getTok().isNot(AsmToken::Identifier), "expected relocation name"))
3176 return true;
3178 SMLoc NameLoc = Lexer.getTok().getLoc();
3179 StringRef Name = Lexer.getTok().getIdentifier();
3180 Lex();
3182 if (Lexer.is(AsmToken::Comma)) {
3183 Lex();
3184 SMLoc ExprLoc = Lexer.getLoc();
3185 if (parseExpression(Expr))
3186 return true;
3188 MCValue Value;
3189 if (!Expr->evaluateAsRelocatable(Value, nullptr, nullptr))
3190 return Error(ExprLoc, "expression must be relocatable");
3193 if (parseEOL())
3194 return true;
3196 const MCTargetAsmParser &MCT = getTargetParser();
3197 const MCSubtargetInfo &STI = MCT.getSTI();
3198 if (std::optional<std::pair<bool, std::string>> Err =
3199 getStreamer().emitRelocDirective(*Offset, Name, Expr, DirectiveLoc,
3200 STI))
3201 return Error(Err->first ? NameLoc : OffsetLoc, Err->second);
3203 return false;
3206 /// parseDirectiveValue
3207 /// ::= (.byte | .short | ... ) [ expression (, expression)* ]
3208 bool AsmParser::parseDirectiveValue(StringRef IDVal, unsigned Size) {
3209 auto parseOp = [&]() -> bool {
3210 const MCExpr *Value;
3211 SMLoc ExprLoc = getLexer().getLoc();
3212 if (checkForValidSection() || parseExpression(Value))
3213 return true;
3214 // Special case constant expressions to match code generator.
3215 if (const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value)) {
3216 assert(Size <= 8 && "Invalid size");
3217 uint64_t IntValue = MCE->getValue();
3218 if (!isUIntN(8 * Size, IntValue) && !isIntN(8 * Size, IntValue))
3219 return Error(ExprLoc, "out of range literal value");
3220 getStreamer().emitIntValue(IntValue, Size);
3221 } else
3222 getStreamer().emitValue(Value, Size, ExprLoc);
3223 return false;
3226 return parseMany(parseOp);
3229 static bool parseHexOcta(AsmParser &Asm, uint64_t &hi, uint64_t &lo) {
3230 if (Asm.getTok().isNot(AsmToken::Integer) &&
3231 Asm.getTok().isNot(AsmToken::BigNum))
3232 return Asm.TokError("unknown token in expression");
3233 SMLoc ExprLoc = Asm.getTok().getLoc();
3234 APInt IntValue = Asm.getTok().getAPIntVal();
3235 Asm.Lex();
3236 if (!IntValue.isIntN(128))
3237 return Asm.Error(ExprLoc, "out of range literal value");
3238 if (!IntValue.isIntN(64)) {
3239 hi = IntValue.getHiBits(IntValue.getBitWidth() - 64).getZExtValue();
3240 lo = IntValue.getLoBits(64).getZExtValue();
3241 } else {
3242 hi = 0;
3243 lo = IntValue.getZExtValue();
3245 return false;
3248 /// ParseDirectiveOctaValue
3249 /// ::= .octa [ hexconstant (, hexconstant)* ]
3251 bool AsmParser::parseDirectiveOctaValue(StringRef IDVal) {
3252 auto parseOp = [&]() -> bool {
3253 if (checkForValidSection())
3254 return true;
3255 uint64_t hi, lo;
3256 if (parseHexOcta(*this, hi, lo))
3257 return true;
3258 if (MAI.isLittleEndian()) {
3259 getStreamer().emitInt64(lo);
3260 getStreamer().emitInt64(hi);
3261 } else {
3262 getStreamer().emitInt64(hi);
3263 getStreamer().emitInt64(lo);
3265 return false;
3268 return parseMany(parseOp);
3271 bool AsmParser::parseRealValue(const fltSemantics &Semantics, APInt &Res) {
3272 // We don't truly support arithmetic on floating point expressions, so we
3273 // have to manually parse unary prefixes.
3274 bool IsNeg = false;
3275 if (getLexer().is(AsmToken::Minus)) {
3276 Lexer.Lex();
3277 IsNeg = true;
3278 } else if (getLexer().is(AsmToken::Plus))
3279 Lexer.Lex();
3281 if (Lexer.is(AsmToken::Error))
3282 return TokError(Lexer.getErr());
3283 if (Lexer.isNot(AsmToken::Integer) && Lexer.isNot(AsmToken::Real) &&
3284 Lexer.isNot(AsmToken::Identifier))
3285 return TokError("unexpected token in directive");
3287 // Convert to an APFloat.
3288 APFloat Value(Semantics);
3289 StringRef IDVal = getTok().getString();
3290 if (getLexer().is(AsmToken::Identifier)) {
3291 if (!IDVal.compare_insensitive("infinity") ||
3292 !IDVal.compare_insensitive("inf"))
3293 Value = APFloat::getInf(Semantics);
3294 else if (!IDVal.compare_insensitive("nan"))
3295 Value = APFloat::getNaN(Semantics, false, ~0);
3296 else
3297 return TokError("invalid floating point literal");
3298 } else if (errorToBool(
3299 Value.convertFromString(IDVal, APFloat::rmNearestTiesToEven)
3300 .takeError()))
3301 return TokError("invalid floating point literal");
3302 if (IsNeg)
3303 Value.changeSign();
3305 // Consume the numeric token.
3306 Lex();
3308 Res = Value.bitcastToAPInt();
3310 return false;
3313 /// parseDirectiveRealValue
3314 /// ::= (.single | .double) [ expression (, expression)* ]
3315 bool AsmParser::parseDirectiveRealValue(StringRef IDVal,
3316 const fltSemantics &Semantics) {
3317 auto parseOp = [&]() -> bool {
3318 APInt AsInt;
3319 if (checkForValidSection() || parseRealValue(Semantics, AsInt))
3320 return true;
3321 getStreamer().emitIntValue(AsInt.getLimitedValue(),
3322 AsInt.getBitWidth() / 8);
3323 return false;
3326 return parseMany(parseOp);
3329 /// parseDirectiveZero
3330 /// ::= .zero expression
3331 bool AsmParser::parseDirectiveZero() {
3332 SMLoc NumBytesLoc = Lexer.getLoc();
3333 const MCExpr *NumBytes;
3334 if (checkForValidSection() || parseExpression(NumBytes))
3335 return true;
3337 int64_t Val = 0;
3338 if (getLexer().is(AsmToken::Comma)) {
3339 Lex();
3340 if (parseAbsoluteExpression(Val))
3341 return true;
3344 if (parseEOL())
3345 return true;
3346 getStreamer().emitFill(*NumBytes, Val, NumBytesLoc);
3348 return false;
3351 /// parseDirectiveFill
3352 /// ::= .fill expression [ , expression [ , expression ] ]
3353 bool AsmParser::parseDirectiveFill() {
3354 SMLoc NumValuesLoc = Lexer.getLoc();
3355 const MCExpr *NumValues;
3356 if (checkForValidSection() || parseExpression(NumValues))
3357 return true;
3359 int64_t FillSize = 1;
3360 int64_t FillExpr = 0;
3362 SMLoc SizeLoc, ExprLoc;
3364 if (parseOptionalToken(AsmToken::Comma)) {
3365 SizeLoc = getTok().getLoc();
3366 if (parseAbsoluteExpression(FillSize))
3367 return true;
3368 if (parseOptionalToken(AsmToken::Comma)) {
3369 ExprLoc = getTok().getLoc();
3370 if (parseAbsoluteExpression(FillExpr))
3371 return true;
3374 if (parseEOL())
3375 return true;
3377 if (FillSize < 0) {
3378 Warning(SizeLoc, "'.fill' directive with negative size has no effect");
3379 return false;
3381 if (FillSize > 8) {
3382 Warning(SizeLoc, "'.fill' directive with size greater than 8 has been truncated to 8");
3383 FillSize = 8;
3386 if (!isUInt<32>(FillExpr) && FillSize > 4)
3387 Warning(ExprLoc, "'.fill' directive pattern has been truncated to 32-bits");
3389 getStreamer().emitFill(*NumValues, FillSize, FillExpr, NumValuesLoc);
3391 return false;
3394 /// parseDirectiveOrg
3395 /// ::= .org expression [ , expression ]
3396 bool AsmParser::parseDirectiveOrg() {
3397 const MCExpr *Offset;
3398 SMLoc OffsetLoc = Lexer.getLoc();
3399 if (checkForValidSection() || parseExpression(Offset))
3400 return true;
3402 // Parse optional fill expression.
3403 int64_t FillExpr = 0;
3404 if (parseOptionalToken(AsmToken::Comma))
3405 if (parseAbsoluteExpression(FillExpr))
3406 return true;
3407 if (parseEOL())
3408 return true;
3410 getStreamer().emitValueToOffset(Offset, FillExpr, OffsetLoc);
3411 return false;
3414 /// parseDirectiveAlign
3415 /// ::= {.align, ...} expression [ , expression [ , expression ]]
3416 bool AsmParser::parseDirectiveAlign(bool IsPow2, unsigned ValueSize) {
3417 SMLoc AlignmentLoc = getLexer().getLoc();
3418 int64_t Alignment;
3419 SMLoc MaxBytesLoc;
3420 bool HasFillExpr = false;
3421 int64_t FillExpr = 0;
3422 int64_t MaxBytesToFill = 0;
3423 SMLoc FillExprLoc;
3425 auto parseAlign = [&]() -> bool {
3426 if (parseAbsoluteExpression(Alignment))
3427 return true;
3428 if (parseOptionalToken(AsmToken::Comma)) {
3429 // The fill expression can be omitted while specifying a maximum number of
3430 // alignment bytes, e.g:
3431 // .align 3,,4
3432 if (getTok().isNot(AsmToken::Comma)) {
3433 HasFillExpr = true;
3434 if (parseTokenLoc(FillExprLoc) || parseAbsoluteExpression(FillExpr))
3435 return true;
3437 if (parseOptionalToken(AsmToken::Comma))
3438 if (parseTokenLoc(MaxBytesLoc) ||
3439 parseAbsoluteExpression(MaxBytesToFill))
3440 return true;
3442 return parseEOL();
3445 if (checkForValidSection())
3446 return true;
3447 // Ignore empty '.p2align' directives for GNU-as compatibility
3448 if (IsPow2 && (ValueSize == 1) && getTok().is(AsmToken::EndOfStatement)) {
3449 Warning(AlignmentLoc, "p2align directive with no operand(s) is ignored");
3450 return parseEOL();
3452 if (parseAlign())
3453 return true;
3455 // Always emit an alignment here even if we thrown an error.
3456 bool ReturnVal = false;
3458 // Compute alignment in bytes.
3459 if (IsPow2) {
3460 // FIXME: Diagnose overflow.
3461 if (Alignment >= 32) {
3462 ReturnVal |= Error(AlignmentLoc, "invalid alignment value");
3463 Alignment = 31;
3466 Alignment = 1ULL << Alignment;
3467 } else {
3468 // Reject alignments that aren't either a power of two or zero,
3469 // for gas compatibility. Alignment of zero is silently rounded
3470 // up to one.
3471 if (Alignment == 0)
3472 Alignment = 1;
3473 else if (!isPowerOf2_64(Alignment)) {
3474 ReturnVal |= Error(AlignmentLoc, "alignment must be a power of 2");
3475 Alignment = llvm::bit_floor<uint64_t>(Alignment);
3477 if (!isUInt<32>(Alignment)) {
3478 ReturnVal |= Error(AlignmentLoc, "alignment must be smaller than 2**32");
3479 Alignment = 1u << 31;
3483 // Diagnose non-sensical max bytes to align.
3484 if (MaxBytesLoc.isValid()) {
3485 if (MaxBytesToFill < 1) {
3486 ReturnVal |= Error(MaxBytesLoc,
3487 "alignment directive can never be satisfied in this "
3488 "many bytes, ignoring maximum bytes expression");
3489 MaxBytesToFill = 0;
3492 if (MaxBytesToFill >= Alignment) {
3493 Warning(MaxBytesLoc, "maximum bytes expression exceeds alignment and "
3494 "has no effect");
3495 MaxBytesToFill = 0;
3499 const MCSection *Section = getStreamer().getCurrentSectionOnly();
3500 assert(Section && "must have section to emit alignment");
3502 if (HasFillExpr && FillExpr != 0 && Section->isVirtualSection()) {
3503 ReturnVal |=
3504 Warning(FillExprLoc, "ignoring non-zero fill value in " +
3505 Section->getVirtualSectionKind() +
3506 " section '" + Section->getName() + "'");
3507 FillExpr = 0;
3510 // Check whether we should use optimal code alignment for this .align
3511 // directive.
3512 if (Section->useCodeAlign() && !HasFillExpr) {
3513 getStreamer().emitCodeAlignment(
3514 Align(Alignment), &getTargetParser().getSTI(), MaxBytesToFill);
3515 } else {
3516 // FIXME: Target specific behavior about how the "extra" bytes are filled.
3517 getStreamer().emitValueToAlignment(Align(Alignment), FillExpr, ValueSize,
3518 MaxBytesToFill);
3521 return ReturnVal;
3524 /// parseDirectiveFile
3525 /// ::= .file filename
3526 /// ::= .file number [directory] filename [md5 checksum] [source source-text]
3527 bool AsmParser::parseDirectiveFile(SMLoc DirectiveLoc) {
3528 // FIXME: I'm not sure what this is.
3529 int64_t FileNumber = -1;
3530 if (getLexer().is(AsmToken::Integer)) {
3531 FileNumber = getTok().getIntVal();
3532 Lex();
3534 if (FileNumber < 0)
3535 return TokError("negative file number");
3538 std::string Path;
3540 // Usually the directory and filename together, otherwise just the directory.
3541 // Allow the strings to have escaped octal character sequence.
3542 if (parseEscapedString(Path))
3543 return true;
3545 StringRef Directory;
3546 StringRef Filename;
3547 std::string FilenameData;
3548 if (getLexer().is(AsmToken::String)) {
3549 if (check(FileNumber == -1,
3550 "explicit path specified, but no file number") ||
3551 parseEscapedString(FilenameData))
3552 return true;
3553 Filename = FilenameData;
3554 Directory = Path;
3555 } else {
3556 Filename = Path;
3559 uint64_t MD5Hi, MD5Lo;
3560 bool HasMD5 = false;
3562 std::optional<StringRef> Source;
3563 bool HasSource = false;
3564 std::string SourceString;
3566 while (!parseOptionalToken(AsmToken::EndOfStatement)) {
3567 StringRef Keyword;
3568 if (check(getTok().isNot(AsmToken::Identifier),
3569 "unexpected token in '.file' directive") ||
3570 parseIdentifier(Keyword))
3571 return true;
3572 if (Keyword == "md5") {
3573 HasMD5 = true;
3574 if (check(FileNumber == -1,
3575 "MD5 checksum specified, but no file number") ||
3576 parseHexOcta(*this, MD5Hi, MD5Lo))
3577 return true;
3578 } else if (Keyword == "source") {
3579 HasSource = true;
3580 if (check(FileNumber == -1,
3581 "source specified, but no file number") ||
3582 check(getTok().isNot(AsmToken::String),
3583 "unexpected token in '.file' directive") ||
3584 parseEscapedString(SourceString))
3585 return true;
3586 } else {
3587 return TokError("unexpected token in '.file' directive");
3591 if (FileNumber == -1) {
3592 // Ignore the directive if there is no number and the target doesn't support
3593 // numberless .file directives. This allows some portability of assembler
3594 // between different object file formats.
3595 if (getContext().getAsmInfo()->hasSingleParameterDotFile())
3596 getStreamer().emitFileDirective(Filename);
3597 } else {
3598 // In case there is a -g option as well as debug info from directive .file,
3599 // we turn off the -g option, directly use the existing debug info instead.
3600 // Throw away any implicit file table for the assembler source.
3601 if (Ctx.getGenDwarfForAssembly()) {
3602 Ctx.getMCDwarfLineTable(0).resetFileTable();
3603 Ctx.setGenDwarfForAssembly(false);
3606 std::optional<MD5::MD5Result> CKMem;
3607 if (HasMD5) {
3608 MD5::MD5Result Sum;
3609 for (unsigned i = 0; i != 8; ++i) {
3610 Sum[i] = uint8_t(MD5Hi >> ((7 - i) * 8));
3611 Sum[i + 8] = uint8_t(MD5Lo >> ((7 - i) * 8));
3613 CKMem = Sum;
3615 if (HasSource) {
3616 char *SourceBuf = static_cast<char *>(Ctx.allocate(SourceString.size()));
3617 memcpy(SourceBuf, SourceString.data(), SourceString.size());
3618 Source = StringRef(SourceBuf, SourceString.size());
3620 if (FileNumber == 0) {
3621 // Upgrade to Version 5 for assembly actions like clang -c a.s.
3622 if (Ctx.getDwarfVersion() < 5)
3623 Ctx.setDwarfVersion(5);
3624 getStreamer().emitDwarfFile0Directive(Directory, Filename, CKMem, Source);
3625 } else {
3626 Expected<unsigned> FileNumOrErr = getStreamer().tryEmitDwarfFileDirective(
3627 FileNumber, Directory, Filename, CKMem, Source);
3628 if (!FileNumOrErr)
3629 return Error(DirectiveLoc, toString(FileNumOrErr.takeError()));
3631 // Alert the user if there are some .file directives with MD5 and some not.
3632 // But only do that once.
3633 if (!ReportedInconsistentMD5 && !Ctx.isDwarfMD5UsageConsistent(0)) {
3634 ReportedInconsistentMD5 = true;
3635 return Warning(DirectiveLoc, "inconsistent use of MD5 checksums");
3639 return false;
3642 /// parseDirectiveLine
3643 /// ::= .line [number]
3644 bool AsmParser::parseDirectiveLine() {
3645 int64_t LineNumber;
3646 if (getLexer().is(AsmToken::Integer)) {
3647 if (parseIntToken(LineNumber, "unexpected token in '.line' directive"))
3648 return true;
3649 (void)LineNumber;
3650 // FIXME: Do something with the .line.
3652 return parseEOL();
3655 /// parseDirectiveLoc
3656 /// ::= .loc FileNumber [LineNumber] [ColumnPos] [basic_block] [prologue_end]
3657 /// [epilogue_begin] [is_stmt VALUE] [isa VALUE]
3658 /// The first number is a file number, must have been previously assigned with
3659 /// a .file directive, the second number is the line number and optionally the
3660 /// third number is a column position (zero if not specified). The remaining
3661 /// optional items are .loc sub-directives.
3662 bool AsmParser::parseDirectiveLoc() {
3663 int64_t FileNumber = 0, LineNumber = 0;
3664 SMLoc Loc = getTok().getLoc();
3665 if (parseIntToken(FileNumber, "unexpected token in '.loc' directive") ||
3666 check(FileNumber < 1 && Ctx.getDwarfVersion() < 5, Loc,
3667 "file number less than one in '.loc' directive") ||
3668 check(!getContext().isValidDwarfFileNumber(FileNumber), Loc,
3669 "unassigned file number in '.loc' directive"))
3670 return true;
3672 // optional
3673 if (getLexer().is(AsmToken::Integer)) {
3674 LineNumber = getTok().getIntVal();
3675 if (LineNumber < 0)
3676 return TokError("line number less than zero in '.loc' directive");
3677 Lex();
3680 int64_t ColumnPos = 0;
3681 if (getLexer().is(AsmToken::Integer)) {
3682 ColumnPos = getTok().getIntVal();
3683 if (ColumnPos < 0)
3684 return TokError("column position less than zero in '.loc' directive");
3685 Lex();
3688 auto PrevFlags = getContext().getCurrentDwarfLoc().getFlags();
3689 unsigned Flags = PrevFlags & DWARF2_FLAG_IS_STMT;
3690 unsigned Isa = 0;
3691 int64_t Discriminator = 0;
3693 auto parseLocOp = [&]() -> bool {
3694 StringRef Name;
3695 SMLoc Loc = getTok().getLoc();
3696 if (parseIdentifier(Name))
3697 return TokError("unexpected token in '.loc' directive");
3699 if (Name == "basic_block")
3700 Flags |= DWARF2_FLAG_BASIC_BLOCK;
3701 else if (Name == "prologue_end")
3702 Flags |= DWARF2_FLAG_PROLOGUE_END;
3703 else if (Name == "epilogue_begin")
3704 Flags |= DWARF2_FLAG_EPILOGUE_BEGIN;
3705 else if (Name == "is_stmt") {
3706 Loc = getTok().getLoc();
3707 const MCExpr *Value;
3708 if (parseExpression(Value))
3709 return true;
3710 // The expression must be the constant 0 or 1.
3711 if (const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value)) {
3712 int Value = MCE->getValue();
3713 if (Value == 0)
3714 Flags &= ~DWARF2_FLAG_IS_STMT;
3715 else if (Value == 1)
3716 Flags |= DWARF2_FLAG_IS_STMT;
3717 else
3718 return Error(Loc, "is_stmt value not 0 or 1");
3719 } else {
3720 return Error(Loc, "is_stmt value not the constant value of 0 or 1");
3722 } else if (Name == "isa") {
3723 Loc = getTok().getLoc();
3724 const MCExpr *Value;
3725 if (parseExpression(Value))
3726 return true;
3727 // The expression must be a constant greater or equal to 0.
3728 if (const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value)) {
3729 int Value = MCE->getValue();
3730 if (Value < 0)
3731 return Error(Loc, "isa number less than zero");
3732 Isa = Value;
3733 } else {
3734 return Error(Loc, "isa number not a constant value");
3736 } else if (Name == "discriminator") {
3737 if (parseAbsoluteExpression(Discriminator))
3738 return true;
3739 } else {
3740 return Error(Loc, "unknown sub-directive in '.loc' directive");
3742 return false;
3745 if (parseMany(parseLocOp, false /*hasComma*/))
3746 return true;
3748 getStreamer().emitDwarfLocDirective(FileNumber, LineNumber, ColumnPos, Flags,
3749 Isa, Discriminator, StringRef());
3751 return false;
3754 /// parseDirectiveLoc
3755 /// ::= .loc_label label
3756 bool AsmParser::parseDirectiveLocLabel(SMLoc DirectiveLoc) {
3757 StringRef Name;
3758 DirectiveLoc = Lexer.getLoc();
3759 if (parseIdentifier(Name))
3760 return TokError("expected identifier");
3761 if (parseEOL())
3762 return true;
3763 getStreamer().emitDwarfLocLabelDirective(DirectiveLoc, Name);
3764 return false;
3767 /// parseDirectiveStabs
3768 /// ::= .stabs string, number, number, number
3769 bool AsmParser::parseDirectiveStabs() {
3770 return TokError("unsupported directive '.stabs'");
3773 /// parseDirectiveCVFile
3774 /// ::= .cv_file number filename [checksum] [checksumkind]
3775 bool AsmParser::parseDirectiveCVFile() {
3776 SMLoc FileNumberLoc = getTok().getLoc();
3777 int64_t FileNumber;
3778 std::string Filename;
3779 std::string Checksum;
3780 int64_t ChecksumKind = 0;
3782 if (parseIntToken(FileNumber,
3783 "expected file number in '.cv_file' directive") ||
3784 check(FileNumber < 1, FileNumberLoc, "file number less than one") ||
3785 check(getTok().isNot(AsmToken::String),
3786 "unexpected token in '.cv_file' directive") ||
3787 parseEscapedString(Filename))
3788 return true;
3789 if (!parseOptionalToken(AsmToken::EndOfStatement)) {
3790 if (check(getTok().isNot(AsmToken::String),
3791 "unexpected token in '.cv_file' directive") ||
3792 parseEscapedString(Checksum) ||
3793 parseIntToken(ChecksumKind,
3794 "expected checksum kind in '.cv_file' directive") ||
3795 parseEOL())
3796 return true;
3799 Checksum = fromHex(Checksum);
3800 void *CKMem = Ctx.allocate(Checksum.size(), 1);
3801 memcpy(CKMem, Checksum.data(), Checksum.size());
3802 ArrayRef<uint8_t> ChecksumAsBytes(reinterpret_cast<const uint8_t *>(CKMem),
3803 Checksum.size());
3805 if (!getStreamer().emitCVFileDirective(FileNumber, Filename, ChecksumAsBytes,
3806 static_cast<uint8_t>(ChecksumKind)))
3807 return Error(FileNumberLoc, "file number already allocated");
3809 return false;
3812 bool AsmParser::parseCVFunctionId(int64_t &FunctionId,
3813 StringRef DirectiveName) {
3814 SMLoc Loc;
3815 return parseTokenLoc(Loc) ||
3816 parseIntToken(FunctionId, "expected function id in '" + DirectiveName +
3817 "' directive") ||
3818 check(FunctionId < 0 || FunctionId >= UINT_MAX, Loc,
3819 "expected function id within range [0, UINT_MAX)");
3822 bool AsmParser::parseCVFileId(int64_t &FileNumber, StringRef DirectiveName) {
3823 SMLoc Loc;
3824 return parseTokenLoc(Loc) ||
3825 parseIntToken(FileNumber, "expected integer in '" + DirectiveName +
3826 "' directive") ||
3827 check(FileNumber < 1, Loc, "file number less than one in '" +
3828 DirectiveName + "' directive") ||
3829 check(!getCVContext().isValidFileNumber(FileNumber), Loc,
3830 "unassigned file number in '" + DirectiveName + "' directive");
3833 /// parseDirectiveCVFuncId
3834 /// ::= .cv_func_id FunctionId
3836 /// Introduces a function ID that can be used with .cv_loc.
3837 bool AsmParser::parseDirectiveCVFuncId() {
3838 SMLoc FunctionIdLoc = getTok().getLoc();
3839 int64_t FunctionId;
3841 if (parseCVFunctionId(FunctionId, ".cv_func_id") || parseEOL())
3842 return true;
3844 if (!getStreamer().emitCVFuncIdDirective(FunctionId))
3845 return Error(FunctionIdLoc, "function id already allocated");
3847 return false;
3850 /// parseDirectiveCVInlineSiteId
3851 /// ::= .cv_inline_site_id FunctionId
3852 /// "within" IAFunc
3853 /// "inlined_at" IAFile IALine [IACol]
3855 /// Introduces a function ID that can be used with .cv_loc. Includes "inlined
3856 /// at" source location information for use in the line table of the caller,
3857 /// whether the caller is a real function or another inlined call site.
3858 bool AsmParser::parseDirectiveCVInlineSiteId() {
3859 SMLoc FunctionIdLoc = getTok().getLoc();
3860 int64_t FunctionId;
3861 int64_t IAFunc;
3862 int64_t IAFile;
3863 int64_t IALine;
3864 int64_t IACol = 0;
3866 // FunctionId
3867 if (parseCVFunctionId(FunctionId, ".cv_inline_site_id"))
3868 return true;
3870 // "within"
3871 if (check((getLexer().isNot(AsmToken::Identifier) ||
3872 getTok().getIdentifier() != "within"),
3873 "expected 'within' identifier in '.cv_inline_site_id' directive"))
3874 return true;
3875 Lex();
3877 // IAFunc
3878 if (parseCVFunctionId(IAFunc, ".cv_inline_site_id"))
3879 return true;
3881 // "inlined_at"
3882 if (check((getLexer().isNot(AsmToken::Identifier) ||
3883 getTok().getIdentifier() != "inlined_at"),
3884 "expected 'inlined_at' identifier in '.cv_inline_site_id' "
3885 "directive") )
3886 return true;
3887 Lex();
3889 // IAFile IALine
3890 if (parseCVFileId(IAFile, ".cv_inline_site_id") ||
3891 parseIntToken(IALine, "expected line number after 'inlined_at'"))
3892 return true;
3894 // [IACol]
3895 if (getLexer().is(AsmToken::Integer)) {
3896 IACol = getTok().getIntVal();
3897 Lex();
3900 if (parseEOL())
3901 return true;
3903 if (!getStreamer().emitCVInlineSiteIdDirective(FunctionId, IAFunc, IAFile,
3904 IALine, IACol, FunctionIdLoc))
3905 return Error(FunctionIdLoc, "function id already allocated");
3907 return false;
3910 /// parseDirectiveCVLoc
3911 /// ::= .cv_loc FunctionId FileNumber [LineNumber] [ColumnPos] [prologue_end]
3912 /// [is_stmt VALUE]
3913 /// The first number is a file number, must have been previously assigned with
3914 /// a .file directive, the second number is the line number and optionally the
3915 /// third number is a column position (zero if not specified). The remaining
3916 /// optional items are .loc sub-directives.
3917 bool AsmParser::parseDirectiveCVLoc() {
3918 SMLoc DirectiveLoc = getTok().getLoc();
3919 int64_t FunctionId, FileNumber;
3920 if (parseCVFunctionId(FunctionId, ".cv_loc") ||
3921 parseCVFileId(FileNumber, ".cv_loc"))
3922 return true;
3924 int64_t LineNumber = 0;
3925 if (getLexer().is(AsmToken::Integer)) {
3926 LineNumber = getTok().getIntVal();
3927 if (LineNumber < 0)
3928 return TokError("line number less than zero in '.cv_loc' directive");
3929 Lex();
3932 int64_t ColumnPos = 0;
3933 if (getLexer().is(AsmToken::Integer)) {
3934 ColumnPos = getTok().getIntVal();
3935 if (ColumnPos < 0)
3936 return TokError("column position less than zero in '.cv_loc' directive");
3937 Lex();
3940 bool PrologueEnd = false;
3941 uint64_t IsStmt = 0;
3943 auto parseOp = [&]() -> bool {
3944 StringRef Name;
3945 SMLoc Loc = getTok().getLoc();
3946 if (parseIdentifier(Name))
3947 return TokError("unexpected token in '.cv_loc' directive");
3948 if (Name == "prologue_end")
3949 PrologueEnd = true;
3950 else if (Name == "is_stmt") {
3951 Loc = getTok().getLoc();
3952 const MCExpr *Value;
3953 if (parseExpression(Value))
3954 return true;
3955 // The expression must be the constant 0 or 1.
3956 IsStmt = ~0ULL;
3957 if (const auto *MCE = dyn_cast<MCConstantExpr>(Value))
3958 IsStmt = MCE->getValue();
3960 if (IsStmt > 1)
3961 return Error(Loc, "is_stmt value not 0 or 1");
3962 } else {
3963 return Error(Loc, "unknown sub-directive in '.cv_loc' directive");
3965 return false;
3968 if (parseMany(parseOp, false /*hasComma*/))
3969 return true;
3971 getStreamer().emitCVLocDirective(FunctionId, FileNumber, LineNumber,
3972 ColumnPos, PrologueEnd, IsStmt, StringRef(),
3973 DirectiveLoc);
3974 return false;
3977 /// parseDirectiveCVLinetable
3978 /// ::= .cv_linetable FunctionId, FnStart, FnEnd
3979 bool AsmParser::parseDirectiveCVLinetable() {
3980 int64_t FunctionId;
3981 StringRef FnStartName, FnEndName;
3982 SMLoc Loc = getTok().getLoc();
3983 if (parseCVFunctionId(FunctionId, ".cv_linetable") || parseComma() ||
3984 parseTokenLoc(Loc) ||
3985 check(parseIdentifier(FnStartName), Loc,
3986 "expected identifier in directive") ||
3987 parseComma() || parseTokenLoc(Loc) ||
3988 check(parseIdentifier(FnEndName), Loc,
3989 "expected identifier in directive"))
3990 return true;
3992 MCSymbol *FnStartSym = getContext().getOrCreateSymbol(FnStartName);
3993 MCSymbol *FnEndSym = getContext().getOrCreateSymbol(FnEndName);
3995 getStreamer().emitCVLinetableDirective(FunctionId, FnStartSym, FnEndSym);
3996 return false;
3999 /// parseDirectiveCVInlineLinetable
4000 /// ::= .cv_inline_linetable PrimaryFunctionId FileId LineNum FnStart FnEnd
4001 bool AsmParser::parseDirectiveCVInlineLinetable() {
4002 int64_t PrimaryFunctionId, SourceFileId, SourceLineNum;
4003 StringRef FnStartName, FnEndName;
4004 SMLoc Loc = getTok().getLoc();
4005 if (parseCVFunctionId(PrimaryFunctionId, ".cv_inline_linetable") ||
4006 parseTokenLoc(Loc) ||
4007 parseIntToken(
4008 SourceFileId,
4009 "expected SourceField in '.cv_inline_linetable' directive") ||
4010 check(SourceFileId <= 0, Loc,
4011 "File id less than zero in '.cv_inline_linetable' directive") ||
4012 parseTokenLoc(Loc) ||
4013 parseIntToken(
4014 SourceLineNum,
4015 "expected SourceLineNum in '.cv_inline_linetable' directive") ||
4016 check(SourceLineNum < 0, Loc,
4017 "Line number less than zero in '.cv_inline_linetable' directive") ||
4018 parseTokenLoc(Loc) || check(parseIdentifier(FnStartName), Loc,
4019 "expected identifier in directive") ||
4020 parseTokenLoc(Loc) || check(parseIdentifier(FnEndName), Loc,
4021 "expected identifier in directive"))
4022 return true;
4024 if (parseEOL())
4025 return true;
4027 MCSymbol *FnStartSym = getContext().getOrCreateSymbol(FnStartName);
4028 MCSymbol *FnEndSym = getContext().getOrCreateSymbol(FnEndName);
4029 getStreamer().emitCVInlineLinetableDirective(PrimaryFunctionId, SourceFileId,
4030 SourceLineNum, FnStartSym,
4031 FnEndSym);
4032 return false;
4035 void AsmParser::initializeCVDefRangeTypeMap() {
4036 CVDefRangeTypeMap["reg"] = CVDR_DEFRANGE_REGISTER;
4037 CVDefRangeTypeMap["frame_ptr_rel"] = CVDR_DEFRANGE_FRAMEPOINTER_REL;
4038 CVDefRangeTypeMap["subfield_reg"] = CVDR_DEFRANGE_SUBFIELD_REGISTER;
4039 CVDefRangeTypeMap["reg_rel"] = CVDR_DEFRANGE_REGISTER_REL;
4042 /// parseDirectiveCVDefRange
4043 /// ::= .cv_def_range RangeStart RangeEnd (GapStart GapEnd)*, bytes*
4044 bool AsmParser::parseDirectiveCVDefRange() {
4045 SMLoc Loc;
4046 std::vector<std::pair<const MCSymbol *, const MCSymbol *>> Ranges;
4047 while (getLexer().is(AsmToken::Identifier)) {
4048 Loc = getLexer().getLoc();
4049 StringRef GapStartName;
4050 if (parseIdentifier(GapStartName))
4051 return Error(Loc, "expected identifier in directive");
4052 MCSymbol *GapStartSym = getContext().getOrCreateSymbol(GapStartName);
4054 Loc = getLexer().getLoc();
4055 StringRef GapEndName;
4056 if (parseIdentifier(GapEndName))
4057 return Error(Loc, "expected identifier in directive");
4058 MCSymbol *GapEndSym = getContext().getOrCreateSymbol(GapEndName);
4060 Ranges.push_back({GapStartSym, GapEndSym});
4063 StringRef CVDefRangeTypeStr;
4064 if (parseToken(
4065 AsmToken::Comma,
4066 "expected comma before def_range type in .cv_def_range directive") ||
4067 parseIdentifier(CVDefRangeTypeStr))
4068 return Error(Loc, "expected def_range type in directive");
4070 StringMap<CVDefRangeType>::const_iterator CVTypeIt =
4071 CVDefRangeTypeMap.find(CVDefRangeTypeStr);
4072 CVDefRangeType CVDRType = (CVTypeIt == CVDefRangeTypeMap.end())
4073 ? CVDR_DEFRANGE
4074 : CVTypeIt->getValue();
4075 switch (CVDRType) {
4076 case CVDR_DEFRANGE_REGISTER: {
4077 int64_t DRRegister;
4078 if (parseToken(AsmToken::Comma, "expected comma before register number in "
4079 ".cv_def_range directive") ||
4080 parseAbsoluteExpression(DRRegister))
4081 return Error(Loc, "expected register number");
4083 codeview::DefRangeRegisterHeader DRHdr;
4084 DRHdr.Register = DRRegister;
4085 DRHdr.MayHaveNoName = 0;
4086 getStreamer().emitCVDefRangeDirective(Ranges, DRHdr);
4087 break;
4089 case CVDR_DEFRANGE_FRAMEPOINTER_REL: {
4090 int64_t DROffset;
4091 if (parseToken(AsmToken::Comma,
4092 "expected comma before offset in .cv_def_range directive") ||
4093 parseAbsoluteExpression(DROffset))
4094 return Error(Loc, "expected offset value");
4096 codeview::DefRangeFramePointerRelHeader DRHdr;
4097 DRHdr.Offset = DROffset;
4098 getStreamer().emitCVDefRangeDirective(Ranges, DRHdr);
4099 break;
4101 case CVDR_DEFRANGE_SUBFIELD_REGISTER: {
4102 int64_t DRRegister;
4103 int64_t DROffsetInParent;
4104 if (parseToken(AsmToken::Comma, "expected comma before register number in "
4105 ".cv_def_range directive") ||
4106 parseAbsoluteExpression(DRRegister))
4107 return Error(Loc, "expected register number");
4108 if (parseToken(AsmToken::Comma,
4109 "expected comma before offset in .cv_def_range directive") ||
4110 parseAbsoluteExpression(DROffsetInParent))
4111 return Error(Loc, "expected offset value");
4113 codeview::DefRangeSubfieldRegisterHeader DRHdr;
4114 DRHdr.Register = DRRegister;
4115 DRHdr.MayHaveNoName = 0;
4116 DRHdr.OffsetInParent = DROffsetInParent;
4117 getStreamer().emitCVDefRangeDirective(Ranges, DRHdr);
4118 break;
4120 case CVDR_DEFRANGE_REGISTER_REL: {
4121 int64_t DRRegister;
4122 int64_t DRFlags;
4123 int64_t DRBasePointerOffset;
4124 if (parseToken(AsmToken::Comma, "expected comma before register number in "
4125 ".cv_def_range directive") ||
4126 parseAbsoluteExpression(DRRegister))
4127 return Error(Loc, "expected register value");
4128 if (parseToken(
4129 AsmToken::Comma,
4130 "expected comma before flag value in .cv_def_range directive") ||
4131 parseAbsoluteExpression(DRFlags))
4132 return Error(Loc, "expected flag value");
4133 if (parseToken(AsmToken::Comma, "expected comma before base pointer offset "
4134 "in .cv_def_range directive") ||
4135 parseAbsoluteExpression(DRBasePointerOffset))
4136 return Error(Loc, "expected base pointer offset value");
4138 codeview::DefRangeRegisterRelHeader DRHdr;
4139 DRHdr.Register = DRRegister;
4140 DRHdr.Flags = DRFlags;
4141 DRHdr.BasePointerOffset = DRBasePointerOffset;
4142 getStreamer().emitCVDefRangeDirective(Ranges, DRHdr);
4143 break;
4145 default:
4146 return Error(Loc, "unexpected def_range type in .cv_def_range directive");
4148 return true;
4151 /// parseDirectiveCVString
4152 /// ::= .cv_stringtable "string"
4153 bool AsmParser::parseDirectiveCVString() {
4154 std::string Data;
4155 if (checkForValidSection() || parseEscapedString(Data))
4156 return true;
4158 // Put the string in the table and emit the offset.
4159 std::pair<StringRef, unsigned> Insertion =
4160 getCVContext().addToStringTable(Data);
4161 getStreamer().emitInt32(Insertion.second);
4162 return false;
4165 /// parseDirectiveCVStringTable
4166 /// ::= .cv_stringtable
4167 bool AsmParser::parseDirectiveCVStringTable() {
4168 getStreamer().emitCVStringTableDirective();
4169 return false;
4172 /// parseDirectiveCVFileChecksums
4173 /// ::= .cv_filechecksums
4174 bool AsmParser::parseDirectiveCVFileChecksums() {
4175 getStreamer().emitCVFileChecksumsDirective();
4176 return false;
4179 /// parseDirectiveCVFileChecksumOffset
4180 /// ::= .cv_filechecksumoffset fileno
4181 bool AsmParser::parseDirectiveCVFileChecksumOffset() {
4182 int64_t FileNo;
4183 if (parseIntToken(FileNo, "expected identifier in directive"))
4184 return true;
4185 if (parseEOL())
4186 return true;
4187 getStreamer().emitCVFileChecksumOffsetDirective(FileNo);
4188 return false;
4191 /// parseDirectiveCVFPOData
4192 /// ::= .cv_fpo_data procsym
4193 bool AsmParser::parseDirectiveCVFPOData() {
4194 SMLoc DirLoc = getLexer().getLoc();
4195 StringRef ProcName;
4196 if (parseIdentifier(ProcName))
4197 return TokError("expected symbol name");
4198 if (parseEOL())
4199 return true;
4200 MCSymbol *ProcSym = getContext().getOrCreateSymbol(ProcName);
4201 getStreamer().emitCVFPOData(ProcSym, DirLoc);
4202 return false;
4205 /// parseDirectiveCFISections
4206 /// ::= .cfi_sections section [, section]
4207 bool AsmParser::parseDirectiveCFISections() {
4208 StringRef Name;
4209 bool EH = false;
4210 bool Debug = false;
4212 if (!parseOptionalToken(AsmToken::EndOfStatement)) {
4213 for (;;) {
4214 if (parseIdentifier(Name))
4215 return TokError("expected .eh_frame or .debug_frame");
4216 if (Name == ".eh_frame")
4217 EH = true;
4218 else if (Name == ".debug_frame")
4219 Debug = true;
4220 if (parseOptionalToken(AsmToken::EndOfStatement))
4221 break;
4222 if (parseComma())
4223 return true;
4226 getStreamer().emitCFISections(EH, Debug);
4227 return false;
4230 /// parseDirectiveCFIStartProc
4231 /// ::= .cfi_startproc [simple]
4232 bool AsmParser::parseDirectiveCFIStartProc() {
4233 CFIStartProcLoc = StartTokLoc;
4235 StringRef Simple;
4236 if (!parseOptionalToken(AsmToken::EndOfStatement)) {
4237 if (check(parseIdentifier(Simple) || Simple != "simple",
4238 "unexpected token") ||
4239 parseEOL())
4240 return true;
4243 // TODO(kristina): Deal with a corner case of incorrect diagnostic context
4244 // being produced if this directive is emitted as part of preprocessor macro
4245 // expansion which can *ONLY* happen if Clang's cc1as is the API consumer.
4246 // Tools like llvm-mc on the other hand are not affected by it, and report
4247 // correct context information.
4248 getStreamer().emitCFIStartProc(!Simple.empty(), Lexer.getLoc());
4249 return false;
4252 /// parseDirectiveCFIEndProc
4253 /// ::= .cfi_endproc
4254 bool AsmParser::parseDirectiveCFIEndProc() {
4255 CFIStartProcLoc = std::nullopt;
4257 if (parseEOL())
4258 return true;
4260 getStreamer().emitCFIEndProc();
4261 return false;
4264 /// parse register name or number.
4265 bool AsmParser::parseRegisterOrRegisterNumber(int64_t &Register,
4266 SMLoc DirectiveLoc) {
4267 MCRegister RegNo;
4269 if (getLexer().isNot(AsmToken::Integer)) {
4270 if (getTargetParser().parseRegister(RegNo, DirectiveLoc, DirectiveLoc))
4271 return true;
4272 Register = getContext().getRegisterInfo()->getDwarfRegNum(RegNo, true);
4273 } else
4274 return parseAbsoluteExpression(Register);
4276 return false;
4279 /// parseDirectiveCFIDefCfa
4280 /// ::= .cfi_def_cfa register, offset
4281 bool AsmParser::parseDirectiveCFIDefCfa(SMLoc DirectiveLoc) {
4282 int64_t Register = 0, Offset = 0;
4283 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc) || parseComma() ||
4284 parseAbsoluteExpression(Offset) || parseEOL())
4285 return true;
4287 getStreamer().emitCFIDefCfa(Register, Offset, DirectiveLoc);
4288 return false;
4291 /// parseDirectiveCFIDefCfaOffset
4292 /// ::= .cfi_def_cfa_offset offset
4293 bool AsmParser::parseDirectiveCFIDefCfaOffset(SMLoc DirectiveLoc) {
4294 int64_t Offset = 0;
4295 if (parseAbsoluteExpression(Offset) || parseEOL())
4296 return true;
4298 getStreamer().emitCFIDefCfaOffset(Offset, DirectiveLoc);
4299 return false;
4302 /// parseDirectiveCFIRegister
4303 /// ::= .cfi_register register, register
4304 bool AsmParser::parseDirectiveCFIRegister(SMLoc DirectiveLoc) {
4305 int64_t Register1 = 0, Register2 = 0;
4306 if (parseRegisterOrRegisterNumber(Register1, DirectiveLoc) || parseComma() ||
4307 parseRegisterOrRegisterNumber(Register2, DirectiveLoc) || parseEOL())
4308 return true;
4310 getStreamer().emitCFIRegister(Register1, Register2, DirectiveLoc);
4311 return false;
4314 /// parseDirectiveCFIWindowSave
4315 /// ::= .cfi_window_save
4316 bool AsmParser::parseDirectiveCFIWindowSave(SMLoc DirectiveLoc) {
4317 if (parseEOL())
4318 return true;
4319 getStreamer().emitCFIWindowSave(DirectiveLoc);
4320 return false;
4323 /// parseDirectiveCFIAdjustCfaOffset
4324 /// ::= .cfi_adjust_cfa_offset adjustment
4325 bool AsmParser::parseDirectiveCFIAdjustCfaOffset(SMLoc DirectiveLoc) {
4326 int64_t Adjustment = 0;
4327 if (parseAbsoluteExpression(Adjustment) || parseEOL())
4328 return true;
4330 getStreamer().emitCFIAdjustCfaOffset(Adjustment, DirectiveLoc);
4331 return false;
4334 /// parseDirectiveCFIDefCfaRegister
4335 /// ::= .cfi_def_cfa_register register
4336 bool AsmParser::parseDirectiveCFIDefCfaRegister(SMLoc DirectiveLoc) {
4337 int64_t Register = 0;
4338 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc) || parseEOL())
4339 return true;
4341 getStreamer().emitCFIDefCfaRegister(Register, DirectiveLoc);
4342 return false;
4345 /// parseDirectiveCFILLVMDefAspaceCfa
4346 /// ::= .cfi_llvm_def_aspace_cfa register, offset, address_space
4347 bool AsmParser::parseDirectiveCFILLVMDefAspaceCfa(SMLoc DirectiveLoc) {
4348 int64_t Register = 0, Offset = 0, AddressSpace = 0;
4349 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc) || parseComma() ||
4350 parseAbsoluteExpression(Offset) || parseComma() ||
4351 parseAbsoluteExpression(AddressSpace) || parseEOL())
4352 return true;
4354 getStreamer().emitCFILLVMDefAspaceCfa(Register, Offset, AddressSpace,
4355 DirectiveLoc);
4356 return false;
4359 /// parseDirectiveCFIOffset
4360 /// ::= .cfi_offset register, offset
4361 bool AsmParser::parseDirectiveCFIOffset(SMLoc DirectiveLoc) {
4362 int64_t Register = 0;
4363 int64_t Offset = 0;
4365 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc) || parseComma() ||
4366 parseAbsoluteExpression(Offset) || parseEOL())
4367 return true;
4369 getStreamer().emitCFIOffset(Register, Offset, DirectiveLoc);
4370 return false;
4373 /// parseDirectiveCFIRelOffset
4374 /// ::= .cfi_rel_offset register, offset
4375 bool AsmParser::parseDirectiveCFIRelOffset(SMLoc DirectiveLoc) {
4376 int64_t Register = 0, Offset = 0;
4378 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc) || parseComma() ||
4379 parseAbsoluteExpression(Offset) || parseEOL())
4380 return true;
4382 getStreamer().emitCFIRelOffset(Register, Offset, DirectiveLoc);
4383 return false;
4386 static bool isValidEncoding(int64_t Encoding) {
4387 if (Encoding & ~0xff)
4388 return false;
4390 if (Encoding == dwarf::DW_EH_PE_omit)
4391 return true;
4393 const unsigned Format = Encoding & 0xf;
4394 if (Format != dwarf::DW_EH_PE_absptr && Format != dwarf::DW_EH_PE_udata2 &&
4395 Format != dwarf::DW_EH_PE_udata4 && Format != dwarf::DW_EH_PE_udata8 &&
4396 Format != dwarf::DW_EH_PE_sdata2 && Format != dwarf::DW_EH_PE_sdata4 &&
4397 Format != dwarf::DW_EH_PE_sdata8 && Format != dwarf::DW_EH_PE_signed)
4398 return false;
4400 const unsigned Application = Encoding & 0x70;
4401 if (Application != dwarf::DW_EH_PE_absptr &&
4402 Application != dwarf::DW_EH_PE_pcrel)
4403 return false;
4405 return true;
4408 /// parseDirectiveCFIPersonalityOrLsda
4409 /// IsPersonality true for cfi_personality, false for cfi_lsda
4410 /// ::= .cfi_personality encoding, [symbol_name]
4411 /// ::= .cfi_lsda encoding, [symbol_name]
4412 bool AsmParser::parseDirectiveCFIPersonalityOrLsda(bool IsPersonality) {
4413 int64_t Encoding = 0;
4414 if (parseAbsoluteExpression(Encoding))
4415 return true;
4416 if (Encoding == dwarf::DW_EH_PE_omit)
4417 return false;
4419 StringRef Name;
4420 if (check(!isValidEncoding(Encoding), "unsupported encoding.") ||
4421 parseComma() ||
4422 check(parseIdentifier(Name), "expected identifier in directive") ||
4423 parseEOL())
4424 return true;
4426 MCSymbol *Sym = getContext().getOrCreateSymbol(Name);
4428 if (IsPersonality)
4429 getStreamer().emitCFIPersonality(Sym, Encoding);
4430 else
4431 getStreamer().emitCFILsda(Sym, Encoding);
4432 return false;
4435 /// parseDirectiveCFIRememberState
4436 /// ::= .cfi_remember_state
4437 bool AsmParser::parseDirectiveCFIRememberState(SMLoc DirectiveLoc) {
4438 if (parseEOL())
4439 return true;
4440 getStreamer().emitCFIRememberState(DirectiveLoc);
4441 return false;
4444 /// parseDirectiveCFIRestoreState
4445 /// ::= .cfi_remember_state
4446 bool AsmParser::parseDirectiveCFIRestoreState(SMLoc DirectiveLoc) {
4447 if (parseEOL())
4448 return true;
4449 getStreamer().emitCFIRestoreState(DirectiveLoc);
4450 return false;
4453 /// parseDirectiveCFISameValue
4454 /// ::= .cfi_same_value register
4455 bool AsmParser::parseDirectiveCFISameValue(SMLoc DirectiveLoc) {
4456 int64_t Register = 0;
4458 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc) || parseEOL())
4459 return true;
4461 getStreamer().emitCFISameValue(Register, DirectiveLoc);
4462 return false;
4465 /// parseDirectiveCFIRestore
4466 /// ::= .cfi_restore register
4467 bool AsmParser::parseDirectiveCFIRestore(SMLoc DirectiveLoc) {
4468 int64_t Register = 0;
4469 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc) || parseEOL())
4470 return true;
4472 getStreamer().emitCFIRestore(Register, DirectiveLoc);
4473 return false;
4476 /// parseDirectiveCFIEscape
4477 /// ::= .cfi_escape expression[,...]
4478 bool AsmParser::parseDirectiveCFIEscape(SMLoc DirectiveLoc) {
4479 std::string Values;
4480 int64_t CurrValue;
4481 if (parseAbsoluteExpression(CurrValue))
4482 return true;
4484 Values.push_back((uint8_t)CurrValue);
4486 while (getLexer().is(AsmToken::Comma)) {
4487 Lex();
4489 if (parseAbsoluteExpression(CurrValue))
4490 return true;
4492 Values.push_back((uint8_t)CurrValue);
4495 getStreamer().emitCFIEscape(Values, DirectiveLoc);
4496 return false;
4499 /// parseDirectiveCFIReturnColumn
4500 /// ::= .cfi_return_column register
4501 bool AsmParser::parseDirectiveCFIReturnColumn(SMLoc DirectiveLoc) {
4502 int64_t Register = 0;
4503 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc) || parseEOL())
4504 return true;
4505 getStreamer().emitCFIReturnColumn(Register);
4506 return false;
4509 /// parseDirectiveCFISignalFrame
4510 /// ::= .cfi_signal_frame
4511 bool AsmParser::parseDirectiveCFISignalFrame(SMLoc DirectiveLoc) {
4512 if (parseEOL())
4513 return true;
4515 getStreamer().emitCFISignalFrame();
4516 return false;
4519 /// parseDirectiveCFIUndefined
4520 /// ::= .cfi_undefined register
4521 bool AsmParser::parseDirectiveCFIUndefined(SMLoc DirectiveLoc) {
4522 int64_t Register = 0;
4524 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc) || parseEOL())
4525 return true;
4527 getStreamer().emitCFIUndefined(Register, DirectiveLoc);
4528 return false;
4531 /// parseDirectiveCFILabel
4532 /// ::= .cfi_label label
4533 bool AsmParser::parseDirectiveCFILabel(SMLoc Loc) {
4534 StringRef Name;
4535 Loc = Lexer.getLoc();
4536 if (parseIdentifier(Name))
4537 return TokError("expected identifier");
4538 if (parseEOL())
4539 return true;
4540 getStreamer().emitCFILabelDirective(Loc, Name);
4541 return false;
4544 /// parseDirectiveCFIValOffset
4545 /// ::= .cfi_val_offset register, offset
4546 bool AsmParser::parseDirectiveCFIValOffset(SMLoc DirectiveLoc) {
4547 int64_t Register = 0;
4548 int64_t Offset = 0;
4550 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc) || parseComma() ||
4551 parseAbsoluteExpression(Offset) || parseEOL())
4552 return true;
4554 getStreamer().emitCFIValOffset(Register, Offset, DirectiveLoc);
4555 return false;
4558 /// parseDirectiveAltmacro
4559 /// ::= .altmacro
4560 /// ::= .noaltmacro
4561 bool AsmParser::parseDirectiveAltmacro(StringRef Directive) {
4562 if (parseEOL())
4563 return true;
4564 AltMacroMode = (Directive == ".altmacro");
4565 return false;
4568 /// parseDirectiveMacrosOnOff
4569 /// ::= .macros_on
4570 /// ::= .macros_off
4571 bool AsmParser::parseDirectiveMacrosOnOff(StringRef Directive) {
4572 if (parseEOL())
4573 return true;
4574 setMacrosEnabled(Directive == ".macros_on");
4575 return false;
4578 /// parseDirectiveMacro
4579 /// ::= .macro name[,] [parameters]
4580 bool AsmParser::parseDirectiveMacro(SMLoc DirectiveLoc) {
4581 StringRef Name;
4582 if (parseIdentifier(Name))
4583 return TokError("expected identifier in '.macro' directive");
4585 if (getLexer().is(AsmToken::Comma))
4586 Lex();
4588 MCAsmMacroParameters Parameters;
4589 while (getLexer().isNot(AsmToken::EndOfStatement)) {
4591 if (!Parameters.empty() && Parameters.back().Vararg)
4592 return Error(Lexer.getLoc(), "vararg parameter '" +
4593 Parameters.back().Name +
4594 "' should be the last parameter");
4596 MCAsmMacroParameter Parameter;
4597 if (parseIdentifier(Parameter.Name))
4598 return TokError("expected identifier in '.macro' directive");
4600 // Emit an error if two (or more) named parameters share the same name
4601 for (const MCAsmMacroParameter& CurrParam : Parameters)
4602 if (CurrParam.Name == Parameter.Name)
4603 return TokError("macro '" + Name + "' has multiple parameters"
4604 " named '" + Parameter.Name + "'");
4606 if (Lexer.is(AsmToken::Colon)) {
4607 Lex(); // consume ':'
4609 SMLoc QualLoc;
4610 StringRef Qualifier;
4612 QualLoc = Lexer.getLoc();
4613 if (parseIdentifier(Qualifier))
4614 return Error(QualLoc, "missing parameter qualifier for "
4615 "'" + Parameter.Name + "' in macro '" + Name + "'");
4617 if (Qualifier == "req")
4618 Parameter.Required = true;
4619 else if (Qualifier == "vararg")
4620 Parameter.Vararg = true;
4621 else
4622 return Error(QualLoc, Qualifier + " is not a valid parameter qualifier "
4623 "for '" + Parameter.Name + "' in macro '" + Name + "'");
4626 if (getLexer().is(AsmToken::Equal)) {
4627 Lex();
4629 SMLoc ParamLoc;
4631 ParamLoc = Lexer.getLoc();
4632 if (parseMacroArgument(Parameter.Value, /*Vararg=*/false ))
4633 return true;
4635 if (Parameter.Required)
4636 Warning(ParamLoc, "pointless default value for required parameter "
4637 "'" + Parameter.Name + "' in macro '" + Name + "'");
4640 Parameters.push_back(std::move(Parameter));
4642 if (getLexer().is(AsmToken::Comma))
4643 Lex();
4646 // Eat just the end of statement.
4647 Lexer.Lex();
4649 // Consuming deferred text, so use Lexer.Lex to ignore Lexing Errors
4650 AsmToken EndToken, StartToken = getTok();
4651 unsigned MacroDepth = 0;
4652 // Lex the macro definition.
4653 while (true) {
4654 // Ignore Lexing errors in macros.
4655 while (Lexer.is(AsmToken::Error)) {
4656 Lexer.Lex();
4659 // Check whether we have reached the end of the file.
4660 if (getLexer().is(AsmToken::Eof))
4661 return Error(DirectiveLoc, "no matching '.endmacro' in definition");
4663 // Otherwise, check whether we have reach the .endmacro or the start of a
4664 // preprocessor line marker.
4665 if (getLexer().is(AsmToken::Identifier)) {
4666 if (getTok().getIdentifier() == ".endm" ||
4667 getTok().getIdentifier() == ".endmacro") {
4668 if (MacroDepth == 0) { // Outermost macro.
4669 EndToken = getTok();
4670 Lexer.Lex();
4671 if (getLexer().isNot(AsmToken::EndOfStatement))
4672 return TokError("unexpected token in '" + EndToken.getIdentifier() +
4673 "' directive");
4674 break;
4675 } else {
4676 // Otherwise we just found the end of an inner macro.
4677 --MacroDepth;
4679 } else if (getTok().getIdentifier() == ".macro") {
4680 // We allow nested macros. Those aren't instantiated until the outermost
4681 // macro is expanded so just ignore them for now.
4682 ++MacroDepth;
4684 } else if (Lexer.is(AsmToken::HashDirective)) {
4685 (void)parseCppHashLineFilenameComment(getLexer().getLoc());
4688 // Otherwise, scan til the end of the statement.
4689 eatToEndOfStatement();
4692 if (getContext().lookupMacro(Name)) {
4693 return Error(DirectiveLoc, "macro '" + Name + "' is already defined");
4696 const char *BodyStart = StartToken.getLoc().getPointer();
4697 const char *BodyEnd = EndToken.getLoc().getPointer();
4698 StringRef Body = StringRef(BodyStart, BodyEnd - BodyStart);
4699 checkForBadMacro(DirectiveLoc, Name, Body, Parameters);
4700 MCAsmMacro Macro(Name, Body, std::move(Parameters));
4701 DEBUG_WITH_TYPE("asm-macros", dbgs() << "Defining new macro:\n";
4702 Macro.dump());
4703 getContext().defineMacro(Name, std::move(Macro));
4704 return false;
4707 /// checkForBadMacro
4709 /// With the support added for named parameters there may be code out there that
4710 /// is transitioning from positional parameters. In versions of gas that did
4711 /// not support named parameters they would be ignored on the macro definition.
4712 /// But to support both styles of parameters this is not possible so if a macro
4713 /// definition has named parameters but does not use them and has what appears
4714 /// to be positional parameters, strings like $1, $2, ... and $n, then issue a
4715 /// warning that the positional parameter found in body which have no effect.
4716 /// Hoping the developer will either remove the named parameters from the macro
4717 /// definition so the positional parameters get used if that was what was
4718 /// intended or change the macro to use the named parameters. It is possible
4719 /// this warning will trigger when the none of the named parameters are used
4720 /// and the strings like $1 are infact to simply to be passed trough unchanged.
4721 void AsmParser::checkForBadMacro(SMLoc DirectiveLoc, StringRef Name,
4722 StringRef Body,
4723 ArrayRef<MCAsmMacroParameter> Parameters) {
4724 // If this macro is not defined with named parameters the warning we are
4725 // checking for here doesn't apply.
4726 unsigned NParameters = Parameters.size();
4727 if (NParameters == 0)
4728 return;
4730 bool NamedParametersFound = false;
4731 bool PositionalParametersFound = false;
4733 // Look at the body of the macro for use of both the named parameters and what
4734 // are likely to be positional parameters. This is what expandMacro() is
4735 // doing when it finds the parameters in the body.
4736 while (!Body.empty()) {
4737 // Scan for the next possible parameter.
4738 std::size_t End = Body.size(), Pos = 0;
4739 for (; Pos != End; ++Pos) {
4740 // Check for a substitution or escape.
4741 // This macro is defined with parameters, look for \foo, \bar, etc.
4742 if (Body[Pos] == '\\' && Pos + 1 != End)
4743 break;
4745 // This macro should have parameters, but look for $0, $1, ..., $n too.
4746 if (Body[Pos] != '$' || Pos + 1 == End)
4747 continue;
4748 char Next = Body[Pos + 1];
4749 if (Next == '$' || Next == 'n' ||
4750 isdigit(static_cast<unsigned char>(Next)))
4751 break;
4754 // Check if we reached the end.
4755 if (Pos == End)
4756 break;
4758 if (Body[Pos] == '$') {
4759 switch (Body[Pos + 1]) {
4760 // $$ => $
4761 case '$':
4762 break;
4764 // $n => number of arguments
4765 case 'n':
4766 PositionalParametersFound = true;
4767 break;
4769 // $[0-9] => argument
4770 default: {
4771 PositionalParametersFound = true;
4772 break;
4775 Pos += 2;
4776 } else {
4777 unsigned I = Pos + 1;
4778 while (isIdentifierChar(Body[I]) && I + 1 != End)
4779 ++I;
4781 const char *Begin = Body.data() + Pos + 1;
4782 StringRef Argument(Begin, I - (Pos + 1));
4783 unsigned Index = 0;
4784 for (; Index < NParameters; ++Index)
4785 if (Parameters[Index].Name == Argument)
4786 break;
4788 if (Index == NParameters) {
4789 if (Body[Pos + 1] == '(' && Body[Pos + 2] == ')')
4790 Pos += 3;
4791 else {
4792 Pos = I;
4794 } else {
4795 NamedParametersFound = true;
4796 Pos += 1 + Argument.size();
4799 // Update the scan point.
4800 Body = Body.substr(Pos);
4803 if (!NamedParametersFound && PositionalParametersFound)
4804 Warning(DirectiveLoc, "macro defined with named parameters which are not "
4805 "used in macro body, possible positional parameter "
4806 "found in body which will have no effect");
4809 /// parseDirectiveExitMacro
4810 /// ::= .exitm
4811 bool AsmParser::parseDirectiveExitMacro(StringRef Directive) {
4812 if (parseEOL())
4813 return true;
4815 if (!isInsideMacroInstantiation())
4816 return TokError("unexpected '" + Directive + "' in file, "
4817 "no current macro definition");
4819 // Exit all conditionals that are active in the current macro.
4820 while (TheCondStack.size() != ActiveMacros.back()->CondStackDepth) {
4821 TheCondState = TheCondStack.back();
4822 TheCondStack.pop_back();
4825 handleMacroExit();
4826 return false;
4829 /// parseDirectiveEndMacro
4830 /// ::= .endm
4831 /// ::= .endmacro
4832 bool AsmParser::parseDirectiveEndMacro(StringRef Directive) {
4833 if (getLexer().isNot(AsmToken::EndOfStatement))
4834 return TokError("unexpected token in '" + Directive + "' directive");
4836 // If we are inside a macro instantiation, terminate the current
4837 // instantiation.
4838 if (isInsideMacroInstantiation()) {
4839 handleMacroExit();
4840 return false;
4843 // Otherwise, this .endmacro is a stray entry in the file; well formed
4844 // .endmacro directives are handled during the macro definition parsing.
4845 return TokError("unexpected '" + Directive + "' in file, "
4846 "no current macro definition");
4849 /// parseDirectivePurgeMacro
4850 /// ::= .purgem name
4851 bool AsmParser::parseDirectivePurgeMacro(SMLoc DirectiveLoc) {
4852 StringRef Name;
4853 SMLoc Loc;
4854 if (parseTokenLoc(Loc) ||
4855 check(parseIdentifier(Name), Loc,
4856 "expected identifier in '.purgem' directive") ||
4857 parseEOL())
4858 return true;
4860 if (!getContext().lookupMacro(Name))
4861 return Error(DirectiveLoc, "macro '" + Name + "' is not defined");
4863 getContext().undefineMacro(Name);
4864 DEBUG_WITH_TYPE("asm-macros", dbgs()
4865 << "Un-defining macro: " << Name << "\n");
4866 return false;
4869 /// parseDirectiveBundleAlignMode
4870 /// ::= {.bundle_align_mode} expression
4871 bool AsmParser::parseDirectiveBundleAlignMode() {
4872 // Expect a single argument: an expression that evaluates to a constant
4873 // in the inclusive range 0-30.
4874 SMLoc ExprLoc = getLexer().getLoc();
4875 int64_t AlignSizePow2;
4876 if (checkForValidSection() || parseAbsoluteExpression(AlignSizePow2) ||
4877 parseEOL() ||
4878 check(AlignSizePow2 < 0 || AlignSizePow2 > 30, ExprLoc,
4879 "invalid bundle alignment size (expected between 0 and 30)"))
4880 return true;
4882 getStreamer().emitBundleAlignMode(Align(1ULL << AlignSizePow2));
4883 return false;
4886 /// parseDirectiveBundleLock
4887 /// ::= {.bundle_lock} [align_to_end]
4888 bool AsmParser::parseDirectiveBundleLock() {
4889 if (checkForValidSection())
4890 return true;
4891 bool AlignToEnd = false;
4893 StringRef Option;
4894 SMLoc Loc = getTok().getLoc();
4895 const char *kInvalidOptionError =
4896 "invalid option for '.bundle_lock' directive";
4898 if (!parseOptionalToken(AsmToken::EndOfStatement)) {
4899 if (check(parseIdentifier(Option), Loc, kInvalidOptionError) ||
4900 check(Option != "align_to_end", Loc, kInvalidOptionError) || parseEOL())
4901 return true;
4902 AlignToEnd = true;
4905 getStreamer().emitBundleLock(AlignToEnd);
4906 return false;
4909 /// parseDirectiveBundleLock
4910 /// ::= {.bundle_lock}
4911 bool AsmParser::parseDirectiveBundleUnlock() {
4912 if (checkForValidSection() || parseEOL())
4913 return true;
4915 getStreamer().emitBundleUnlock();
4916 return false;
4919 /// parseDirectiveSpace
4920 /// ::= (.skip | .space) expression [ , expression ]
4921 bool AsmParser::parseDirectiveSpace(StringRef IDVal) {
4922 SMLoc NumBytesLoc = Lexer.getLoc();
4923 const MCExpr *NumBytes;
4924 if (checkForValidSection() || parseExpression(NumBytes))
4925 return true;
4927 int64_t FillExpr = 0;
4928 if (parseOptionalToken(AsmToken::Comma))
4929 if (parseAbsoluteExpression(FillExpr))
4930 return true;
4931 if (parseEOL())
4932 return true;
4934 // FIXME: Sometimes the fill expr is 'nop' if it isn't supplied, instead of 0.
4935 getStreamer().emitFill(*NumBytes, FillExpr, NumBytesLoc);
4937 return false;
4940 /// parseDirectiveDCB
4941 /// ::= .dcb.{b, l, w} expression, expression
4942 bool AsmParser::parseDirectiveDCB(StringRef IDVal, unsigned Size) {
4943 SMLoc NumValuesLoc = Lexer.getLoc();
4944 int64_t NumValues;
4945 if (checkForValidSection() || parseAbsoluteExpression(NumValues))
4946 return true;
4948 if (NumValues < 0) {
4949 Warning(NumValuesLoc, "'" + Twine(IDVal) + "' directive with negative repeat count has no effect");
4950 return false;
4953 if (parseComma())
4954 return true;
4956 const MCExpr *Value;
4957 SMLoc ExprLoc = getLexer().getLoc();
4958 if (parseExpression(Value))
4959 return true;
4961 // Special case constant expressions to match code generator.
4962 if (const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value)) {
4963 assert(Size <= 8 && "Invalid size");
4964 uint64_t IntValue = MCE->getValue();
4965 if (!isUIntN(8 * Size, IntValue) && !isIntN(8 * Size, IntValue))
4966 return Error(ExprLoc, "literal value out of range for directive");
4967 for (uint64_t i = 0, e = NumValues; i != e; ++i)
4968 getStreamer().emitIntValue(IntValue, Size);
4969 } else {
4970 for (uint64_t i = 0, e = NumValues; i != e; ++i)
4971 getStreamer().emitValue(Value, Size, ExprLoc);
4974 return parseEOL();
4977 /// parseDirectiveRealDCB
4978 /// ::= .dcb.{d, s} expression, expression
4979 bool AsmParser::parseDirectiveRealDCB(StringRef IDVal, const fltSemantics &Semantics) {
4980 SMLoc NumValuesLoc = Lexer.getLoc();
4981 int64_t NumValues;
4982 if (checkForValidSection() || parseAbsoluteExpression(NumValues))
4983 return true;
4985 if (NumValues < 0) {
4986 Warning(NumValuesLoc, "'" + Twine(IDVal) + "' directive with negative repeat count has no effect");
4987 return false;
4990 if (parseComma())
4991 return true;
4993 APInt AsInt;
4994 if (parseRealValue(Semantics, AsInt) || parseEOL())
4995 return true;
4997 for (uint64_t i = 0, e = NumValues; i != e; ++i)
4998 getStreamer().emitIntValue(AsInt.getLimitedValue(),
4999 AsInt.getBitWidth() / 8);
5001 return false;
5004 /// parseDirectiveDS
5005 /// ::= .ds.{b, d, l, p, s, w, x} expression
5006 bool AsmParser::parseDirectiveDS(StringRef IDVal, unsigned Size) {
5007 SMLoc NumValuesLoc = Lexer.getLoc();
5008 int64_t NumValues;
5009 if (checkForValidSection() || parseAbsoluteExpression(NumValues) ||
5010 parseEOL())
5011 return true;
5013 if (NumValues < 0) {
5014 Warning(NumValuesLoc, "'" + Twine(IDVal) + "' directive with negative repeat count has no effect");
5015 return false;
5018 for (uint64_t i = 0, e = NumValues; i != e; ++i)
5019 getStreamer().emitFill(Size, 0);
5021 return false;
5024 /// parseDirectiveLEB128
5025 /// ::= (.sleb128 | .uleb128) [ expression (, expression)* ]
5026 bool AsmParser::parseDirectiveLEB128(bool Signed) {
5027 if (checkForValidSection())
5028 return true;
5030 auto parseOp = [&]() -> bool {
5031 const MCExpr *Value;
5032 if (parseExpression(Value))
5033 return true;
5034 if (Signed)
5035 getStreamer().emitSLEB128Value(Value);
5036 else
5037 getStreamer().emitULEB128Value(Value);
5038 return false;
5041 return parseMany(parseOp);
5044 /// parseDirectiveSymbolAttribute
5045 /// ::= { ".globl", ".weak", ... } [ identifier ( , identifier )* ]
5046 bool AsmParser::parseDirectiveSymbolAttribute(MCSymbolAttr Attr) {
5047 auto parseOp = [&]() -> bool {
5048 StringRef Name;
5049 SMLoc Loc = getTok().getLoc();
5050 if (parseIdentifier(Name))
5051 return Error(Loc, "expected identifier");
5053 if (discardLTOSymbol(Name))
5054 return false;
5056 MCSymbol *Sym = getContext().getOrCreateSymbol(Name);
5058 // Assembler local symbols don't make any sense here, except for directives
5059 // that the symbol should be tagged.
5060 if (Sym->isTemporary() && Attr != MCSA_Memtag)
5061 return Error(Loc, "non-local symbol required");
5063 if (!getStreamer().emitSymbolAttribute(Sym, Attr))
5064 return Error(Loc, "unable to emit symbol attribute");
5065 return false;
5068 return parseMany(parseOp);
5071 /// parseDirectiveComm
5072 /// ::= ( .comm | .lcomm ) identifier , size_expression [ , align_expression ]
5073 bool AsmParser::parseDirectiveComm(bool IsLocal) {
5074 if (checkForValidSection())
5075 return true;
5077 SMLoc IDLoc = getLexer().getLoc();
5078 StringRef Name;
5079 if (parseIdentifier(Name))
5080 return TokError("expected identifier in directive");
5082 // Handle the identifier as the key symbol.
5083 MCSymbol *Sym = getContext().getOrCreateSymbol(Name);
5085 if (parseComma())
5086 return true;
5088 int64_t Size;
5089 SMLoc SizeLoc = getLexer().getLoc();
5090 if (parseAbsoluteExpression(Size))
5091 return true;
5093 int64_t Pow2Alignment = 0;
5094 SMLoc Pow2AlignmentLoc;
5095 if (getLexer().is(AsmToken::Comma)) {
5096 Lex();
5097 Pow2AlignmentLoc = getLexer().getLoc();
5098 if (parseAbsoluteExpression(Pow2Alignment))
5099 return true;
5101 LCOMM::LCOMMType LCOMM = Lexer.getMAI().getLCOMMDirectiveAlignmentType();
5102 if (IsLocal && LCOMM == LCOMM::NoAlignment)
5103 return Error(Pow2AlignmentLoc, "alignment not supported on this target");
5105 // If this target takes alignments in bytes (not log) validate and convert.
5106 if ((!IsLocal && Lexer.getMAI().getCOMMDirectiveAlignmentIsInBytes()) ||
5107 (IsLocal && LCOMM == LCOMM::ByteAlignment)) {
5108 if (!isPowerOf2_64(Pow2Alignment))
5109 return Error(Pow2AlignmentLoc, "alignment must be a power of 2");
5110 Pow2Alignment = Log2_64(Pow2Alignment);
5114 if (parseEOL())
5115 return true;
5117 // NOTE: a size of zero for a .comm should create a undefined symbol
5118 // but a size of .lcomm creates a bss symbol of size zero.
5119 if (Size < 0)
5120 return Error(SizeLoc, "size must be non-negative");
5122 Sym->redefineIfPossible();
5123 if (!Sym->isUndefined())
5124 return Error(IDLoc, "invalid symbol redefinition");
5126 // Create the Symbol as a common or local common with Size and Pow2Alignment
5127 if (IsLocal) {
5128 getStreamer().emitLocalCommonSymbol(Sym, Size,
5129 Align(1ULL << Pow2Alignment));
5130 return false;
5133 getStreamer().emitCommonSymbol(Sym, Size, Align(1ULL << Pow2Alignment));
5134 return false;
5137 /// parseDirectiveAbort
5138 /// ::= .abort [... message ...]
5139 bool AsmParser::parseDirectiveAbort(SMLoc DirectiveLoc) {
5140 StringRef Str = parseStringToEndOfStatement();
5141 if (parseEOL())
5142 return true;
5144 if (Str.empty())
5145 return Error(DirectiveLoc, ".abort detected. Assembly stopping");
5147 // FIXME: Actually abort assembly here.
5148 return Error(DirectiveLoc,
5149 ".abort '" + Str + "' detected. Assembly stopping");
5152 /// parseDirectiveInclude
5153 /// ::= .include "filename"
5154 bool AsmParser::parseDirectiveInclude() {
5155 // Allow the strings to have escaped octal character sequence.
5156 std::string Filename;
5157 SMLoc IncludeLoc = getTok().getLoc();
5159 if (check(getTok().isNot(AsmToken::String),
5160 "expected string in '.include' directive") ||
5161 parseEscapedString(Filename) ||
5162 check(getTok().isNot(AsmToken::EndOfStatement),
5163 "unexpected token in '.include' directive") ||
5164 // Attempt to switch the lexer to the included file before consuming the
5165 // end of statement to avoid losing it when we switch.
5166 check(enterIncludeFile(Filename), IncludeLoc,
5167 "Could not find include file '" + Filename + "'"))
5168 return true;
5170 return false;
5173 /// parseDirectiveIncbin
5174 /// ::= .incbin "filename" [ , skip [ , count ] ]
5175 bool AsmParser::parseDirectiveIncbin() {
5176 // Allow the strings to have escaped octal character sequence.
5177 std::string Filename;
5178 SMLoc IncbinLoc = getTok().getLoc();
5179 if (check(getTok().isNot(AsmToken::String),
5180 "expected string in '.incbin' directive") ||
5181 parseEscapedString(Filename))
5182 return true;
5184 int64_t Skip = 0;
5185 const MCExpr *Count = nullptr;
5186 SMLoc SkipLoc, CountLoc;
5187 if (parseOptionalToken(AsmToken::Comma)) {
5188 // The skip expression can be omitted while specifying the count, e.g:
5189 // .incbin "filename",,4
5190 if (getTok().isNot(AsmToken::Comma)) {
5191 if (parseTokenLoc(SkipLoc) || parseAbsoluteExpression(Skip))
5192 return true;
5194 if (parseOptionalToken(AsmToken::Comma)) {
5195 CountLoc = getTok().getLoc();
5196 if (parseExpression(Count))
5197 return true;
5201 if (parseEOL())
5202 return true;
5204 if (check(Skip < 0, SkipLoc, "skip is negative"))
5205 return true;
5207 // Attempt to process the included file.
5208 if (processIncbinFile(Filename, Skip, Count, CountLoc))
5209 return Error(IncbinLoc, "Could not find incbin file '" + Filename + "'");
5210 return false;
5213 /// parseDirectiveIf
5214 /// ::= .if{,eq,ge,gt,le,lt,ne} expression
5215 bool AsmParser::parseDirectiveIf(SMLoc DirectiveLoc, DirectiveKind DirKind) {
5216 TheCondStack.push_back(TheCondState);
5217 TheCondState.TheCond = AsmCond::IfCond;
5218 if (TheCondState.Ignore) {
5219 eatToEndOfStatement();
5220 } else {
5221 int64_t ExprValue;
5222 if (parseAbsoluteExpression(ExprValue) || parseEOL())
5223 return true;
5225 switch (DirKind) {
5226 default:
5227 llvm_unreachable("unsupported directive");
5228 case DK_IF:
5229 case DK_IFNE:
5230 break;
5231 case DK_IFEQ:
5232 ExprValue = ExprValue == 0;
5233 break;
5234 case DK_IFGE:
5235 ExprValue = ExprValue >= 0;
5236 break;
5237 case DK_IFGT:
5238 ExprValue = ExprValue > 0;
5239 break;
5240 case DK_IFLE:
5241 ExprValue = ExprValue <= 0;
5242 break;
5243 case DK_IFLT:
5244 ExprValue = ExprValue < 0;
5245 break;
5248 TheCondState.CondMet = ExprValue;
5249 TheCondState.Ignore = !TheCondState.CondMet;
5252 return false;
5255 /// parseDirectiveIfb
5256 /// ::= .ifb string
5257 bool AsmParser::parseDirectiveIfb(SMLoc DirectiveLoc, bool ExpectBlank) {
5258 TheCondStack.push_back(TheCondState);
5259 TheCondState.TheCond = AsmCond::IfCond;
5261 if (TheCondState.Ignore) {
5262 eatToEndOfStatement();
5263 } else {
5264 StringRef Str = parseStringToEndOfStatement();
5266 if (parseEOL())
5267 return true;
5269 TheCondState.CondMet = ExpectBlank == Str.empty();
5270 TheCondState.Ignore = !TheCondState.CondMet;
5273 return false;
5276 /// parseDirectiveIfc
5277 /// ::= .ifc string1, string2
5278 /// ::= .ifnc string1, string2
5279 bool AsmParser::parseDirectiveIfc(SMLoc DirectiveLoc, bool ExpectEqual) {
5280 TheCondStack.push_back(TheCondState);
5281 TheCondState.TheCond = AsmCond::IfCond;
5283 if (TheCondState.Ignore) {
5284 eatToEndOfStatement();
5285 } else {
5286 StringRef Str1 = parseStringToComma();
5288 if (parseComma())
5289 return true;
5291 StringRef Str2 = parseStringToEndOfStatement();
5293 if (parseEOL())
5294 return true;
5296 TheCondState.CondMet = ExpectEqual == (Str1.trim() == Str2.trim());
5297 TheCondState.Ignore = !TheCondState.CondMet;
5300 return false;
5303 /// parseDirectiveIfeqs
5304 /// ::= .ifeqs string1, string2
5305 bool AsmParser::parseDirectiveIfeqs(SMLoc DirectiveLoc, bool ExpectEqual) {
5306 if (Lexer.isNot(AsmToken::String)) {
5307 if (ExpectEqual)
5308 return TokError("expected string parameter for '.ifeqs' directive");
5309 return TokError("expected string parameter for '.ifnes' directive");
5312 StringRef String1 = getTok().getStringContents();
5313 Lex();
5315 if (Lexer.isNot(AsmToken::Comma)) {
5316 if (ExpectEqual)
5317 return TokError(
5318 "expected comma after first string for '.ifeqs' directive");
5319 return TokError("expected comma after first string for '.ifnes' directive");
5322 Lex();
5324 if (Lexer.isNot(AsmToken::String)) {
5325 if (ExpectEqual)
5326 return TokError("expected string parameter for '.ifeqs' directive");
5327 return TokError("expected string parameter for '.ifnes' directive");
5330 StringRef String2 = getTok().getStringContents();
5331 Lex();
5333 TheCondStack.push_back(TheCondState);
5334 TheCondState.TheCond = AsmCond::IfCond;
5335 TheCondState.CondMet = ExpectEqual == (String1 == String2);
5336 TheCondState.Ignore = !TheCondState.CondMet;
5338 return false;
5341 /// parseDirectiveIfdef
5342 /// ::= .ifdef symbol
5343 bool AsmParser::parseDirectiveIfdef(SMLoc DirectiveLoc, bool expect_defined) {
5344 StringRef Name;
5345 TheCondStack.push_back(TheCondState);
5346 TheCondState.TheCond = AsmCond::IfCond;
5348 if (TheCondState.Ignore) {
5349 eatToEndOfStatement();
5350 } else {
5351 if (check(parseIdentifier(Name), "expected identifier after '.ifdef'") ||
5352 parseEOL())
5353 return true;
5355 MCSymbol *Sym = getContext().lookupSymbol(Name);
5357 if (expect_defined)
5358 TheCondState.CondMet = (Sym && !Sym->isUndefined(false));
5359 else
5360 TheCondState.CondMet = (!Sym || Sym->isUndefined(false));
5361 TheCondState.Ignore = !TheCondState.CondMet;
5364 return false;
5367 /// parseDirectiveElseIf
5368 /// ::= .elseif expression
5369 bool AsmParser::parseDirectiveElseIf(SMLoc DirectiveLoc) {
5370 if (TheCondState.TheCond != AsmCond::IfCond &&
5371 TheCondState.TheCond != AsmCond::ElseIfCond)
5372 return Error(DirectiveLoc, "Encountered a .elseif that doesn't follow an"
5373 " .if or an .elseif");
5374 TheCondState.TheCond = AsmCond::ElseIfCond;
5376 bool LastIgnoreState = false;
5377 if (!TheCondStack.empty())
5378 LastIgnoreState = TheCondStack.back().Ignore;
5379 if (LastIgnoreState || TheCondState.CondMet) {
5380 TheCondState.Ignore = true;
5381 eatToEndOfStatement();
5382 } else {
5383 int64_t ExprValue;
5384 if (parseAbsoluteExpression(ExprValue))
5385 return true;
5387 if (parseEOL())
5388 return true;
5390 TheCondState.CondMet = ExprValue;
5391 TheCondState.Ignore = !TheCondState.CondMet;
5394 return false;
5397 /// parseDirectiveElse
5398 /// ::= .else
5399 bool AsmParser::parseDirectiveElse(SMLoc DirectiveLoc) {
5400 if (parseEOL())
5401 return true;
5403 if (TheCondState.TheCond != AsmCond::IfCond &&
5404 TheCondState.TheCond != AsmCond::ElseIfCond)
5405 return Error(DirectiveLoc, "Encountered a .else that doesn't follow "
5406 " an .if or an .elseif");
5407 TheCondState.TheCond = AsmCond::ElseCond;
5408 bool LastIgnoreState = false;
5409 if (!TheCondStack.empty())
5410 LastIgnoreState = TheCondStack.back().Ignore;
5411 if (LastIgnoreState || TheCondState.CondMet)
5412 TheCondState.Ignore = true;
5413 else
5414 TheCondState.Ignore = false;
5416 return false;
5419 /// parseDirectiveEnd
5420 /// ::= .end
5421 bool AsmParser::parseDirectiveEnd(SMLoc DirectiveLoc) {
5422 if (parseEOL())
5423 return true;
5425 while (Lexer.isNot(AsmToken::Eof))
5426 Lexer.Lex();
5428 return false;
5431 /// parseDirectiveError
5432 /// ::= .err
5433 /// ::= .error [string]
5434 bool AsmParser::parseDirectiveError(SMLoc L, bool WithMessage) {
5435 if (!TheCondStack.empty()) {
5436 if (TheCondStack.back().Ignore) {
5437 eatToEndOfStatement();
5438 return false;
5442 if (!WithMessage)
5443 return Error(L, ".err encountered");
5445 StringRef Message = ".error directive invoked in source file";
5446 if (Lexer.isNot(AsmToken::EndOfStatement)) {
5447 if (Lexer.isNot(AsmToken::String))
5448 return TokError(".error argument must be a string");
5450 Message = getTok().getStringContents();
5451 Lex();
5454 return Error(L, Message);
5457 /// parseDirectiveWarning
5458 /// ::= .warning [string]
5459 bool AsmParser::parseDirectiveWarning(SMLoc L) {
5460 if (!TheCondStack.empty()) {
5461 if (TheCondStack.back().Ignore) {
5462 eatToEndOfStatement();
5463 return false;
5467 StringRef Message = ".warning directive invoked in source file";
5469 if (!parseOptionalToken(AsmToken::EndOfStatement)) {
5470 if (Lexer.isNot(AsmToken::String))
5471 return TokError(".warning argument must be a string");
5473 Message = getTok().getStringContents();
5474 Lex();
5475 if (parseEOL())
5476 return true;
5479 return Warning(L, Message);
5482 /// parseDirectiveEndIf
5483 /// ::= .endif
5484 bool AsmParser::parseDirectiveEndIf(SMLoc DirectiveLoc) {
5485 if (parseEOL())
5486 return true;
5488 if ((TheCondState.TheCond == AsmCond::NoCond) || TheCondStack.empty())
5489 return Error(DirectiveLoc, "Encountered a .endif that doesn't follow "
5490 "an .if or .else");
5491 if (!TheCondStack.empty()) {
5492 TheCondState = TheCondStack.back();
5493 TheCondStack.pop_back();
5496 return false;
5499 void AsmParser::initializeDirectiveKindMap() {
5500 /* Lookup will be done with the directive
5501 * converted to lower case, so all these
5502 * keys should be lower case.
5503 * (target specific directives are handled
5504 * elsewhere)
5506 DirectiveKindMap[".set"] = DK_SET;
5507 DirectiveKindMap[".equ"] = DK_EQU;
5508 DirectiveKindMap[".equiv"] = DK_EQUIV;
5509 DirectiveKindMap[".ascii"] = DK_ASCII;
5510 DirectiveKindMap[".asciz"] = DK_ASCIZ;
5511 DirectiveKindMap[".string"] = DK_STRING;
5512 DirectiveKindMap[".byte"] = DK_BYTE;
5513 DirectiveKindMap[".short"] = DK_SHORT;
5514 DirectiveKindMap[".value"] = DK_VALUE;
5515 DirectiveKindMap[".2byte"] = DK_2BYTE;
5516 DirectiveKindMap[".long"] = DK_LONG;
5517 DirectiveKindMap[".int"] = DK_INT;
5518 DirectiveKindMap[".4byte"] = DK_4BYTE;
5519 DirectiveKindMap[".quad"] = DK_QUAD;
5520 DirectiveKindMap[".8byte"] = DK_8BYTE;
5521 DirectiveKindMap[".octa"] = DK_OCTA;
5522 DirectiveKindMap[".single"] = DK_SINGLE;
5523 DirectiveKindMap[".float"] = DK_FLOAT;
5524 DirectiveKindMap[".double"] = DK_DOUBLE;
5525 DirectiveKindMap[".align"] = DK_ALIGN;
5526 DirectiveKindMap[".align32"] = DK_ALIGN32;
5527 DirectiveKindMap[".balign"] = DK_BALIGN;
5528 DirectiveKindMap[".balignw"] = DK_BALIGNW;
5529 DirectiveKindMap[".balignl"] = DK_BALIGNL;
5530 DirectiveKindMap[".p2align"] = DK_P2ALIGN;
5531 DirectiveKindMap[".p2alignw"] = DK_P2ALIGNW;
5532 DirectiveKindMap[".p2alignl"] = DK_P2ALIGNL;
5533 DirectiveKindMap[".org"] = DK_ORG;
5534 DirectiveKindMap[".fill"] = DK_FILL;
5535 DirectiveKindMap[".zero"] = DK_ZERO;
5536 DirectiveKindMap[".extern"] = DK_EXTERN;
5537 DirectiveKindMap[".globl"] = DK_GLOBL;
5538 DirectiveKindMap[".global"] = DK_GLOBAL;
5539 DirectiveKindMap[".lazy_reference"] = DK_LAZY_REFERENCE;
5540 DirectiveKindMap[".no_dead_strip"] = DK_NO_DEAD_STRIP;
5541 DirectiveKindMap[".symbol_resolver"] = DK_SYMBOL_RESOLVER;
5542 DirectiveKindMap[".private_extern"] = DK_PRIVATE_EXTERN;
5543 DirectiveKindMap[".reference"] = DK_REFERENCE;
5544 DirectiveKindMap[".weak_definition"] = DK_WEAK_DEFINITION;
5545 DirectiveKindMap[".weak_reference"] = DK_WEAK_REFERENCE;
5546 DirectiveKindMap[".weak_def_can_be_hidden"] = DK_WEAK_DEF_CAN_BE_HIDDEN;
5547 DirectiveKindMap[".cold"] = DK_COLD;
5548 DirectiveKindMap[".comm"] = DK_COMM;
5549 DirectiveKindMap[".common"] = DK_COMMON;
5550 DirectiveKindMap[".lcomm"] = DK_LCOMM;
5551 DirectiveKindMap[".abort"] = DK_ABORT;
5552 DirectiveKindMap[".include"] = DK_INCLUDE;
5553 DirectiveKindMap[".incbin"] = DK_INCBIN;
5554 DirectiveKindMap[".code16"] = DK_CODE16;
5555 DirectiveKindMap[".code16gcc"] = DK_CODE16GCC;
5556 DirectiveKindMap[".rept"] = DK_REPT;
5557 DirectiveKindMap[".rep"] = DK_REPT;
5558 DirectiveKindMap[".irp"] = DK_IRP;
5559 DirectiveKindMap[".irpc"] = DK_IRPC;
5560 DirectiveKindMap[".endr"] = DK_ENDR;
5561 DirectiveKindMap[".bundle_align_mode"] = DK_BUNDLE_ALIGN_MODE;
5562 DirectiveKindMap[".bundle_lock"] = DK_BUNDLE_LOCK;
5563 DirectiveKindMap[".bundle_unlock"] = DK_BUNDLE_UNLOCK;
5564 DirectiveKindMap[".if"] = DK_IF;
5565 DirectiveKindMap[".ifeq"] = DK_IFEQ;
5566 DirectiveKindMap[".ifge"] = DK_IFGE;
5567 DirectiveKindMap[".ifgt"] = DK_IFGT;
5568 DirectiveKindMap[".ifle"] = DK_IFLE;
5569 DirectiveKindMap[".iflt"] = DK_IFLT;
5570 DirectiveKindMap[".ifne"] = DK_IFNE;
5571 DirectiveKindMap[".ifb"] = DK_IFB;
5572 DirectiveKindMap[".ifnb"] = DK_IFNB;
5573 DirectiveKindMap[".ifc"] = DK_IFC;
5574 DirectiveKindMap[".ifeqs"] = DK_IFEQS;
5575 DirectiveKindMap[".ifnc"] = DK_IFNC;
5576 DirectiveKindMap[".ifnes"] = DK_IFNES;
5577 DirectiveKindMap[".ifdef"] = DK_IFDEF;
5578 DirectiveKindMap[".ifndef"] = DK_IFNDEF;
5579 DirectiveKindMap[".ifnotdef"] = DK_IFNOTDEF;
5580 DirectiveKindMap[".elseif"] = DK_ELSEIF;
5581 DirectiveKindMap[".else"] = DK_ELSE;
5582 DirectiveKindMap[".end"] = DK_END;
5583 DirectiveKindMap[".endif"] = DK_ENDIF;
5584 DirectiveKindMap[".skip"] = DK_SKIP;
5585 DirectiveKindMap[".space"] = DK_SPACE;
5586 DirectiveKindMap[".file"] = DK_FILE;
5587 DirectiveKindMap[".line"] = DK_LINE;
5588 DirectiveKindMap[".loc"] = DK_LOC;
5589 DirectiveKindMap[".loc_label"] = DK_LOC_LABEL;
5590 DirectiveKindMap[".stabs"] = DK_STABS;
5591 DirectiveKindMap[".cv_file"] = DK_CV_FILE;
5592 DirectiveKindMap[".cv_func_id"] = DK_CV_FUNC_ID;
5593 DirectiveKindMap[".cv_loc"] = DK_CV_LOC;
5594 DirectiveKindMap[".cv_linetable"] = DK_CV_LINETABLE;
5595 DirectiveKindMap[".cv_inline_linetable"] = DK_CV_INLINE_LINETABLE;
5596 DirectiveKindMap[".cv_inline_site_id"] = DK_CV_INLINE_SITE_ID;
5597 DirectiveKindMap[".cv_def_range"] = DK_CV_DEF_RANGE;
5598 DirectiveKindMap[".cv_string"] = DK_CV_STRING;
5599 DirectiveKindMap[".cv_stringtable"] = DK_CV_STRINGTABLE;
5600 DirectiveKindMap[".cv_filechecksums"] = DK_CV_FILECHECKSUMS;
5601 DirectiveKindMap[".cv_filechecksumoffset"] = DK_CV_FILECHECKSUM_OFFSET;
5602 DirectiveKindMap[".cv_fpo_data"] = DK_CV_FPO_DATA;
5603 DirectiveKindMap[".sleb128"] = DK_SLEB128;
5604 DirectiveKindMap[".uleb128"] = DK_ULEB128;
5605 DirectiveKindMap[".cfi_sections"] = DK_CFI_SECTIONS;
5606 DirectiveKindMap[".cfi_startproc"] = DK_CFI_STARTPROC;
5607 DirectiveKindMap[".cfi_endproc"] = DK_CFI_ENDPROC;
5608 DirectiveKindMap[".cfi_def_cfa"] = DK_CFI_DEF_CFA;
5609 DirectiveKindMap[".cfi_def_cfa_offset"] = DK_CFI_DEF_CFA_OFFSET;
5610 DirectiveKindMap[".cfi_adjust_cfa_offset"] = DK_CFI_ADJUST_CFA_OFFSET;
5611 DirectiveKindMap[".cfi_def_cfa_register"] = DK_CFI_DEF_CFA_REGISTER;
5612 DirectiveKindMap[".cfi_llvm_def_aspace_cfa"] = DK_CFI_LLVM_DEF_ASPACE_CFA;
5613 DirectiveKindMap[".cfi_offset"] = DK_CFI_OFFSET;
5614 DirectiveKindMap[".cfi_rel_offset"] = DK_CFI_REL_OFFSET;
5615 DirectiveKindMap[".cfi_personality"] = DK_CFI_PERSONALITY;
5616 DirectiveKindMap[".cfi_lsda"] = DK_CFI_LSDA;
5617 DirectiveKindMap[".cfi_remember_state"] = DK_CFI_REMEMBER_STATE;
5618 DirectiveKindMap[".cfi_restore_state"] = DK_CFI_RESTORE_STATE;
5619 DirectiveKindMap[".cfi_same_value"] = DK_CFI_SAME_VALUE;
5620 DirectiveKindMap[".cfi_restore"] = DK_CFI_RESTORE;
5621 DirectiveKindMap[".cfi_escape"] = DK_CFI_ESCAPE;
5622 DirectiveKindMap[".cfi_return_column"] = DK_CFI_RETURN_COLUMN;
5623 DirectiveKindMap[".cfi_signal_frame"] = DK_CFI_SIGNAL_FRAME;
5624 DirectiveKindMap[".cfi_undefined"] = DK_CFI_UNDEFINED;
5625 DirectiveKindMap[".cfi_register"] = DK_CFI_REGISTER;
5626 DirectiveKindMap[".cfi_window_save"] = DK_CFI_WINDOW_SAVE;
5627 DirectiveKindMap[".cfi_label"] = DK_CFI_LABEL;
5628 DirectiveKindMap[".cfi_b_key_frame"] = DK_CFI_B_KEY_FRAME;
5629 DirectiveKindMap[".cfi_mte_tagged_frame"] = DK_CFI_MTE_TAGGED_FRAME;
5630 DirectiveKindMap[".cfi_val_offset"] = DK_CFI_VAL_OFFSET;
5631 DirectiveKindMap[".macros_on"] = DK_MACROS_ON;
5632 DirectiveKindMap[".macros_off"] = DK_MACROS_OFF;
5633 DirectiveKindMap[".macro"] = DK_MACRO;
5634 DirectiveKindMap[".exitm"] = DK_EXITM;
5635 DirectiveKindMap[".endm"] = DK_ENDM;
5636 DirectiveKindMap[".endmacro"] = DK_ENDMACRO;
5637 DirectiveKindMap[".purgem"] = DK_PURGEM;
5638 DirectiveKindMap[".err"] = DK_ERR;
5639 DirectiveKindMap[".error"] = DK_ERROR;
5640 DirectiveKindMap[".warning"] = DK_WARNING;
5641 DirectiveKindMap[".altmacro"] = DK_ALTMACRO;
5642 DirectiveKindMap[".noaltmacro"] = DK_NOALTMACRO;
5643 DirectiveKindMap[".reloc"] = DK_RELOC;
5644 DirectiveKindMap[".dc"] = DK_DC;
5645 DirectiveKindMap[".dc.a"] = DK_DC_A;
5646 DirectiveKindMap[".dc.b"] = DK_DC_B;
5647 DirectiveKindMap[".dc.d"] = DK_DC_D;
5648 DirectiveKindMap[".dc.l"] = DK_DC_L;
5649 DirectiveKindMap[".dc.s"] = DK_DC_S;
5650 DirectiveKindMap[".dc.w"] = DK_DC_W;
5651 DirectiveKindMap[".dc.x"] = DK_DC_X;
5652 DirectiveKindMap[".dcb"] = DK_DCB;
5653 DirectiveKindMap[".dcb.b"] = DK_DCB_B;
5654 DirectiveKindMap[".dcb.d"] = DK_DCB_D;
5655 DirectiveKindMap[".dcb.l"] = DK_DCB_L;
5656 DirectiveKindMap[".dcb.s"] = DK_DCB_S;
5657 DirectiveKindMap[".dcb.w"] = DK_DCB_W;
5658 DirectiveKindMap[".dcb.x"] = DK_DCB_X;
5659 DirectiveKindMap[".ds"] = DK_DS;
5660 DirectiveKindMap[".ds.b"] = DK_DS_B;
5661 DirectiveKindMap[".ds.d"] = DK_DS_D;
5662 DirectiveKindMap[".ds.l"] = DK_DS_L;
5663 DirectiveKindMap[".ds.p"] = DK_DS_P;
5664 DirectiveKindMap[".ds.s"] = DK_DS_S;
5665 DirectiveKindMap[".ds.w"] = DK_DS_W;
5666 DirectiveKindMap[".ds.x"] = DK_DS_X;
5667 DirectiveKindMap[".print"] = DK_PRINT;
5668 DirectiveKindMap[".addrsig"] = DK_ADDRSIG;
5669 DirectiveKindMap[".addrsig_sym"] = DK_ADDRSIG_SYM;
5670 DirectiveKindMap[".pseudoprobe"] = DK_PSEUDO_PROBE;
5671 DirectiveKindMap[".lto_discard"] = DK_LTO_DISCARD;
5672 DirectiveKindMap[".lto_set_conditional"] = DK_LTO_SET_CONDITIONAL;
5673 DirectiveKindMap[".memtag"] = DK_MEMTAG;
5676 MCAsmMacro *AsmParser::parseMacroLikeBody(SMLoc DirectiveLoc) {
5677 AsmToken EndToken, StartToken = getTok();
5679 unsigned NestLevel = 0;
5680 while (true) {
5681 // Check whether we have reached the end of the file.
5682 if (getLexer().is(AsmToken::Eof)) {
5683 printError(DirectiveLoc, "no matching '.endr' in definition");
5684 return nullptr;
5687 if (Lexer.is(AsmToken::Identifier)) {
5688 StringRef Ident = getTok().getIdentifier();
5689 if (Ident == ".rep" || Ident == ".rept" || Ident == ".irp" ||
5690 Ident == ".irpc") {
5691 ++NestLevel;
5692 } else if (Ident == ".endr") {
5693 if (NestLevel == 0) {
5694 EndToken = getTok();
5695 Lex();
5696 if (Lexer.is(AsmToken::EndOfStatement))
5697 break;
5698 printError(getTok().getLoc(), "expected newline");
5699 return nullptr;
5701 --NestLevel;
5705 // Otherwise, scan till the end of the statement.
5706 eatToEndOfStatement();
5709 const char *BodyStart = StartToken.getLoc().getPointer();
5710 const char *BodyEnd = EndToken.getLoc().getPointer();
5711 StringRef Body = StringRef(BodyStart, BodyEnd - BodyStart);
5713 // We Are Anonymous.
5714 MacroLikeBodies.emplace_back(StringRef(), Body, MCAsmMacroParameters());
5715 return &MacroLikeBodies.back();
5718 void AsmParser::instantiateMacroLikeBody(MCAsmMacro *M, SMLoc DirectiveLoc,
5719 raw_svector_ostream &OS) {
5720 OS << ".endr\n";
5722 std::unique_ptr<MemoryBuffer> Instantiation =
5723 MemoryBuffer::getMemBufferCopy(OS.str(), "<instantiation>");
5725 // Create the macro instantiation object and add to the current macro
5726 // instantiation stack.
5727 MacroInstantiation *MI = new MacroInstantiation{
5728 DirectiveLoc, CurBuffer, getTok().getLoc(), TheCondStack.size()};
5729 ActiveMacros.push_back(MI);
5731 // Jump to the macro instantiation and prime the lexer.
5732 CurBuffer = SrcMgr.AddNewSourceBuffer(std::move(Instantiation), SMLoc());
5733 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer());
5734 Lex();
5737 /// parseDirectiveRept
5738 /// ::= .rep | .rept count
5739 bool AsmParser::parseDirectiveRept(SMLoc DirectiveLoc, StringRef Dir) {
5740 const MCExpr *CountExpr;
5741 SMLoc CountLoc = getTok().getLoc();
5742 if (parseExpression(CountExpr))
5743 return true;
5745 int64_t Count;
5746 if (!CountExpr->evaluateAsAbsolute(Count, getStreamer().getAssemblerPtr())) {
5747 return Error(CountLoc, "unexpected token in '" + Dir + "' directive");
5750 if (check(Count < 0, CountLoc, "Count is negative") || parseEOL())
5751 return true;
5753 // Lex the rept definition.
5754 MCAsmMacro *M = parseMacroLikeBody(DirectiveLoc);
5755 if (!M)
5756 return true;
5758 // Macro instantiation is lexical, unfortunately. We construct a new buffer
5759 // to hold the macro body with substitutions.
5760 SmallString<256> Buf;
5761 raw_svector_ostream OS(Buf);
5762 while (Count--) {
5763 // Note that the AtPseudoVariable is disabled for instantiations of .rep(t).
5764 if (expandMacro(OS, *M, {}, {}, false))
5765 return true;
5767 instantiateMacroLikeBody(M, DirectiveLoc, OS);
5769 return false;
5772 /// parseDirectiveIrp
5773 /// ::= .irp symbol,values
5774 bool AsmParser::parseDirectiveIrp(SMLoc DirectiveLoc) {
5775 MCAsmMacroParameter Parameter;
5776 MCAsmMacroArguments A;
5777 if (check(parseIdentifier(Parameter.Name),
5778 "expected identifier in '.irp' directive") ||
5779 parseComma() || parseMacroArguments(nullptr, A) || parseEOL())
5780 return true;
5782 // Lex the irp definition.
5783 MCAsmMacro *M = parseMacroLikeBody(DirectiveLoc);
5784 if (!M)
5785 return true;
5787 // Macro instantiation is lexical, unfortunately. We construct a new buffer
5788 // to hold the macro body with substitutions.
5789 SmallString<256> Buf;
5790 raw_svector_ostream OS(Buf);
5792 for (const MCAsmMacroArgument &Arg : A) {
5793 // Note that the AtPseudoVariable is enabled for instantiations of .irp.
5794 // This is undocumented, but GAS seems to support it.
5795 if (expandMacro(OS, *M, Parameter, Arg, true))
5796 return true;
5799 instantiateMacroLikeBody(M, DirectiveLoc, OS);
5801 return false;
5804 /// parseDirectiveIrpc
5805 /// ::= .irpc symbol,values
5806 bool AsmParser::parseDirectiveIrpc(SMLoc DirectiveLoc) {
5807 MCAsmMacroParameter Parameter;
5808 MCAsmMacroArguments A;
5810 if (check(parseIdentifier(Parameter.Name),
5811 "expected identifier in '.irpc' directive") ||
5812 parseComma() || parseMacroArguments(nullptr, A))
5813 return true;
5815 if (A.size() != 1 || A.front().size() != 1)
5816 return TokError("unexpected token in '.irpc' directive");
5817 if (parseEOL())
5818 return true;
5820 // Lex the irpc definition.
5821 MCAsmMacro *M = parseMacroLikeBody(DirectiveLoc);
5822 if (!M)
5823 return true;
5825 // Macro instantiation is lexical, unfortunately. We construct a new buffer
5826 // to hold the macro body with substitutions.
5827 SmallString<256> Buf;
5828 raw_svector_ostream OS(Buf);
5830 StringRef Values = A[0][0].is(AsmToken::String) ? A[0][0].getStringContents()
5831 : A[0][0].getString();
5832 for (std::size_t I = 0, End = Values.size(); I != End; ++I) {
5833 MCAsmMacroArgument Arg;
5834 Arg.emplace_back(AsmToken::Identifier, Values.substr(I, 1));
5836 // Note that the AtPseudoVariable is enabled for instantiations of .irpc.
5837 // This is undocumented, but GAS seems to support it.
5838 if (expandMacro(OS, *M, Parameter, Arg, true))
5839 return true;
5842 instantiateMacroLikeBody(M, DirectiveLoc, OS);
5844 return false;
5847 bool AsmParser::parseDirectiveEndr(SMLoc DirectiveLoc) {
5848 if (ActiveMacros.empty())
5849 return TokError("unmatched '.endr' directive");
5851 // The only .repl that should get here are the ones created by
5852 // instantiateMacroLikeBody.
5853 assert(getLexer().is(AsmToken::EndOfStatement));
5855 handleMacroExit();
5856 return false;
5859 bool AsmParser::parseDirectiveMSEmit(SMLoc IDLoc, ParseStatementInfo &Info,
5860 size_t Len) {
5861 const MCExpr *Value;
5862 SMLoc ExprLoc = getLexer().getLoc();
5863 if (parseExpression(Value))
5864 return true;
5865 const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value);
5866 if (!MCE)
5867 return Error(ExprLoc, "unexpected expression in _emit");
5868 uint64_t IntValue = MCE->getValue();
5869 if (!isUInt<8>(IntValue) && !isInt<8>(IntValue))
5870 return Error(ExprLoc, "literal value out of range for directive");
5872 Info.AsmRewrites->emplace_back(AOK_Emit, IDLoc, Len);
5873 return false;
5876 bool AsmParser::parseDirectiveMSAlign(SMLoc IDLoc, ParseStatementInfo &Info) {
5877 const MCExpr *Value;
5878 SMLoc ExprLoc = getLexer().getLoc();
5879 if (parseExpression(Value))
5880 return true;
5881 const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value);
5882 if (!MCE)
5883 return Error(ExprLoc, "unexpected expression in align");
5884 uint64_t IntValue = MCE->getValue();
5885 if (!isPowerOf2_64(IntValue))
5886 return Error(ExprLoc, "literal value not a power of two greater then zero");
5888 Info.AsmRewrites->emplace_back(AOK_Align, IDLoc, 5, Log2_64(IntValue));
5889 return false;
5892 bool AsmParser::parseDirectivePrint(SMLoc DirectiveLoc) {
5893 const AsmToken StrTok = getTok();
5894 Lex();
5895 if (StrTok.isNot(AsmToken::String) || StrTok.getString().front() != '"')
5896 return Error(DirectiveLoc, "expected double quoted string after .print");
5897 if (parseEOL())
5898 return true;
5899 llvm::outs() << StrTok.getStringContents() << '\n';
5900 return false;
5903 bool AsmParser::parseDirectiveAddrsig() {
5904 if (parseEOL())
5905 return true;
5906 getStreamer().emitAddrsig();
5907 return false;
5910 bool AsmParser::parseDirectiveAddrsigSym() {
5911 StringRef Name;
5912 if (check(parseIdentifier(Name), "expected identifier") || parseEOL())
5913 return true;
5914 MCSymbol *Sym = getContext().getOrCreateSymbol(Name);
5915 getStreamer().emitAddrsigSym(Sym);
5916 return false;
5919 bool AsmParser::parseDirectivePseudoProbe() {
5920 int64_t Guid;
5921 int64_t Index;
5922 int64_t Type;
5923 int64_t Attr;
5924 int64_t Discriminator = 0;
5926 if (parseIntToken(Guid, "unexpected token in '.pseudoprobe' directive"))
5927 return true;
5929 if (parseIntToken(Index, "unexpected token in '.pseudoprobe' directive"))
5930 return true;
5932 if (parseIntToken(Type, "unexpected token in '.pseudoprobe' directive"))
5933 return true;
5935 if (parseIntToken(Attr, "unexpected token in '.pseudoprobe' directive"))
5936 return true;
5938 if (hasDiscriminator(Attr)) {
5939 if (parseIntToken(Discriminator,
5940 "unexpected token in '.pseudoprobe' directive"))
5941 return true;
5944 // Parse inline stack like @ GUID:11:12 @ GUID:1:11 @ GUID:3:21
5945 MCPseudoProbeInlineStack InlineStack;
5947 while (getLexer().is(AsmToken::At)) {
5948 // eat @
5949 Lex();
5951 int64_t CallerGuid = 0;
5952 if (getLexer().is(AsmToken::Integer)) {
5953 if (parseIntToken(CallerGuid,
5954 "unexpected token in '.pseudoprobe' directive"))
5955 return true;
5958 // eat colon
5959 if (getLexer().is(AsmToken::Colon))
5960 Lex();
5962 int64_t CallerProbeId = 0;
5963 if (getLexer().is(AsmToken::Integer)) {
5964 if (parseIntToken(CallerProbeId,
5965 "unexpected token in '.pseudoprobe' directive"))
5966 return true;
5969 InlineSite Site(CallerGuid, CallerProbeId);
5970 InlineStack.push_back(Site);
5973 // Parse function entry name
5974 StringRef FnName;
5975 if (parseIdentifier(FnName))
5976 return Error(getLexer().getLoc(), "unexpected token in '.pseudoprobe' directive");
5977 MCSymbol *FnSym = getContext().lookupSymbol(FnName);
5979 if (parseEOL())
5980 return true;
5982 getStreamer().emitPseudoProbe(Guid, Index, Type, Attr, Discriminator,
5983 InlineStack, FnSym);
5984 return false;
5987 /// parseDirectiveLTODiscard
5988 /// ::= ".lto_discard" [ identifier ( , identifier )* ]
5989 /// The LTO library emits this directive to discard non-prevailing symbols.
5990 /// We ignore symbol assignments and attribute changes for the specified
5991 /// symbols.
5992 bool AsmParser::parseDirectiveLTODiscard() {
5993 auto ParseOp = [&]() -> bool {
5994 StringRef Name;
5995 SMLoc Loc = getTok().getLoc();
5996 if (parseIdentifier(Name))
5997 return Error(Loc, "expected identifier");
5998 LTODiscardSymbols.insert(Name);
5999 return false;
6002 LTODiscardSymbols.clear();
6003 return parseMany(ParseOp);
6006 // We are comparing pointers, but the pointers are relative to a single string.
6007 // Thus, this should always be deterministic.
6008 static int rewritesSort(const AsmRewrite *AsmRewriteA,
6009 const AsmRewrite *AsmRewriteB) {
6010 if (AsmRewriteA->Loc.getPointer() < AsmRewriteB->Loc.getPointer())
6011 return -1;
6012 if (AsmRewriteB->Loc.getPointer() < AsmRewriteA->Loc.getPointer())
6013 return 1;
6015 // It's possible to have a SizeDirective, Imm/ImmPrefix and an Input/Output
6016 // rewrite to the same location. Make sure the SizeDirective rewrite is
6017 // performed first, then the Imm/ImmPrefix and finally the Input/Output. This
6018 // ensures the sort algorithm is stable.
6019 if (AsmRewritePrecedence[AsmRewriteA->Kind] >
6020 AsmRewritePrecedence[AsmRewriteB->Kind])
6021 return -1;
6023 if (AsmRewritePrecedence[AsmRewriteA->Kind] <
6024 AsmRewritePrecedence[AsmRewriteB->Kind])
6025 return 1;
6026 llvm_unreachable("Unstable rewrite sort.");
6029 bool AsmParser::parseMSInlineAsm(
6030 std::string &AsmString, unsigned &NumOutputs, unsigned &NumInputs,
6031 SmallVectorImpl<std::pair<void *, bool>> &OpDecls,
6032 SmallVectorImpl<std::string> &Constraints,
6033 SmallVectorImpl<std::string> &Clobbers, const MCInstrInfo *MII,
6034 MCInstPrinter *IP, MCAsmParserSemaCallback &SI) {
6035 SmallVector<void *, 4> InputDecls;
6036 SmallVector<void *, 4> OutputDecls;
6037 SmallVector<bool, 4> InputDeclsAddressOf;
6038 SmallVector<bool, 4> OutputDeclsAddressOf;
6039 SmallVector<std::string, 4> InputConstraints;
6040 SmallVector<std::string, 4> OutputConstraints;
6041 SmallVector<MCRegister, 4> ClobberRegs;
6043 SmallVector<AsmRewrite, 4> AsmStrRewrites;
6045 // Prime the lexer.
6046 Lex();
6048 // While we have input, parse each statement.
6049 unsigned InputIdx = 0;
6050 unsigned OutputIdx = 0;
6051 while (getLexer().isNot(AsmToken::Eof)) {
6052 // Parse curly braces marking block start/end
6053 if (parseCurlyBlockScope(AsmStrRewrites))
6054 continue;
6056 ParseStatementInfo Info(&AsmStrRewrites);
6057 bool StatementErr = parseStatement(Info, &SI);
6059 if (StatementErr || Info.ParseError) {
6060 // Emit pending errors if any exist.
6061 printPendingErrors();
6062 return true;
6065 // No pending error should exist here.
6066 assert(!hasPendingError() && "unexpected error from parseStatement");
6068 if (Info.Opcode == ~0U)
6069 continue;
6071 const MCInstrDesc &Desc = MII->get(Info.Opcode);
6073 // Build the list of clobbers, outputs and inputs.
6074 for (unsigned i = 1, e = Info.ParsedOperands.size(); i != e; ++i) {
6075 MCParsedAsmOperand &Operand = *Info.ParsedOperands[i];
6077 // Register operand.
6078 if (Operand.isReg() && !Operand.needAddressOf() &&
6079 !getTargetParser().omitRegisterFromClobberLists(Operand.getReg())) {
6080 unsigned NumDefs = Desc.getNumDefs();
6081 // Clobber.
6082 if (NumDefs && Operand.getMCOperandNum() < NumDefs)
6083 ClobberRegs.push_back(Operand.getReg());
6084 continue;
6087 // Expr/Input or Output.
6088 StringRef SymName = Operand.getSymName();
6089 if (SymName.empty())
6090 continue;
6092 void *OpDecl = Operand.getOpDecl();
6093 if (!OpDecl)
6094 continue;
6096 StringRef Constraint = Operand.getConstraint();
6097 if (Operand.isImm()) {
6098 // Offset as immediate
6099 if (Operand.isOffsetOfLocal())
6100 Constraint = "r";
6101 else
6102 Constraint = "i";
6105 bool isOutput = (i == 1) && Desc.mayStore();
6106 bool Restricted = Operand.isMemUseUpRegs();
6107 SMLoc Start = SMLoc::getFromPointer(SymName.data());
6108 if (isOutput) {
6109 ++InputIdx;
6110 OutputDecls.push_back(OpDecl);
6111 OutputDeclsAddressOf.push_back(Operand.needAddressOf());
6112 OutputConstraints.push_back(("=" + Constraint).str());
6113 AsmStrRewrites.emplace_back(AOK_Output, Start, SymName.size(), 0,
6114 Restricted);
6115 } else {
6116 InputDecls.push_back(OpDecl);
6117 InputDeclsAddressOf.push_back(Operand.needAddressOf());
6118 InputConstraints.push_back(Constraint.str());
6119 if (Desc.operands()[i - 1].isBranchTarget())
6120 AsmStrRewrites.emplace_back(AOK_CallInput, Start, SymName.size(), 0,
6121 Restricted);
6122 else
6123 AsmStrRewrites.emplace_back(AOK_Input, Start, SymName.size(), 0,
6124 Restricted);
6128 // Consider implicit defs to be clobbers. Think of cpuid and push.
6129 llvm::append_range(ClobberRegs, Desc.implicit_defs());
6132 // Set the number of Outputs and Inputs.
6133 NumOutputs = OutputDecls.size();
6134 NumInputs = InputDecls.size();
6136 // Set the unique clobbers.
6137 array_pod_sort(ClobberRegs.begin(), ClobberRegs.end());
6138 ClobberRegs.erase(llvm::unique(ClobberRegs), ClobberRegs.end());
6139 Clobbers.assign(ClobberRegs.size(), std::string());
6140 for (unsigned I = 0, E = ClobberRegs.size(); I != E; ++I) {
6141 raw_string_ostream OS(Clobbers[I]);
6142 IP->printRegName(OS, ClobberRegs[I]);
6145 // Merge the various outputs and inputs. Output are expected first.
6146 if (NumOutputs || NumInputs) {
6147 unsigned NumExprs = NumOutputs + NumInputs;
6148 OpDecls.resize(NumExprs);
6149 Constraints.resize(NumExprs);
6150 for (unsigned i = 0; i < NumOutputs; ++i) {
6151 OpDecls[i] = std::make_pair(OutputDecls[i], OutputDeclsAddressOf[i]);
6152 Constraints[i] = OutputConstraints[i];
6154 for (unsigned i = 0, j = NumOutputs; i < NumInputs; ++i, ++j) {
6155 OpDecls[j] = std::make_pair(InputDecls[i], InputDeclsAddressOf[i]);
6156 Constraints[j] = InputConstraints[i];
6160 // Build the IR assembly string.
6161 std::string AsmStringIR;
6162 raw_string_ostream OS(AsmStringIR);
6163 StringRef ASMString =
6164 SrcMgr.getMemoryBuffer(SrcMgr.getMainFileID())->getBuffer();
6165 const char *AsmStart = ASMString.begin();
6166 const char *AsmEnd = ASMString.end();
6167 array_pod_sort(AsmStrRewrites.begin(), AsmStrRewrites.end(), rewritesSort);
6168 for (auto I = AsmStrRewrites.begin(), E = AsmStrRewrites.end(); I != E; ++I) {
6169 const AsmRewrite &AR = *I;
6170 // Check if this has already been covered by another rewrite...
6171 if (AR.Done)
6172 continue;
6173 AsmRewriteKind Kind = AR.Kind;
6175 const char *Loc = AR.Loc.getPointer();
6176 assert(Loc >= AsmStart && "Expected Loc to be at or after Start!");
6178 // Emit everything up to the immediate/expression.
6179 if (unsigned Len = Loc - AsmStart)
6180 OS << StringRef(AsmStart, Len);
6182 // Skip the original expression.
6183 if (Kind == AOK_Skip) {
6184 AsmStart = Loc + AR.Len;
6185 continue;
6188 unsigned AdditionalSkip = 0;
6189 // Rewrite expressions in $N notation.
6190 switch (Kind) {
6191 default:
6192 break;
6193 case AOK_IntelExpr:
6194 assert(AR.IntelExp.isValid() && "cannot write invalid intel expression");
6195 if (AR.IntelExp.NeedBracs)
6196 OS << "[";
6197 if (AR.IntelExp.hasBaseReg())
6198 OS << AR.IntelExp.BaseReg;
6199 if (AR.IntelExp.hasIndexReg())
6200 OS << (AR.IntelExp.hasBaseReg() ? " + " : "")
6201 << AR.IntelExp.IndexReg;
6202 if (AR.IntelExp.Scale > 1)
6203 OS << " * $$" << AR.IntelExp.Scale;
6204 if (AR.IntelExp.hasOffset()) {
6205 if (AR.IntelExp.hasRegs())
6206 OS << " + ";
6207 // Fuse this rewrite with a rewrite of the offset name, if present.
6208 StringRef OffsetName = AR.IntelExp.OffsetName;
6209 SMLoc OffsetLoc = SMLoc::getFromPointer(AR.IntelExp.OffsetName.data());
6210 size_t OffsetLen = OffsetName.size();
6211 auto rewrite_it = std::find_if(
6212 I, AsmStrRewrites.end(), [&](const AsmRewrite &FusingAR) {
6213 return FusingAR.Loc == OffsetLoc && FusingAR.Len == OffsetLen &&
6214 (FusingAR.Kind == AOK_Input ||
6215 FusingAR.Kind == AOK_CallInput);
6217 if (rewrite_it == AsmStrRewrites.end()) {
6218 OS << "offset " << OffsetName;
6219 } else if (rewrite_it->Kind == AOK_CallInput) {
6220 OS << "${" << InputIdx++ << ":P}";
6221 rewrite_it->Done = true;
6222 } else {
6223 OS << '$' << InputIdx++;
6224 rewrite_it->Done = true;
6227 if (AR.IntelExp.Imm || AR.IntelExp.emitImm())
6228 OS << (AR.IntelExp.emitImm() ? "$$" : " + $$") << AR.IntelExp.Imm;
6229 if (AR.IntelExp.NeedBracs)
6230 OS << "]";
6231 break;
6232 case AOK_Label:
6233 OS << Ctx.getAsmInfo()->getPrivateLabelPrefix() << AR.Label;
6234 break;
6235 case AOK_Input:
6236 if (AR.IntelExpRestricted)
6237 OS << "${" << InputIdx++ << ":P}";
6238 else
6239 OS << '$' << InputIdx++;
6240 break;
6241 case AOK_CallInput:
6242 OS << "${" << InputIdx++ << ":P}";
6243 break;
6244 case AOK_Output:
6245 if (AR.IntelExpRestricted)
6246 OS << "${" << OutputIdx++ << ":P}";
6247 else
6248 OS << '$' << OutputIdx++;
6249 break;
6250 case AOK_SizeDirective:
6251 switch (AR.Val) {
6252 default: break;
6253 case 8: OS << "byte ptr "; break;
6254 case 16: OS << "word ptr "; break;
6255 case 32: OS << "dword ptr "; break;
6256 case 64: OS << "qword ptr "; break;
6257 case 80: OS << "xword ptr "; break;
6258 case 128: OS << "xmmword ptr "; break;
6259 case 256: OS << "ymmword ptr "; break;
6261 break;
6262 case AOK_Emit:
6263 OS << ".byte";
6264 break;
6265 case AOK_Align: {
6266 // MS alignment directives are measured in bytes. If the native assembler
6267 // measures alignment in bytes, we can pass it straight through.
6268 OS << ".align";
6269 if (getContext().getAsmInfo()->getAlignmentIsInBytes())
6270 break;
6272 // Alignment is in log2 form, so print that instead and skip the original
6273 // immediate.
6274 unsigned Val = AR.Val;
6275 OS << ' ' << Val;
6276 assert(Val < 10 && "Expected alignment less then 2^10.");
6277 AdditionalSkip = (Val < 4) ? 2 : Val < 7 ? 3 : 4;
6278 break;
6280 case AOK_EVEN:
6281 OS << ".even";
6282 break;
6283 case AOK_EndOfStatement:
6284 OS << "\n\t";
6285 break;
6288 // Skip the original expression.
6289 AsmStart = Loc + AR.Len + AdditionalSkip;
6292 // Emit the remainder of the asm string.
6293 if (AsmStart != AsmEnd)
6294 OS << StringRef(AsmStart, AsmEnd - AsmStart);
6296 AsmString = std::move(AsmStringIR);
6297 return false;
6300 bool HLASMAsmParser::parseAsHLASMLabel(ParseStatementInfo &Info,
6301 MCAsmParserSemaCallback *SI) {
6302 AsmToken LabelTok = getTok();
6303 SMLoc LabelLoc = LabelTok.getLoc();
6304 StringRef LabelVal;
6306 if (parseIdentifier(LabelVal))
6307 return Error(LabelLoc, "The HLASM Label has to be an Identifier");
6309 // We have validated whether the token is an Identifier.
6310 // Now we have to validate whether the token is a
6311 // valid HLASM Label.
6312 if (!getTargetParser().isLabel(LabelTok) || checkForValidSection())
6313 return true;
6315 // Lex leading spaces to get to the next operand.
6316 lexLeadingSpaces();
6318 // We shouldn't emit the label if there is nothing else after the label.
6319 // i.e asm("<token>\n")
6320 if (getTok().is(AsmToken::EndOfStatement))
6321 return Error(LabelLoc,
6322 "Cannot have just a label for an HLASM inline asm statement");
6324 MCSymbol *Sym = getContext().getOrCreateSymbol(
6325 getContext().getAsmInfo()->shouldEmitLabelsInUpperCase()
6326 ? LabelVal.upper()
6327 : LabelVal);
6329 getTargetParser().doBeforeLabelEmit(Sym, LabelLoc);
6331 // Emit the label.
6332 Out.emitLabel(Sym, LabelLoc);
6334 // If we are generating dwarf for assembly source files then gather the
6335 // info to make a dwarf label entry for this label if needed.
6336 if (enabledGenDwarfForAssembly())
6337 MCGenDwarfLabelEntry::Make(Sym, &getStreamer(), getSourceManager(),
6338 LabelLoc);
6340 getTargetParser().onLabelParsed(Sym);
6342 return false;
6345 bool HLASMAsmParser::parseAsMachineInstruction(ParseStatementInfo &Info,
6346 MCAsmParserSemaCallback *SI) {
6347 AsmToken OperationEntryTok = Lexer.getTok();
6348 SMLoc OperationEntryLoc = OperationEntryTok.getLoc();
6349 StringRef OperationEntryVal;
6351 // Attempt to parse the first token as an Identifier
6352 if (parseIdentifier(OperationEntryVal))
6353 return Error(OperationEntryLoc, "unexpected token at start of statement");
6355 // Once we've parsed the operation entry successfully, lex
6356 // any spaces to get to the OperandEntries.
6357 lexLeadingSpaces();
6359 return parseAndMatchAndEmitTargetInstruction(
6360 Info, OperationEntryVal, OperationEntryTok, OperationEntryLoc);
6363 bool HLASMAsmParser::parseStatement(ParseStatementInfo &Info,
6364 MCAsmParserSemaCallback *SI) {
6365 assert(!hasPendingError() && "parseStatement started with pending error");
6367 // Should the first token be interpreted as a HLASM Label.
6368 bool ShouldParseAsHLASMLabel = false;
6370 // If a Name Entry exists, it should occur at the very
6371 // start of the string. In this case, we should parse the
6372 // first non-space token as a Label.
6373 // If the Name entry is missing (i.e. there's some other
6374 // token), then we attempt to parse the first non-space
6375 // token as a Machine Instruction.
6376 if (getTok().isNot(AsmToken::Space))
6377 ShouldParseAsHLASMLabel = true;
6379 // If we have an EndOfStatement (which includes the target's comment
6380 // string) we can appropriately lex it early on)
6381 if (Lexer.is(AsmToken::EndOfStatement)) {
6382 // if this is a line comment we can drop it safely
6383 if (getTok().getString().empty() || getTok().getString().front() == '\r' ||
6384 getTok().getString().front() == '\n')
6385 Out.addBlankLine();
6386 Lex();
6387 return false;
6390 // We have established how to parse the inline asm statement.
6391 // Now we can safely lex any leading spaces to get to the
6392 // first token.
6393 lexLeadingSpaces();
6395 // If we see a new line or carriage return as the first operand,
6396 // after lexing leading spaces, emit the new line and lex the
6397 // EndOfStatement token.
6398 if (Lexer.is(AsmToken::EndOfStatement)) {
6399 if (getTok().getString().front() == '\n' ||
6400 getTok().getString().front() == '\r') {
6401 Out.addBlankLine();
6402 Lex();
6403 return false;
6407 // Handle the label first if we have to before processing the rest
6408 // of the tokens as a machine instruction.
6409 if (ShouldParseAsHLASMLabel) {
6410 // If there were any errors while handling and emitting the label,
6411 // early return.
6412 if (parseAsHLASMLabel(Info, SI)) {
6413 // If we know we've failed in parsing, simply eat until end of the
6414 // statement. This ensures that we don't process any other statements.
6415 eatToEndOfStatement();
6416 return true;
6420 return parseAsMachineInstruction(Info, SI);
6423 namespace llvm {
6424 namespace MCParserUtils {
6426 bool parseAssignmentExpression(StringRef Name, bool allow_redef,
6427 MCAsmParser &Parser, MCSymbol *&Sym,
6428 const MCExpr *&Value) {
6430 // FIXME: Use better location, we should use proper tokens.
6431 SMLoc EqualLoc = Parser.getTok().getLoc();
6432 if (Parser.parseExpression(Value))
6433 return Parser.TokError("missing expression");
6435 // Note: we don't count b as used in "a = b". This is to allow
6436 // a = b
6437 // b = c
6439 if (Parser.parseEOL())
6440 return true;
6442 // Validate that the LHS is allowed to be a variable (either it has not been
6443 // used as a symbol, or it is an absolute symbol).
6444 Sym = Parser.getContext().lookupSymbol(Name);
6445 if (Sym) {
6446 // Diagnose assignment to a label.
6448 // FIXME: Diagnostics. Note the location of the definition as a label.
6449 // FIXME: Diagnose assignment to protected identifier (e.g., register name).
6450 if (Value->isSymbolUsedInExpression(Sym))
6451 return Parser.Error(EqualLoc, "Recursive use of '" + Name + "'");
6452 else if (Sym->isUndefined(/*SetUsed*/ false) && !Sym->isUsed() &&
6453 !Sym->isVariable())
6454 ; // Allow redefinitions of undefined symbols only used in directives.
6455 else if (Sym->isVariable() && !Sym->isUsed() && allow_redef)
6456 ; // Allow redefinitions of variables that haven't yet been used.
6457 else if (!Sym->isUndefined() && (!Sym->isVariable() || !allow_redef))
6458 return Parser.Error(EqualLoc, "redefinition of '" + Name + "'");
6459 else if (!Sym->isVariable())
6460 return Parser.Error(EqualLoc, "invalid assignment to '" + Name + "'");
6461 else if (!isa<MCConstantExpr>(Sym->getVariableValue()))
6462 return Parser.Error(EqualLoc,
6463 "invalid reassignment of non-absolute variable '" +
6464 Name + "'");
6465 } else if (Name == ".") {
6466 Parser.getStreamer().emitValueToOffset(Value, 0, EqualLoc);
6467 return false;
6468 } else
6469 Sym = Parser.getContext().getOrCreateSymbol(Name);
6471 Sym->setRedefinable(allow_redef);
6473 return false;
6476 } // end namespace MCParserUtils
6477 } // end namespace llvm
6479 /// Create an MCAsmParser instance.
6480 MCAsmParser *llvm::createMCAsmParser(SourceMgr &SM, MCContext &C,
6481 MCStreamer &Out, const MCAsmInfo &MAI,
6482 unsigned CB) {
6483 if (C.getTargetTriple().isSystemZ() && C.getTargetTriple().isOSzOS())
6484 return new HLASMAsmParser(SM, C, Out, MAI, CB);
6486 return new AsmParser(SM, C, Out, MAI, CB);