1 //===- AsmParser.cpp - Parser for Assembly Files --------------------------===//
2 //
3 // The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // This class implements the parser for assembly files.
11 //
12 //===----------------------------------------------------------------------===//
13
14 #include "llvm/ADT/APFloat.h"
15 #include "llvm/ADT/STLExtras.h"
16 #include "llvm/ADT/SmallString.h"
17 #include "llvm/ADT/StringMap.h"
18 #include "llvm/ADT/Twine.h"
19 #include "llvm/MC/MCAsmInfo.h"
20 #include "llvm/MC/MCContext.h"
21 #include "llvm/MC/MCDwarf.h"
22 #include "llvm/MC/MCExpr.h"
23 #include "llvm/MC/MCInstPrinter.h"
24 #include "llvm/MC/MCInstrInfo.h"
25 #include "llvm/MC/MCObjectFileInfo.h"
26 #include "llvm/MC/MCParser/AsmCond.h"
27 #include "llvm/MC/MCParser/AsmLexer.h"
28 #include "llvm/MC/MCParser/MCAsmParser.h"
29 #include "llvm/MC/MCParser/MCAsmParserUtils.h"
30 #include "llvm/MC/MCParser/MCParsedAsmOperand.h"
31 #include "llvm/MC/MCRegisterInfo.h"
32 #include "llvm/MC/MCSectionMachO.h"
33 #include "llvm/MC/MCStreamer.h"
34 #include "llvm/MC/MCSymbol.h"
35 #include "llvm/MC/MCTargetAsmParser.h"
36 #include "llvm/Support/CommandLine.h"
37 #include "llvm/Support/ErrorHandling.h"
38 #include "llvm/Support/MathExtras.h"
39 #include "llvm/Support/MemoryBuffer.h"
40 #include "llvm/Support/SourceMgr.h"
41 #include "llvm/Support/raw_ostream.h"
42 #include <cctype>
43 #include <deque>
44 #include <set>
45 #include <string>
46 #include <vector>
47 using namespace llvm;
48
~MCAsmParserSemaCallback()49 MCAsmParserSemaCallback::~MCAsmParserSemaCallback() {}
50
51 namespace {
52 /// \brief Helper types for tracking macro definitions.
53 typedef std::vector<AsmToken> MCAsmMacroArgument;
54 typedef std::vector<MCAsmMacroArgument> MCAsmMacroArguments;
55
56 struct MCAsmMacroParameter {
57 StringRef Name;
58 MCAsmMacroArgument Value;
59 bool Required;
60 bool Vararg;
61
MCAsmMacroParameter__anona80e9bb80111::MCAsmMacroParameter62 MCAsmMacroParameter() : Required(false), Vararg(false) {}
63 };
64
65 typedef std::vector<MCAsmMacroParameter> MCAsmMacroParameters;
66
67 struct MCAsmMacro {
68 StringRef Name;
69 StringRef Body;
70 MCAsmMacroParameters Parameters;
71
72 public:
MCAsmMacro__anona80e9bb80111::MCAsmMacro73 MCAsmMacro(StringRef N, StringRef B, MCAsmMacroParameters P)
74 : Name(N), Body(B), Parameters(std::move(P)) {}
75 };
76
77 /// \brief Helper class for storing information about an active macro
78 /// instantiation.
79 struct MacroInstantiation {
80 /// The location of the instantiation.
81 SMLoc InstantiationLoc;
82
83 /// The buffer where parsing should resume upon instantiation completion.
84 int ExitBuffer;
85
86 /// The location where parsing should resume upon instantiation completion.
87 SMLoc ExitLoc;
88
89 /// The depth of TheCondStack at the start of the instantiation.
90 size_t CondStackDepth;
91
92 public:
93 MacroInstantiation(SMLoc IL, int EB, SMLoc EL, size_t CondStackDepth);
94 };
95
96 struct ParseStatementInfo {
97 /// \brief The parsed operands from the last parsed statement.
98 SmallVector<std::unique_ptr<MCParsedAsmOperand>, 8> ParsedOperands;
99
100 /// \brief The opcode from the last parsed instruction.
101 unsigned Opcode;
102
103 /// \brief Was there an error parsing the inline assembly?
104 bool ParseError;
105
106 SmallVectorImpl<AsmRewrite> *AsmRewrites;
107
ParseStatementInfo__anona80e9bb80111::ParseStatementInfo108 ParseStatementInfo() : Opcode(~0U), ParseError(false), AsmRewrites(nullptr) {}
ParseStatementInfo__anona80e9bb80111::ParseStatementInfo109 ParseStatementInfo(SmallVectorImpl<AsmRewrite> *rewrites)
110 : Opcode(~0), ParseError(false), AsmRewrites(rewrites) {}
111 };
112
113 /// \brief The concrete assembly parser instance.
114 class AsmParser : public MCAsmParser {
115 AsmParser(const AsmParser &) = delete;
116 void operator=(const AsmParser &) = delete;
117 private:
118 AsmLexer Lexer;
119 MCContext &Ctx;
120 MCStreamer &Out;
121 const MCAsmInfo &MAI;
122 SourceMgr &SrcMgr;
123 SourceMgr::DiagHandlerTy SavedDiagHandler;
124 void *SavedDiagContext;
125 std::unique_ptr<MCAsmParserExtension> PlatformParser;
126
127 /// This is the current buffer index we're lexing from as managed by the
128 /// SourceMgr object.
129 unsigned CurBuffer;
130
131 AsmCond TheCondState;
132 std::vector<AsmCond> TheCondStack;
133
134 /// \brief maps directive names to handler methods in parser
135 /// extensions. Extensions register themselves in this map by calling
136 /// addDirectiveHandler.
137 StringMap<ExtensionDirectiveHandler> ExtensionDirectiveMap;
138
139 /// \brief Map of currently defined macros.
140 StringMap<MCAsmMacro> MacroMap;
141
142 /// \brief Stack of active macro instantiations.
143 std::vector<MacroInstantiation*> ActiveMacros;
144
145 /// \brief List of bodies of anonymous macros.
146 std::deque<MCAsmMacro> MacroLikeBodies;
147
148 /// Boolean tracking whether macro substitution is enabled.
149 unsigned MacrosEnabledFlag : 1;
150
151 /// \brief Keeps track of how many .macro's have been instantiated.
152 unsigned NumOfMacroInstantiations;
153
154 /// Flag tracking whether any errors have been encountered.
155 unsigned HadError : 1;
156
157 /// The values from the last parsed cpp hash file line comment if any.
158 StringRef CppHashFilename;
159 int64_t CppHashLineNumber;
160 SMLoc CppHashLoc;
161 unsigned CppHashBuf;
162 /// When generating dwarf for assembly source files we need to calculate the
163 /// logical line number based on the last parsed cpp hash file line comment
164 /// and current line. Since this is slow and messes up the SourceMgr's
165 /// cache we save the last info we queried with SrcMgr.FindLineNumber().
166 SMLoc LastQueryIDLoc;
167 unsigned LastQueryBuffer;
168 unsigned LastQueryLine;
169
170 /// AssemblerDialect. ~OU means unset value and use value provided by MAI.
171 unsigned AssemblerDialect;
172
173 /// \brief is Darwin compatibility enabled?
174 bool IsDarwin;
175
176 /// \brief Are we parsing ms-style inline assembly?
177 bool ParsingInlineAsm;
178
179 public:
180 AsmParser(SourceMgr &SM, MCContext &Ctx, MCStreamer &Out,
181 const MCAsmInfo &MAI);
182 ~AsmParser() override;
183
184 bool Run(bool NoInitialTextSection, bool NoFinalize = false) override;
185
addDirectiveHandler(StringRef Directive,ExtensionDirectiveHandler Handler)186 void addDirectiveHandler(StringRef Directive,
187 ExtensionDirectiveHandler Handler) override {
188 ExtensionDirectiveMap[Directive] = Handler;
189 }
190
addAliasForDirective(StringRef Directive,StringRef Alias)191 void addAliasForDirective(StringRef Directive, StringRef Alias) override {
192 DirectiveKindMap[Directive] = DirectiveKindMap[Alias];
193 }
194
195 public:
196 /// @name MCAsmParser Interface
197 /// {
198
getSourceManager()199 SourceMgr &getSourceManager() override { return SrcMgr; }
getLexer()200 MCAsmLexer &getLexer() override { return Lexer; }
getContext()201 MCContext &getContext() override { return Ctx; }
getStreamer()202 MCStreamer &getStreamer() override { return Out; }
getAssemblerDialect()203 unsigned getAssemblerDialect() override {
204 if (AssemblerDialect == ~0U)
205 return MAI.getAssemblerDialect();
206 else
207 return AssemblerDialect;
208 }
setAssemblerDialect(unsigned i)209 void setAssemblerDialect(unsigned i) override {
210 AssemblerDialect = i;
211 }
212
213 void Note(SMLoc L, const Twine &Msg,
214 ArrayRef<SMRange> Ranges = None) override;
215 bool Warning(SMLoc L, const Twine &Msg,
216 ArrayRef<SMRange> Ranges = None) override;
217 bool Error(SMLoc L, const Twine &Msg,
218 ArrayRef<SMRange> Ranges = None) override;
219
220 const AsmToken &Lex() override;
221
setParsingInlineAsm(bool V)222 void setParsingInlineAsm(bool V) override { ParsingInlineAsm = V; }
isParsingInlineAsm()223 bool isParsingInlineAsm() override { return ParsingInlineAsm; }
224
225 bool parseMSInlineAsm(void *AsmLoc, std::string &AsmString,
226 unsigned &NumOutputs, unsigned &NumInputs,
227 SmallVectorImpl<std::pair<void *,bool> > &OpDecls,
228 SmallVectorImpl<std::string> &Constraints,
229 SmallVectorImpl<std::string> &Clobbers,
230 const MCInstrInfo *MII, const MCInstPrinter *IP,
231 MCAsmParserSemaCallback &SI) override;
232
233 bool parseExpression(const MCExpr *&Res);
234 bool parseExpression(const MCExpr *&Res, SMLoc &EndLoc) override;
235 bool parsePrimaryExpr(const MCExpr *&Res, SMLoc &EndLoc) override;
236 bool parseParenExpression(const MCExpr *&Res, SMLoc &EndLoc) override;
237 bool parseParenExprOfDepth(unsigned ParenDepth, const MCExpr *&Res,
238 SMLoc &EndLoc) override;
239 bool parseAbsoluteExpression(int64_t &Res) override;
240
241 /// \brief Parse an identifier or string (as a quoted identifier)
242 /// and set \p Res to the identifier contents.
243 bool parseIdentifier(StringRef &Res) override;
244 void eatToEndOfStatement() override;
245
246 void checkForValidSection() override;
247 /// }
248
249 private:
250
251 bool parseStatement(ParseStatementInfo &Info,
252 MCAsmParserSemaCallback *SI);
253 void eatToEndOfLine();
254 bool parseCppHashLineFilenameComment(const SMLoc &L);
255
256 void checkForBadMacro(SMLoc DirectiveLoc, StringRef Name, StringRef Body,
257 ArrayRef<MCAsmMacroParameter> Parameters);
258 bool expandMacro(raw_svector_ostream &OS, StringRef Body,
259 ArrayRef<MCAsmMacroParameter> Parameters,
260 ArrayRef<MCAsmMacroArgument> A, bool EnableAtPseudoVariable,
261 const SMLoc &L);
262
263 /// \brief Are macros enabled in the parser?
areMacrosEnabled()264 bool areMacrosEnabled() {return MacrosEnabledFlag;}
265
266 /// \brief Control a flag in the parser that enables or disables macros.
setMacrosEnabled(bool Flag)267 void setMacrosEnabled(bool Flag) {MacrosEnabledFlag = Flag;}
268
269 /// \brief Lookup a previously defined macro.
270 /// \param Name Macro name.
271 /// \returns Pointer to macro. NULL if no such macro was defined.
272 const MCAsmMacro* lookupMacro(StringRef Name);
273
274 /// \brief Define a new macro with the given name and information.
275 void defineMacro(StringRef Name, MCAsmMacro Macro);
276
277 /// \brief Undefine a macro. If no such macro was defined, it's a no-op.
278 void undefineMacro(StringRef Name);
279
280 /// \brief Are we inside a macro instantiation?
isInsideMacroInstantiation()281 bool isInsideMacroInstantiation() {return !ActiveMacros.empty();}
282
283 /// \brief Handle entry to macro instantiation.
284 ///
285 /// \param M The macro.
286 /// \param NameLoc Instantiation location.
287 bool handleMacroEntry(const MCAsmMacro *M, SMLoc NameLoc);
288
289 /// \brief Handle exit from macro instantiation.
290 void handleMacroExit();
291
292 /// \brief Extract AsmTokens for a macro argument.
293 bool parseMacroArgument(MCAsmMacroArgument &MA, bool Vararg);
294
295 /// \brief Parse all macro arguments for a given macro.
296 bool parseMacroArguments(const MCAsmMacro *M, MCAsmMacroArguments &A);
297
298 void printMacroInstantiations();
printMessage(SMLoc Loc,SourceMgr::DiagKind Kind,const Twine & Msg,ArrayRef<SMRange> Ranges=None) const299 void printMessage(SMLoc Loc, SourceMgr::DiagKind Kind, const Twine &Msg,
300 ArrayRef<SMRange> Ranges = None) const {
301 SrcMgr.PrintMessage(Loc, Kind, Msg, Ranges);
302 }
303 static void DiagHandler(const SMDiagnostic &Diag, void *Context);
304
305 /// \brief Enter the specified file. This returns true on failure.
306 bool enterIncludeFile(const std::string &Filename);
307
308 /// \brief Process the specified file for the .incbin directive.
309 /// This returns true on failure.
310 bool processIncbinFile(const std::string &Filename);
311
312 /// \brief Reset the current lexer position to that given by \p Loc. The
313 /// current token is not set; clients should ensure Lex() is called
314 /// subsequently.
315 ///
316 /// \param InBuffer If not 0, should be the known buffer id that contains the
317 /// location.
318 void jumpToLoc(SMLoc Loc, unsigned InBuffer = 0);
319
320 /// \brief Parse up to the end of statement and a return the contents from the
321 /// current token until the end of the statement; the current token on exit
322 /// will be either the EndOfStatement or EOF.
323 StringRef parseStringToEndOfStatement() override;
324
325 /// \brief Parse until the end of a statement or a comma is encountered,
326 /// return the contents from the current token up to the end or comma.
327 StringRef parseStringToComma();
328
329 bool parseAssignment(StringRef Name, bool allow_redef,
330 bool NoDeadStrip = false);
331
332 unsigned getBinOpPrecedence(AsmToken::TokenKind K,
333 MCBinaryExpr::Opcode &Kind);
334
335 bool parseBinOpRHS(unsigned Precedence, const MCExpr *&Res, SMLoc &EndLoc);
336 bool parseParenExpr(const MCExpr *&Res, SMLoc &EndLoc);
337 bool parseBracketExpr(const MCExpr *&Res, SMLoc &EndLoc);
338
339 bool parseRegisterOrRegisterNumber(int64_t &Register, SMLoc DirectiveLoc);
340
341 // Generic (target and platform independent) directive parsing.
342 enum DirectiveKind {
343 DK_NO_DIRECTIVE, // Placeholder
344 DK_SET, DK_EQU, DK_EQUIV, DK_ASCII, DK_ASCIZ, DK_STRING, DK_BYTE, DK_SHORT,
345 DK_VALUE, DK_2BYTE, DK_LONG, DK_INT, DK_4BYTE, DK_QUAD, DK_8BYTE, DK_OCTA,
346 DK_SINGLE, DK_FLOAT, DK_DOUBLE, DK_ALIGN, DK_ALIGN32, DK_BALIGN, DK_BALIGNW,
347 DK_BALIGNL, DK_P2ALIGN, DK_P2ALIGNW, DK_P2ALIGNL, DK_ORG, DK_FILL, DK_ENDR,
348 DK_BUNDLE_ALIGN_MODE, DK_BUNDLE_LOCK, DK_BUNDLE_UNLOCK,
349 DK_ZERO, DK_EXTERN, DK_GLOBL, DK_GLOBAL,
350 DK_LAZY_REFERENCE, DK_NO_DEAD_STRIP, DK_SYMBOL_RESOLVER, DK_PRIVATE_EXTERN,
351 DK_REFERENCE, DK_WEAK_DEFINITION, DK_WEAK_REFERENCE,
352 DK_WEAK_DEF_CAN_BE_HIDDEN, DK_COMM, DK_COMMON, DK_LCOMM, DK_ABORT,
353 DK_INCLUDE, DK_INCBIN, DK_CODE16, DK_CODE16GCC, DK_REPT, DK_IRP, DK_IRPC,
354 DK_IF, DK_IFEQ, DK_IFGE, DK_IFGT, DK_IFLE, DK_IFLT, DK_IFNE, DK_IFB,
355 DK_IFNB, DK_IFC, DK_IFEQS, DK_IFNC, DK_IFNES, DK_IFDEF, DK_IFNDEF,
356 DK_IFNOTDEF, DK_ELSEIF, DK_ELSE, DK_ENDIF,
357 DK_SPACE, DK_SKIP, DK_FILE, DK_LINE, DK_LOC, DK_STABS,
358 DK_CFI_SECTIONS, DK_CFI_STARTPROC, DK_CFI_ENDPROC, DK_CFI_DEF_CFA,
359 DK_CFI_DEF_CFA_OFFSET, DK_CFI_ADJUST_CFA_OFFSET, DK_CFI_DEF_CFA_REGISTER,
360 DK_CFI_OFFSET, DK_CFI_REL_OFFSET, DK_CFI_PERSONALITY, DK_CFI_LSDA,
361 DK_CFI_REMEMBER_STATE, DK_CFI_RESTORE_STATE, DK_CFI_SAME_VALUE,
362 DK_CFI_RESTORE, DK_CFI_ESCAPE, DK_CFI_SIGNAL_FRAME, DK_CFI_UNDEFINED,
363 DK_CFI_REGISTER, DK_CFI_WINDOW_SAVE,
364 DK_MACROS_ON, DK_MACROS_OFF,
365 DK_MACRO, DK_EXITM, DK_ENDM, DK_ENDMACRO, DK_PURGEM,
366 DK_SLEB128, DK_ULEB128,
367 DK_ERR, DK_ERROR, DK_WARNING,
368 DK_END
369 };
370
371 /// \brief Maps directive name --> DirectiveKind enum, for
372 /// directives parsed by this class.
373 StringMap<DirectiveKind> DirectiveKindMap;
374
375 // ".ascii", ".asciz", ".string"
376 bool parseDirectiveAscii(StringRef IDVal, bool ZeroTerminated);
377 bool parseDirectiveValue(unsigned Size); // ".byte", ".long", ...
378 bool parseDirectiveOctaValue(); // ".octa"
379 bool parseDirectiveRealValue(const fltSemantics &); // ".single", ...
380 bool parseDirectiveFill(); // ".fill"
381 bool parseDirectiveZero(); // ".zero"
382 // ".set", ".equ", ".equiv"
383 bool parseDirectiveSet(StringRef IDVal, bool allow_redef);
384 bool parseDirectiveOrg(); // ".org"
385 // ".align{,32}", ".p2align{,w,l}"
386 bool parseDirectiveAlign(bool IsPow2, unsigned ValueSize);
387
388 // ".file", ".line", ".loc", ".stabs"
389 bool parseDirectiveFile(SMLoc DirectiveLoc);
390 bool parseDirectiveLine();
391 bool parseDirectiveLoc();
392 bool parseDirectiveStabs();
393
394 // .cfi directives
395 bool parseDirectiveCFIRegister(SMLoc DirectiveLoc);
396 bool parseDirectiveCFIWindowSave();
397 bool parseDirectiveCFISections();
398 bool parseDirectiveCFIStartProc();
399 bool parseDirectiveCFIEndProc();
400 bool parseDirectiveCFIDefCfaOffset();
401 bool parseDirectiveCFIDefCfa(SMLoc DirectiveLoc);
402 bool parseDirectiveCFIAdjustCfaOffset();
403 bool parseDirectiveCFIDefCfaRegister(SMLoc DirectiveLoc);
404 bool parseDirectiveCFIOffset(SMLoc DirectiveLoc);
405 bool parseDirectiveCFIRelOffset(SMLoc DirectiveLoc);
406 bool parseDirectiveCFIPersonalityOrLsda(bool IsPersonality);
407 bool parseDirectiveCFIRememberState();
408 bool parseDirectiveCFIRestoreState();
409 bool parseDirectiveCFISameValue(SMLoc DirectiveLoc);
410 bool parseDirectiveCFIRestore(SMLoc DirectiveLoc);
411 bool parseDirectiveCFIEscape();
412 bool parseDirectiveCFISignalFrame();
413 bool parseDirectiveCFIUndefined(SMLoc DirectiveLoc);
414
415 // macro directives
416 bool parseDirectivePurgeMacro(SMLoc DirectiveLoc);
417 bool parseDirectiveExitMacro(StringRef Directive);
418 bool parseDirectiveEndMacro(StringRef Directive);
419 bool parseDirectiveMacro(SMLoc DirectiveLoc);
420 bool parseDirectiveMacrosOnOff(StringRef Directive);
421
422 // ".bundle_align_mode"
423 bool parseDirectiveBundleAlignMode();
424 // ".bundle_lock"
425 bool parseDirectiveBundleLock();
426 // ".bundle_unlock"
427 bool parseDirectiveBundleUnlock();
428
429 // ".space", ".skip"
430 bool parseDirectiveSpace(StringRef IDVal);
431
432 // .sleb128 (Signed=true) and .uleb128 (Signed=false)
433 bool parseDirectiveLEB128(bool Signed);
434
435 /// \brief Parse a directive like ".globl" which
436 /// accepts a single symbol (which should be a label or an external).
437 bool parseDirectiveSymbolAttribute(MCSymbolAttr Attr);
438
439 bool parseDirectiveComm(bool IsLocal); // ".comm" and ".lcomm"
440
441 bool parseDirectiveAbort(); // ".abort"
442 bool parseDirectiveInclude(); // ".include"
443 bool parseDirectiveIncbin(); // ".incbin"
444
445 // ".if", ".ifeq", ".ifge", ".ifgt" , ".ifle", ".iflt" or ".ifne"
446 bool parseDirectiveIf(SMLoc DirectiveLoc, DirectiveKind DirKind);
447 // ".ifb" or ".ifnb", depending on ExpectBlank.
448 bool parseDirectiveIfb(SMLoc DirectiveLoc, bool ExpectBlank);
449 // ".ifc" or ".ifnc", depending on ExpectEqual.
450 bool parseDirectiveIfc(SMLoc DirectiveLoc, bool ExpectEqual);
451 // ".ifeqs" or ".ifnes", depending on ExpectEqual.
452 bool parseDirectiveIfeqs(SMLoc DirectiveLoc, bool ExpectEqual);
453 // ".ifdef" or ".ifndef", depending on expect_defined
454 bool parseDirectiveIfdef(SMLoc DirectiveLoc, bool expect_defined);
455 bool parseDirectiveElseIf(SMLoc DirectiveLoc); // ".elseif"
456 bool parseDirectiveElse(SMLoc DirectiveLoc); // ".else"
457 bool parseDirectiveEndIf(SMLoc DirectiveLoc); // .endif
458 bool parseEscapedString(std::string &Data) override;
459
460 const MCExpr *applyModifierToExpr(const MCExpr *E,
461 MCSymbolRefExpr::VariantKind Variant);
462
463 // Macro-like directives
464 MCAsmMacro *parseMacroLikeBody(SMLoc DirectiveLoc);
465 void instantiateMacroLikeBody(MCAsmMacro *M, SMLoc DirectiveLoc,
466 raw_svector_ostream &OS);
467 bool parseDirectiveRept(SMLoc DirectiveLoc, StringRef Directive);
468 bool parseDirectiveIrp(SMLoc DirectiveLoc); // ".irp"
469 bool parseDirectiveIrpc(SMLoc DirectiveLoc); // ".irpc"
470 bool parseDirectiveEndr(SMLoc DirectiveLoc); // ".endr"
471
472 // "_emit" or "__emit"
473 bool parseDirectiveMSEmit(SMLoc DirectiveLoc, ParseStatementInfo &Info,
474 size_t Len);
475
476 // "align"
477 bool parseDirectiveMSAlign(SMLoc DirectiveLoc, ParseStatementInfo &Info);
478
479 // "end"
480 bool parseDirectiveEnd(SMLoc DirectiveLoc);
481
482 // ".err" or ".error"
483 bool parseDirectiveError(SMLoc DirectiveLoc, bool WithMessage);
484
485 // ".warning"
486 bool parseDirectiveWarning(SMLoc DirectiveLoc);
487
488 void initializeDirectiveKindMap();
489 };
490 }
491
492 namespace llvm {
493
494 extern MCAsmParserExtension *createDarwinAsmParser();
495 extern MCAsmParserExtension *createELFAsmParser();
496 extern MCAsmParserExtension *createCOFFAsmParser();
497
498 }
499
500 enum { DEFAULT_ADDRSPACE = 0 };
501
AsmParser(SourceMgr & SM,MCContext & Ctx,MCStreamer & Out,const MCAsmInfo & MAI)502 AsmParser::AsmParser(SourceMgr &SM, MCContext &Ctx, MCStreamer &Out,
503 const MCAsmInfo &MAI)
504 : Lexer(MAI), Ctx(Ctx), Out(Out), MAI(MAI), SrcMgr(SM),
505 PlatformParser(nullptr), CurBuffer(SM.getMainFileID()),
506 MacrosEnabledFlag(true), HadError(false), CppHashLineNumber(0),
507 AssemblerDialect(~0U), IsDarwin(false), ParsingInlineAsm(false) {
508 // Save the old handler.
509 SavedDiagHandler = SrcMgr.getDiagHandler();
510 SavedDiagContext = SrcMgr.getDiagContext();
511 // Set our own handler which calls the saved handler.
512 SrcMgr.setDiagHandler(DiagHandler, this);
513 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer());
514
515 // Initialize the platform / file format parser.
516 switch (Ctx.getObjectFileInfo()->getObjectFileType()) {
517 case MCObjectFileInfo::IsCOFF:
518 PlatformParser.reset(createCOFFAsmParser());
519 break;
520 case MCObjectFileInfo::IsMachO:
521 PlatformParser.reset(createDarwinAsmParser());
522 IsDarwin = true;
523 break;
524 case MCObjectFileInfo::IsELF:
525 PlatformParser.reset(createELFAsmParser());
526 break;
527 }
528
529 PlatformParser->Initialize(*this);
530 initializeDirectiveKindMap();
531
532 NumOfMacroInstantiations = 0;
533 }
534
~AsmParser()535 AsmParser::~AsmParser() {
536 assert((HadError || ActiveMacros.empty()) &&
537 "Unexpected active macro instantiation!");
538 }
539
printMacroInstantiations()540 void AsmParser::printMacroInstantiations() {
541 // Print the active macro instantiation stack.
542 for (std::vector<MacroInstantiation *>::const_reverse_iterator
543 it = ActiveMacros.rbegin(),
544 ie = ActiveMacros.rend();
545 it != ie; ++it)
546 printMessage((*it)->InstantiationLoc, SourceMgr::DK_Note,
547 "while in macro instantiation");
548 }
549
Note(SMLoc L,const Twine & Msg,ArrayRef<SMRange> Ranges)550 void AsmParser::Note(SMLoc L, const Twine &Msg, ArrayRef<SMRange> Ranges) {
551 printMessage(L, SourceMgr::DK_Note, Msg, Ranges);
552 printMacroInstantiations();
553 }
554
Warning(SMLoc L,const Twine & Msg,ArrayRef<SMRange> Ranges)555 bool AsmParser::Warning(SMLoc L, const Twine &Msg, ArrayRef<SMRange> Ranges) {
556 if (getTargetParser().getTargetOptions().MCFatalWarnings)
557 return Error(L, Msg, Ranges);
558 printMessage(L, SourceMgr::DK_Warning, Msg, Ranges);
559 printMacroInstantiations();
560 return false;
561 }
562
Error(SMLoc L,const Twine & Msg,ArrayRef<SMRange> Ranges)563 bool AsmParser::Error(SMLoc L, const Twine &Msg, ArrayRef<SMRange> Ranges) {
564 HadError = true;
565 printMessage(L, SourceMgr::DK_Error, Msg, Ranges);
566 printMacroInstantiations();
567 return true;
568 }
569
enterIncludeFile(const std::string & Filename)570 bool AsmParser::enterIncludeFile(const std::string &Filename) {
571 std::string IncludedFile;
572 unsigned NewBuf =
573 SrcMgr.AddIncludeFile(Filename, Lexer.getLoc(), IncludedFile);
574 if (!NewBuf)
575 return true;
576
577 CurBuffer = NewBuf;
578 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer());
579 return false;
580 }
581
582 /// Process the specified .incbin file by searching for it in the include paths
583 /// then just emitting the byte contents of the file to the streamer. This
584 /// returns true on failure.
processIncbinFile(const std::string & Filename)585 bool AsmParser::processIncbinFile(const std::string &Filename) {
586 std::string IncludedFile;
587 unsigned NewBuf =
588 SrcMgr.AddIncludeFile(Filename, Lexer.getLoc(), IncludedFile);
589 if (!NewBuf)
590 return true;
591
592 // Pick up the bytes from the file and emit them.
593 getStreamer().EmitBytes(SrcMgr.getMemoryBuffer(NewBuf)->getBuffer());
594 return false;
595 }
596
jumpToLoc(SMLoc Loc,unsigned InBuffer)597 void AsmParser::jumpToLoc(SMLoc Loc, unsigned InBuffer) {
598 CurBuffer = InBuffer ? InBuffer : SrcMgr.FindBufferContainingLoc(Loc);
599 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer(),
600 Loc.getPointer());
601 }
602
Lex()603 const AsmToken &AsmParser::Lex() {
604 const AsmToken *tok = &Lexer.Lex();
605
606 if (tok->is(AsmToken::Eof)) {
607 // If this is the end of an included file, pop the parent file off the
608 // include stack.
609 SMLoc ParentIncludeLoc = SrcMgr.getParentIncludeLoc(CurBuffer);
610 if (ParentIncludeLoc != SMLoc()) {
611 jumpToLoc(ParentIncludeLoc);
612 tok = &Lexer.Lex();
613 }
614 }
615
616 if (tok->is(AsmToken::Error))
617 Error(Lexer.getErrLoc(), Lexer.getErr());
618
619 return *tok;
620 }
621
Run(bool NoInitialTextSection,bool NoFinalize)622 bool AsmParser::Run(bool NoInitialTextSection, bool NoFinalize) {
623 // Create the initial section, if requested.
624 if (!NoInitialTextSection)
625 Out.InitSections(false);
626
627 // Prime the lexer.
628 Lex();
629
630 HadError = false;
631 AsmCond StartingCondState = TheCondState;
632
633 // If we are generating dwarf for assembly source files save the initial text
634 // section and generate a .file directive.
635 if (getContext().getGenDwarfForAssembly()) {
636 MCSection *Sec = getStreamer().getCurrentSection().first;
637 if (!Sec->getBeginSymbol()) {
638 MCSymbol *SectionStartSym = getContext().createTempSymbol();
639 getStreamer().EmitLabel(SectionStartSym);
640 Sec->setBeginSymbol(SectionStartSym);
641 }
642 bool InsertResult = getContext().addGenDwarfSection(Sec);
643 assert(InsertResult && ".text section should not have debug info yet");
644 (void)InsertResult;
645 getContext().setGenDwarfFileNumber(getStreamer().EmitDwarfFileDirective(
646 0, StringRef(), getContext().getMainFileName()));
647 }
648
649 // While we have input, parse each statement.
650 while (Lexer.isNot(AsmToken::Eof)) {
651 ParseStatementInfo Info;
652 if (!parseStatement(Info, nullptr))
653 continue;
654
655 // We had an error, validate that one was emitted and recover by skipping to
656 // the next line.
657 assert(HadError && "Parse statement returned an error, but none emitted!");
658 eatToEndOfStatement();
659 }
660
661 if (TheCondState.TheCond != StartingCondState.TheCond ||
662 TheCondState.Ignore != StartingCondState.Ignore)
663 return TokError("unmatched .ifs or .elses");
664
665 // Check to see there are no empty DwarfFile slots.
666 const auto &LineTables = getContext().getMCDwarfLineTables();
667 if (!LineTables.empty()) {
668 unsigned Index = 0;
669 for (const auto &File : LineTables.begin()->second.getMCDwarfFiles()) {
670 if (File.Name.empty() && Index != 0)
671 TokError("unassigned file number: " + Twine(Index) +
672 " for .file directives");
673 ++Index;
674 }
675 }
676
677 // Check to see that all assembler local symbols were actually defined.
678 // Targets that don't do subsections via symbols may not want this, though,
679 // so conservatively exclude them. Only do this if we're finalizing, though,
680 // as otherwise we won't necessarilly have seen everything yet.
681 if (!NoFinalize && MAI.hasSubsectionsViaSymbols()) {
682 const MCContext::SymbolTable &Symbols = getContext().getSymbols();
683 for (MCContext::SymbolTable::const_iterator i = Symbols.begin(),
684 e = Symbols.end();
685 i != e; ++i) {
686 MCSymbol *Sym = i->getValue();
687 // Variable symbols may not be marked as defined, so check those
688 // explicitly. If we know it's a variable, we have a definition for
689 // the purposes of this check.
690 if (Sym->isTemporary() && !Sym->isVariable() && !Sym->isDefined())
691 // FIXME: We would really like to refer back to where the symbol was
692 // first referenced for a source location. We need to add something
693 // to track that. Currently, we just point to the end of the file.
694 printMessage(
695 getLexer().getLoc(), SourceMgr::DK_Error,
696 "assembler local symbol '" + Sym->getName() + "' not defined");
697 }
698 }
699
700 // Finalize the output stream if there are no errors and if the client wants
701 // us to.
702 if (!HadError && !NoFinalize)
703 Out.Finish();
704
705 return HadError;
706 }
707
checkForValidSection()708 void AsmParser::checkForValidSection() {
709 if (!ParsingInlineAsm && !getStreamer().getCurrentSection().first) {
710 TokError("expected section directive before assembly directive");
711 Out.InitSections(false);
712 }
713 }
714
715 /// \brief Throw away the rest of the line for testing purposes.
eatToEndOfStatement()716 void AsmParser::eatToEndOfStatement() {
717 while (Lexer.isNot(AsmToken::EndOfStatement) && Lexer.isNot(AsmToken::Eof))
718 Lex();
719
720 // Eat EOL.
721 if (Lexer.is(AsmToken::EndOfStatement))
722 Lex();
723 }
724
parseStringToEndOfStatement()725 StringRef AsmParser::parseStringToEndOfStatement() {
726 const char *Start = getTok().getLoc().getPointer();
727
728 while (Lexer.isNot(AsmToken::EndOfStatement) && Lexer.isNot(AsmToken::Eof))
729 Lex();
730
731 const char *End = getTok().getLoc().getPointer();
732 return StringRef(Start, End - Start);
733 }
734
parseStringToComma()735 StringRef AsmParser::parseStringToComma() {
736 const char *Start = getTok().getLoc().getPointer();
737
738 while (Lexer.isNot(AsmToken::EndOfStatement) &&
739 Lexer.isNot(AsmToken::Comma) && Lexer.isNot(AsmToken::Eof))
740 Lex();
741
742 const char *End = getTok().getLoc().getPointer();
743 return StringRef(Start, End - Start);
744 }
745
746 /// \brief Parse a paren expression and return it.
747 /// NOTE: This assumes the leading '(' has already been consumed.
748 ///
749 /// parenexpr ::= expr)
750 ///
parseParenExpr(const MCExpr * & Res,SMLoc & EndLoc)751 bool AsmParser::parseParenExpr(const MCExpr *&Res, SMLoc &EndLoc) {
752 if (parseExpression(Res))
753 return true;
754 if (Lexer.isNot(AsmToken::RParen))
755 return TokError("expected ')' in parentheses expression");
756 EndLoc = Lexer.getTok().getEndLoc();
757 Lex();
758 return false;
759 }
760
761 /// \brief Parse a bracket expression and return it.
762 /// NOTE: This assumes the leading '[' has already been consumed.
763 ///
764 /// bracketexpr ::= expr]
765 ///
parseBracketExpr(const MCExpr * & Res,SMLoc & EndLoc)766 bool AsmParser::parseBracketExpr(const MCExpr *&Res, SMLoc &EndLoc) {
767 if (parseExpression(Res))
768 return true;
769 if (Lexer.isNot(AsmToken::RBrac))
770 return TokError("expected ']' in brackets expression");
771 EndLoc = Lexer.getTok().getEndLoc();
772 Lex();
773 return false;
774 }
775
776 /// \brief Parse a primary expression and return it.
777 /// primaryexpr ::= (parenexpr
778 /// primaryexpr ::= symbol
779 /// primaryexpr ::= number
780 /// primaryexpr ::= '.'
781 /// primaryexpr ::= ~,+,- primaryexpr
parsePrimaryExpr(const MCExpr * & Res,SMLoc & EndLoc)782 bool AsmParser::parsePrimaryExpr(const MCExpr *&Res, SMLoc &EndLoc) {
783 SMLoc FirstTokenLoc = getLexer().getLoc();
784 AsmToken::TokenKind FirstTokenKind = Lexer.getKind();
785 switch (FirstTokenKind) {
786 default:
787 return TokError("unknown token in expression");
788 // If we have an error assume that we've already handled it.
789 case AsmToken::Error:
790 return true;
791 case AsmToken::Exclaim:
792 Lex(); // Eat the operator.
793 if (parsePrimaryExpr(Res, EndLoc))
794 return true;
795 Res = MCUnaryExpr::createLNot(Res, getContext());
796 return false;
797 case AsmToken::Dollar:
798 case AsmToken::At:
799 case AsmToken::String:
800 case AsmToken::Identifier: {
801 StringRef Identifier;
802 if (parseIdentifier(Identifier)) {
803 if (FirstTokenKind == AsmToken::Dollar) {
804 if (Lexer.getMAI().getDollarIsPC()) {
805 // This is a '$' reference, which references the current PC. Emit a
806 // temporary label to the streamer and refer to it.
807 MCSymbol *Sym = Ctx.createTempSymbol();
808 Out.EmitLabel(Sym);
809 Res = MCSymbolRefExpr::create(Sym, MCSymbolRefExpr::VK_None,
810 getContext());
811 EndLoc = FirstTokenLoc;
812 return false;
813 }
814 return Error(FirstTokenLoc, "invalid token in expression");
815 }
816 }
817 // Parse symbol variant
818 std::pair<StringRef, StringRef> Split;
819 if (!MAI.useParensForSymbolVariant()) {
820 if (FirstTokenKind == AsmToken::String) {
821 if (Lexer.is(AsmToken::At)) {
822 Lexer.Lex(); // eat @
823 SMLoc AtLoc = getLexer().getLoc();
824 StringRef VName;
825 if (parseIdentifier(VName))
826 return Error(AtLoc, "expected symbol variant after '@'");
827
828 Split = std::make_pair(Identifier, VName);
829 }
830 } else {
831 Split = Identifier.split('@');
832 }
833 } else if (Lexer.is(AsmToken::LParen)) {
834 Lexer.Lex(); // eat (
835 StringRef VName;
836 parseIdentifier(VName);
837 if (Lexer.isNot(AsmToken::RParen)) {
838 return Error(Lexer.getTok().getLoc(),
839 "unexpected token in variant, expected ')'");
840 }
841 Lexer.Lex(); // eat )
842 Split = std::make_pair(Identifier, VName);
843 }
844
845 EndLoc = SMLoc::getFromPointer(Identifier.end());
846
847 // This is a symbol reference.
848 StringRef SymbolName = Identifier;
849 MCSymbolRefExpr::VariantKind Variant = MCSymbolRefExpr::VK_None;
850
851 // Lookup the symbol variant if used.
852 if (Split.second.size()) {
853 Variant = MCSymbolRefExpr::getVariantKindForName(Split.second);
854 if (Variant != MCSymbolRefExpr::VK_Invalid) {
855 SymbolName = Split.first;
856 } else if (MAI.doesAllowAtInName() && !MAI.useParensForSymbolVariant()) {
857 Variant = MCSymbolRefExpr::VK_None;
858 } else {
859 return Error(SMLoc::getFromPointer(Split.second.begin()),
860 "invalid variant '" + Split.second + "'");
861 }
862 }
863
864 MCSymbol *Sym = getContext().getOrCreateSymbol(SymbolName);
865
866 // If this is an absolute variable reference, substitute it now to preserve
867 // semantics in the face of reassignment.
868 if (Sym->isVariable() && isa<MCConstantExpr>(Sym->getVariableValue())) {
869 if (Variant)
870 return Error(EndLoc, "unexpected modifier on variable reference");
871
872 Res = Sym->getVariableValue();
873 return false;
874 }
875
876 // Otherwise create a symbol ref.
877 Res = MCSymbolRefExpr::create(Sym, Variant, getContext());
878 return false;
879 }
880 case AsmToken::BigNum:
881 return TokError("literal value out of range for directive");
882 case AsmToken::Integer: {
883 SMLoc Loc = getTok().getLoc();
884 int64_t IntVal = getTok().getIntVal();
885 Res = MCConstantExpr::create(IntVal, getContext());
886 EndLoc = Lexer.getTok().getEndLoc();
887 Lex(); // Eat token.
888 // Look for 'b' or 'f' following an Integer as a directional label
889 if (Lexer.getKind() == AsmToken::Identifier) {
890 StringRef IDVal = getTok().getString();
891 // Lookup the symbol variant if used.
892 std::pair<StringRef, StringRef> Split = IDVal.split('@');
893 MCSymbolRefExpr::VariantKind Variant = MCSymbolRefExpr::VK_None;
894 if (Split.first.size() != IDVal.size()) {
895 Variant = MCSymbolRefExpr::getVariantKindForName(Split.second);
896 if (Variant == MCSymbolRefExpr::VK_Invalid)
897 return TokError("invalid variant '" + Split.second + "'");
898 IDVal = Split.first;
899 }
900 if (IDVal == "f" || IDVal == "b") {
901 MCSymbol *Sym =
902 Ctx.getDirectionalLocalSymbol(IntVal, IDVal == "b");
903 Res = MCSymbolRefExpr::create(Sym, Variant, getContext());
904 if (IDVal == "b" && Sym->isUndefined())
905 return Error(Loc, "invalid reference to undefined symbol");
906 EndLoc = Lexer.getTok().getEndLoc();
907 Lex(); // Eat identifier.
908 }
909 }
910 return false;
911 }
912 case AsmToken::Real: {
913 APFloat RealVal(APFloat::IEEEdouble, getTok().getString());
914 uint64_t IntVal = RealVal.bitcastToAPInt().getZExtValue();
915 Res = MCConstantExpr::create(IntVal, getContext());
916 EndLoc = Lexer.getTok().getEndLoc();
917 Lex(); // Eat token.
918 return false;
919 }
920 case AsmToken::Dot: {
921 // This is a '.' reference, which references the current PC. Emit a
922 // temporary label to the streamer and refer to it.
923 MCSymbol *Sym = Ctx.createTempSymbol();
924 Out.EmitLabel(Sym);
925 Res = MCSymbolRefExpr::create(Sym, MCSymbolRefExpr::VK_None, getContext());
926 EndLoc = Lexer.getTok().getEndLoc();
927 Lex(); // Eat identifier.
928 return false;
929 }
930 case AsmToken::LParen:
931 Lex(); // Eat the '('.
932 return parseParenExpr(Res, EndLoc);
933 case AsmToken::LBrac:
934 if (!PlatformParser->HasBracketExpressions())
935 return TokError("brackets expression not supported on this target");
936 Lex(); // Eat the '['.
937 return parseBracketExpr(Res, EndLoc);
938 case AsmToken::Minus:
939 Lex(); // Eat the operator.
940 if (parsePrimaryExpr(Res, EndLoc))
941 return true;
942 Res = MCUnaryExpr::createMinus(Res, getContext());
943 return false;
944 case AsmToken::Plus:
945 Lex(); // Eat the operator.
946 if (parsePrimaryExpr(Res, EndLoc))
947 return true;
948 Res = MCUnaryExpr::createPlus(Res, getContext());
949 return false;
950 case AsmToken::Tilde:
951 Lex(); // Eat the operator.
952 if (parsePrimaryExpr(Res, EndLoc))
953 return true;
954 Res = MCUnaryExpr::createNot(Res, getContext());
955 return false;
956 }
957 }
958
parseExpression(const MCExpr * & Res)959 bool AsmParser::parseExpression(const MCExpr *&Res) {
960 SMLoc EndLoc;
961 return parseExpression(Res, EndLoc);
962 }
963
964 const MCExpr *
applyModifierToExpr(const MCExpr * E,MCSymbolRefExpr::VariantKind Variant)965 AsmParser::applyModifierToExpr(const MCExpr *E,
966 MCSymbolRefExpr::VariantKind Variant) {
967 // Ask the target implementation about this expression first.
968 const MCExpr *NewE = getTargetParser().applyModifierToExpr(E, Variant, Ctx);
969 if (NewE)
970 return NewE;
971 // Recurse over the given expression, rebuilding it to apply the given variant
972 // if there is exactly one symbol.
973 switch (E->getKind()) {
974 case MCExpr::Target:
975 case MCExpr::Constant:
976 return nullptr;
977
978 case MCExpr::SymbolRef: {
979 const MCSymbolRefExpr *SRE = cast<MCSymbolRefExpr>(E);
980
981 if (SRE->getKind() != MCSymbolRefExpr::VK_None) {
982 TokError("invalid variant on expression '" + getTok().getIdentifier() +
983 "' (already modified)");
984 return E;
985 }
986
987 return MCSymbolRefExpr::create(&SRE->getSymbol(), Variant, getContext());
988 }
989
990 case MCExpr::Unary: {
991 const MCUnaryExpr *UE = cast<MCUnaryExpr>(E);
992 const MCExpr *Sub = applyModifierToExpr(UE->getSubExpr(), Variant);
993 if (!Sub)
994 return nullptr;
995 return MCUnaryExpr::create(UE->getOpcode(), Sub, getContext());
996 }
997
998 case MCExpr::Binary: {
999 const MCBinaryExpr *BE = cast<MCBinaryExpr>(E);
1000 const MCExpr *LHS = applyModifierToExpr(BE->getLHS(), Variant);
1001 const MCExpr *RHS = applyModifierToExpr(BE->getRHS(), Variant);
1002
1003 if (!LHS && !RHS)
1004 return nullptr;
1005
1006 if (!LHS)
1007 LHS = BE->getLHS();
1008 if (!RHS)
1009 RHS = BE->getRHS();
1010
1011 return MCBinaryExpr::create(BE->getOpcode(), LHS, RHS, getContext());
1012 }
1013 }
1014
1015 llvm_unreachable("Invalid expression kind!");
1016 }
1017
1018 /// \brief Parse an expression and return it.
1019 ///
1020 /// expr ::= expr &&,|| expr -> lowest.
1021 /// expr ::= expr |,^,&,! expr
1022 /// expr ::= expr ==,!=,<>,<,<=,>,>= expr
1023 /// expr ::= expr <<,>> expr
1024 /// expr ::= expr +,- expr
1025 /// expr ::= expr *,/,% expr -> highest.
1026 /// expr ::= primaryexpr
1027 ///
parseExpression(const MCExpr * & Res,SMLoc & EndLoc)1028 bool AsmParser::parseExpression(const MCExpr *&Res, SMLoc &EndLoc) {
1029 // Parse the expression.
1030 Res = nullptr;
1031 if (parsePrimaryExpr(Res, EndLoc) || parseBinOpRHS(1, Res, EndLoc))
1032 return true;
1033
1034 // As a special case, we support 'a op b @ modifier' by rewriting the
1035 // expression to include the modifier. This is inefficient, but in general we
1036 // expect users to use 'a@modifier op b'.
1037 if (Lexer.getKind() == AsmToken::At) {
1038 Lex();
1039
1040 if (Lexer.isNot(AsmToken::Identifier))
1041 return TokError("unexpected symbol modifier following '@'");
1042
1043 MCSymbolRefExpr::VariantKind Variant =
1044 MCSymbolRefExpr::getVariantKindForName(getTok().getIdentifier());
1045 if (Variant == MCSymbolRefExpr::VK_Invalid)
1046 return TokError("invalid variant '" + getTok().getIdentifier() + "'");
1047
1048 const MCExpr *ModifiedRes = applyModifierToExpr(Res, Variant);
1049 if (!ModifiedRes) {
1050 return TokError("invalid modifier '" + getTok().getIdentifier() +
1051 "' (no symbols present)");
1052 }
1053
1054 Res = ModifiedRes;
1055 Lex();
1056 }
1057
1058 // Try to constant fold it up front, if possible.
1059 int64_t Value;
1060 if (Res->evaluateAsAbsolute(Value))
1061 Res = MCConstantExpr::create(Value, getContext());
1062
1063 return false;
1064 }
1065
parseParenExpression(const MCExpr * & Res,SMLoc & EndLoc)1066 bool AsmParser::parseParenExpression(const MCExpr *&Res, SMLoc &EndLoc) {
1067 Res = nullptr;
1068 return parseParenExpr(Res, EndLoc) || parseBinOpRHS(1, Res, EndLoc);
1069 }
1070
parseParenExprOfDepth(unsigned ParenDepth,const MCExpr * & Res,SMLoc & EndLoc)1071 bool AsmParser::parseParenExprOfDepth(unsigned ParenDepth, const MCExpr *&Res,
1072 SMLoc &EndLoc) {
1073 if (parseParenExpr(Res, EndLoc))
1074 return true;
1075
1076 for (; ParenDepth > 0; --ParenDepth) {
1077 if (parseBinOpRHS(1, Res, EndLoc))
1078 return true;
1079
1080 // We don't Lex() the last RParen.
1081 // This is the same behavior as parseParenExpression().
1082 if (ParenDepth - 1 > 0) {
1083 if (Lexer.isNot(AsmToken::RParen))
1084 return TokError("expected ')' in parentheses expression");
1085 EndLoc = Lexer.getTok().getEndLoc();
1086 Lex();
1087 }
1088 }
1089 return false;
1090 }
1091
parseAbsoluteExpression(int64_t & Res)1092 bool AsmParser::parseAbsoluteExpression(int64_t &Res) {
1093 const MCExpr *Expr;
1094
1095 SMLoc StartLoc = Lexer.getLoc();
1096 if (parseExpression(Expr))
1097 return true;
1098
1099 if (!Expr->evaluateAsAbsolute(Res))
1100 return Error(StartLoc, "expected absolute expression");
1101
1102 return false;
1103 }
1104
getBinOpPrecedence(AsmToken::TokenKind K,MCBinaryExpr::Opcode & Kind)1105 unsigned AsmParser::getBinOpPrecedence(AsmToken::TokenKind K,
1106 MCBinaryExpr::Opcode &Kind) {
1107 switch (K) {
1108 default:
1109 return 0; // not a binop.
1110
1111 // Lowest Precedence: &&, ||
1112 case AsmToken::AmpAmp:
1113 Kind = MCBinaryExpr::LAnd;
1114 return 1;
1115 case AsmToken::PipePipe:
1116 Kind = MCBinaryExpr::LOr;
1117 return 1;
1118
1119 // Low Precedence: |, &, ^
1120 //
1121 // FIXME: gas seems to support '!' as an infix operator?
1122 case AsmToken::Pipe:
1123 Kind = MCBinaryExpr::Or;
1124 return 2;
1125 case AsmToken::Caret:
1126 Kind = MCBinaryExpr::Xor;
1127 return 2;
1128 case AsmToken::Amp:
1129 Kind = MCBinaryExpr::And;
1130 return 2;
1131
1132 // Low Intermediate Precedence: ==, !=, <>, <, <=, >, >=
1133 case AsmToken::EqualEqual:
1134 Kind = MCBinaryExpr::EQ;
1135 return 3;
1136 case AsmToken::ExclaimEqual:
1137 case AsmToken::LessGreater:
1138 Kind = MCBinaryExpr::NE;
1139 return 3;
1140 case AsmToken::Less:
1141 Kind = MCBinaryExpr::LT;
1142 return 3;
1143 case AsmToken::LessEqual:
1144 Kind = MCBinaryExpr::LTE;
1145 return 3;
1146 case AsmToken::Greater:
1147 Kind = MCBinaryExpr::GT;
1148 return 3;
1149 case AsmToken::GreaterEqual:
1150 Kind = MCBinaryExpr::GTE;
1151 return 3;
1152
1153 // Intermediate Precedence: <<, >>
1154 case AsmToken::LessLess:
1155 Kind = MCBinaryExpr::Shl;
1156 return 4;
1157 case AsmToken::GreaterGreater:
1158 Kind = MAI.shouldUseLogicalShr() ? MCBinaryExpr::LShr : MCBinaryExpr::AShr;
1159 return 4;
1160
1161 // High Intermediate Precedence: +, -
1162 case AsmToken::Plus:
1163 Kind = MCBinaryExpr::Add;
1164 return 5;
1165 case AsmToken::Minus:
1166 Kind = MCBinaryExpr::Sub;
1167 return 5;
1168
1169 // Highest Precedence: *, /, %
1170 case AsmToken::Star:
1171 Kind = MCBinaryExpr::Mul;
1172 return 6;
1173 case AsmToken::Slash:
1174 Kind = MCBinaryExpr::Div;
1175 return 6;
1176 case AsmToken::Percent:
1177 Kind = MCBinaryExpr::Mod;
1178 return 6;
1179 }
1180 }
1181
1182 /// \brief Parse all binary operators with precedence >= 'Precedence'.
1183 /// Res contains the LHS of the expression on input.
parseBinOpRHS(unsigned Precedence,const MCExpr * & Res,SMLoc & EndLoc)1184 bool AsmParser::parseBinOpRHS(unsigned Precedence, const MCExpr *&Res,
1185 SMLoc &EndLoc) {
1186 while (1) {
1187 MCBinaryExpr::Opcode Kind = MCBinaryExpr::Add;
1188 unsigned TokPrec = getBinOpPrecedence(Lexer.getKind(), Kind);
1189
1190 // If the next token is lower precedence than we are allowed to eat, return
1191 // successfully with what we ate already.
1192 if (TokPrec < Precedence)
1193 return false;
1194
1195 Lex();
1196
1197 // Eat the next primary expression.
1198 const MCExpr *RHS;
1199 if (parsePrimaryExpr(RHS, EndLoc))
1200 return true;
1201
1202 // If BinOp binds less tightly with RHS than the operator after RHS, let
1203 // the pending operator take RHS as its LHS.
1204 MCBinaryExpr::Opcode Dummy;
1205 unsigned NextTokPrec = getBinOpPrecedence(Lexer.getKind(), Dummy);
1206 if (TokPrec < NextTokPrec && parseBinOpRHS(TokPrec + 1, RHS, EndLoc))
1207 return true;
1208
1209 // Merge LHS and RHS according to operator.
1210 Res = MCBinaryExpr::create(Kind, Res, RHS, getContext());
1211 }
1212 }
1213
1214 /// ParseStatement:
1215 /// ::= EndOfStatement
1216 /// ::= Label* Directive ...Operands... EndOfStatement
1217 /// ::= Label* Identifier OperandList* EndOfStatement
parseStatement(ParseStatementInfo & Info,MCAsmParserSemaCallback * SI)1218 bool AsmParser::parseStatement(ParseStatementInfo &Info,
1219 MCAsmParserSemaCallback *SI) {
1220 if (Lexer.is(AsmToken::EndOfStatement)) {
1221 Out.AddBlankLine();
1222 Lex();
1223 return false;
1224 }
1225
1226 // Statements always start with an identifier or are a full line comment.
1227 AsmToken ID = getTok();
1228 SMLoc IDLoc = ID.getLoc();
1229 StringRef IDVal;
1230 int64_t LocalLabelVal = -1;
1231 // A full line comment is a '#' as the first token.
1232 if (Lexer.is(AsmToken::Hash))
1233 return parseCppHashLineFilenameComment(IDLoc);
1234
1235 // Allow an integer followed by a ':' as a directional local label.
1236 if (Lexer.is(AsmToken::Integer)) {
1237 LocalLabelVal = getTok().getIntVal();
1238 if (LocalLabelVal < 0) {
1239 if (!TheCondState.Ignore)
1240 return TokError("unexpected token at start of statement");
1241 IDVal = "";
1242 } else {
1243 IDVal = getTok().getString();
1244 Lex(); // Consume the integer token to be used as an identifier token.
1245 if (Lexer.getKind() != AsmToken::Colon) {
1246 if (!TheCondState.Ignore)
1247 return TokError("unexpected token at start of statement");
1248 }
1249 }
1250 } else if (Lexer.is(AsmToken::Dot)) {
1251 // Treat '.' as a valid identifier in this context.
1252 Lex();
1253 IDVal = ".";
1254 } else if (parseIdentifier(IDVal)) {
1255 if (!TheCondState.Ignore)
1256 return TokError("unexpected token at start of statement");
1257 IDVal = "";
1258 }
1259
1260 // Handle conditional assembly here before checking for skipping. We
1261 // have to do this so that .endif isn't skipped in a ".if 0" block for
1262 // example.
1263 StringMap<DirectiveKind>::const_iterator DirKindIt =
1264 DirectiveKindMap.find(IDVal);
1265 DirectiveKind DirKind = (DirKindIt == DirectiveKindMap.end())
1266 ? DK_NO_DIRECTIVE
1267 : DirKindIt->getValue();
1268 switch (DirKind) {
1269 default:
1270 break;
1271 case DK_IF:
1272 case DK_IFEQ:
1273 case DK_IFGE:
1274 case DK_IFGT:
1275 case DK_IFLE:
1276 case DK_IFLT:
1277 case DK_IFNE:
1278 return parseDirectiveIf(IDLoc, DirKind);
1279 case DK_IFB:
1280 return parseDirectiveIfb(IDLoc, true);
1281 case DK_IFNB:
1282 return parseDirectiveIfb(IDLoc, false);
1283 case DK_IFC:
1284 return parseDirectiveIfc(IDLoc, true);
1285 case DK_IFEQS:
1286 return parseDirectiveIfeqs(IDLoc, true);
1287 case DK_IFNC:
1288 return parseDirectiveIfc(IDLoc, false);
1289 case DK_IFNES:
1290 return parseDirectiveIfeqs(IDLoc, false);
1291 case DK_IFDEF:
1292 return parseDirectiveIfdef(IDLoc, true);
1293 case DK_IFNDEF:
1294 case DK_IFNOTDEF:
1295 return parseDirectiveIfdef(IDLoc, false);
1296 case DK_ELSEIF:
1297 return parseDirectiveElseIf(IDLoc);
1298 case DK_ELSE:
1299 return parseDirectiveElse(IDLoc);
1300 case DK_ENDIF:
1301 return parseDirectiveEndIf(IDLoc);
1302 }
1303
1304 // Ignore the statement if in the middle of inactive conditional
1305 // (e.g. ".if 0").
1306 if (TheCondState.Ignore) {
1307 eatToEndOfStatement();
1308 return false;
1309 }
1310
1311 // FIXME: Recurse on local labels?
1312
1313 // See what kind of statement we have.
1314 switch (Lexer.getKind()) {
1315 case AsmToken::Colon: {
1316 checkForValidSection();
1317
1318 // identifier ':' -> Label.
1319 Lex();
1320
1321 // Diagnose attempt to use '.' as a label.
1322 if (IDVal == ".")
1323 return Error(IDLoc, "invalid use of pseudo-symbol '.' as a label");
1324
1325 // Diagnose attempt to use a variable as a label.
1326 //
1327 // FIXME: Diagnostics. Note the location of the definition as a label.
1328 // FIXME: This doesn't diagnose assignment to a symbol which has been
1329 // implicitly marked as external.
1330 MCSymbol *Sym;
1331 if (LocalLabelVal == -1) {
1332 if (ParsingInlineAsm && SI) {
1333 StringRef RewrittenLabel =
1334 SI->LookupInlineAsmLabel(IDVal, getSourceManager(), IDLoc, true);
1335 assert(RewrittenLabel.size() &&
1336 "We should have an internal name here.");
1337 Info.AsmRewrites->push_back(AsmRewrite(AOK_Label, IDLoc,
1338 IDVal.size(), RewrittenLabel));
1339 IDVal = RewrittenLabel;
1340 }
1341 Sym = getContext().getOrCreateSymbol(IDVal);
1342 } else
1343 Sym = Ctx.createDirectionalLocalSymbol(LocalLabelVal);
1344
1345 Sym->redefineIfPossible();
1346
1347 if (!Sym->isUndefined() || Sym->isVariable())
1348 return Error(IDLoc, "invalid symbol redefinition");
1349
1350 // Emit the label.
1351 if (!ParsingInlineAsm)
1352 Out.EmitLabel(Sym);
1353
1354 // If we are generating dwarf for assembly source files then gather the
1355 // info to make a dwarf label entry for this label if needed.
1356 if (getContext().getGenDwarfForAssembly())
1357 MCGenDwarfLabelEntry::Make(Sym, &getStreamer(), getSourceManager(),
1358 IDLoc);
1359
1360 getTargetParser().onLabelParsed(Sym);
1361
1362 // Consume any end of statement token, if present, to avoid spurious
1363 // AddBlankLine calls().
1364 if (Lexer.is(AsmToken::EndOfStatement)) {
1365 Lex();
1366 if (Lexer.is(AsmToken::Eof))
1367 return false;
1368 }
1369
1370 return false;
1371 }
1372
1373 case AsmToken::Equal:
1374 // identifier '=' ... -> assignment statement
1375 Lex();
1376
1377 return parseAssignment(IDVal, true);
1378
1379 default: // Normal instruction or directive.
1380 break;
1381 }
1382
1383 // If macros are enabled, check to see if this is a macro instantiation.
1384 if (areMacrosEnabled())
1385 if (const MCAsmMacro *M = lookupMacro(IDVal)) {
1386 return handleMacroEntry(M, IDLoc);
1387 }
1388
1389 // Otherwise, we have a normal instruction or directive.
1390
1391 // Directives start with "."
1392 if (IDVal[0] == '.' && IDVal != ".") {
1393 // There are several entities interested in parsing directives:
1394 //
1395 // 1. The target-specific assembly parser. Some directives are target
1396 // specific or may potentially behave differently on certain targets.
1397 // 2. Asm parser extensions. For example, platform-specific parsers
1398 // (like the ELF parser) register themselves as extensions.
1399 // 3. The generic directive parser implemented by this class. These are
1400 // all the directives that behave in a target and platform independent
1401 // manner, or at least have a default behavior that's shared between
1402 // all targets and platforms.
1403
1404 // First query the target-specific parser. It will return 'true' if it
1405 // isn't interested in this directive.
1406 if (!getTargetParser().ParseDirective(ID))
1407 return false;
1408
1409 // Next, check the extension directive map to see if any extension has
1410 // registered itself to parse this directive.
1411 std::pair<MCAsmParserExtension *, DirectiveHandler> Handler =
1412 ExtensionDirectiveMap.lookup(IDVal);
1413 if (Handler.first)
1414 return (*Handler.second)(Handler.first, IDVal, IDLoc);
1415
1416 // Finally, if no one else is interested in this directive, it must be
1417 // generic and familiar to this class.
1418 switch (DirKind) {
1419 default:
1420 break;
1421 case DK_SET:
1422 case DK_EQU:
1423 return parseDirectiveSet(IDVal, true);
1424 case DK_EQUIV:
1425 return parseDirectiveSet(IDVal, false);
1426 case DK_ASCII:
1427 return parseDirectiveAscii(IDVal, false);
1428 case DK_ASCIZ:
1429 case DK_STRING:
1430 return parseDirectiveAscii(IDVal, true);
1431 case DK_BYTE:
1432 return parseDirectiveValue(1);
1433 case DK_SHORT:
1434 case DK_VALUE:
1435 case DK_2BYTE:
1436 return parseDirectiveValue(2);
1437 case DK_LONG:
1438 case DK_INT:
1439 case DK_4BYTE:
1440 return parseDirectiveValue(4);
1441 case DK_QUAD:
1442 case DK_8BYTE:
1443 return parseDirectiveValue(8);
1444 case DK_OCTA:
1445 return parseDirectiveOctaValue();
1446 case DK_SINGLE:
1447 case DK_FLOAT:
1448 return parseDirectiveRealValue(APFloat::IEEEsingle);
1449 case DK_DOUBLE:
1450 return parseDirectiveRealValue(APFloat::IEEEdouble);
1451 case DK_ALIGN: {
1452 bool IsPow2 = !getContext().getAsmInfo()->getAlignmentIsInBytes();
1453 return parseDirectiveAlign(IsPow2, /*ExprSize=*/1);
1454 }
1455 case DK_ALIGN32: {
1456 bool IsPow2 = !getContext().getAsmInfo()->getAlignmentIsInBytes();
1457 return parseDirectiveAlign(IsPow2, /*ExprSize=*/4);
1458 }
1459 case DK_BALIGN:
1460 return parseDirectiveAlign(/*IsPow2=*/false, /*ExprSize=*/1);
1461 case DK_BALIGNW:
1462 return parseDirectiveAlign(/*IsPow2=*/false, /*ExprSize=*/2);
1463 case DK_BALIGNL:
1464 return parseDirectiveAlign(/*IsPow2=*/false, /*ExprSize=*/4);
1465 case DK_P2ALIGN:
1466 return parseDirectiveAlign(/*IsPow2=*/true, /*ExprSize=*/1);
1467 case DK_P2ALIGNW:
1468 return parseDirectiveAlign(/*IsPow2=*/true, /*ExprSize=*/2);
1469 case DK_P2ALIGNL:
1470 return parseDirectiveAlign(/*IsPow2=*/true, /*ExprSize=*/4);
1471 case DK_ORG:
1472 return parseDirectiveOrg();
1473 case DK_FILL:
1474 return parseDirectiveFill();
1475 case DK_ZERO:
1476 return parseDirectiveZero();
1477 case DK_EXTERN:
1478 eatToEndOfStatement(); // .extern is the default, ignore it.
1479 return false;
1480 case DK_GLOBL:
1481 case DK_GLOBAL:
1482 return parseDirectiveSymbolAttribute(MCSA_Global);
1483 case DK_LAZY_REFERENCE:
1484 return parseDirectiveSymbolAttribute(MCSA_LazyReference);
1485 case DK_NO_DEAD_STRIP:
1486 return parseDirectiveSymbolAttribute(MCSA_NoDeadStrip);
1487 case DK_SYMBOL_RESOLVER:
1488 return parseDirectiveSymbolAttribute(MCSA_SymbolResolver);
1489 case DK_PRIVATE_EXTERN:
1490 return parseDirectiveSymbolAttribute(MCSA_PrivateExtern);
1491 case DK_REFERENCE:
1492 return parseDirectiveSymbolAttribute(MCSA_Reference);
1493 case DK_WEAK_DEFINITION:
1494 return parseDirectiveSymbolAttribute(MCSA_WeakDefinition);
1495 case DK_WEAK_REFERENCE:
1496 return parseDirectiveSymbolAttribute(MCSA_WeakReference);
1497 case DK_WEAK_DEF_CAN_BE_HIDDEN:
1498 return parseDirectiveSymbolAttribute(MCSA_WeakDefAutoPrivate);
1499 case DK_COMM:
1500 case DK_COMMON:
1501 return parseDirectiveComm(/*IsLocal=*/false);
1502 case DK_LCOMM:
1503 return parseDirectiveComm(/*IsLocal=*/true);
1504 case DK_ABORT:
1505 return parseDirectiveAbort();
1506 case DK_INCLUDE:
1507 return parseDirectiveInclude();
1508 case DK_INCBIN:
1509 return parseDirectiveIncbin();
1510 case DK_CODE16:
1511 case DK_CODE16GCC:
1512 return TokError(Twine(IDVal) + " not supported yet");
1513 case DK_REPT:
1514 return parseDirectiveRept(IDLoc, IDVal);
1515 case DK_IRP:
1516 return parseDirectiveIrp(IDLoc);
1517 case DK_IRPC:
1518 return parseDirectiveIrpc(IDLoc);
1519 case DK_ENDR:
1520 return parseDirectiveEndr(IDLoc);
1521 case DK_BUNDLE_ALIGN_MODE:
1522 return parseDirectiveBundleAlignMode();
1523 case DK_BUNDLE_LOCK:
1524 return parseDirectiveBundleLock();
1525 case DK_BUNDLE_UNLOCK:
1526 return parseDirectiveBundleUnlock();
1527 case DK_SLEB128:
1528 return parseDirectiveLEB128(true);
1529 case DK_ULEB128:
1530 return parseDirectiveLEB128(false);
1531 case DK_SPACE:
1532 case DK_SKIP:
1533 return parseDirectiveSpace(IDVal);
1534 case DK_FILE:
1535 return parseDirectiveFile(IDLoc);
1536 case DK_LINE:
1537 return parseDirectiveLine();
1538 case DK_LOC:
1539 return parseDirectiveLoc();
1540 case DK_STABS:
1541 return parseDirectiveStabs();
1542 case DK_CFI_SECTIONS:
1543 return parseDirectiveCFISections();
1544 case DK_CFI_STARTPROC:
1545 return parseDirectiveCFIStartProc();
1546 case DK_CFI_ENDPROC:
1547 return parseDirectiveCFIEndProc();
1548 case DK_CFI_DEF_CFA:
1549 return parseDirectiveCFIDefCfa(IDLoc);
1550 case DK_CFI_DEF_CFA_OFFSET:
1551 return parseDirectiveCFIDefCfaOffset();
1552 case DK_CFI_ADJUST_CFA_OFFSET:
1553 return parseDirectiveCFIAdjustCfaOffset();
1554 case DK_CFI_DEF_CFA_REGISTER:
1555 return parseDirectiveCFIDefCfaRegister(IDLoc);
1556 case DK_CFI_OFFSET:
1557 return parseDirectiveCFIOffset(IDLoc);
1558 case DK_CFI_REL_OFFSET:
1559 return parseDirectiveCFIRelOffset(IDLoc);
1560 case DK_CFI_PERSONALITY:
1561 return parseDirectiveCFIPersonalityOrLsda(true);
1562 case DK_CFI_LSDA:
1563 return parseDirectiveCFIPersonalityOrLsda(false);
1564 case DK_CFI_REMEMBER_STATE:
1565 return parseDirectiveCFIRememberState();
1566 case DK_CFI_RESTORE_STATE:
1567 return parseDirectiveCFIRestoreState();
1568 case DK_CFI_SAME_VALUE:
1569 return parseDirectiveCFISameValue(IDLoc);
1570 case DK_CFI_RESTORE:
1571 return parseDirectiveCFIRestore(IDLoc);
1572 case DK_CFI_ESCAPE:
1573 return parseDirectiveCFIEscape();
1574 case DK_CFI_SIGNAL_FRAME:
1575 return parseDirectiveCFISignalFrame();
1576 case DK_CFI_UNDEFINED:
1577 return parseDirectiveCFIUndefined(IDLoc);
1578 case DK_CFI_REGISTER:
1579 return parseDirectiveCFIRegister(IDLoc);
1580 case DK_CFI_WINDOW_SAVE:
1581 return parseDirectiveCFIWindowSave();
1582 case DK_MACROS_ON:
1583 case DK_MACROS_OFF:
1584 return parseDirectiveMacrosOnOff(IDVal);
1585 case DK_MACRO:
1586 return parseDirectiveMacro(IDLoc);
1587 case DK_EXITM:
1588 return parseDirectiveExitMacro(IDVal);
1589 case DK_ENDM:
1590 case DK_ENDMACRO:
1591 return parseDirectiveEndMacro(IDVal);
1592 case DK_PURGEM:
1593 return parseDirectivePurgeMacro(IDLoc);
1594 case DK_END:
1595 return parseDirectiveEnd(IDLoc);
1596 case DK_ERR:
1597 return parseDirectiveError(IDLoc, false);
1598 case DK_ERROR:
1599 return parseDirectiveError(IDLoc, true);
1600 case DK_WARNING:
1601 return parseDirectiveWarning(IDLoc);
1602 }
1603
1604 return Error(IDLoc, "unknown directive");
1605 }
1606
1607 // __asm _emit or __asm __emit
1608 if (ParsingInlineAsm && (IDVal == "_emit" || IDVal == "__emit" ||
1609 IDVal == "_EMIT" || IDVal == "__EMIT"))
1610 return parseDirectiveMSEmit(IDLoc, Info, IDVal.size());
1611
1612 // __asm align
1613 if (ParsingInlineAsm && (IDVal == "align" || IDVal == "ALIGN"))
1614 return parseDirectiveMSAlign(IDLoc, Info);
1615
1616 checkForValidSection();
1617
1618 // Canonicalize the opcode to lower case.
1619 std::string OpcodeStr = IDVal.lower();
1620 ParseInstructionInfo IInfo(Info.AsmRewrites);
1621 bool HadError = getTargetParser().ParseInstruction(IInfo, OpcodeStr, IDLoc,
1622 Info.ParsedOperands);
1623 Info.ParseError = HadError;
1624
1625 // Dump the parsed representation, if requested.
1626 if (getShowParsedOperands()) {
1627 SmallString<256> Str;
1628 raw_svector_ostream OS(Str);
1629 OS << "parsed instruction: [";
1630 for (unsigned i = 0; i != Info.ParsedOperands.size(); ++i) {
1631 if (i != 0)
1632 OS << ", ";
1633 Info.ParsedOperands[i]->print(OS);
1634 }
1635 OS << "]";
1636
1637 printMessage(IDLoc, SourceMgr::DK_Note, OS.str());
1638 }
1639
1640 // If we are generating dwarf for the current section then generate a .loc
1641 // directive for the instruction.
1642 if (!HadError && getContext().getGenDwarfForAssembly() &&
1643 getContext().getGenDwarfSectionSyms().count(
1644 getStreamer().getCurrentSection().first)) {
1645 unsigned Line;
1646 if (ActiveMacros.empty())
1647 Line = SrcMgr.FindLineNumber(IDLoc, CurBuffer);
1648 else
1649 Line = SrcMgr.FindLineNumber(ActiveMacros.front()->InstantiationLoc,
1650 ActiveMacros.front()->ExitBuffer);
1651
1652 // If we previously parsed a cpp hash file line comment then make sure the
1653 // current Dwarf File is for the CppHashFilename if not then emit the
1654 // Dwarf File table for it and adjust the line number for the .loc.
1655 if (CppHashFilename.size()) {
1656 unsigned FileNumber = getStreamer().EmitDwarfFileDirective(
1657 0, StringRef(), CppHashFilename);
1658 getContext().setGenDwarfFileNumber(FileNumber);
1659
1660 // Since SrcMgr.FindLineNumber() is slow and messes up the SourceMgr's
1661 // cache with the different Loc from the call above we save the last
1662 // info we queried here with SrcMgr.FindLineNumber().
1663 unsigned CppHashLocLineNo;
1664 if (LastQueryIDLoc == CppHashLoc && LastQueryBuffer == CppHashBuf)
1665 CppHashLocLineNo = LastQueryLine;
1666 else {
1667 CppHashLocLineNo = SrcMgr.FindLineNumber(CppHashLoc, CppHashBuf);
1668 LastQueryLine = CppHashLocLineNo;
1669 LastQueryIDLoc = CppHashLoc;
1670 LastQueryBuffer = CppHashBuf;
1671 }
1672 Line = CppHashLineNumber - 1 + (Line - CppHashLocLineNo);
1673 }
1674
1675 getStreamer().EmitDwarfLocDirective(
1676 getContext().getGenDwarfFileNumber(), Line, 0,
1677 DWARF2_LINE_DEFAULT_IS_STMT ? DWARF2_FLAG_IS_STMT : 0, 0, 0,
1678 StringRef());
1679 }
1680
1681 // If parsing succeeded, match the instruction.
1682 if (!HadError) {
1683 uint64_t ErrorInfo;
1684 getTargetParser().MatchAndEmitInstruction(IDLoc, Info.Opcode,
1685 Info.ParsedOperands, Out,
1686 ErrorInfo, ParsingInlineAsm);
1687 }
1688
1689 // Don't skip the rest of the line, the instruction parser is responsible for
1690 // that.
1691 return false;
1692 }
1693
1694 /// eatToEndOfLine uses the Lexer to eat the characters to the end of the line
1695 /// since they may not be able to be tokenized to get to the end of line token.
eatToEndOfLine()1696 void AsmParser::eatToEndOfLine() {
1697 if (!Lexer.is(AsmToken::EndOfStatement))
1698 Lexer.LexUntilEndOfLine();
1699 // Eat EOL.
1700 Lex();
1701 }
1702
1703 /// parseCppHashLineFilenameComment as this:
1704 /// ::= # number "filename"
1705 /// or just as a full line comment if it doesn't have a number and a string.
parseCppHashLineFilenameComment(const SMLoc & L)1706 bool AsmParser::parseCppHashLineFilenameComment(const SMLoc &L) {
1707 Lex(); // Eat the hash token.
1708
1709 if (getLexer().isNot(AsmToken::Integer)) {
1710 // Consume the line since in cases it is not a well-formed line directive,
1711 // as if were simply a full line comment.
1712 eatToEndOfLine();
1713 return false;
1714 }
1715
1716 int64_t LineNumber = getTok().getIntVal();
1717 Lex();
1718
1719 if (getLexer().isNot(AsmToken::String)) {
1720 eatToEndOfLine();
1721 return false;
1722 }
1723
1724 StringRef Filename = getTok().getString();
1725 // Get rid of the enclosing quotes.
1726 Filename = Filename.substr(1, Filename.size() - 2);
1727
1728 // Save the SMLoc, Filename and LineNumber for later use by diagnostics.
1729 CppHashLoc = L;
1730 CppHashFilename = Filename;
1731 CppHashLineNumber = LineNumber;
1732 CppHashBuf = CurBuffer;
1733
1734 // Ignore any trailing characters, they're just comment.
1735 eatToEndOfLine();
1736 return false;
1737 }
1738
1739 /// \brief will use the last parsed cpp hash line filename comment
1740 /// for the Filename and LineNo if any in the diagnostic.
DiagHandler(const SMDiagnostic & Diag,void * Context)1741 void AsmParser::DiagHandler(const SMDiagnostic &Diag, void *Context) {
1742 const AsmParser *Parser = static_cast<const AsmParser *>(Context);
1743 raw_ostream &OS = errs();
1744
1745 const SourceMgr &DiagSrcMgr = *Diag.getSourceMgr();
1746 const SMLoc &DiagLoc = Diag.getLoc();
1747 unsigned DiagBuf = DiagSrcMgr.FindBufferContainingLoc(DiagLoc);
1748 unsigned CppHashBuf =
1749 Parser->SrcMgr.FindBufferContainingLoc(Parser->CppHashLoc);
1750
1751 // Like SourceMgr::printMessage() we need to print the include stack if any
1752 // before printing the message.
1753 unsigned DiagCurBuffer = DiagSrcMgr.FindBufferContainingLoc(DiagLoc);
1754 if (!Parser->SavedDiagHandler && DiagCurBuffer &&
1755 DiagCurBuffer != DiagSrcMgr.getMainFileID()) {
1756 SMLoc ParentIncludeLoc = DiagSrcMgr.getParentIncludeLoc(DiagCurBuffer);
1757 DiagSrcMgr.PrintIncludeStack(ParentIncludeLoc, OS);
1758 }
1759
1760 // If we have not parsed a cpp hash line filename comment or the source
1761 // manager changed or buffer changed (like in a nested include) then just
1762 // print the normal diagnostic using its Filename and LineNo.
1763 if (!Parser->CppHashLineNumber || &DiagSrcMgr != &Parser->SrcMgr ||
1764 DiagBuf != CppHashBuf) {
1765 if (Parser->SavedDiagHandler)
1766 Parser->SavedDiagHandler(Diag, Parser->SavedDiagContext);
1767 else
1768 Diag.print(nullptr, OS);
1769 return;
1770 }
1771
1772 // Use the CppHashFilename and calculate a line number based on the
1773 // CppHashLoc and CppHashLineNumber relative to this Diag's SMLoc for
1774 // the diagnostic.
1775 const std::string &Filename = Parser->CppHashFilename;
1776
1777 int DiagLocLineNo = DiagSrcMgr.FindLineNumber(DiagLoc, DiagBuf);
1778 int CppHashLocLineNo =
1779 Parser->SrcMgr.FindLineNumber(Parser->CppHashLoc, CppHashBuf);
1780 int LineNo =
1781 Parser->CppHashLineNumber - 1 + (DiagLocLineNo - CppHashLocLineNo);
1782
1783 SMDiagnostic NewDiag(*Diag.getSourceMgr(), Diag.getLoc(), Filename, LineNo,
1784 Diag.getColumnNo(), Diag.getKind(), Diag.getMessage(),
1785 Diag.getLineContents(), Diag.getRanges());
1786
1787 if (Parser->SavedDiagHandler)
1788 Parser->SavedDiagHandler(NewDiag, Parser->SavedDiagContext);
1789 else
1790 NewDiag.print(nullptr, OS);
1791 }
1792
1793 // FIXME: This is mostly duplicated from the function in AsmLexer.cpp. The
1794 // difference being that that function accepts '@' as part of identifiers and
1795 // we can't do that. AsmLexer.cpp should probably be changed to handle
1796 // '@' as a special case when needed.
isIdentifierChar(char c)1797 static bool isIdentifierChar(char c) {
1798 return isalnum(static_cast<unsigned char>(c)) || c == '_' || c == '$' ||
1799 c == '.';
1800 }
1801
expandMacro(raw_svector_ostream & OS,StringRef Body,ArrayRef<MCAsmMacroParameter> Parameters,ArrayRef<MCAsmMacroArgument> A,bool EnableAtPseudoVariable,const SMLoc & L)1802 bool AsmParser::expandMacro(raw_svector_ostream &OS, StringRef Body,
1803 ArrayRef<MCAsmMacroParameter> Parameters,
1804 ArrayRef<MCAsmMacroArgument> A,
1805 bool EnableAtPseudoVariable, const SMLoc &L) {
1806 unsigned NParameters = Parameters.size();
1807 bool HasVararg = NParameters ? Parameters.back().Vararg : false;
1808 if ((!IsDarwin || NParameters != 0) && NParameters != A.size())
1809 return Error(L, "Wrong number of arguments");
1810
1811 // A macro without parameters is handled differently on Darwin:
1812 // gas accepts no arguments and does no substitutions
1813 while (!Body.empty()) {
1814 // Scan for the next substitution.
1815 std::size_t End = Body.size(), Pos = 0;
1816 for (; Pos != End; ++Pos) {
1817 // Check for a substitution or escape.
1818 if (IsDarwin && !NParameters) {
1819 // This macro has no parameters, look for $0, $1, etc.
1820 if (Body[Pos] != '$' || Pos + 1 == End)
1821 continue;
1822
1823 char Next = Body[Pos + 1];
1824 if (Next == '$' || Next == 'n' ||
1825 isdigit(static_cast<unsigned char>(Next)))
1826 break;
1827 } else {
1828 // This macro has parameters, look for \foo, \bar, etc.
1829 if (Body[Pos] == '\\' && Pos + 1 != End)
1830 break;
1831 }
1832 }
1833
1834 // Add the prefix.
1835 OS << Body.slice(0, Pos);
1836
1837 // Check if we reached the end.
1838 if (Pos == End)
1839 break;
1840
1841 if (IsDarwin && !NParameters) {
1842 switch (Body[Pos + 1]) {
1843 // $$ => $
1844 case '$':
1845 OS << '$';
1846 break;
1847
1848 // $n => number of arguments
1849 case 'n':
1850 OS << A.size();
1851 break;
1852
1853 // $[0-9] => argument
1854 default: {
1855 // Missing arguments are ignored.
1856 unsigned Index = Body[Pos + 1] - '0';
1857 if (Index >= A.size())
1858 break;
1859
1860 // Otherwise substitute with the token values, with spaces eliminated.
1861 for (MCAsmMacroArgument::const_iterator it = A[Index].begin(),
1862 ie = A[Index].end();
1863 it != ie; ++it)
1864 OS << it->getString();
1865 break;
1866 }
1867 }
1868 Pos += 2;
1869 } else {
1870 unsigned I = Pos + 1;
1871
1872 // Check for the \@ pseudo-variable.
1873 if (EnableAtPseudoVariable && Body[I] == '@' && I + 1 != End)
1874 ++I;
1875 else
1876 while (isIdentifierChar(Body[I]) && I + 1 != End)
1877 ++I;
1878
1879 const char *Begin = Body.data() + Pos + 1;
1880 StringRef Argument(Begin, I - (Pos + 1));
1881 unsigned Index = 0;
1882
1883 if (Argument == "@") {
1884 OS << NumOfMacroInstantiations;
1885 Pos += 2;
1886 } else {
1887 for (; Index < NParameters; ++Index)
1888 if (Parameters[Index].Name == Argument)
1889 break;
1890
1891 if (Index == NParameters) {
1892 if (Body[Pos + 1] == '(' && Body[Pos + 2] == ')')
1893 Pos += 3;
1894 else {
1895 OS << '\\' << Argument;
1896 Pos = I;
1897 }
1898 } else {
1899 bool VarargParameter = HasVararg && Index == (NParameters - 1);
1900 for (MCAsmMacroArgument::const_iterator it = A[Index].begin(),
1901 ie = A[Index].end();
1902 it != ie; ++it)
1903 // We expect no quotes around the string's contents when
1904 // parsing for varargs.
1905 if (it->getKind() != AsmToken::String || VarargParameter)
1906 OS << it->getString();
1907 else
1908 OS << it->getStringContents();
1909
1910 Pos += 1 + Argument.size();
1911 }
1912 }
1913 }
1914 // Update the scan point.
1915 Body = Body.substr(Pos);
1916 }
1917
1918 return false;
1919 }
1920
MacroInstantiation(SMLoc IL,int EB,SMLoc EL,size_t CondStackDepth)1921 MacroInstantiation::MacroInstantiation(SMLoc IL, int EB, SMLoc EL,
1922 size_t CondStackDepth)
1923 : InstantiationLoc(IL), ExitBuffer(EB), ExitLoc(EL),
1924 CondStackDepth(CondStackDepth) {}
1925
isOperator(AsmToken::TokenKind kind)1926 static bool isOperator(AsmToken::TokenKind kind) {
1927 switch (kind) {
1928 default:
1929 return false;
1930 case AsmToken::Plus:
1931 case AsmToken::Minus:
1932 case AsmToken::Tilde:
1933 case AsmToken::Slash:
1934 case AsmToken::Star:
1935 case AsmToken::Dot:
1936 case AsmToken::Equal:
1937 case AsmToken::EqualEqual:
1938 case AsmToken::Pipe:
1939 case AsmToken::PipePipe:
1940 case AsmToken::Caret:
1941 case AsmToken::Amp:
1942 case AsmToken::AmpAmp:
1943 case AsmToken::Exclaim:
1944 case AsmToken::ExclaimEqual:
1945 case AsmToken::Percent:
1946 case AsmToken::Less:
1947 case AsmToken::LessEqual:
1948 case AsmToken::LessLess:
1949 case AsmToken::LessGreater:
1950 case AsmToken::Greater:
1951 case AsmToken::GreaterEqual:
1952 case AsmToken::GreaterGreater:
1953 return true;
1954 }
1955 }
1956
1957 namespace {
1958 class AsmLexerSkipSpaceRAII {
1959 public:
AsmLexerSkipSpaceRAII(AsmLexer & Lexer,bool SkipSpace)1960 AsmLexerSkipSpaceRAII(AsmLexer &Lexer, bool SkipSpace) : Lexer(Lexer) {
1961 Lexer.setSkipSpace(SkipSpace);
1962 }
1963
~AsmLexerSkipSpaceRAII()1964 ~AsmLexerSkipSpaceRAII() {
1965 Lexer.setSkipSpace(true);
1966 }
1967
1968 private:
1969 AsmLexer &Lexer;
1970 };
1971 }
1972
parseMacroArgument(MCAsmMacroArgument & MA,bool Vararg)1973 bool AsmParser::parseMacroArgument(MCAsmMacroArgument &MA, bool Vararg) {
1974
1975 if (Vararg) {
1976 if (Lexer.isNot(AsmToken::EndOfStatement)) {
1977 StringRef Str = parseStringToEndOfStatement();
1978 MA.emplace_back(AsmToken::String, Str);
1979 }
1980 return false;
1981 }
1982
1983 unsigned ParenLevel = 0;
1984 unsigned AddTokens = 0;
1985
1986 // Darwin doesn't use spaces to delmit arguments.
1987 AsmLexerSkipSpaceRAII ScopedSkipSpace(Lexer, IsDarwin);
1988
1989 for (;;) {
1990 if (Lexer.is(AsmToken::Eof) || Lexer.is(AsmToken::Equal))
1991 return TokError("unexpected token in macro instantiation");
1992
1993 if (ParenLevel == 0 && Lexer.is(AsmToken::Comma))
1994 break;
1995
1996 if (Lexer.is(AsmToken::Space)) {
1997 Lex(); // Eat spaces
1998
1999 // Spaces can delimit parameters, but could also be part an expression.
2000 // If the token after a space is an operator, add the token and the next
2001 // one into this argument
2002 if (!IsDarwin) {
2003 if (isOperator(Lexer.getKind())) {
2004 // Check to see whether the token is used as an operator,
2005 // or part of an identifier
2006 const char *NextChar = getTok().getEndLoc().getPointer();
2007 if (*NextChar == ' ')
2008 AddTokens = 2;
2009 }
2010
2011 if (!AddTokens && ParenLevel == 0) {
2012 break;
2013 }
2014 }
2015 }
2016
2017 // handleMacroEntry relies on not advancing the lexer here
2018 // to be able to fill in the remaining default parameter values
2019 if (Lexer.is(AsmToken::EndOfStatement))
2020 break;
2021
2022 // Adjust the current parentheses level.
2023 if (Lexer.is(AsmToken::LParen))
2024 ++ParenLevel;
2025 else if (Lexer.is(AsmToken::RParen) && ParenLevel)
2026 --ParenLevel;
2027
2028 // Append the token to the current argument list.
2029 MA.push_back(getTok());
2030 if (AddTokens)
2031 AddTokens--;
2032 Lex();
2033 }
2034
2035 if (ParenLevel != 0)
2036 return TokError("unbalanced parentheses in macro argument");
2037 return false;
2038 }
2039
2040 // Parse the macro instantiation arguments.
parseMacroArguments(const MCAsmMacro * M,MCAsmMacroArguments & A)2041 bool AsmParser::parseMacroArguments(const MCAsmMacro *M,
2042 MCAsmMacroArguments &A) {
2043 const unsigned NParameters = M ? M->Parameters.size() : 0;
2044 bool NamedParametersFound = false;
2045 SmallVector<SMLoc, 4> FALocs;
2046
2047 A.resize(NParameters);
2048 FALocs.resize(NParameters);
2049
2050 // Parse two kinds of macro invocations:
2051 // - macros defined without any parameters accept an arbitrary number of them
2052 // - macros defined with parameters accept at most that many of them
2053 bool HasVararg = NParameters ? M->Parameters.back().Vararg : false;
2054 for (unsigned Parameter = 0; !NParameters || Parameter < NParameters;
2055 ++Parameter) {
2056 SMLoc IDLoc = Lexer.getLoc();
2057 MCAsmMacroParameter FA;
2058
2059 if (Lexer.is(AsmToken::Identifier) && Lexer.peekTok().is(AsmToken::Equal)) {
2060 if (parseIdentifier(FA.Name)) {
2061 Error(IDLoc, "invalid argument identifier for formal argument");
2062 eatToEndOfStatement();
2063 return true;
2064 }
2065
2066 if (!Lexer.is(AsmToken::Equal)) {
2067 TokError("expected '=' after formal parameter identifier");
2068 eatToEndOfStatement();
2069 return true;
2070 }
2071 Lex();
2072
2073 NamedParametersFound = true;
2074 }
2075
2076 if (NamedParametersFound && FA.Name.empty()) {
2077 Error(IDLoc, "cannot mix positional and keyword arguments");
2078 eatToEndOfStatement();
2079 return true;
2080 }
2081
2082 bool Vararg = HasVararg && Parameter == (NParameters - 1);
2083 if (parseMacroArgument(FA.Value, Vararg))
2084 return true;
2085
2086 unsigned PI = Parameter;
2087 if (!FA.Name.empty()) {
2088 unsigned FAI = 0;
2089 for (FAI = 0; FAI < NParameters; ++FAI)
2090 if (M->Parameters[FAI].Name == FA.Name)
2091 break;
2092
2093 if (FAI >= NParameters) {
2094 assert(M && "expected macro to be defined");
2095 Error(IDLoc,
2096 "parameter named '" + FA.Name + "' does not exist for macro '" +
2097 M->Name + "'");
2098 return true;
2099 }
2100 PI = FAI;
2101 }
2102
2103 if (!FA.Value.empty()) {
2104 if (A.size() <= PI)
2105 A.resize(PI + 1);
2106 A[PI] = FA.Value;
2107
2108 if (FALocs.size() <= PI)
2109 FALocs.resize(PI + 1);
2110
2111 FALocs[PI] = Lexer.getLoc();
2112 }
2113
2114 // At the end of the statement, fill in remaining arguments that have
2115 // default values. If there aren't any, then the next argument is
2116 // required but missing
2117 if (Lexer.is(AsmToken::EndOfStatement)) {
2118 bool Failure = false;
2119 for (unsigned FAI = 0; FAI < NParameters; ++FAI) {
2120 if (A[FAI].empty()) {
2121 if (M->Parameters[FAI].Required) {
2122 Error(FALocs[FAI].isValid() ? FALocs[FAI] : Lexer.getLoc(),
2123 "missing value for required parameter "
2124 "'" + M->Parameters[FAI].Name + "' in macro '" + M->Name + "'");
2125 Failure = true;
2126 }
2127
2128 if (!M->Parameters[FAI].Value.empty())
2129 A[FAI] = M->Parameters[FAI].Value;
2130 }
2131 }
2132 return Failure;
2133 }
2134
2135 if (Lexer.is(AsmToken::Comma))
2136 Lex();
2137 }
2138
2139 return TokError("too many positional arguments");
2140 }
2141
lookupMacro(StringRef Name)2142 const MCAsmMacro *AsmParser::lookupMacro(StringRef Name) {
2143 StringMap<MCAsmMacro>::iterator I = MacroMap.find(Name);
2144 return (I == MacroMap.end()) ? nullptr : &I->getValue();
2145 }
2146
defineMacro(StringRef Name,MCAsmMacro Macro)2147 void AsmParser::defineMacro(StringRef Name, MCAsmMacro Macro) {
2148 MacroMap.insert(std::make_pair(Name, std::move(Macro)));
2149 }
2150
undefineMacro(StringRef Name)2151 void AsmParser::undefineMacro(StringRef Name) { MacroMap.erase(Name); }
2152
handleMacroEntry(const MCAsmMacro * M,SMLoc NameLoc)2153 bool AsmParser::handleMacroEntry(const MCAsmMacro *M, SMLoc NameLoc) {
2154 // Arbitrarily limit macro nesting depth, to match 'as'. We can eliminate
2155 // this, although we should protect against infinite loops.
2156 if (ActiveMacros.size() == 20)
2157 return TokError("macros cannot be nested more than 20 levels deep");
2158
2159 MCAsmMacroArguments A;
2160 if (parseMacroArguments(M, A))
2161 return true;
2162
2163 // Macro instantiation is lexical, unfortunately. We construct a new buffer
2164 // to hold the macro body with substitutions.
2165 SmallString<256> Buf;
2166 StringRef Body = M->Body;
2167 raw_svector_ostream OS(Buf);
2168
2169 if (expandMacro(OS, Body, M->Parameters, A, true, getTok().getLoc()))
2170 return true;
2171
2172 // We include the .endmacro in the buffer as our cue to exit the macro
2173 // instantiation.
2174 OS << ".endmacro\n";
2175
2176 std::unique_ptr<MemoryBuffer> Instantiation =
2177 MemoryBuffer::getMemBufferCopy(OS.str(), "<instantiation>");
2178
2179 // Create the macro instantiation object and add to the current macro
2180 // instantiation stack.
2181 MacroInstantiation *MI = new MacroInstantiation(
2182 NameLoc, CurBuffer, getTok().getLoc(), TheCondStack.size());
2183 ActiveMacros.push_back(MI);
2184
2185 ++NumOfMacroInstantiations;
2186
2187 // Jump to the macro instantiation and prime the lexer.
2188 CurBuffer = SrcMgr.AddNewSourceBuffer(std::move(Instantiation), SMLoc());
2189 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer());
2190 Lex();
2191
2192 return false;
2193 }
2194
handleMacroExit()2195 void AsmParser::handleMacroExit() {
2196 // Jump to the EndOfStatement we should return to, and consume it.
2197 jumpToLoc(ActiveMacros.back()->ExitLoc, ActiveMacros.back()->ExitBuffer);
2198 Lex();
2199
2200 // Pop the instantiation entry.
2201 delete ActiveMacros.back();
2202 ActiveMacros.pop_back();
2203 }
2204
parseAssignment(StringRef Name,bool allow_redef,bool NoDeadStrip)2205 bool AsmParser::parseAssignment(StringRef Name, bool allow_redef,
2206 bool NoDeadStrip) {
2207 MCSymbol *Sym;
2208 const MCExpr *Value;
2209 if (MCParserUtils::parseAssignmentExpression(Name, allow_redef, *this, Sym,
2210 Value))
2211 return true;
2212
2213 if (!Sym) {
2214 // In the case where we parse an expression starting with a '.', we will
2215 // not generate an error, nor will we create a symbol. In this case we
2216 // should just return out.
2217 return false;
2218 }
2219
2220 // Do the assignment.
2221 Out.EmitAssignment(Sym, Value);
2222 if (NoDeadStrip)
2223 Out.EmitSymbolAttribute(Sym, MCSA_NoDeadStrip);
2224
2225 return false;
2226 }
2227
2228 /// parseIdentifier:
2229 /// ::= identifier
2230 /// ::= string
parseIdentifier(StringRef & Res)2231 bool AsmParser::parseIdentifier(StringRef &Res) {
2232 // The assembler has relaxed rules for accepting identifiers, in particular we
2233 // allow things like '.globl $foo' and '.def @feat.00', which would normally be
2234 // separate tokens. At this level, we have already lexed so we cannot (currently)
2235 // handle this as a context dependent token, instead we detect adjacent tokens
2236 // and return the combined identifier.
2237 if (Lexer.is(AsmToken::Dollar) || Lexer.is(AsmToken::At)) {
2238 SMLoc PrefixLoc = getLexer().getLoc();
2239
2240 // Consume the prefix character, and check for a following identifier.
2241 Lex();
2242 if (Lexer.isNot(AsmToken::Identifier))
2243 return true;
2244
2245 // We have a '$' or '@' followed by an identifier, make sure they are adjacent.
2246 if (PrefixLoc.getPointer() + 1 != getTok().getLoc().getPointer())
2247 return true;
2248
2249 // Construct the joined identifier and consume the token.
2250 Res =
2251 StringRef(PrefixLoc.getPointer(), getTok().getIdentifier().size() + 1);
2252 Lex();
2253 return false;
2254 }
2255
2256 if (Lexer.isNot(AsmToken::Identifier) && Lexer.isNot(AsmToken::String))
2257 return true;
2258
2259 Res = getTok().getIdentifier();
2260
2261 Lex(); // Consume the identifier token.
2262
2263 return false;
2264 }
2265
2266 /// parseDirectiveSet:
2267 /// ::= .equ identifier ',' expression
2268 /// ::= .equiv identifier ',' expression
2269 /// ::= .set identifier ',' expression
parseDirectiveSet(StringRef IDVal,bool allow_redef)2270 bool AsmParser::parseDirectiveSet(StringRef IDVal, bool allow_redef) {
2271 StringRef Name;
2272
2273 if (parseIdentifier(Name))
2274 return TokError("expected identifier after '" + Twine(IDVal) + "'");
2275
2276 if (getLexer().isNot(AsmToken::Comma))
2277 return TokError("unexpected token in '" + Twine(IDVal) + "'");
2278 Lex();
2279
2280 return parseAssignment(Name, allow_redef, true);
2281 }
2282
parseEscapedString(std::string & Data)2283 bool AsmParser::parseEscapedString(std::string &Data) {
2284 assert(getLexer().is(AsmToken::String) && "Unexpected current token!");
2285
2286 Data = "";
2287 StringRef Str = getTok().getStringContents();
2288 for (unsigned i = 0, e = Str.size(); i != e; ++i) {
2289 if (Str[i] != '\\') {
2290 Data += Str[i];
2291 continue;
2292 }
2293
2294 // Recognize escaped characters. Note that this escape semantics currently
2295 // loosely follows Darwin 'as'. Notably, it doesn't support hex escapes.
2296 ++i;
2297 if (i == e)
2298 return TokError("unexpected backslash at end of string");
2299
2300 // Recognize octal sequences.
2301 if ((unsigned)(Str[i] - '0') <= 7) {
2302 // Consume up to three octal characters.
2303 unsigned Value = Str[i] - '0';
2304
2305 if (i + 1 != e && ((unsigned)(Str[i + 1] - '0')) <= 7) {
2306 ++i;
2307 Value = Value * 8 + (Str[i] - '0');
2308
2309 if (i + 1 != e && ((unsigned)(Str[i + 1] - '0')) <= 7) {
2310 ++i;
2311 Value = Value * 8 + (Str[i] - '0');
2312 }
2313 }
2314
2315 if (Value > 255)
2316 return TokError("invalid octal escape sequence (out of range)");
2317
2318 Data += (unsigned char)Value;
2319 continue;
2320 }
2321
2322 // Otherwise recognize individual escapes.
2323 switch (Str[i]) {
2324 default:
2325 // Just reject invalid escape sequences for now.
2326 return TokError("invalid escape sequence (unrecognized character)");
2327
2328 case 'b': Data += '\b'; break;
2329 case 'f': Data += '\f'; break;
2330 case 'n': Data += '\n'; break;
2331 case 'r': Data += '\r'; break;
2332 case 't': Data += '\t'; break;
2333 case '"': Data += '"'; break;
2334 case '\\': Data += '\\'; break;
2335 }
2336 }
2337
2338 return false;
2339 }
2340
2341 /// parseDirectiveAscii:
2342 /// ::= ( .ascii | .asciz | .string ) [ "string" ( , "string" )* ]
parseDirectiveAscii(StringRef IDVal,bool ZeroTerminated)2343 bool AsmParser::parseDirectiveAscii(StringRef IDVal, bool ZeroTerminated) {
2344 if (getLexer().isNot(AsmToken::EndOfStatement)) {
2345 checkForValidSection();
2346
2347 for (;;) {
2348 if (getLexer().isNot(AsmToken::String))
2349 return TokError("expected string in '" + Twine(IDVal) + "' directive");
2350
2351 std::string Data;
2352 if (parseEscapedString(Data))
2353 return true;
2354
2355 getStreamer().EmitBytes(Data);
2356 if (ZeroTerminated)
2357 getStreamer().EmitBytes(StringRef("\0", 1));
2358
2359 Lex();
2360
2361 if (getLexer().is(AsmToken::EndOfStatement))
2362 break;
2363
2364 if (getLexer().isNot(AsmToken::Comma))
2365 return TokError("unexpected token in '" + Twine(IDVal) + "' directive");
2366 Lex();
2367 }
2368 }
2369
2370 Lex();
2371 return false;
2372 }
2373
2374 /// parseDirectiveValue
2375 /// ::= (.byte | .short | ... ) [ expression (, expression)* ]
parseDirectiveValue(unsigned Size)2376 bool AsmParser::parseDirectiveValue(unsigned Size) {
2377 if (getLexer().isNot(AsmToken::EndOfStatement)) {
2378 checkForValidSection();
2379
2380 for (;;) {
2381 const MCExpr *Value;
2382 SMLoc ExprLoc = getLexer().getLoc();
2383 if (parseExpression(Value))
2384 return true;
2385
2386 // Special case constant expressions to match code generator.
2387 if (const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value)) {
2388 assert(Size <= 8 && "Invalid size");
2389 uint64_t IntValue = MCE->getValue();
2390 if (!isUIntN(8 * Size, IntValue) && !isIntN(8 * Size, IntValue))
2391 return Error(ExprLoc, "literal value out of range for directive");
2392 getStreamer().EmitIntValue(IntValue, Size);
2393 } else
2394 getStreamer().EmitValue(Value, Size, ExprLoc);
2395
2396 if (getLexer().is(AsmToken::EndOfStatement))
2397 break;
2398
2399 // FIXME: Improve diagnostic.
2400 if (getLexer().isNot(AsmToken::Comma))
2401 return TokError("unexpected token in directive");
2402 Lex();
2403 }
2404 }
2405
2406 Lex();
2407 return false;
2408 }
2409
2410 /// ParseDirectiveOctaValue
2411 /// ::= .octa [ hexconstant (, hexconstant)* ]
parseDirectiveOctaValue()2412 bool AsmParser::parseDirectiveOctaValue() {
2413 if (getLexer().isNot(AsmToken::EndOfStatement)) {
2414 checkForValidSection();
2415
2416 for (;;) {
2417 if (Lexer.getKind() == AsmToken::Error)
2418 return true;
2419 if (Lexer.getKind() != AsmToken::Integer &&
2420 Lexer.getKind() != AsmToken::BigNum)
2421 return TokError("unknown token in expression");
2422
2423 SMLoc ExprLoc = getLexer().getLoc();
2424 APInt IntValue = getTok().getAPIntVal();
2425 Lex();
2426
2427 uint64_t hi, lo;
2428 if (IntValue.isIntN(64)) {
2429 hi = 0;
2430 lo = IntValue.getZExtValue();
2431 } else if (IntValue.isIntN(128)) {
2432 // It might actually have more than 128 bits, but the top ones are zero.
2433 hi = IntValue.getHiBits(IntValue.getBitWidth() - 64).getZExtValue();
2434 lo = IntValue.getLoBits(64).getZExtValue();
2435 } else
2436 return Error(ExprLoc, "literal value out of range for directive");
2437
2438 if (MAI.isLittleEndian()) {
2439 getStreamer().EmitIntValue(lo, 8);
2440 getStreamer().EmitIntValue(hi, 8);
2441 } else {
2442 getStreamer().EmitIntValue(hi, 8);
2443 getStreamer().EmitIntValue(lo, 8);
2444 }
2445
2446 if (getLexer().is(AsmToken::EndOfStatement))
2447 break;
2448
2449 // FIXME: Improve diagnostic.
2450 if (getLexer().isNot(AsmToken::Comma))
2451 return TokError("unexpected token in directive");
2452 Lex();
2453 }
2454 }
2455
2456 Lex();
2457 return false;
2458 }
2459
2460 /// parseDirectiveRealValue
2461 /// ::= (.single | .double) [ expression (, expression)* ]
parseDirectiveRealValue(const fltSemantics & Semantics)2462 bool AsmParser::parseDirectiveRealValue(const fltSemantics &Semantics) {
2463 if (getLexer().isNot(AsmToken::EndOfStatement)) {
2464 checkForValidSection();
2465
2466 for (;;) {
2467 // We don't truly support arithmetic on floating point expressions, so we
2468 // have to manually parse unary prefixes.
2469 bool IsNeg = false;
2470 if (getLexer().is(AsmToken::Minus)) {
2471 Lex();
2472 IsNeg = true;
2473 } else if (getLexer().is(AsmToken::Plus))
2474 Lex();
2475
2476 if (getLexer().isNot(AsmToken::Integer) &&
2477 getLexer().isNot(AsmToken::Real) &&
2478 getLexer().isNot(AsmToken::Identifier))
2479 return TokError("unexpected token in directive");
2480
2481 // Convert to an APFloat.
2482 APFloat Value(Semantics);
2483 StringRef IDVal = getTok().getString();
2484 if (getLexer().is(AsmToken::Identifier)) {
2485 if (!IDVal.compare_lower("infinity") || !IDVal.compare_lower("inf"))
2486 Value = APFloat::getInf(Semantics);
2487 else if (!IDVal.compare_lower("nan"))
2488 Value = APFloat::getNaN(Semantics, false, ~0);
2489 else
2490 return TokError("invalid floating point literal");
2491 } else if (Value.convertFromString(IDVal, APFloat::rmNearestTiesToEven) ==
2492 APFloat::opInvalidOp)
2493 return TokError("invalid floating point literal");
2494 if (IsNeg)
2495 Value.changeSign();
2496
2497 // Consume the numeric token.
2498 Lex();
2499
2500 // Emit the value as an integer.
2501 APInt AsInt = Value.bitcastToAPInt();
2502 getStreamer().EmitIntValue(AsInt.getLimitedValue(),
2503 AsInt.getBitWidth() / 8);
2504
2505 if (getLexer().is(AsmToken::EndOfStatement))
2506 break;
2507
2508 if (getLexer().isNot(AsmToken::Comma))
2509 return TokError("unexpected token in directive");
2510 Lex();
2511 }
2512 }
2513
2514 Lex();
2515 return false;
2516 }
2517
2518 /// parseDirectiveZero
2519 /// ::= .zero expression
parseDirectiveZero()2520 bool AsmParser::parseDirectiveZero() {
2521 checkForValidSection();
2522
2523 int64_t NumBytes;
2524 if (parseAbsoluteExpression(NumBytes))
2525 return true;
2526
2527 int64_t Val = 0;
2528 if (getLexer().is(AsmToken::Comma)) {
2529 Lex();
2530 if (parseAbsoluteExpression(Val))
2531 return true;
2532 }
2533
2534 if (getLexer().isNot(AsmToken::EndOfStatement))
2535 return TokError("unexpected token in '.zero' directive");
2536
2537 Lex();
2538
2539 getStreamer().EmitFill(NumBytes, Val);
2540
2541 return false;
2542 }
2543
2544 /// parseDirectiveFill
2545 /// ::= .fill expression [ , expression [ , expression ] ]
parseDirectiveFill()2546 bool AsmParser::parseDirectiveFill() {
2547 checkForValidSection();
2548
2549 SMLoc RepeatLoc = getLexer().getLoc();
2550 int64_t NumValues;
2551 if (parseAbsoluteExpression(NumValues))
2552 return true;
2553
2554 if (NumValues < 0) {
2555 Warning(RepeatLoc,
2556 "'.fill' directive with negative repeat count has no effect");
2557 NumValues = 0;
2558 }
2559
2560 int64_t FillSize = 1;
2561 int64_t FillExpr = 0;
2562
2563 SMLoc SizeLoc, ExprLoc;
2564 if (getLexer().isNot(AsmToken::EndOfStatement)) {
2565 if (getLexer().isNot(AsmToken::Comma))
2566 return TokError("unexpected token in '.fill' directive");
2567 Lex();
2568
2569 SizeLoc = getLexer().getLoc();
2570 if (parseAbsoluteExpression(FillSize))
2571 return true;
2572
2573 if (getLexer().isNot(AsmToken::EndOfStatement)) {
2574 if (getLexer().isNot(AsmToken::Comma))
2575 return TokError("unexpected token in '.fill' directive");
2576 Lex();
2577
2578 ExprLoc = getLexer().getLoc();
2579 if (parseAbsoluteExpression(FillExpr))
2580 return true;
2581
2582 if (getLexer().isNot(AsmToken::EndOfStatement))
2583 return TokError("unexpected token in '.fill' directive");
2584
2585 Lex();
2586 }
2587 }
2588
2589 if (FillSize < 0) {
2590 Warning(SizeLoc, "'.fill' directive with negative size has no effect");
2591 NumValues = 0;
2592 }
2593 if (FillSize > 8) {
2594 Warning(SizeLoc, "'.fill' directive with size greater than 8 has been truncated to 8");
2595 FillSize = 8;
2596 }
2597
2598 if (!isUInt<32>(FillExpr) && FillSize > 4)
2599 Warning(ExprLoc, "'.fill' directive pattern has been truncated to 32-bits");
2600
2601 if (NumValues > 0) {
2602 int64_t NonZeroFillSize = FillSize > 4 ? 4 : FillSize;
2603 FillExpr &= ~0ULL >> (64 - NonZeroFillSize * 8);
2604 for (uint64_t i = 0, e = NumValues; i != e; ++i) {
2605 getStreamer().EmitIntValue(FillExpr, NonZeroFillSize);
2606 if (NonZeroFillSize < FillSize)
2607 getStreamer().EmitIntValue(0, FillSize - NonZeroFillSize);
2608 }
2609 }
2610
2611 return false;
2612 }
2613
2614 /// parseDirectiveOrg
2615 /// ::= .org expression [ , expression ]
parseDirectiveOrg()2616 bool AsmParser::parseDirectiveOrg() {
2617 checkForValidSection();
2618
2619 const MCExpr *Offset;
2620 SMLoc Loc = getTok().getLoc();
2621 if (parseExpression(Offset))
2622 return true;
2623
2624 // Parse optional fill expression.
2625 int64_t FillExpr = 0;
2626 if (getLexer().isNot(AsmToken::EndOfStatement)) {
2627 if (getLexer().isNot(AsmToken::Comma))
2628 return TokError("unexpected token in '.org' directive");
2629 Lex();
2630
2631 if (parseAbsoluteExpression(FillExpr))
2632 return true;
2633
2634 if (getLexer().isNot(AsmToken::EndOfStatement))
2635 return TokError("unexpected token in '.org' directive");
2636 }
2637
2638 Lex();
2639
2640 // Only limited forms of relocatable expressions are accepted here, it
2641 // has to be relative to the current section. The streamer will return
2642 // 'true' if the expression wasn't evaluatable.
2643 if (getStreamer().EmitValueToOffset(Offset, FillExpr))
2644 return Error(Loc, "expected assembly-time absolute expression");
2645
2646 return false;
2647 }
2648
2649 /// parseDirectiveAlign
2650 /// ::= {.align, ...} expression [ , expression [ , expression ]]
parseDirectiveAlign(bool IsPow2,unsigned ValueSize)2651 bool AsmParser::parseDirectiveAlign(bool IsPow2, unsigned ValueSize) {
2652 checkForValidSection();
2653
2654 SMLoc AlignmentLoc = getLexer().getLoc();
2655 int64_t Alignment;
2656 if (parseAbsoluteExpression(Alignment))
2657 return true;
2658
2659 SMLoc MaxBytesLoc;
2660 bool HasFillExpr = false;
2661 int64_t FillExpr = 0;
2662 int64_t MaxBytesToFill = 0;
2663 if (getLexer().isNot(AsmToken::EndOfStatement)) {
2664 if (getLexer().isNot(AsmToken::Comma))
2665 return TokError("unexpected token in directive");
2666 Lex();
2667
2668 // The fill expression can be omitted while specifying a maximum number of
2669 // alignment bytes, e.g:
2670 // .align 3,,4
2671 if (getLexer().isNot(AsmToken::Comma)) {
2672 HasFillExpr = true;
2673 if (parseAbsoluteExpression(FillExpr))
2674 return true;
2675 }
2676
2677 if (getLexer().isNot(AsmToken::EndOfStatement)) {
2678 if (getLexer().isNot(AsmToken::Comma))
2679 return TokError("unexpected token in directive");
2680 Lex();
2681
2682 MaxBytesLoc = getLexer().getLoc();
2683 if (parseAbsoluteExpression(MaxBytesToFill))
2684 return true;
2685
2686 if (getLexer().isNot(AsmToken::EndOfStatement))
2687 return TokError("unexpected token in directive");
2688 }
2689 }
2690
2691 Lex();
2692
2693 if (!HasFillExpr)
2694 FillExpr = 0;
2695
2696 // Compute alignment in bytes.
2697 if (IsPow2) {
2698 // FIXME: Diagnose overflow.
2699 if (Alignment >= 32) {
2700 Error(AlignmentLoc, "invalid alignment value");
2701 Alignment = 31;
2702 }
2703
2704 Alignment = 1ULL << Alignment;
2705 } else {
2706 // Reject alignments that aren't a power of two, for gas compatibility.
2707 if (!isPowerOf2_64(Alignment))
2708 Error(AlignmentLoc, "alignment must be a power of 2");
2709 }
2710
2711 // Diagnose non-sensical max bytes to align.
2712 if (MaxBytesLoc.isValid()) {
2713 if (MaxBytesToFill < 1) {
2714 Error(MaxBytesLoc, "alignment directive can never be satisfied in this "
2715 "many bytes, ignoring maximum bytes expression");
2716 MaxBytesToFill = 0;
2717 }
2718
2719 if (MaxBytesToFill >= Alignment) {
2720 Warning(MaxBytesLoc, "maximum bytes expression exceeds alignment and "
2721 "has no effect");
2722 MaxBytesToFill = 0;
2723 }
2724 }
2725
2726 // Check whether we should use optimal code alignment for this .align
2727 // directive.
2728 const MCSection *Section = getStreamer().getCurrentSection().first;
2729 assert(Section && "must have section to emit alignment");
2730 bool UseCodeAlign = Section->UseCodeAlign();
2731 if ((!HasFillExpr || Lexer.getMAI().getTextAlignFillValue() == FillExpr) &&
2732 ValueSize == 1 && UseCodeAlign) {
2733 getStreamer().EmitCodeAlignment(Alignment, MaxBytesToFill);
2734 } else {
2735 // FIXME: Target specific behavior about how the "extra" bytes are filled.
2736 getStreamer().EmitValueToAlignment(Alignment, FillExpr, ValueSize,
2737 MaxBytesToFill);
2738 }
2739
2740 return false;
2741 }
2742
2743 /// parseDirectiveFile
2744 /// ::= .file [number] filename
2745 /// ::= .file number directory filename
parseDirectiveFile(SMLoc DirectiveLoc)2746 bool AsmParser::parseDirectiveFile(SMLoc DirectiveLoc) {
2747 // FIXME: I'm not sure what this is.
2748 int64_t FileNumber = -1;
2749 SMLoc FileNumberLoc = getLexer().getLoc();
2750 if (getLexer().is(AsmToken::Integer)) {
2751 FileNumber = getTok().getIntVal();
2752 Lex();
2753
2754 if (FileNumber < 1)
2755 return TokError("file number less than one");
2756 }
2757
2758 if (getLexer().isNot(AsmToken::String))
2759 return TokError("unexpected token in '.file' directive");
2760
2761 // Usually the directory and filename together, otherwise just the directory.
2762 // Allow the strings to have escaped octal character sequence.
2763 std::string Path = getTok().getString();
2764 if (parseEscapedString(Path))
2765 return true;
2766 Lex();
2767
2768 StringRef Directory;
2769 StringRef Filename;
2770 std::string FilenameData;
2771 if (getLexer().is(AsmToken::String)) {
2772 if (FileNumber == -1)
2773 return TokError("explicit path specified, but no file number");
2774 if (parseEscapedString(FilenameData))
2775 return true;
2776 Filename = FilenameData;
2777 Directory = Path;
2778 Lex();
2779 } else {
2780 Filename = Path;
2781 }
2782
2783 if (getLexer().isNot(AsmToken::EndOfStatement))
2784 return TokError("unexpected token in '.file' directive");
2785
2786 if (FileNumber == -1)
2787 getStreamer().EmitFileDirective(Filename);
2788 else {
2789 if (getContext().getGenDwarfForAssembly())
2790 Error(DirectiveLoc,
2791 "input can't have .file dwarf directives when -g is "
2792 "used to generate dwarf debug info for assembly code");
2793
2794 if (getStreamer().EmitDwarfFileDirective(FileNumber, Directory, Filename) ==
2795 0)
2796 Error(FileNumberLoc, "file number already allocated");
2797 }
2798
2799 return false;
2800 }
2801
2802 /// parseDirectiveLine
2803 /// ::= .line [number]
parseDirectiveLine()2804 bool AsmParser::parseDirectiveLine() {
2805 if (getLexer().isNot(AsmToken::EndOfStatement)) {
2806 if (getLexer().isNot(AsmToken::Integer))
2807 return TokError("unexpected token in '.line' directive");
2808
2809 int64_t LineNumber = getTok().getIntVal();
2810 (void)LineNumber;
2811 Lex();
2812
2813 // FIXME: Do something with the .line.
2814 }
2815
2816 if (getLexer().isNot(AsmToken::EndOfStatement))
2817 return TokError("unexpected token in '.line' directive");
2818
2819 return false;
2820 }
2821
2822 /// parseDirectiveLoc
2823 /// ::= .loc FileNumber [LineNumber] [ColumnPos] [basic_block] [prologue_end]
2824 /// [epilogue_begin] [is_stmt VALUE] [isa VALUE]
2825 /// The first number is a file number, must have been previously assigned with
2826 /// a .file directive, the second number is the line number and optionally the
2827 /// third number is a column position (zero if not specified). The remaining
2828 /// optional items are .loc sub-directives.
parseDirectiveLoc()2829 bool AsmParser::parseDirectiveLoc() {
2830 if (getLexer().isNot(AsmToken::Integer))
2831 return TokError("unexpected token in '.loc' directive");
2832 int64_t FileNumber = getTok().getIntVal();
2833 if (FileNumber < 1)
2834 return TokError("file number less than one in '.loc' directive");
2835 if (!getContext().isValidDwarfFileNumber(FileNumber))
2836 return TokError("unassigned file number in '.loc' directive");
2837 Lex();
2838
2839 int64_t LineNumber = 0;
2840 if (getLexer().is(AsmToken::Integer)) {
2841 LineNumber = getTok().getIntVal();
2842 if (LineNumber < 0)
2843 return TokError("line number less than zero in '.loc' directive");
2844 Lex();
2845 }
2846
2847 int64_t ColumnPos = 0;
2848 if (getLexer().is(AsmToken::Integer)) {
2849 ColumnPos = getTok().getIntVal();
2850 if (ColumnPos < 0)
2851 return TokError("column position less than zero in '.loc' directive");
2852 Lex();
2853 }
2854
2855 unsigned Flags = DWARF2_LINE_DEFAULT_IS_STMT ? DWARF2_FLAG_IS_STMT : 0;
2856 unsigned Isa = 0;
2857 int64_t Discriminator = 0;
2858 if (getLexer().isNot(AsmToken::EndOfStatement)) {
2859 for (;;) {
2860 if (getLexer().is(AsmToken::EndOfStatement))
2861 break;
2862
2863 StringRef Name;
2864 SMLoc Loc = getTok().getLoc();
2865 if (parseIdentifier(Name))
2866 return TokError("unexpected token in '.loc' directive");
2867
2868 if (Name == "basic_block")
2869 Flags |= DWARF2_FLAG_BASIC_BLOCK;
2870 else if (Name == "prologue_end")
2871 Flags |= DWARF2_FLAG_PROLOGUE_END;
2872 else if (Name == "epilogue_begin")
2873 Flags |= DWARF2_FLAG_EPILOGUE_BEGIN;
2874 else if (Name == "is_stmt") {
2875 Loc = getTok().getLoc();
2876 const MCExpr *Value;
2877 if (parseExpression(Value))
2878 return true;
2879 // The expression must be the constant 0 or 1.
2880 if (const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value)) {
2881 int Value = MCE->getValue();
2882 if (Value == 0)
2883 Flags &= ~DWARF2_FLAG_IS_STMT;
2884 else if (Value == 1)
2885 Flags |= DWARF2_FLAG_IS_STMT;
2886 else
2887 return Error(Loc, "is_stmt value not 0 or 1");
2888 } else {
2889 return Error(Loc, "is_stmt value not the constant value of 0 or 1");
2890 }
2891 } else if (Name == "isa") {
2892 Loc = getTok().getLoc();
2893 const MCExpr *Value;
2894 if (parseExpression(Value))
2895 return true;
2896 // The expression must be a constant greater or equal to 0.
2897 if (const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value)) {
2898 int Value = MCE->getValue();
2899 if (Value < 0)
2900 return Error(Loc, "isa number less than zero");
2901 Isa = Value;
2902 } else {
2903 return Error(Loc, "isa number not a constant value");
2904 }
2905 } else if (Name == "discriminator") {
2906 if (parseAbsoluteExpression(Discriminator))
2907 return true;
2908 } else {
2909 return Error(Loc, "unknown sub-directive in '.loc' directive");
2910 }
2911
2912 if (getLexer().is(AsmToken::EndOfStatement))
2913 break;
2914 }
2915 }
2916
2917 getStreamer().EmitDwarfLocDirective(FileNumber, LineNumber, ColumnPos, Flags,
2918 Isa, Discriminator, StringRef());
2919
2920 return false;
2921 }
2922
2923 /// parseDirectiveStabs
2924 /// ::= .stabs string, number, number, number
parseDirectiveStabs()2925 bool AsmParser::parseDirectiveStabs() {
2926 return TokError("unsupported directive '.stabs'");
2927 }
2928
2929 /// parseDirectiveCFISections
2930 /// ::= .cfi_sections section [, section]
parseDirectiveCFISections()2931 bool AsmParser::parseDirectiveCFISections() {
2932 StringRef Name;
2933 bool EH = false;
2934 bool Debug = false;
2935
2936 if (parseIdentifier(Name))
2937 return TokError("Expected an identifier");
2938
2939 if (Name == ".eh_frame")
2940 EH = true;
2941 else if (Name == ".debug_frame")
2942 Debug = true;
2943
2944 if (getLexer().is(AsmToken::Comma)) {
2945 Lex();
2946
2947 if (parseIdentifier(Name))
2948 return TokError("Expected an identifier");
2949
2950 if (Name == ".eh_frame")
2951 EH = true;
2952 else if (Name == ".debug_frame")
2953 Debug = true;
2954 }
2955
2956 getStreamer().EmitCFISections(EH, Debug);
2957 return false;
2958 }
2959
2960 /// parseDirectiveCFIStartProc
2961 /// ::= .cfi_startproc [simple]
parseDirectiveCFIStartProc()2962 bool AsmParser::parseDirectiveCFIStartProc() {
2963 StringRef Simple;
2964 if (getLexer().isNot(AsmToken::EndOfStatement))
2965 if (parseIdentifier(Simple) || Simple != "simple")
2966 return TokError("unexpected token in .cfi_startproc directive");
2967
2968 getStreamer().EmitCFIStartProc(!Simple.empty());
2969 return false;
2970 }
2971
2972 /// parseDirectiveCFIEndProc
2973 /// ::= .cfi_endproc
parseDirectiveCFIEndProc()2974 bool AsmParser::parseDirectiveCFIEndProc() {
2975 getStreamer().EmitCFIEndProc();
2976 return false;
2977 }
2978
2979 /// \brief parse register name or number.
parseRegisterOrRegisterNumber(int64_t & Register,SMLoc DirectiveLoc)2980 bool AsmParser::parseRegisterOrRegisterNumber(int64_t &Register,
2981 SMLoc DirectiveLoc) {
2982 unsigned RegNo;
2983
2984 if (getLexer().isNot(AsmToken::Integer)) {
2985 if (getTargetParser().ParseRegister(RegNo, DirectiveLoc, DirectiveLoc))
2986 return true;
2987 Register = getContext().getRegisterInfo()->getDwarfRegNum(RegNo, true);
2988 } else
2989 return parseAbsoluteExpression(Register);
2990
2991 return false;
2992 }
2993
2994 /// parseDirectiveCFIDefCfa
2995 /// ::= .cfi_def_cfa register, offset
parseDirectiveCFIDefCfa(SMLoc DirectiveLoc)2996 bool AsmParser::parseDirectiveCFIDefCfa(SMLoc DirectiveLoc) {
2997 int64_t Register = 0;
2998 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc))
2999 return true;
3000
3001 if (getLexer().isNot(AsmToken::Comma))
3002 return TokError("unexpected token in directive");
3003 Lex();
3004
3005 int64_t Offset = 0;
3006 if (parseAbsoluteExpression(Offset))
3007 return true;
3008
3009 getStreamer().EmitCFIDefCfa(Register, Offset);
3010 return false;
3011 }
3012
3013 /// parseDirectiveCFIDefCfaOffset
3014 /// ::= .cfi_def_cfa_offset offset
parseDirectiveCFIDefCfaOffset()3015 bool AsmParser::parseDirectiveCFIDefCfaOffset() {
3016 int64_t Offset = 0;
3017 if (parseAbsoluteExpression(Offset))
3018 return true;
3019
3020 getStreamer().EmitCFIDefCfaOffset(Offset);
3021 return false;
3022 }
3023
3024 /// parseDirectiveCFIRegister
3025 /// ::= .cfi_register register, register
parseDirectiveCFIRegister(SMLoc DirectiveLoc)3026 bool AsmParser::parseDirectiveCFIRegister(SMLoc DirectiveLoc) {
3027 int64_t Register1 = 0;
3028 if (parseRegisterOrRegisterNumber(Register1, DirectiveLoc))
3029 return true;
3030
3031 if (getLexer().isNot(AsmToken::Comma))
3032 return TokError("unexpected token in directive");
3033 Lex();
3034
3035 int64_t Register2 = 0;
3036 if (parseRegisterOrRegisterNumber(Register2, DirectiveLoc))
3037 return true;
3038
3039 getStreamer().EmitCFIRegister(Register1, Register2);
3040 return false;
3041 }
3042
3043 /// parseDirectiveCFIWindowSave
3044 /// ::= .cfi_window_save
parseDirectiveCFIWindowSave()3045 bool AsmParser::parseDirectiveCFIWindowSave() {
3046 getStreamer().EmitCFIWindowSave();
3047 return false;
3048 }
3049
3050 /// parseDirectiveCFIAdjustCfaOffset
3051 /// ::= .cfi_adjust_cfa_offset adjustment
parseDirectiveCFIAdjustCfaOffset()3052 bool AsmParser::parseDirectiveCFIAdjustCfaOffset() {
3053 int64_t Adjustment = 0;
3054 if (parseAbsoluteExpression(Adjustment))
3055 return true;
3056
3057 getStreamer().EmitCFIAdjustCfaOffset(Adjustment);
3058 return false;
3059 }
3060
3061 /// parseDirectiveCFIDefCfaRegister
3062 /// ::= .cfi_def_cfa_register register
parseDirectiveCFIDefCfaRegister(SMLoc DirectiveLoc)3063 bool AsmParser::parseDirectiveCFIDefCfaRegister(SMLoc DirectiveLoc) {
3064 int64_t Register = 0;
3065 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc))
3066 return true;
3067
3068 getStreamer().EmitCFIDefCfaRegister(Register);
3069 return false;
3070 }
3071
3072 /// parseDirectiveCFIOffset
3073 /// ::= .cfi_offset register, offset
parseDirectiveCFIOffset(SMLoc DirectiveLoc)3074 bool AsmParser::parseDirectiveCFIOffset(SMLoc DirectiveLoc) {
3075 int64_t Register = 0;
3076 int64_t Offset = 0;
3077
3078 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc))
3079 return true;
3080
3081 if (getLexer().isNot(AsmToken::Comma))
3082 return TokError("unexpected token in directive");
3083 Lex();
3084
3085 if (parseAbsoluteExpression(Offset))
3086 return true;
3087
3088 getStreamer().EmitCFIOffset(Register, Offset);
3089 return false;
3090 }
3091
3092 /// parseDirectiveCFIRelOffset
3093 /// ::= .cfi_rel_offset register, offset
parseDirectiveCFIRelOffset(SMLoc DirectiveLoc)3094 bool AsmParser::parseDirectiveCFIRelOffset(SMLoc DirectiveLoc) {
3095 int64_t Register = 0;
3096
3097 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc))
3098 return true;
3099
3100 if (getLexer().isNot(AsmToken::Comma))
3101 return TokError("unexpected token in directive");
3102 Lex();
3103
3104 int64_t Offset = 0;
3105 if (parseAbsoluteExpression(Offset))
3106 return true;
3107
3108 getStreamer().EmitCFIRelOffset(Register, Offset);
3109 return false;
3110 }
3111
isValidEncoding(int64_t Encoding)3112 static bool isValidEncoding(int64_t Encoding) {
3113 if (Encoding & ~0xff)
3114 return false;
3115
3116 if (Encoding == dwarf::DW_EH_PE_omit)
3117 return true;
3118
3119 const unsigned Format = Encoding & 0xf;
3120 if (Format != dwarf::DW_EH_PE_absptr && Format != dwarf::DW_EH_PE_udata2 &&
3121 Format != dwarf::DW_EH_PE_udata4 && Format != dwarf::DW_EH_PE_udata8 &&
3122 Format != dwarf::DW_EH_PE_sdata2 && Format != dwarf::DW_EH_PE_sdata4 &&
3123 Format != dwarf::DW_EH_PE_sdata8 && Format != dwarf::DW_EH_PE_signed)
3124 return false;
3125
3126 const unsigned Application = Encoding & 0x70;
3127 if (Application != dwarf::DW_EH_PE_absptr &&
3128 Application != dwarf::DW_EH_PE_pcrel)
3129 return false;
3130
3131 return true;
3132 }
3133
3134 /// parseDirectiveCFIPersonalityOrLsda
3135 /// IsPersonality true for cfi_personality, false for cfi_lsda
3136 /// ::= .cfi_personality encoding, [symbol_name]
3137 /// ::= .cfi_lsda encoding, [symbol_name]
parseDirectiveCFIPersonalityOrLsda(bool IsPersonality)3138 bool AsmParser::parseDirectiveCFIPersonalityOrLsda(bool IsPersonality) {
3139 int64_t Encoding = 0;
3140 if (parseAbsoluteExpression(Encoding))
3141 return true;
3142 if (Encoding == dwarf::DW_EH_PE_omit)
3143 return false;
3144
3145 if (!isValidEncoding(Encoding))
3146 return TokError("unsupported encoding.");
3147
3148 if (getLexer().isNot(AsmToken::Comma))
3149 return TokError("unexpected token in directive");
3150 Lex();
3151
3152 StringRef Name;
3153 if (parseIdentifier(Name))
3154 return TokError("expected identifier in directive");
3155
3156 MCSymbol *Sym = getContext().getOrCreateSymbol(Name);
3157
3158 if (IsPersonality)
3159 getStreamer().EmitCFIPersonality(Sym, Encoding);
3160 else
3161 getStreamer().EmitCFILsda(Sym, Encoding);
3162 return false;
3163 }
3164
3165 /// parseDirectiveCFIRememberState
3166 /// ::= .cfi_remember_state
parseDirectiveCFIRememberState()3167 bool AsmParser::parseDirectiveCFIRememberState() {
3168 getStreamer().EmitCFIRememberState();
3169 return false;
3170 }
3171
3172 /// parseDirectiveCFIRestoreState
3173 /// ::= .cfi_remember_state
parseDirectiveCFIRestoreState()3174 bool AsmParser::parseDirectiveCFIRestoreState() {
3175 getStreamer().EmitCFIRestoreState();
3176 return false;
3177 }
3178
3179 /// parseDirectiveCFISameValue
3180 /// ::= .cfi_same_value register
parseDirectiveCFISameValue(SMLoc DirectiveLoc)3181 bool AsmParser::parseDirectiveCFISameValue(SMLoc DirectiveLoc) {
3182 int64_t Register = 0;
3183
3184 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc))
3185 return true;
3186
3187 getStreamer().EmitCFISameValue(Register);
3188 return false;
3189 }
3190
3191 /// parseDirectiveCFIRestore
3192 /// ::= .cfi_restore register
parseDirectiveCFIRestore(SMLoc DirectiveLoc)3193 bool AsmParser::parseDirectiveCFIRestore(SMLoc DirectiveLoc) {
3194 int64_t Register = 0;
3195 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc))
3196 return true;
3197
3198 getStreamer().EmitCFIRestore(Register);
3199 return false;
3200 }
3201
3202 /// parseDirectiveCFIEscape
3203 /// ::= .cfi_escape expression[,...]
parseDirectiveCFIEscape()3204 bool AsmParser::parseDirectiveCFIEscape() {
3205 std::string Values;
3206 int64_t CurrValue;
3207 if (parseAbsoluteExpression(CurrValue))
3208 return true;
3209
3210 Values.push_back((uint8_t)CurrValue);
3211
3212 while (getLexer().is(AsmToken::Comma)) {
3213 Lex();
3214
3215 if (parseAbsoluteExpression(CurrValue))
3216 return true;
3217
3218 Values.push_back((uint8_t)CurrValue);
3219 }
3220
3221 getStreamer().EmitCFIEscape(Values);
3222 return false;
3223 }
3224
3225 /// parseDirectiveCFISignalFrame
3226 /// ::= .cfi_signal_frame
parseDirectiveCFISignalFrame()3227 bool AsmParser::parseDirectiveCFISignalFrame() {
3228 if (getLexer().isNot(AsmToken::EndOfStatement))
3229 return Error(getLexer().getLoc(),
3230 "unexpected token in '.cfi_signal_frame'");
3231
3232 getStreamer().EmitCFISignalFrame();
3233 return false;
3234 }
3235
3236 /// parseDirectiveCFIUndefined
3237 /// ::= .cfi_undefined register
parseDirectiveCFIUndefined(SMLoc DirectiveLoc)3238 bool AsmParser::parseDirectiveCFIUndefined(SMLoc DirectiveLoc) {
3239 int64_t Register = 0;
3240
3241 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc))
3242 return true;
3243
3244 getStreamer().EmitCFIUndefined(Register);
3245 return false;
3246 }
3247
3248 /// parseDirectiveMacrosOnOff
3249 /// ::= .macros_on
3250 /// ::= .macros_off
parseDirectiveMacrosOnOff(StringRef Directive)3251 bool AsmParser::parseDirectiveMacrosOnOff(StringRef Directive) {
3252 if (getLexer().isNot(AsmToken::EndOfStatement))
3253 return Error(getLexer().getLoc(),
3254 "unexpected token in '" + Directive + "' directive");
3255
3256 setMacrosEnabled(Directive == ".macros_on");
3257 return false;
3258 }
3259
3260 /// parseDirectiveMacro
3261 /// ::= .macro name[,] [parameters]
parseDirectiveMacro(SMLoc DirectiveLoc)3262 bool AsmParser::parseDirectiveMacro(SMLoc DirectiveLoc) {
3263 StringRef Name;
3264 if (parseIdentifier(Name))
3265 return TokError("expected identifier in '.macro' directive");
3266
3267 if (getLexer().is(AsmToken::Comma))
3268 Lex();
3269
3270 MCAsmMacroParameters Parameters;
3271 while (getLexer().isNot(AsmToken::EndOfStatement)) {
3272
3273 if (!Parameters.empty() && Parameters.back().Vararg)
3274 return Error(Lexer.getLoc(),
3275 "Vararg parameter '" + Parameters.back().Name +
3276 "' should be last one in the list of parameters.");
3277
3278 MCAsmMacroParameter Parameter;
3279 if (parseIdentifier(Parameter.Name))
3280 return TokError("expected identifier in '.macro' directive");
3281
3282 if (Lexer.is(AsmToken::Colon)) {
3283 Lex(); // consume ':'
3284
3285 SMLoc QualLoc;
3286 StringRef Qualifier;
3287
3288 QualLoc = Lexer.getLoc();
3289 if (parseIdentifier(Qualifier))
3290 return Error(QualLoc, "missing parameter qualifier for "
3291 "'" + Parameter.Name + "' in macro '" + Name + "'");
3292
3293 if (Qualifier == "req")
3294 Parameter.Required = true;
3295 else if (Qualifier == "vararg")
3296 Parameter.Vararg = true;
3297 else
3298 return Error(QualLoc, Qualifier + " is not a valid parameter qualifier "
3299 "for '" + Parameter.Name + "' in macro '" + Name + "'");
3300 }
3301
3302 if (getLexer().is(AsmToken::Equal)) {
3303 Lex();
3304
3305 SMLoc ParamLoc;
3306
3307 ParamLoc = Lexer.getLoc();
3308 if (parseMacroArgument(Parameter.Value, /*Vararg=*/false ))
3309 return true;
3310
3311 if (Parameter.Required)
3312 Warning(ParamLoc, "pointless default value for required parameter "
3313 "'" + Parameter.Name + "' in macro '" + Name + "'");
3314 }
3315
3316 Parameters.push_back(std::move(Parameter));
3317
3318 if (getLexer().is(AsmToken::Comma))
3319 Lex();
3320 }
3321
3322 // Eat the end of statement.
3323 Lex();
3324
3325 AsmToken EndToken, StartToken = getTok();
3326 unsigned MacroDepth = 0;
3327
3328 // Lex the macro definition.
3329 for (;;) {
3330 // Check whether we have reached the end of the file.
3331 if (getLexer().is(AsmToken::Eof))
3332 return Error(DirectiveLoc, "no matching '.endmacro' in definition");
3333
3334 // Otherwise, check whether we have reach the .endmacro.
3335 if (getLexer().is(AsmToken::Identifier)) {
3336 if (getTok().getIdentifier() == ".endm" ||
3337 getTok().getIdentifier() == ".endmacro") {
3338 if (MacroDepth == 0) { // Outermost macro.
3339 EndToken = getTok();
3340 Lex();
3341 if (getLexer().isNot(AsmToken::EndOfStatement))
3342 return TokError("unexpected token in '" + EndToken.getIdentifier() +
3343 "' directive");
3344 break;
3345 } else {
3346 // Otherwise we just found the end of an inner macro.
3347 --MacroDepth;
3348 }
3349 } else if (getTok().getIdentifier() == ".macro") {
3350 // We allow nested macros. Those aren't instantiated until the outermost
3351 // macro is expanded so just ignore them for now.
3352 ++MacroDepth;
3353 }
3354 }
3355
3356 // Otherwise, scan til the end of the statement.
3357 eatToEndOfStatement();
3358 }
3359
3360 if (lookupMacro(Name)) {
3361 return Error(DirectiveLoc, "macro '" + Name + "' is already defined");
3362 }
3363
3364 const char *BodyStart = StartToken.getLoc().getPointer();
3365 const char *BodyEnd = EndToken.getLoc().getPointer();
3366 StringRef Body = StringRef(BodyStart, BodyEnd - BodyStart);
3367 checkForBadMacro(DirectiveLoc, Name, Body, Parameters);
3368 defineMacro(Name, MCAsmMacro(Name, Body, std::move(Parameters)));
3369 return false;
3370 }
3371
3372 /// checkForBadMacro
3373 ///
3374 /// With the support added for named parameters there may be code out there that
3375 /// is transitioning from positional parameters. In versions of gas that did
3376 /// not support named parameters they would be ignored on the macro definition.
3377 /// But to support both styles of parameters this is not possible so if a macro
3378 /// definition has named parameters but does not use them and has what appears
3379 /// to be positional parameters, strings like $1, $2, ... and $n, then issue a
3380 /// warning that the positional parameter found in body which have no effect.
3381 /// Hoping the developer will either remove the named parameters from the macro
3382 /// definition so the positional parameters get used if that was what was
3383 /// intended or change the macro to use the named parameters. It is possible
3384 /// this warning will trigger when the none of the named parameters are used
3385 /// and the strings like $1 are infact to simply to be passed trough unchanged.
checkForBadMacro(SMLoc DirectiveLoc,StringRef Name,StringRef Body,ArrayRef<MCAsmMacroParameter> Parameters)3386 void AsmParser::checkForBadMacro(SMLoc DirectiveLoc, StringRef Name,
3387 StringRef Body,
3388 ArrayRef<MCAsmMacroParameter> Parameters) {
3389 // If this macro is not defined with named parameters the warning we are
3390 // checking for here doesn't apply.
3391 unsigned NParameters = Parameters.size();
3392 if (NParameters == 0)
3393 return;
3394
3395 bool NamedParametersFound = false;
3396 bool PositionalParametersFound = false;
3397
3398 // Look at the body of the macro for use of both the named parameters and what
3399 // are likely to be positional parameters. This is what expandMacro() is
3400 // doing when it finds the parameters in the body.
3401 while (!Body.empty()) {
3402 // Scan for the next possible parameter.
3403 std::size_t End = Body.size(), Pos = 0;
3404 for (; Pos != End; ++Pos) {
3405 // Check for a substitution or escape.
3406 // This macro is defined with parameters, look for \foo, \bar, etc.
3407 if (Body[Pos] == '\\' && Pos + 1 != End)
3408 break;
3409
3410 // This macro should have parameters, but look for $0, $1, ..., $n too.
3411 if (Body[Pos] != '$' || Pos + 1 == End)
3412 continue;
3413 char Next = Body[Pos + 1];
3414 if (Next == '$' || Next == 'n' ||
3415 isdigit(static_cast<unsigned char>(Next)))
3416 break;
3417 }
3418
3419 // Check if we reached the end.
3420 if (Pos == End)
3421 break;
3422
3423 if (Body[Pos] == '$') {
3424 switch (Body[Pos + 1]) {
3425 // $$ => $
3426 case '$':
3427 break;
3428
3429 // $n => number of arguments
3430 case 'n':
3431 PositionalParametersFound = true;
3432 break;
3433
3434 // $[0-9] => argument
3435 default: {
3436 PositionalParametersFound = true;
3437 break;
3438 }
3439 }
3440 Pos += 2;
3441 } else {
3442 unsigned I = Pos + 1;
3443 while (isIdentifierChar(Body[I]) && I + 1 != End)
3444 ++I;
3445
3446 const char *Begin = Body.data() + Pos + 1;
3447 StringRef Argument(Begin, I - (Pos + 1));
3448 unsigned Index = 0;
3449 for (; Index < NParameters; ++Index)
3450 if (Parameters[Index].Name == Argument)
3451 break;
3452
3453 if (Index == NParameters) {
3454 if (Body[Pos + 1] == '(' && Body[Pos + 2] == ')')
3455 Pos += 3;
3456 else {
3457 Pos = I;
3458 }
3459 } else {
3460 NamedParametersFound = true;
3461 Pos += 1 + Argument.size();
3462 }
3463 }
3464 // Update the scan point.
3465 Body = Body.substr(Pos);
3466 }
3467
3468 if (!NamedParametersFound && PositionalParametersFound)
3469 Warning(DirectiveLoc, "macro defined with named parameters which are not "
3470 "used in macro body, possible positional parameter "
3471 "found in body which will have no effect");
3472 }
3473
3474 /// parseDirectiveExitMacro
3475 /// ::= .exitm
parseDirectiveExitMacro(StringRef Directive)3476 bool AsmParser::parseDirectiveExitMacro(StringRef Directive) {
3477 if (getLexer().isNot(AsmToken::EndOfStatement))
3478 return TokError("unexpected token in '" + Directive + "' directive");
3479
3480 if (!isInsideMacroInstantiation())
3481 return TokError("unexpected '" + Directive + "' in file, "
3482 "no current macro definition");
3483
3484 // Exit all conditionals that are active in the current macro.
3485 while (TheCondStack.size() != ActiveMacros.back()->CondStackDepth) {
3486 TheCondState = TheCondStack.back();
3487 TheCondStack.pop_back();
3488 }
3489
3490 handleMacroExit();
3491 return false;
3492 }
3493
3494 /// parseDirectiveEndMacro
3495 /// ::= .endm
3496 /// ::= .endmacro
parseDirectiveEndMacro(StringRef Directive)3497 bool AsmParser::parseDirectiveEndMacro(StringRef Directive) {
3498 if (getLexer().isNot(AsmToken::EndOfStatement))
3499 return TokError("unexpected token in '" + Directive + "' directive");
3500
3501 // If we are inside a macro instantiation, terminate the current
3502 // instantiation.
3503 if (isInsideMacroInstantiation()) {
3504 handleMacroExit();
3505 return false;
3506 }
3507
3508 // Otherwise, this .endmacro is a stray entry in the file; well formed
3509 // .endmacro directives are handled during the macro definition parsing.
3510 return TokError("unexpected '" + Directive + "' in file, "
3511 "no current macro definition");
3512 }
3513
3514 /// parseDirectivePurgeMacro
3515 /// ::= .purgem
parseDirectivePurgeMacro(SMLoc DirectiveLoc)3516 bool AsmParser::parseDirectivePurgeMacro(SMLoc DirectiveLoc) {
3517 StringRef Name;
3518 if (parseIdentifier(Name))
3519 return TokError("expected identifier in '.purgem' directive");
3520
3521 if (getLexer().isNot(AsmToken::EndOfStatement))
3522 return TokError("unexpected token in '.purgem' directive");
3523
3524 if (!lookupMacro(Name))
3525 return Error(DirectiveLoc, "macro '" + Name + "' is not defined");
3526
3527 undefineMacro(Name);
3528 return false;
3529 }
3530
3531 /// parseDirectiveBundleAlignMode
3532 /// ::= {.bundle_align_mode} expression
parseDirectiveBundleAlignMode()3533 bool AsmParser::parseDirectiveBundleAlignMode() {
3534 checkForValidSection();
3535
3536 // Expect a single argument: an expression that evaluates to a constant
3537 // in the inclusive range 0-30.
3538 SMLoc ExprLoc = getLexer().getLoc();
3539 int64_t AlignSizePow2;
3540 if (parseAbsoluteExpression(AlignSizePow2))
3541 return true;
3542 else if (getLexer().isNot(AsmToken::EndOfStatement))
3543 return TokError("unexpected token after expression in"
3544 " '.bundle_align_mode' directive");
3545 else if (AlignSizePow2 < 0 || AlignSizePow2 > 30)
3546 return Error(ExprLoc,
3547 "invalid bundle alignment size (expected between 0 and 30)");
3548
3549 Lex();
3550
3551 // Because of AlignSizePow2's verified range we can safely truncate it to
3552 // unsigned.
3553 getStreamer().EmitBundleAlignMode(static_cast<unsigned>(AlignSizePow2));
3554 return false;
3555 }
3556
3557 /// parseDirectiveBundleLock
3558 /// ::= {.bundle_lock} [align_to_end]
parseDirectiveBundleLock()3559 bool AsmParser::parseDirectiveBundleLock() {
3560 checkForValidSection();
3561 bool AlignToEnd = false;
3562
3563 if (getLexer().isNot(AsmToken::EndOfStatement)) {
3564 StringRef Option;
3565 SMLoc Loc = getTok().getLoc();
3566 const char *kInvalidOptionError =
3567 "invalid option for '.bundle_lock' directive";
3568
3569 if (parseIdentifier(Option))
3570 return Error(Loc, kInvalidOptionError);
3571
3572 if (Option != "align_to_end")
3573 return Error(Loc, kInvalidOptionError);
3574 else if (getLexer().isNot(AsmToken::EndOfStatement))
3575 return Error(Loc,
3576 "unexpected token after '.bundle_lock' directive option");
3577 AlignToEnd = true;
3578 }
3579
3580 Lex();
3581
3582 getStreamer().EmitBundleLock(AlignToEnd);
3583 return false;
3584 }
3585
3586 /// parseDirectiveBundleLock
3587 /// ::= {.bundle_lock}
parseDirectiveBundleUnlock()3588 bool AsmParser::parseDirectiveBundleUnlock() {
3589 checkForValidSection();
3590
3591 if (getLexer().isNot(AsmToken::EndOfStatement))
3592 return TokError("unexpected token in '.bundle_unlock' directive");
3593 Lex();
3594
3595 getStreamer().EmitBundleUnlock();
3596 return false;
3597 }
3598
3599 /// parseDirectiveSpace
3600 /// ::= (.skip | .space) expression [ , expression ]
parseDirectiveSpace(StringRef IDVal)3601 bool AsmParser::parseDirectiveSpace(StringRef IDVal) {
3602 checkForValidSection();
3603
3604 int64_t NumBytes;
3605 if (parseAbsoluteExpression(NumBytes))
3606 return true;
3607
3608 int64_t FillExpr = 0;
3609 if (getLexer().isNot(AsmToken::EndOfStatement)) {
3610 if (getLexer().isNot(AsmToken::Comma))
3611 return TokError("unexpected token in '" + Twine(IDVal) + "' directive");
3612 Lex();
3613
3614 if (parseAbsoluteExpression(FillExpr))
3615 return true;
3616
3617 if (getLexer().isNot(AsmToken::EndOfStatement))
3618 return TokError("unexpected token in '" + Twine(IDVal) + "' directive");
3619 }
3620
3621 Lex();
3622
3623 if (NumBytes <= 0)
3624 return TokError("invalid number of bytes in '" + Twine(IDVal) +
3625 "' directive");
3626
3627 // FIXME: Sometimes the fill expr is 'nop' if it isn't supplied, instead of 0.
3628 getStreamer().EmitFill(NumBytes, FillExpr);
3629
3630 return false;
3631 }
3632
3633 /// parseDirectiveLEB128
3634 /// ::= (.sleb128 | .uleb128) [ expression (, expression)* ]
parseDirectiveLEB128(bool Signed)3635 bool AsmParser::parseDirectiveLEB128(bool Signed) {
3636 checkForValidSection();
3637 const MCExpr *Value;
3638
3639 for (;;) {
3640 if (parseExpression(Value))
3641 return true;
3642
3643 if (Signed)
3644 getStreamer().EmitSLEB128Value(Value);
3645 else
3646 getStreamer().EmitULEB128Value(Value);
3647
3648 if (getLexer().is(AsmToken::EndOfStatement))
3649 break;
3650
3651 if (getLexer().isNot(AsmToken::Comma))
3652 return TokError("unexpected token in directive");
3653 Lex();
3654 }
3655
3656 return false;
3657 }
3658
3659 /// parseDirectiveSymbolAttribute
3660 /// ::= { ".globl", ".weak", ... } [ identifier ( , identifier )* ]
parseDirectiveSymbolAttribute(MCSymbolAttr Attr)3661 bool AsmParser::parseDirectiveSymbolAttribute(MCSymbolAttr Attr) {
3662 if (getLexer().isNot(AsmToken::EndOfStatement)) {
3663 for (;;) {
3664 StringRef Name;
3665 SMLoc Loc = getTok().getLoc();
3666
3667 if (parseIdentifier(Name))
3668 return Error(Loc, "expected identifier in directive");
3669
3670 MCSymbol *Sym = getContext().getOrCreateSymbol(Name);
3671
3672 // Assembler local symbols don't make any sense here. Complain loudly.
3673 if (Sym->isTemporary())
3674 return Error(Loc, "non-local symbol required in directive");
3675
3676 if (!getStreamer().EmitSymbolAttribute(Sym, Attr))
3677 return Error(Loc, "unable to emit symbol attribute");
3678
3679 if (getLexer().is(AsmToken::EndOfStatement))
3680 break;
3681
3682 if (getLexer().isNot(AsmToken::Comma))
3683 return TokError("unexpected token in directive");
3684 Lex();
3685 }
3686 }
3687
3688 Lex();
3689 return false;
3690 }
3691
3692 /// parseDirectiveComm
3693 /// ::= ( .comm | .lcomm ) identifier , size_expression [ , align_expression ]
parseDirectiveComm(bool IsLocal)3694 bool AsmParser::parseDirectiveComm(bool IsLocal) {
3695 checkForValidSection();
3696
3697 SMLoc IDLoc = getLexer().getLoc();
3698 StringRef Name;
3699 if (parseIdentifier(Name))
3700 return TokError("expected identifier in directive");
3701
3702 // Handle the identifier as the key symbol.
3703 MCSymbol *Sym = getContext().getOrCreateSymbol(Name);
3704
3705 if (getLexer().isNot(AsmToken::Comma))
3706 return TokError("unexpected token in directive");
3707 Lex();
3708
3709 int64_t Size;
3710 SMLoc SizeLoc = getLexer().getLoc();
3711 if (parseAbsoluteExpression(Size))
3712 return true;
3713
3714 int64_t Pow2Alignment = 0;
3715 SMLoc Pow2AlignmentLoc;
3716 if (getLexer().is(AsmToken::Comma)) {
3717 Lex();
3718 Pow2AlignmentLoc = getLexer().getLoc();
3719 if (parseAbsoluteExpression(Pow2Alignment))
3720 return true;
3721
3722 LCOMM::LCOMMType LCOMM = Lexer.getMAI().getLCOMMDirectiveAlignmentType();
3723 if (IsLocal && LCOMM == LCOMM::NoAlignment)
3724 return Error(Pow2AlignmentLoc, "alignment not supported on this target");
3725
3726 // If this target takes alignments in bytes (not log) validate and convert.
3727 if ((!IsLocal && Lexer.getMAI().getCOMMDirectiveAlignmentIsInBytes()) ||
3728 (IsLocal && LCOMM == LCOMM::ByteAlignment)) {
3729 if (!isPowerOf2_64(Pow2Alignment))
3730 return Error(Pow2AlignmentLoc, "alignment must be a power of 2");
3731 Pow2Alignment = Log2_64(Pow2Alignment);
3732 }
3733 }
3734
3735 if (getLexer().isNot(AsmToken::EndOfStatement))
3736 return TokError("unexpected token in '.comm' or '.lcomm' directive");
3737
3738 Lex();
3739
3740 // NOTE: a size of zero for a .comm should create a undefined symbol
3741 // but a size of .lcomm creates a bss symbol of size zero.
3742 if (Size < 0)
3743 return Error(SizeLoc, "invalid '.comm' or '.lcomm' directive size, can't "
3744 "be less than zero");
3745
3746 // NOTE: The alignment in the directive is a power of 2 value, the assembler
3747 // may internally end up wanting an alignment in bytes.
3748 // FIXME: Diagnose overflow.
3749 if (Pow2Alignment < 0)
3750 return Error(Pow2AlignmentLoc, "invalid '.comm' or '.lcomm' directive "
3751 "alignment, can't be less than zero");
3752
3753 if (!Sym->isUndefined())
3754 return Error(IDLoc, "invalid symbol redefinition");
3755
3756 // Create the Symbol as a common or local common with Size and Pow2Alignment
3757 if (IsLocal) {
3758 getStreamer().EmitLocalCommonSymbol(Sym, Size, 1 << Pow2Alignment);
3759 return false;
3760 }
3761
3762 getStreamer().EmitCommonSymbol(Sym, Size, 1 << Pow2Alignment);
3763 return false;
3764 }
3765
3766 /// parseDirectiveAbort
3767 /// ::= .abort [... message ...]
parseDirectiveAbort()3768 bool AsmParser::parseDirectiveAbort() {
3769 // FIXME: Use loc from directive.
3770 SMLoc Loc = getLexer().getLoc();
3771
3772 StringRef Str = parseStringToEndOfStatement();
3773 if (getLexer().isNot(AsmToken::EndOfStatement))
3774 return TokError("unexpected token in '.abort' directive");
3775
3776 Lex();
3777
3778 if (Str.empty())
3779 Error(Loc, ".abort detected. Assembly stopping.");
3780 else
3781 Error(Loc, ".abort '" + Str + "' detected. Assembly stopping.");
3782 // FIXME: Actually abort assembly here.
3783
3784 return false;
3785 }
3786
3787 /// parseDirectiveInclude
3788 /// ::= .include "filename"
parseDirectiveInclude()3789 bool AsmParser::parseDirectiveInclude() {
3790 if (getLexer().isNot(AsmToken::String))
3791 return TokError("expected string in '.include' directive");
3792
3793 // Allow the strings to have escaped octal character sequence.
3794 std::string Filename;
3795 if (parseEscapedString(Filename))
3796 return true;
3797 SMLoc IncludeLoc = getLexer().getLoc();
3798 Lex();
3799
3800 if (getLexer().isNot(AsmToken::EndOfStatement))
3801 return TokError("unexpected token in '.include' directive");
3802
3803 // Attempt to switch the lexer to the included file before consuming the end
3804 // of statement to avoid losing it when we switch.
3805 if (enterIncludeFile(Filename)) {
3806 Error(IncludeLoc, "Could not find include file '" + Filename + "'");
3807 return true;
3808 }
3809
3810 return false;
3811 }
3812
3813 /// parseDirectiveIncbin
3814 /// ::= .incbin "filename"
parseDirectiveIncbin()3815 bool AsmParser::parseDirectiveIncbin() {
3816 if (getLexer().isNot(AsmToken::String))
3817 return TokError("expected string in '.incbin' directive");
3818
3819 // Allow the strings to have escaped octal character sequence.
3820 std::string Filename;
3821 if (parseEscapedString(Filename))
3822 return true;
3823 SMLoc IncbinLoc = getLexer().getLoc();
3824 Lex();
3825
3826 if (getLexer().isNot(AsmToken::EndOfStatement))
3827 return TokError("unexpected token in '.incbin' directive");
3828
3829 // Attempt to process the included file.
3830 if (processIncbinFile(Filename)) {
3831 Error(IncbinLoc, "Could not find incbin file '" + Filename + "'");
3832 return true;
3833 }
3834
3835 return false;
3836 }
3837
3838 /// parseDirectiveIf
3839 /// ::= .if{,eq,ge,gt,le,lt,ne} expression
parseDirectiveIf(SMLoc DirectiveLoc,DirectiveKind DirKind)3840 bool AsmParser::parseDirectiveIf(SMLoc DirectiveLoc, DirectiveKind DirKind) {
3841 TheCondStack.push_back(TheCondState);
3842 TheCondState.TheCond = AsmCond::IfCond;
3843 if (TheCondState.Ignore) {
3844 eatToEndOfStatement();
3845 } else {
3846 int64_t ExprValue;
3847 if (parseAbsoluteExpression(ExprValue))
3848 return true;
3849
3850 if (getLexer().isNot(AsmToken::EndOfStatement))
3851 return TokError("unexpected token in '.if' directive");
3852
3853 Lex();
3854
3855 switch (DirKind) {
3856 default:
3857 llvm_unreachable("unsupported directive");
3858 case DK_IF:
3859 case DK_IFNE:
3860 break;
3861 case DK_IFEQ:
3862 ExprValue = ExprValue == 0;
3863 break;
3864 case DK_IFGE:
3865 ExprValue = ExprValue >= 0;
3866 break;
3867 case DK_IFGT:
3868 ExprValue = ExprValue > 0;
3869 break;
3870 case DK_IFLE:
3871 ExprValue = ExprValue <= 0;
3872 break;
3873 case DK_IFLT:
3874 ExprValue = ExprValue < 0;
3875 break;
3876 }
3877
3878 TheCondState.CondMet = ExprValue;
3879 TheCondState.Ignore = !TheCondState.CondMet;
3880 }
3881
3882 return false;
3883 }
3884
3885 /// parseDirectiveIfb
3886 /// ::= .ifb string
parseDirectiveIfb(SMLoc DirectiveLoc,bool ExpectBlank)3887 bool AsmParser::parseDirectiveIfb(SMLoc DirectiveLoc, bool ExpectBlank) {
3888 TheCondStack.push_back(TheCondState);
3889 TheCondState.TheCond = AsmCond::IfCond;
3890
3891 if (TheCondState.Ignore) {
3892 eatToEndOfStatement();
3893 } else {
3894 StringRef Str = parseStringToEndOfStatement();
3895
3896 if (getLexer().isNot(AsmToken::EndOfStatement))
3897 return TokError("unexpected token in '.ifb' directive");
3898
3899 Lex();
3900
3901 TheCondState.CondMet = ExpectBlank == Str.empty();
3902 TheCondState.Ignore = !TheCondState.CondMet;
3903 }
3904
3905 return false;
3906 }
3907
3908 /// parseDirectiveIfc
3909 /// ::= .ifc string1, string2
3910 /// ::= .ifnc string1, string2
parseDirectiveIfc(SMLoc DirectiveLoc,bool ExpectEqual)3911 bool AsmParser::parseDirectiveIfc(SMLoc DirectiveLoc, bool ExpectEqual) {
3912 TheCondStack.push_back(TheCondState);
3913 TheCondState.TheCond = AsmCond::IfCond;
3914
3915 if (TheCondState.Ignore) {
3916 eatToEndOfStatement();
3917 } else {
3918 StringRef Str1 = parseStringToComma();
3919
3920 if (getLexer().isNot(AsmToken::Comma))
3921 return TokError("unexpected token in '.ifc' directive");
3922
3923 Lex();
3924
3925 StringRef Str2 = parseStringToEndOfStatement();
3926
3927 if (getLexer().isNot(AsmToken::EndOfStatement))
3928 return TokError("unexpected token in '.ifc' directive");
3929
3930 Lex();
3931
3932 TheCondState.CondMet = ExpectEqual == (Str1.trim() == Str2.trim());
3933 TheCondState.Ignore = !TheCondState.CondMet;
3934 }
3935
3936 return false;
3937 }
3938
3939 /// parseDirectiveIfeqs
3940 /// ::= .ifeqs string1, string2
parseDirectiveIfeqs(SMLoc DirectiveLoc,bool ExpectEqual)3941 bool AsmParser::parseDirectiveIfeqs(SMLoc DirectiveLoc, bool ExpectEqual) {
3942 if (Lexer.isNot(AsmToken::String)) {
3943 if (ExpectEqual)
3944 TokError("expected string parameter for '.ifeqs' directive");
3945 else
3946 TokError("expected string parameter for '.ifnes' directive");
3947 eatToEndOfStatement();
3948 return true;
3949 }
3950
3951 StringRef String1 = getTok().getStringContents();
3952 Lex();
3953
3954 if (Lexer.isNot(AsmToken::Comma)) {
3955 if (ExpectEqual)
3956 TokError("expected comma after first string for '.ifeqs' directive");
3957 else
3958 TokError("expected comma after first string for '.ifnes' directive");
3959 eatToEndOfStatement();
3960 return true;
3961 }
3962
3963 Lex();
3964
3965 if (Lexer.isNot(AsmToken::String)) {
3966 if (ExpectEqual)
3967 TokError("expected string parameter for '.ifeqs' directive");
3968 else
3969 TokError("expected string parameter for '.ifnes' directive");
3970 eatToEndOfStatement();
3971 return true;
3972 }
3973
3974 StringRef String2 = getTok().getStringContents();
3975 Lex();
3976
3977 TheCondStack.push_back(TheCondState);
3978 TheCondState.TheCond = AsmCond::IfCond;
3979 TheCondState.CondMet = ExpectEqual == (String1 == String2);
3980 TheCondState.Ignore = !TheCondState.CondMet;
3981
3982 return false;
3983 }
3984
3985 /// parseDirectiveIfdef
3986 /// ::= .ifdef symbol
parseDirectiveIfdef(SMLoc DirectiveLoc,bool expect_defined)3987 bool AsmParser::parseDirectiveIfdef(SMLoc DirectiveLoc, bool expect_defined) {
3988 StringRef Name;
3989 TheCondStack.push_back(TheCondState);
3990 TheCondState.TheCond = AsmCond::IfCond;
3991
3992 if (TheCondState.Ignore) {
3993 eatToEndOfStatement();
3994 } else {
3995 if (parseIdentifier(Name))
3996 return TokError("expected identifier after '.ifdef'");
3997
3998 Lex();
3999
4000 MCSymbol *Sym = getContext().lookupSymbol(Name);
4001
4002 if (expect_defined)
4003 TheCondState.CondMet = (Sym && !Sym->isUndefined());
4004 else
4005 TheCondState.CondMet = (!Sym || Sym->isUndefined());
4006 TheCondState.Ignore = !TheCondState.CondMet;
4007 }
4008
4009 return false;
4010 }
4011
4012 /// parseDirectiveElseIf
4013 /// ::= .elseif expression
parseDirectiveElseIf(SMLoc DirectiveLoc)4014 bool AsmParser::parseDirectiveElseIf(SMLoc DirectiveLoc) {
4015 if (TheCondState.TheCond != AsmCond::IfCond &&
4016 TheCondState.TheCond != AsmCond::ElseIfCond)
4017 Error(DirectiveLoc, "Encountered a .elseif that doesn't follow a .if or "
4018 " an .elseif");
4019 TheCondState.TheCond = AsmCond::ElseIfCond;
4020
4021 bool LastIgnoreState = false;
4022 if (!TheCondStack.empty())
4023 LastIgnoreState = TheCondStack.back().Ignore;
4024 if (LastIgnoreState || TheCondState.CondMet) {
4025 TheCondState.Ignore = true;
4026 eatToEndOfStatement();
4027 } else {
4028 int64_t ExprValue;
4029 if (parseAbsoluteExpression(ExprValue))
4030 return true;
4031
4032 if (getLexer().isNot(AsmToken::EndOfStatement))
4033 return TokError("unexpected token in '.elseif' directive");
4034
4035 Lex();
4036 TheCondState.CondMet = ExprValue;
4037 TheCondState.Ignore = !TheCondState.CondMet;
4038 }
4039
4040 return false;
4041 }
4042
4043 /// parseDirectiveElse
4044 /// ::= .else
parseDirectiveElse(SMLoc DirectiveLoc)4045 bool AsmParser::parseDirectiveElse(SMLoc DirectiveLoc) {
4046 if (getLexer().isNot(AsmToken::EndOfStatement))
4047 return TokError("unexpected token in '.else' directive");
4048
4049 Lex();
4050
4051 if (TheCondState.TheCond != AsmCond::IfCond &&
4052 TheCondState.TheCond != AsmCond::ElseIfCond)
4053 Error(DirectiveLoc, "Encountered a .else that doesn't follow a .if or an "
4054 ".elseif");
4055 TheCondState.TheCond = AsmCond::ElseCond;
4056 bool LastIgnoreState = false;
4057 if (!TheCondStack.empty())
4058 LastIgnoreState = TheCondStack.back().Ignore;
4059 if (LastIgnoreState || TheCondState.CondMet)
4060 TheCondState.Ignore = true;
4061 else
4062 TheCondState.Ignore = false;
4063
4064 return false;
4065 }
4066
4067 /// parseDirectiveEnd
4068 /// ::= .end
parseDirectiveEnd(SMLoc DirectiveLoc)4069 bool AsmParser::parseDirectiveEnd(SMLoc DirectiveLoc) {
4070 if (getLexer().isNot(AsmToken::EndOfStatement))
4071 return TokError("unexpected token in '.end' directive");
4072
4073 Lex();
4074
4075 while (Lexer.isNot(AsmToken::Eof))
4076 Lex();
4077
4078 return false;
4079 }
4080
4081 /// parseDirectiveError
4082 /// ::= .err
4083 /// ::= .error [string]
parseDirectiveError(SMLoc L,bool WithMessage)4084 bool AsmParser::parseDirectiveError(SMLoc L, bool WithMessage) {
4085 if (!TheCondStack.empty()) {
4086 if (TheCondStack.back().Ignore) {
4087 eatToEndOfStatement();
4088 return false;
4089 }
4090 }
4091
4092 if (!WithMessage)
4093 return Error(L, ".err encountered");
4094
4095 StringRef Message = ".error directive invoked in source file";
4096 if (Lexer.isNot(AsmToken::EndOfStatement)) {
4097 if (Lexer.isNot(AsmToken::String)) {
4098 TokError(".error argument must be a string");
4099 eatToEndOfStatement();
4100 return true;
4101 }
4102
4103 Message = getTok().getStringContents();
4104 Lex();
4105 }
4106
4107 Error(L, Message);
4108 return true;
4109 }
4110
4111 /// parseDirectiveWarning
4112 /// ::= .warning [string]
parseDirectiveWarning(SMLoc L)4113 bool AsmParser::parseDirectiveWarning(SMLoc L) {
4114 if (!TheCondStack.empty()) {
4115 if (TheCondStack.back().Ignore) {
4116 eatToEndOfStatement();
4117 return false;
4118 }
4119 }
4120
4121 StringRef Message = ".warning directive invoked in source file";
4122 if (Lexer.isNot(AsmToken::EndOfStatement)) {
4123 if (Lexer.isNot(AsmToken::String)) {
4124 TokError(".warning argument must be a string");
4125 eatToEndOfStatement();
4126 return true;
4127 }
4128
4129 Message = getTok().getStringContents();
4130 Lex();
4131 }
4132
4133 Warning(L, Message);
4134 return false;
4135 }
4136
4137 /// parseDirectiveEndIf
4138 /// ::= .endif
parseDirectiveEndIf(SMLoc DirectiveLoc)4139 bool AsmParser::parseDirectiveEndIf(SMLoc DirectiveLoc) {
4140 if (getLexer().isNot(AsmToken::EndOfStatement))
4141 return TokError("unexpected token in '.endif' directive");
4142
4143 Lex();
4144
4145 if ((TheCondState.TheCond == AsmCond::NoCond) || TheCondStack.empty())
4146 Error(DirectiveLoc, "Encountered a .endif that doesn't follow a .if or "
4147 ".else");
4148 if (!TheCondStack.empty()) {
4149 TheCondState = TheCondStack.back();
4150 TheCondStack.pop_back();
4151 }
4152
4153 return false;
4154 }
4155
initializeDirectiveKindMap()4156 void AsmParser::initializeDirectiveKindMap() {
4157 DirectiveKindMap[".set"] = DK_SET;
4158 DirectiveKindMap[".equ"] = DK_EQU;
4159 DirectiveKindMap[".equiv"] = DK_EQUIV;
4160 DirectiveKindMap[".ascii"] = DK_ASCII;
4161 DirectiveKindMap[".asciz"] = DK_ASCIZ;
4162 DirectiveKindMap[".string"] = DK_STRING;
4163 DirectiveKindMap[".byte"] = DK_BYTE;
4164 DirectiveKindMap[".short"] = DK_SHORT;
4165 DirectiveKindMap[".value"] = DK_VALUE;
4166 DirectiveKindMap[".2byte"] = DK_2BYTE;
4167 DirectiveKindMap[".long"] = DK_LONG;
4168 DirectiveKindMap[".int"] = DK_INT;
4169 DirectiveKindMap[".4byte"] = DK_4BYTE;
4170 DirectiveKindMap[".quad"] = DK_QUAD;
4171 DirectiveKindMap[".8byte"] = DK_8BYTE;
4172 DirectiveKindMap[".octa"] = DK_OCTA;
4173 DirectiveKindMap[".single"] = DK_SINGLE;
4174 DirectiveKindMap[".float"] = DK_FLOAT;
4175 DirectiveKindMap[".double"] = DK_DOUBLE;
4176 DirectiveKindMap[".align"] = DK_ALIGN;
4177 DirectiveKindMap[".align32"] = DK_ALIGN32;
4178 DirectiveKindMap[".balign"] = DK_BALIGN;
4179 DirectiveKindMap[".balignw"] = DK_BALIGNW;
4180 DirectiveKindMap[".balignl"] = DK_BALIGNL;
4181 DirectiveKindMap[".p2align"] = DK_P2ALIGN;
4182 DirectiveKindMap[".p2alignw"] = DK_P2ALIGNW;
4183 DirectiveKindMap[".p2alignl"] = DK_P2ALIGNL;
4184 DirectiveKindMap[".org"] = DK_ORG;
4185 DirectiveKindMap[".fill"] = DK_FILL;
4186 DirectiveKindMap[".zero"] = DK_ZERO;
4187 DirectiveKindMap[".extern"] = DK_EXTERN;
4188 DirectiveKindMap[".globl"] = DK_GLOBL;
4189 DirectiveKindMap[".global"] = DK_GLOBAL;
4190 DirectiveKindMap[".lazy_reference"] = DK_LAZY_REFERENCE;
4191 DirectiveKindMap[".no_dead_strip"] = DK_NO_DEAD_STRIP;
4192 DirectiveKindMap[".symbol_resolver"] = DK_SYMBOL_RESOLVER;
4193 DirectiveKindMap[".private_extern"] = DK_PRIVATE_EXTERN;
4194 DirectiveKindMap[".reference"] = DK_REFERENCE;
4195 DirectiveKindMap[".weak_definition"] = DK_WEAK_DEFINITION;
4196 DirectiveKindMap[".weak_reference"] = DK_WEAK_REFERENCE;
4197 DirectiveKindMap[".weak_def_can_be_hidden"] = DK_WEAK_DEF_CAN_BE_HIDDEN;
4198 DirectiveKindMap[".comm"] = DK_COMM;
4199 DirectiveKindMap[".common"] = DK_COMMON;
4200 DirectiveKindMap[".lcomm"] = DK_LCOMM;
4201 DirectiveKindMap[".abort"] = DK_ABORT;
4202 DirectiveKindMap[".include"] = DK_INCLUDE;
4203 DirectiveKindMap[".incbin"] = DK_INCBIN;
4204 DirectiveKindMap[".code16"] = DK_CODE16;
4205 DirectiveKindMap[".code16gcc"] = DK_CODE16GCC;
4206 DirectiveKindMap[".rept"] = DK_REPT;
4207 DirectiveKindMap[".rep"] = DK_REPT;
4208 DirectiveKindMap[".irp"] = DK_IRP;
4209 DirectiveKindMap[".irpc"] = DK_IRPC;
4210 DirectiveKindMap[".endr"] = DK_ENDR;
4211 DirectiveKindMap[".bundle_align_mode"] = DK_BUNDLE_ALIGN_MODE;
4212 DirectiveKindMap[".bundle_lock"] = DK_BUNDLE_LOCK;
4213 DirectiveKindMap[".bundle_unlock"] = DK_BUNDLE_UNLOCK;
4214 DirectiveKindMap[".if"] = DK_IF;
4215 DirectiveKindMap[".ifeq"] = DK_IFEQ;
4216 DirectiveKindMap[".ifge"] = DK_IFGE;
4217 DirectiveKindMap[".ifgt"] = DK_IFGT;
4218 DirectiveKindMap[".ifle"] = DK_IFLE;
4219 DirectiveKindMap[".iflt"] = DK_IFLT;
4220 DirectiveKindMap[".ifne"] = DK_IFNE;
4221 DirectiveKindMap[".ifb"] = DK_IFB;
4222 DirectiveKindMap[".ifnb"] = DK_IFNB;
4223 DirectiveKindMap[".ifc"] = DK_IFC;
4224 DirectiveKindMap[".ifeqs"] = DK_IFEQS;
4225 DirectiveKindMap[".ifnc"] = DK_IFNC;
4226 DirectiveKindMap[".ifnes"] = DK_IFNES;
4227 DirectiveKindMap[".ifdef"] = DK_IFDEF;
4228 DirectiveKindMap[".ifndef"] = DK_IFNDEF;
4229 DirectiveKindMap[".ifnotdef"] = DK_IFNOTDEF;
4230 DirectiveKindMap[".elseif"] = DK_ELSEIF;
4231 DirectiveKindMap[".else"] = DK_ELSE;
4232 DirectiveKindMap[".end"] = DK_END;
4233 DirectiveKindMap[".endif"] = DK_ENDIF;
4234 DirectiveKindMap[".skip"] = DK_SKIP;
4235 DirectiveKindMap[".space"] = DK_SPACE;
4236 DirectiveKindMap[".file"] = DK_FILE;
4237 DirectiveKindMap[".line"] = DK_LINE;
4238 DirectiveKindMap[".loc"] = DK_LOC;
4239 DirectiveKindMap[".stabs"] = DK_STABS;
4240 DirectiveKindMap[".sleb128"] = DK_SLEB128;
4241 DirectiveKindMap[".uleb128"] = DK_ULEB128;
4242 DirectiveKindMap[".cfi_sections"] = DK_CFI_SECTIONS;
4243 DirectiveKindMap[".cfi_startproc"] = DK_CFI_STARTPROC;
4244 DirectiveKindMap[".cfi_endproc"] = DK_CFI_ENDPROC;
4245 DirectiveKindMap[".cfi_def_cfa"] = DK_CFI_DEF_CFA;
4246 DirectiveKindMap[".cfi_def_cfa_offset"] = DK_CFI_DEF_CFA_OFFSET;
4247 DirectiveKindMap[".cfi_adjust_cfa_offset"] = DK_CFI_ADJUST_CFA_OFFSET;
4248 DirectiveKindMap[".cfi_def_cfa_register"] = DK_CFI_DEF_CFA_REGISTER;
4249 DirectiveKindMap[".cfi_offset"] = DK_CFI_OFFSET;
4250 DirectiveKindMap[".cfi_rel_offset"] = DK_CFI_REL_OFFSET;
4251 DirectiveKindMap[".cfi_personality"] = DK_CFI_PERSONALITY;
4252 DirectiveKindMap[".cfi_lsda"] = DK_CFI_LSDA;
4253 DirectiveKindMap[".cfi_remember_state"] = DK_CFI_REMEMBER_STATE;
4254 DirectiveKindMap[".cfi_restore_state"] = DK_CFI_RESTORE_STATE;
4255 DirectiveKindMap[".cfi_same_value"] = DK_CFI_SAME_VALUE;
4256 DirectiveKindMap[".cfi_restore"] = DK_CFI_RESTORE;
4257 DirectiveKindMap[".cfi_escape"] = DK_CFI_ESCAPE;
4258 DirectiveKindMap[".cfi_signal_frame"] = DK_CFI_SIGNAL_FRAME;
4259 DirectiveKindMap[".cfi_undefined"] = DK_CFI_UNDEFINED;
4260 DirectiveKindMap[".cfi_register"] = DK_CFI_REGISTER;
4261 DirectiveKindMap[".cfi_window_save"] = DK_CFI_WINDOW_SAVE;
4262 DirectiveKindMap[".macros_on"] = DK_MACROS_ON;
4263 DirectiveKindMap[".macros_off"] = DK_MACROS_OFF;
4264 DirectiveKindMap[".macro"] = DK_MACRO;
4265 DirectiveKindMap[".exitm"] = DK_EXITM;
4266 DirectiveKindMap[".endm"] = DK_ENDM;
4267 DirectiveKindMap[".endmacro"] = DK_ENDMACRO;
4268 DirectiveKindMap[".purgem"] = DK_PURGEM;
4269 DirectiveKindMap[".err"] = DK_ERR;
4270 DirectiveKindMap[".error"] = DK_ERROR;
4271 DirectiveKindMap[".warning"] = DK_WARNING;
4272 }
4273
parseMacroLikeBody(SMLoc DirectiveLoc)4274 MCAsmMacro *AsmParser::parseMacroLikeBody(SMLoc DirectiveLoc) {
4275 AsmToken EndToken, StartToken = getTok();
4276
4277 unsigned NestLevel = 0;
4278 for (;;) {
4279 // Check whether we have reached the end of the file.
4280 if (getLexer().is(AsmToken::Eof)) {
4281 Error(DirectiveLoc, "no matching '.endr' in definition");
4282 return nullptr;
4283 }
4284
4285 if (Lexer.is(AsmToken::Identifier) &&
4286 (getTok().getIdentifier() == ".rept")) {
4287 ++NestLevel;
4288 }
4289
4290 // Otherwise, check whether we have reached the .endr.
4291 if (Lexer.is(AsmToken::Identifier) && getTok().getIdentifier() == ".endr") {
4292 if (NestLevel == 0) {
4293 EndToken = getTok();
4294 Lex();
4295 if (Lexer.isNot(AsmToken::EndOfStatement)) {
4296 TokError("unexpected token in '.endr' directive");
4297 return nullptr;
4298 }
4299 break;
4300 }
4301 --NestLevel;
4302 }
4303
4304 // Otherwise, scan till the end of the statement.
4305 eatToEndOfStatement();
4306 }
4307
4308 const char *BodyStart = StartToken.getLoc().getPointer();
4309 const char *BodyEnd = EndToken.getLoc().getPointer();
4310 StringRef Body = StringRef(BodyStart, BodyEnd - BodyStart);
4311
4312 // We Are Anonymous.
4313 MacroLikeBodies.emplace_back(StringRef(), Body, MCAsmMacroParameters());
4314 return &MacroLikeBodies.back();
4315 }
4316
instantiateMacroLikeBody(MCAsmMacro * M,SMLoc DirectiveLoc,raw_svector_ostream & OS)4317 void AsmParser::instantiateMacroLikeBody(MCAsmMacro *M, SMLoc DirectiveLoc,
4318 raw_svector_ostream &OS) {
4319 OS << ".endr\n";
4320
4321 std::unique_ptr<MemoryBuffer> Instantiation =
4322 MemoryBuffer::getMemBufferCopy(OS.str(), "<instantiation>");
4323
4324 // Create the macro instantiation object and add to the current macro
4325 // instantiation stack.
4326 MacroInstantiation *MI = new MacroInstantiation(
4327 DirectiveLoc, CurBuffer, getTok().getLoc(), TheCondStack.size());
4328 ActiveMacros.push_back(MI);
4329
4330 // Jump to the macro instantiation and prime the lexer.
4331 CurBuffer = SrcMgr.AddNewSourceBuffer(std::move(Instantiation), SMLoc());
4332 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer());
4333 Lex();
4334 }
4335
4336 /// parseDirectiveRept
4337 /// ::= .rep | .rept count
parseDirectiveRept(SMLoc DirectiveLoc,StringRef Dir)4338 bool AsmParser::parseDirectiveRept(SMLoc DirectiveLoc, StringRef Dir) {
4339 const MCExpr *CountExpr;
4340 SMLoc CountLoc = getTok().getLoc();
4341 if (parseExpression(CountExpr))
4342 return true;
4343
4344 int64_t Count;
4345 if (!CountExpr->evaluateAsAbsolute(Count)) {
4346 eatToEndOfStatement();
4347 return Error(CountLoc, "unexpected token in '" + Dir + "' directive");
4348 }
4349
4350 if (Count < 0)
4351 return Error(CountLoc, "Count is negative");
4352
4353 if (Lexer.isNot(AsmToken::EndOfStatement))
4354 return TokError("unexpected token in '" + Dir + "' directive");
4355
4356 // Eat the end of statement.
4357 Lex();
4358
4359 // Lex the rept definition.
4360 MCAsmMacro *M = parseMacroLikeBody(DirectiveLoc);
4361 if (!M)
4362 return true;
4363
4364 // Macro instantiation is lexical, unfortunately. We construct a new buffer
4365 // to hold the macro body with substitutions.
4366 SmallString<256> Buf;
4367 raw_svector_ostream OS(Buf);
4368 while (Count--) {
4369 // Note that the AtPseudoVariable is disabled for instantiations of .rep(t).
4370 if (expandMacro(OS, M->Body, None, None, false, getTok().getLoc()))
4371 return true;
4372 }
4373 instantiateMacroLikeBody(M, DirectiveLoc, OS);
4374
4375 return false;
4376 }
4377
4378 /// parseDirectiveIrp
4379 /// ::= .irp symbol,values
parseDirectiveIrp(SMLoc DirectiveLoc)4380 bool AsmParser::parseDirectiveIrp(SMLoc DirectiveLoc) {
4381 MCAsmMacroParameter Parameter;
4382
4383 if (parseIdentifier(Parameter.Name))
4384 return TokError("expected identifier in '.irp' directive");
4385
4386 if (Lexer.isNot(AsmToken::Comma))
4387 return TokError("expected comma in '.irp' directive");
4388
4389 Lex();
4390
4391 MCAsmMacroArguments A;
4392 if (parseMacroArguments(nullptr, A))
4393 return true;
4394
4395 // Eat the end of statement.
4396 Lex();
4397
4398 // Lex the irp definition.
4399 MCAsmMacro *M = parseMacroLikeBody(DirectiveLoc);
4400 if (!M)
4401 return true;
4402
4403 // Macro instantiation is lexical, unfortunately. We construct a new buffer
4404 // to hold the macro body with substitutions.
4405 SmallString<256> Buf;
4406 raw_svector_ostream OS(Buf);
4407
4408 for (MCAsmMacroArguments::iterator i = A.begin(), e = A.end(); i != e; ++i) {
4409 // Note that the AtPseudoVariable is enabled for instantiations of .irp.
4410 // This is undocumented, but GAS seems to support it.
4411 if (expandMacro(OS, M->Body, Parameter, *i, true, getTok().getLoc()))
4412 return true;
4413 }
4414
4415 instantiateMacroLikeBody(M, DirectiveLoc, OS);
4416
4417 return false;
4418 }
4419
4420 /// parseDirectiveIrpc
4421 /// ::= .irpc symbol,values
parseDirectiveIrpc(SMLoc DirectiveLoc)4422 bool AsmParser::parseDirectiveIrpc(SMLoc DirectiveLoc) {
4423 MCAsmMacroParameter Parameter;
4424
4425 if (parseIdentifier(Parameter.Name))
4426 return TokError("expected identifier in '.irpc' directive");
4427
4428 if (Lexer.isNot(AsmToken::Comma))
4429 return TokError("expected comma in '.irpc' directive");
4430
4431 Lex();
4432
4433 MCAsmMacroArguments A;
4434 if (parseMacroArguments(nullptr, A))
4435 return true;
4436
4437 if (A.size() != 1 || A.front().size() != 1)
4438 return TokError("unexpected token in '.irpc' directive");
4439
4440 // Eat the end of statement.
4441 Lex();
4442
4443 // Lex the irpc definition.
4444 MCAsmMacro *M = parseMacroLikeBody(DirectiveLoc);
4445 if (!M)
4446 return true;
4447
4448 // Macro instantiation is lexical, unfortunately. We construct a new buffer
4449 // to hold the macro body with substitutions.
4450 SmallString<256> Buf;
4451 raw_svector_ostream OS(Buf);
4452
4453 StringRef Values = A.front().front().getString();
4454 for (std::size_t I = 0, End = Values.size(); I != End; ++I) {
4455 MCAsmMacroArgument Arg;
4456 Arg.emplace_back(AsmToken::Identifier, Values.slice(I, I + 1));
4457
4458 // Note that the AtPseudoVariable is enabled for instantiations of .irpc.
4459 // This is undocumented, but GAS seems to support it.
4460 if (expandMacro(OS, M->Body, Parameter, Arg, true, getTok().getLoc()))
4461 return true;
4462 }
4463
4464 instantiateMacroLikeBody(M, DirectiveLoc, OS);
4465
4466 return false;
4467 }
4468
parseDirectiveEndr(SMLoc DirectiveLoc)4469 bool AsmParser::parseDirectiveEndr(SMLoc DirectiveLoc) {
4470 if (ActiveMacros.empty())
4471 return TokError("unmatched '.endr' directive");
4472
4473 // The only .repl that should get here are the ones created by
4474 // instantiateMacroLikeBody.
4475 assert(getLexer().is(AsmToken::EndOfStatement));
4476
4477 handleMacroExit();
4478 return false;
4479 }
4480
parseDirectiveMSEmit(SMLoc IDLoc,ParseStatementInfo & Info,size_t Len)4481 bool AsmParser::parseDirectiveMSEmit(SMLoc IDLoc, ParseStatementInfo &Info,
4482 size_t Len) {
4483 const MCExpr *Value;
4484 SMLoc ExprLoc = getLexer().getLoc();
4485 if (parseExpression(Value))
4486 return true;
4487 const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value);
4488 if (!MCE)
4489 return Error(ExprLoc, "unexpected expression in _emit");
4490 uint64_t IntValue = MCE->getValue();
4491 if (!isUIntN(8, IntValue) && !isIntN(8, IntValue))
4492 return Error(ExprLoc, "literal value out of range for directive");
4493
4494 Info.AsmRewrites->push_back(AsmRewrite(AOK_Emit, IDLoc, Len));
4495 return false;
4496 }
4497
parseDirectiveMSAlign(SMLoc IDLoc,ParseStatementInfo & Info)4498 bool AsmParser::parseDirectiveMSAlign(SMLoc IDLoc, ParseStatementInfo &Info) {
4499 const MCExpr *Value;
4500 SMLoc ExprLoc = getLexer().getLoc();
4501 if (parseExpression(Value))
4502 return true;
4503 const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value);
4504 if (!MCE)
4505 return Error(ExprLoc, "unexpected expression in align");
4506 uint64_t IntValue = MCE->getValue();
4507 if (!isPowerOf2_64(IntValue))
4508 return Error(ExprLoc, "literal value not a power of two greater then zero");
4509
4510 Info.AsmRewrites->push_back(
4511 AsmRewrite(AOK_Align, IDLoc, 5, Log2_64(IntValue)));
4512 return false;
4513 }
4514
4515 // We are comparing pointers, but the pointers are relative to a single string.
4516 // Thus, this should always be deterministic.
rewritesSort(const AsmRewrite * AsmRewriteA,const AsmRewrite * AsmRewriteB)4517 static int rewritesSort(const AsmRewrite *AsmRewriteA,
4518 const AsmRewrite *AsmRewriteB) {
4519 if (AsmRewriteA->Loc.getPointer() < AsmRewriteB->Loc.getPointer())
4520 return -1;
4521 if (AsmRewriteB->Loc.getPointer() < AsmRewriteA->Loc.getPointer())
4522 return 1;
4523
4524 // It's possible to have a SizeDirective, Imm/ImmPrefix and an Input/Output
4525 // rewrite to the same location. Make sure the SizeDirective rewrite is
4526 // performed first, then the Imm/ImmPrefix and finally the Input/Output. This
4527 // ensures the sort algorithm is stable.
4528 if (AsmRewritePrecedence[AsmRewriteA->Kind] >
4529 AsmRewritePrecedence[AsmRewriteB->Kind])
4530 return -1;
4531
4532 if (AsmRewritePrecedence[AsmRewriteA->Kind] <
4533 AsmRewritePrecedence[AsmRewriteB->Kind])
4534 return 1;
4535 llvm_unreachable("Unstable rewrite sort.");
4536 }
4537
parseMSInlineAsm(void * AsmLoc,std::string & AsmString,unsigned & NumOutputs,unsigned & NumInputs,SmallVectorImpl<std::pair<void *,bool>> & OpDecls,SmallVectorImpl<std::string> & Constraints,SmallVectorImpl<std::string> & Clobbers,const MCInstrInfo * MII,const MCInstPrinter * IP,MCAsmParserSemaCallback & SI)4538 bool AsmParser::parseMSInlineAsm(
4539 void *AsmLoc, std::string &AsmString, unsigned &NumOutputs,
4540 unsigned &NumInputs, SmallVectorImpl<std::pair<void *, bool> > &OpDecls,
4541 SmallVectorImpl<std::string> &Constraints,
4542 SmallVectorImpl<std::string> &Clobbers, const MCInstrInfo *MII,
4543 const MCInstPrinter *IP, MCAsmParserSemaCallback &SI) {
4544 SmallVector<void *, 4> InputDecls;
4545 SmallVector<void *, 4> OutputDecls;
4546 SmallVector<bool, 4> InputDeclsAddressOf;
4547 SmallVector<bool, 4> OutputDeclsAddressOf;
4548 SmallVector<std::string, 4> InputConstraints;
4549 SmallVector<std::string, 4> OutputConstraints;
4550 SmallVector<unsigned, 4> ClobberRegs;
4551
4552 SmallVector<AsmRewrite, 4> AsmStrRewrites;
4553
4554 // Prime the lexer.
4555 Lex();
4556
4557 // While we have input, parse each statement.
4558 unsigned InputIdx = 0;
4559 unsigned OutputIdx = 0;
4560 while (getLexer().isNot(AsmToken::Eof)) {
4561 ParseStatementInfo Info(&AsmStrRewrites);
4562 if (parseStatement(Info, &SI))
4563 return true;
4564
4565 if (Info.ParseError)
4566 return true;
4567
4568 if (Info.Opcode == ~0U)
4569 continue;
4570
4571 const MCInstrDesc &Desc = MII->get(Info.Opcode);
4572
4573 // Build the list of clobbers, outputs and inputs.
4574 for (unsigned i = 1, e = Info.ParsedOperands.size(); i != e; ++i) {
4575 MCParsedAsmOperand &Operand = *Info.ParsedOperands[i];
4576
4577 // Immediate.
4578 if (Operand.isImm())
4579 continue;
4580
4581 // Register operand.
4582 if (Operand.isReg() && !Operand.needAddressOf() &&
4583 !getTargetParser().OmitRegisterFromClobberLists(Operand.getReg())) {
4584 unsigned NumDefs = Desc.getNumDefs();
4585 // Clobber.
4586 if (NumDefs && Operand.getMCOperandNum() < NumDefs)
4587 ClobberRegs.push_back(Operand.getReg());
4588 continue;
4589 }
4590
4591 // Expr/Input or Output.
4592 StringRef SymName = Operand.getSymName();
4593 if (SymName.empty())
4594 continue;
4595
4596 void *OpDecl = Operand.getOpDecl();
4597 if (!OpDecl)
4598 continue;
4599
4600 bool isOutput = (i == 1) && Desc.mayStore();
4601 SMLoc Start = SMLoc::getFromPointer(SymName.data());
4602 if (isOutput) {
4603 ++InputIdx;
4604 OutputDecls.push_back(OpDecl);
4605 OutputDeclsAddressOf.push_back(Operand.needAddressOf());
4606 OutputConstraints.push_back(("=" + Operand.getConstraint()).str());
4607 AsmStrRewrites.push_back(AsmRewrite(AOK_Output, Start, SymName.size()));
4608 } else {
4609 InputDecls.push_back(OpDecl);
4610 InputDeclsAddressOf.push_back(Operand.needAddressOf());
4611 InputConstraints.push_back(Operand.getConstraint().str());
4612 AsmStrRewrites.push_back(AsmRewrite(AOK_Input, Start, SymName.size()));
4613 }
4614 }
4615
4616 // Consider implicit defs to be clobbers. Think of cpuid and push.
4617 ArrayRef<uint16_t> ImpDefs(Desc.getImplicitDefs(),
4618 Desc.getNumImplicitDefs());
4619 ClobberRegs.insert(ClobberRegs.end(), ImpDefs.begin(), ImpDefs.end());
4620 }
4621
4622 // Set the number of Outputs and Inputs.
4623 NumOutputs = OutputDecls.size();
4624 NumInputs = InputDecls.size();
4625
4626 // Set the unique clobbers.
4627 array_pod_sort(ClobberRegs.begin(), ClobberRegs.end());
4628 ClobberRegs.erase(std::unique(ClobberRegs.begin(), ClobberRegs.end()),
4629 ClobberRegs.end());
4630 Clobbers.assign(ClobberRegs.size(), std::string());
4631 for (unsigned I = 0, E = ClobberRegs.size(); I != E; ++I) {
4632 raw_string_ostream OS(Clobbers[I]);
4633 IP->printRegName(OS, ClobberRegs[I]);
4634 }
4635
4636 // Merge the various outputs and inputs. Output are expected first.
4637 if (NumOutputs || NumInputs) {
4638 unsigned NumExprs = NumOutputs + NumInputs;
4639 OpDecls.resize(NumExprs);
4640 Constraints.resize(NumExprs);
4641 for (unsigned i = 0; i < NumOutputs; ++i) {
4642 OpDecls[i] = std::make_pair(OutputDecls[i], OutputDeclsAddressOf[i]);
4643 Constraints[i] = OutputConstraints[i];
4644 }
4645 for (unsigned i = 0, j = NumOutputs; i < NumInputs; ++i, ++j) {
4646 OpDecls[j] = std::make_pair(InputDecls[i], InputDeclsAddressOf[i]);
4647 Constraints[j] = InputConstraints[i];
4648 }
4649 }
4650
4651 // Build the IR assembly string.
4652 std::string AsmStringIR;
4653 raw_string_ostream OS(AsmStringIR);
4654 StringRef ASMString =
4655 SrcMgr.getMemoryBuffer(SrcMgr.getMainFileID())->getBuffer();
4656 const char *AsmStart = ASMString.begin();
4657 const char *AsmEnd = ASMString.end();
4658 array_pod_sort(AsmStrRewrites.begin(), AsmStrRewrites.end(), rewritesSort);
4659 for (const AsmRewrite &AR : AsmStrRewrites) {
4660 AsmRewriteKind Kind = AR.Kind;
4661 if (Kind == AOK_Delete)
4662 continue;
4663
4664 const char *Loc = AR.Loc.getPointer();
4665 assert(Loc >= AsmStart && "Expected Loc to be at or after Start!");
4666
4667 // Emit everything up to the immediate/expression.
4668 if (unsigned Len = Loc - AsmStart)
4669 OS << StringRef(AsmStart, Len);
4670
4671 // Skip the original expression.
4672 if (Kind == AOK_Skip) {
4673 AsmStart = Loc + AR.Len;
4674 continue;
4675 }
4676
4677 unsigned AdditionalSkip = 0;
4678 // Rewrite expressions in $N notation.
4679 switch (Kind) {
4680 default:
4681 break;
4682 case AOK_Imm:
4683 OS << "$$" << AR.Val;
4684 break;
4685 case AOK_ImmPrefix:
4686 OS << "$$";
4687 break;
4688 case AOK_Label:
4689 OS << Ctx.getAsmInfo()->getPrivateLabelPrefix() << AR.Label;
4690 break;
4691 case AOK_Input:
4692 OS << '$' << InputIdx++;
4693 break;
4694 case AOK_Output:
4695 OS << '$' << OutputIdx++;
4696 break;
4697 case AOK_SizeDirective:
4698 switch (AR.Val) {
4699 default: break;
4700 case 8: OS << "byte ptr "; break;
4701 case 16: OS << "word ptr "; break;
4702 case 32: OS << "dword ptr "; break;
4703 case 64: OS << "qword ptr "; break;
4704 case 80: OS << "xword ptr "; break;
4705 case 128: OS << "xmmword ptr "; break;
4706 case 256: OS << "ymmword ptr "; break;
4707 }
4708 break;
4709 case AOK_Emit:
4710 OS << ".byte";
4711 break;
4712 case AOK_Align: {
4713 unsigned Val = AR.Val;
4714 OS << ".align " << Val;
4715
4716 // Skip the original immediate.
4717 assert(Val < 10 && "Expected alignment less then 2^10.");
4718 AdditionalSkip = (Val < 4) ? 2 : Val < 7 ? 3 : 4;
4719 break;
4720 }
4721 case AOK_DotOperator:
4722 // Insert the dot if the user omitted it.
4723 OS.flush();
4724 if (AsmStringIR.back() != '.')
4725 OS << '.';
4726 OS << AR.Val;
4727 break;
4728 }
4729
4730 // Skip the original expression.
4731 AsmStart = Loc + AR.Len + AdditionalSkip;
4732 }
4733
4734 // Emit the remainder of the asm string.
4735 if (AsmStart != AsmEnd)
4736 OS << StringRef(AsmStart, AsmEnd - AsmStart);
4737
4738 AsmString = OS.str();
4739 return false;
4740 }
4741
4742 namespace llvm {
4743 namespace MCParserUtils {
4744
4745 /// Returns whether the given symbol is used anywhere in the given expression,
4746 /// or subexpressions.
isSymbolUsedInExpression(const MCSymbol * Sym,const MCExpr * Value)4747 static bool isSymbolUsedInExpression(const MCSymbol *Sym, const MCExpr *Value) {
4748 switch (Value->getKind()) {
4749 case MCExpr::Binary: {
4750 const MCBinaryExpr *BE = static_cast<const MCBinaryExpr *>(Value);
4751 return isSymbolUsedInExpression(Sym, BE->getLHS()) ||
4752 isSymbolUsedInExpression(Sym, BE->getRHS());
4753 }
4754 case MCExpr::Target:
4755 case MCExpr::Constant:
4756 return false;
4757 case MCExpr::SymbolRef: {
4758 const MCSymbol &S =
4759 static_cast<const MCSymbolRefExpr *>(Value)->getSymbol();
4760 if (S.isVariable())
4761 return isSymbolUsedInExpression(Sym, S.getVariableValue());
4762 return &S == Sym;
4763 }
4764 case MCExpr::Unary:
4765 return isSymbolUsedInExpression(
4766 Sym, static_cast<const MCUnaryExpr *>(Value)->getSubExpr());
4767 }
4768
4769 llvm_unreachable("Unknown expr kind!");
4770 }
4771
parseAssignmentExpression(StringRef Name,bool allow_redef,MCAsmParser & Parser,MCSymbol * & Sym,const MCExpr * & Value)4772 bool parseAssignmentExpression(StringRef Name, bool allow_redef,
4773 MCAsmParser &Parser, MCSymbol *&Sym,
4774 const MCExpr *&Value) {
4775 MCAsmLexer &Lexer = Parser.getLexer();
4776
4777 // FIXME: Use better location, we should use proper tokens.
4778 SMLoc EqualLoc = Lexer.getLoc();
4779
4780 if (Parser.parseExpression(Value)) {
4781 Parser.TokError("missing expression");
4782 Parser.eatToEndOfStatement();
4783 return true;
4784 }
4785
4786 // Note: we don't count b as used in "a = b". This is to allow
4787 // a = b
4788 // b = c
4789
4790 if (Lexer.isNot(AsmToken::EndOfStatement))
4791 return Parser.TokError("unexpected token in assignment");
4792
4793 // Eat the end of statement marker.
4794 Parser.Lex();
4795
4796 // Validate that the LHS is allowed to be a variable (either it has not been
4797 // used as a symbol, or it is an absolute symbol).
4798 Sym = Parser.getContext().lookupSymbol(Name);
4799 if (Sym) {
4800 // Diagnose assignment to a label.
4801 //
4802 // FIXME: Diagnostics. Note the location of the definition as a label.
4803 // FIXME: Diagnose assignment to protected identifier (e.g., register name).
4804 if (isSymbolUsedInExpression(Sym, Value))
4805 return Parser.Error(EqualLoc, "Recursive use of '" + Name + "'");
4806 else if (Sym->isUndefined() && !Sym->isUsed() && !Sym->isVariable())
4807 ; // Allow redefinitions of undefined symbols only used in directives.
4808 else if (Sym->isVariable() && !Sym->isUsed() && allow_redef)
4809 ; // Allow redefinitions of variables that haven't yet been used.
4810 else if (!Sym->isUndefined() && (!Sym->isVariable() || !allow_redef))
4811 return Parser.Error(EqualLoc, "redefinition of '" + Name + "'");
4812 else if (!Sym->isVariable())
4813 return Parser.Error(EqualLoc, "invalid assignment to '" + Name + "'");
4814 else if (!isa<MCConstantExpr>(Sym->getVariableValue()))
4815 return Parser.Error(EqualLoc,
4816 "invalid reassignment of non-absolute variable '" +
4817 Name + "'");
4818
4819 // Don't count these checks as uses.
4820 Sym->setUsed(false);
4821 } else if (Name == ".") {
4822 if (Parser.getStreamer().EmitValueToOffset(Value, 0)) {
4823 Parser.Error(EqualLoc, "expected absolute expression");
4824 Parser.eatToEndOfStatement();
4825 return true;
4826 }
4827 return false;
4828 } else
4829 Sym = Parser.getContext().getOrCreateSymbol(Name);
4830
4831 Sym->setRedefinable(allow_redef);
4832
4833 return false;
4834 }
4835
4836 } // namespace MCParserUtils
4837 } // namespace llvm
4838
4839 /// \brief Create an MCAsmParser instance.
createMCAsmParser(SourceMgr & SM,MCContext & C,MCStreamer & Out,const MCAsmInfo & MAI)4840 MCAsmParser *llvm::createMCAsmParser(SourceMgr &SM, MCContext &C,
4841 MCStreamer &Out, const MCAsmInfo &MAI) {
4842 return new AsmParser(SM, C, Out, MAI);
4843 }
4844