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/SmallString.h"
16 #include "llvm/ADT/STLExtras.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/MCParser/AsmCond.h"
26 #include "llvm/MC/MCParser/AsmLexer.h"
27 #include "llvm/MC/MCParser/MCAsmParser.h"
28 #include "llvm/MC/MCParser/MCParsedAsmOperand.h"
29 #include "llvm/MC/MCRegisterInfo.h"
30 #include "llvm/MC/MCSectionMachO.h"
31 #include "llvm/MC/MCStreamer.h"
32 #include "llvm/MC/MCSymbol.h"
33 #include "llvm/MC/MCTargetAsmParser.h"
34 #include "llvm/Support/CommandLine.h"
35 #include "llvm/Support/ErrorHandling.h"
36 #include "llvm/Support/MathExtras.h"
37 #include "llvm/Support/MemoryBuffer.h"
38 #include "llvm/Support/SourceMgr.h"
39 #include "llvm/Support/raw_ostream.h"
40 #include <cctype>
41 #include <set>
42 #include <string>
43 #include <vector>
44 using namespace llvm;
45
46 static cl::opt<bool>
47 FatalAssemblerWarnings("fatal-assembler-warnings",
48 cl::desc("Consider warnings as error"));
49
~MCAsmParserSemaCallback()50 MCAsmParserSemaCallback::~MCAsmParserSemaCallback() {}
51
52 namespace {
53
54 /// \brief Helper types for tracking macro definitions.
55 typedef std::vector<AsmToken> MCAsmMacroArgument;
56 typedef std::vector<MCAsmMacroArgument> MCAsmMacroArguments;
57 typedef std::pair<StringRef, MCAsmMacroArgument> MCAsmMacroParameter;
58 typedef std::vector<MCAsmMacroParameter> MCAsmMacroParameters;
59
60 struct MCAsmMacro {
61 StringRef Name;
62 StringRef Body;
63 MCAsmMacroParameters Parameters;
64
65 public:
MCAsmMacro__anona135d2a90111::MCAsmMacro66 MCAsmMacro(StringRef N, StringRef B, const MCAsmMacroParameters &P) :
67 Name(N), Body(B), Parameters(P) {}
68
MCAsmMacro__anona135d2a90111::MCAsmMacro69 MCAsmMacro(const MCAsmMacro& Other)
70 : Name(Other.Name), Body(Other.Body), Parameters(Other.Parameters) {}
71 };
72
73 /// \brief Helper class for storing information about an active macro
74 /// instantiation.
75 struct MacroInstantiation {
76 /// The macro being instantiated.
77 const MCAsmMacro *TheMacro;
78
79 /// The macro instantiation with substitutions.
80 MemoryBuffer *Instantiation;
81
82 /// The location of the instantiation.
83 SMLoc InstantiationLoc;
84
85 /// The buffer where parsing should resume upon instantiation completion.
86 int ExitBuffer;
87
88 /// The location where parsing should resume upon instantiation completion.
89 SMLoc ExitLoc;
90
91 public:
92 MacroInstantiation(const MCAsmMacro *M, SMLoc IL, int EB, SMLoc EL,
93 MemoryBuffer *I);
94 };
95
96 struct ParseStatementInfo {
97 /// \brief The parsed operands from the last parsed statement.
98 SmallVector<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__anona135d2a90111::ParseStatementInfo108 ParseStatementInfo() : Opcode(~0U), ParseError(false), AsmRewrites(0) {}
ParseStatementInfo__anona135d2a90111::ParseStatementInfo109 ParseStatementInfo(SmallVectorImpl<AsmRewrite> *rewrites)
110 : Opcode(~0), ParseError(false), AsmRewrites(rewrites) {}
111
~ParseStatementInfo__anona135d2a90111::ParseStatementInfo112 ~ParseStatementInfo() {
113 // Free any parsed operands.
114 for (unsigned i = 0, e = ParsedOperands.size(); i != e; ++i)
115 delete ParsedOperands[i];
116 ParsedOperands.clear();
117 }
118 };
119
120 /// \brief The concrete assembly parser instance.
121 class AsmParser : public MCAsmParser {
122 AsmParser(const AsmParser &) LLVM_DELETED_FUNCTION;
123 void operator=(const AsmParser &) LLVM_DELETED_FUNCTION;
124 private:
125 AsmLexer Lexer;
126 MCContext &Ctx;
127 MCStreamer &Out;
128 const MCAsmInfo &MAI;
129 SourceMgr &SrcMgr;
130 SourceMgr::DiagHandlerTy SavedDiagHandler;
131 void *SavedDiagContext;
132 MCAsmParserExtension *PlatformParser;
133
134 /// This is the current buffer index we're lexing from as managed by the
135 /// SourceMgr object.
136 int CurBuffer;
137
138 AsmCond TheCondState;
139 std::vector<AsmCond> TheCondStack;
140
141 /// \brief maps directive names to handler methods in parser
142 /// extensions. Extensions register themselves in this map by calling
143 /// addDirectiveHandler.
144 StringMap<ExtensionDirectiveHandler> ExtensionDirectiveMap;
145
146 /// \brief Map of currently defined macros.
147 StringMap<MCAsmMacro*> MacroMap;
148
149 /// \brief Stack of active macro instantiations.
150 std::vector<MacroInstantiation*> ActiveMacros;
151
152 /// \brief List of bodies of anonymous macros.
153 std::deque<MCAsmMacro> MacroLikeBodies;
154
155 /// Boolean tracking whether macro substitution is enabled.
156 unsigned MacrosEnabledFlag : 1;
157
158 /// Flag tracking whether any errors have been encountered.
159 unsigned HadError : 1;
160
161 /// The values from the last parsed cpp hash file line comment if any.
162 StringRef CppHashFilename;
163 int64_t CppHashLineNumber;
164 SMLoc CppHashLoc;
165 int CppHashBuf;
166 /// When generating dwarf for assembly source files we need to calculate the
167 /// logical line number based on the last parsed cpp hash file line comment
168 /// and current line. Since this is slow and messes up the SourceMgr's
169 /// cache we save the last info we queried with SrcMgr.FindLineNumber().
170 SMLoc LastQueryIDLoc;
171 int LastQueryBuffer;
172 unsigned LastQueryLine;
173
174 /// AssemblerDialect. ~OU means unset value and use value provided by MAI.
175 unsigned AssemblerDialect;
176
177 /// \brief is Darwin compatibility enabled?
178 bool IsDarwin;
179
180 /// \brief Are we parsing ms-style inline assembly?
181 bool ParsingInlineAsm;
182
183 public:
184 AsmParser(SourceMgr &SM, MCContext &Ctx, MCStreamer &Out,
185 const MCAsmInfo &MAI);
186 virtual ~AsmParser();
187
188 virtual bool Run(bool NoInitialTextSection, bool NoFinalize = false);
189
addDirectiveHandler(StringRef Directive,ExtensionDirectiveHandler Handler)190 virtual void addDirectiveHandler(StringRef Directive,
191 ExtensionDirectiveHandler Handler) {
192 ExtensionDirectiveMap[Directive] = Handler;
193 }
194
195 public:
196 /// @name MCAsmParser Interface
197 /// {
198
getSourceManager()199 virtual SourceMgr &getSourceManager() { return SrcMgr; }
getLexer()200 virtual MCAsmLexer &getLexer() { return Lexer; }
getContext()201 virtual MCContext &getContext() { return Ctx; }
getStreamer()202 virtual MCStreamer &getStreamer() { return Out; }
getAssemblerDialect()203 virtual unsigned getAssemblerDialect() {
204 if (AssemblerDialect == ~0U)
205 return MAI.getAssemblerDialect();
206 else
207 return AssemblerDialect;
208 }
setAssemblerDialect(unsigned i)209 virtual void setAssemblerDialect(unsigned i) {
210 AssemblerDialect = i;
211 }
212
213 virtual bool Warning(SMLoc L, const Twine &Msg,
214 ArrayRef<SMRange> Ranges = None);
215 virtual bool Error(SMLoc L, const Twine &Msg,
216 ArrayRef<SMRange> Ranges = None);
217
218 virtual const AsmToken &Lex();
219
setParsingInlineAsm(bool V)220 void setParsingInlineAsm(bool V) { ParsingInlineAsm = V; }
isParsingInlineAsm()221 bool isParsingInlineAsm() { return ParsingInlineAsm; }
222
223 bool parseMSInlineAsm(void *AsmLoc, std::string &AsmString,
224 unsigned &NumOutputs, unsigned &NumInputs,
225 SmallVectorImpl<std::pair<void *,bool> > &OpDecls,
226 SmallVectorImpl<std::string> &Constraints,
227 SmallVectorImpl<std::string> &Clobbers,
228 const MCInstrInfo *MII,
229 const MCInstPrinter *IP,
230 MCAsmParserSemaCallback &SI);
231
232 bool parseExpression(const MCExpr *&Res);
233 virtual bool parseExpression(const MCExpr *&Res, SMLoc &EndLoc);
234 virtual bool parsePrimaryExpr(const MCExpr *&Res, SMLoc &EndLoc);
235 virtual bool parseParenExpression(const MCExpr *&Res, SMLoc &EndLoc);
236 virtual bool parseAbsoluteExpression(int64_t &Res);
237
238 /// \brief Parse an identifier or string (as a quoted identifier)
239 /// and set \p Res to the identifier contents.
240 virtual bool parseIdentifier(StringRef &Res);
241 virtual void eatToEndOfStatement();
242
243 virtual void checkForValidSection();
244 /// }
245
246 private:
247
248 bool parseStatement(ParseStatementInfo &Info);
249 void eatToEndOfLine();
250 bool parseCppHashLineFilenameComment(const SMLoc &L);
251
252 void checkForBadMacro(SMLoc DirectiveLoc, StringRef Name, StringRef Body,
253 MCAsmMacroParameters Parameters);
254 bool expandMacro(raw_svector_ostream &OS, StringRef Body,
255 const MCAsmMacroParameters &Parameters,
256 const MCAsmMacroArguments &A,
257 const SMLoc &L);
258
259 /// \brief Are macros enabled in the parser?
areMacrosEnabled()260 bool areMacrosEnabled() {return MacrosEnabledFlag;}
261
262 /// \brief Control a flag in the parser that enables or disables macros.
setMacrosEnabled(bool Flag)263 void setMacrosEnabled(bool Flag) {MacrosEnabledFlag = Flag;}
264
265 /// \brief Lookup a previously defined macro.
266 /// \param Name Macro name.
267 /// \returns Pointer to macro. NULL if no such macro was defined.
268 const MCAsmMacro* lookupMacro(StringRef Name);
269
270 /// \brief Define a new macro with the given name and information.
271 void defineMacro(StringRef Name, const MCAsmMacro& Macro);
272
273 /// \brief Undefine a macro. If no such macro was defined, it's a no-op.
274 void undefineMacro(StringRef Name);
275
276 /// \brief Are we inside a macro instantiation?
isInsideMacroInstantiation()277 bool isInsideMacroInstantiation() {return !ActiveMacros.empty();}
278
279 /// \brief Handle entry to macro instantiation.
280 ///
281 /// \param M The macro.
282 /// \param NameLoc Instantiation location.
283 bool handleMacroEntry(const MCAsmMacro *M, SMLoc NameLoc);
284
285 /// \brief Handle exit from macro instantiation.
286 void handleMacroExit();
287
288 /// \brief Extract AsmTokens for a macro argument. If the argument delimiter
289 /// is initially unknown, set it to AsmToken::Eof. It will be set to the
290 /// correct delimiter by the method.
291 bool parseMacroArgument(MCAsmMacroArgument &MA,
292 AsmToken::TokenKind &ArgumentDelimiter);
293
294 /// \brief Parse all macro arguments for a given macro.
295 bool parseMacroArguments(const MCAsmMacro *M, MCAsmMacroArguments &A);
296
297 void printMacroInstantiations();
printMessage(SMLoc Loc,SourceMgr::DiagKind Kind,const Twine & Msg,ArrayRef<SMRange> Ranges=None) const298 void printMessage(SMLoc Loc, SourceMgr::DiagKind Kind, const Twine &Msg,
299 ArrayRef<SMRange> Ranges = None) const {
300 SrcMgr.PrintMessage(Loc, Kind, Msg, Ranges);
301 }
302 static void DiagHandler(const SMDiagnostic &Diag, void *Context);
303
304 /// \brief Enter the specified file. This returns true on failure.
305 bool enterIncludeFile(const std::string &Filename);
306
307 /// \brief Process the specified file for the .incbin directive.
308 /// This returns true on failure.
309 bool processIncbinFile(const std::string &Filename);
310
311 /// \brief Reset the current lexer position to that given by \p Loc. The
312 /// current token is not set; clients should ensure Lex() is called
313 /// subsequently.
314 ///
315 /// \param InBuffer If not -1, should be the known buffer id that contains the
316 /// location.
317 void jumpToLoc(SMLoc Loc, int InBuffer=-1);
318
319 /// \brief Parse up to the end of statement and a return the contents from the
320 /// current token until the end of the statement; the current token on exit
321 /// will be either the EndOfStatement or EOF.
322 virtual StringRef parseStringToEndOfStatement();
323
324 /// \brief Parse until the end of a statement or a comma is encountered,
325 /// return the contents from the current token up to the end or comma.
326 StringRef parseStringToComma();
327
328 bool parseAssignment(StringRef Name, bool allow_redef,
329 bool NoDeadStrip = false);
330
331 bool parseBinOpRHS(unsigned Precedence, const MCExpr *&Res, SMLoc &EndLoc);
332 bool parseParenExpr(const MCExpr *&Res, SMLoc &EndLoc);
333 bool parseBracketExpr(const MCExpr *&Res, SMLoc &EndLoc);
334
335 bool parseRegisterOrRegisterNumber(int64_t &Register, SMLoc DirectiveLoc);
336
337 // Generic (target and platform independent) directive parsing.
338 enum DirectiveKind {
339 DK_NO_DIRECTIVE, // Placeholder
340 DK_SET, DK_EQU, DK_EQUIV, DK_ASCII, DK_ASCIZ, DK_STRING, DK_BYTE, DK_SHORT,
341 DK_VALUE, DK_2BYTE, DK_LONG, DK_INT, DK_4BYTE, DK_QUAD, DK_8BYTE, DK_SINGLE,
342 DK_FLOAT, DK_DOUBLE, DK_ALIGN, DK_ALIGN32, DK_BALIGN, DK_BALIGNW,
343 DK_BALIGNL, DK_P2ALIGN, DK_P2ALIGNW, DK_P2ALIGNL, DK_ORG, DK_FILL, DK_ENDR,
344 DK_BUNDLE_ALIGN_MODE, DK_BUNDLE_LOCK, DK_BUNDLE_UNLOCK,
345 DK_ZERO, DK_EXTERN, DK_GLOBL, DK_GLOBAL,
346 DK_LAZY_REFERENCE, DK_NO_DEAD_STRIP, DK_SYMBOL_RESOLVER, DK_PRIVATE_EXTERN,
347 DK_REFERENCE, DK_WEAK_DEFINITION, DK_WEAK_REFERENCE,
348 DK_WEAK_DEF_CAN_BE_HIDDEN, DK_COMM, DK_COMMON, DK_LCOMM, DK_ABORT,
349 DK_INCLUDE, DK_INCBIN, DK_CODE16, DK_CODE16GCC, DK_REPT, DK_IRP, DK_IRPC,
350 DK_IF, DK_IFB, DK_IFNB, DK_IFC, DK_IFNC, DK_IFDEF, DK_IFNDEF, DK_IFNOTDEF,
351 DK_ELSEIF, DK_ELSE, DK_ENDIF,
352 DK_SPACE, DK_SKIP, DK_FILE, DK_LINE, DK_LOC, DK_STABS,
353 DK_CFI_SECTIONS, DK_CFI_STARTPROC, DK_CFI_ENDPROC, DK_CFI_DEF_CFA,
354 DK_CFI_DEF_CFA_OFFSET, DK_CFI_ADJUST_CFA_OFFSET, DK_CFI_DEF_CFA_REGISTER,
355 DK_CFI_OFFSET, DK_CFI_REL_OFFSET, DK_CFI_PERSONALITY, DK_CFI_LSDA,
356 DK_CFI_REMEMBER_STATE, DK_CFI_RESTORE_STATE, DK_CFI_SAME_VALUE,
357 DK_CFI_RESTORE, DK_CFI_ESCAPE, DK_CFI_SIGNAL_FRAME, DK_CFI_UNDEFINED,
358 DK_CFI_REGISTER, DK_CFI_WINDOW_SAVE,
359 DK_MACROS_ON, DK_MACROS_OFF, DK_MACRO, DK_ENDM, DK_ENDMACRO, DK_PURGEM,
360 DK_SLEB128, DK_ULEB128
361 };
362
363 /// \brief Maps directive name --> DirectiveKind enum, for
364 /// directives parsed by this class.
365 StringMap<DirectiveKind> DirectiveKindMap;
366
367 // ".ascii", ".asciz", ".string"
368 bool parseDirectiveAscii(StringRef IDVal, bool ZeroTerminated);
369 bool parseDirectiveValue(unsigned Size); // ".byte", ".long", ...
370 bool parseDirectiveRealValue(const fltSemantics &); // ".single", ...
371 bool parseDirectiveFill(); // ".fill"
372 bool parseDirectiveZero(); // ".zero"
373 // ".set", ".equ", ".equiv"
374 bool parseDirectiveSet(StringRef IDVal, bool allow_redef);
375 bool parseDirectiveOrg(); // ".org"
376 // ".align{,32}", ".p2align{,w,l}"
377 bool parseDirectiveAlign(bool IsPow2, unsigned ValueSize);
378
379 // ".file", ".line", ".loc", ".stabs"
380 bool parseDirectiveFile(SMLoc DirectiveLoc);
381 bool parseDirectiveLine();
382 bool parseDirectiveLoc();
383 bool parseDirectiveStabs();
384
385 // .cfi directives
386 bool parseDirectiveCFIRegister(SMLoc DirectiveLoc);
387 bool parseDirectiveCFIWindowSave();
388 bool parseDirectiveCFISections();
389 bool parseDirectiveCFIStartProc();
390 bool parseDirectiveCFIEndProc();
391 bool parseDirectiveCFIDefCfaOffset();
392 bool parseDirectiveCFIDefCfa(SMLoc DirectiveLoc);
393 bool parseDirectiveCFIAdjustCfaOffset();
394 bool parseDirectiveCFIDefCfaRegister(SMLoc DirectiveLoc);
395 bool parseDirectiveCFIOffset(SMLoc DirectiveLoc);
396 bool parseDirectiveCFIRelOffset(SMLoc DirectiveLoc);
397 bool parseDirectiveCFIPersonalityOrLsda(bool IsPersonality);
398 bool parseDirectiveCFIRememberState();
399 bool parseDirectiveCFIRestoreState();
400 bool parseDirectiveCFISameValue(SMLoc DirectiveLoc);
401 bool parseDirectiveCFIRestore(SMLoc DirectiveLoc);
402 bool parseDirectiveCFIEscape();
403 bool parseDirectiveCFISignalFrame();
404 bool parseDirectiveCFIUndefined(SMLoc DirectiveLoc);
405
406 // macro directives
407 bool parseDirectivePurgeMacro(SMLoc DirectiveLoc);
408 bool parseDirectiveEndMacro(StringRef Directive);
409 bool parseDirectiveMacro(SMLoc DirectiveLoc);
410 bool parseDirectiveMacrosOnOff(StringRef Directive);
411
412 // ".bundle_align_mode"
413 bool parseDirectiveBundleAlignMode();
414 // ".bundle_lock"
415 bool parseDirectiveBundleLock();
416 // ".bundle_unlock"
417 bool parseDirectiveBundleUnlock();
418
419 // ".space", ".skip"
420 bool parseDirectiveSpace(StringRef IDVal);
421
422 // .sleb128 (Signed=true) and .uleb128 (Signed=false)
423 bool parseDirectiveLEB128(bool Signed);
424
425 /// \brief Parse a directive like ".globl" which
426 /// accepts a single symbol (which should be a label or an external).
427 bool parseDirectiveSymbolAttribute(MCSymbolAttr Attr);
428
429 bool parseDirectiveComm(bool IsLocal); // ".comm" and ".lcomm"
430
431 bool parseDirectiveAbort(); // ".abort"
432 bool parseDirectiveInclude(); // ".include"
433 bool parseDirectiveIncbin(); // ".incbin"
434
435 bool parseDirectiveIf(SMLoc DirectiveLoc); // ".if"
436 // ".ifb" or ".ifnb", depending on ExpectBlank.
437 bool parseDirectiveIfb(SMLoc DirectiveLoc, bool ExpectBlank);
438 // ".ifc" or ".ifnc", depending on ExpectEqual.
439 bool parseDirectiveIfc(SMLoc DirectiveLoc, bool ExpectEqual);
440 // ".ifdef" or ".ifndef", depending on expect_defined
441 bool parseDirectiveIfdef(SMLoc DirectiveLoc, bool expect_defined);
442 bool parseDirectiveElseIf(SMLoc DirectiveLoc); // ".elseif"
443 bool parseDirectiveElse(SMLoc DirectiveLoc); // ".else"
444 bool parseDirectiveEndIf(SMLoc DirectiveLoc); // .endif
445 virtual bool parseEscapedString(std::string &Data);
446
447 const MCExpr *applyModifierToExpr(const MCExpr *E,
448 MCSymbolRefExpr::VariantKind Variant);
449
450 // Macro-like directives
451 MCAsmMacro *parseMacroLikeBody(SMLoc DirectiveLoc);
452 void instantiateMacroLikeBody(MCAsmMacro *M, SMLoc DirectiveLoc,
453 raw_svector_ostream &OS);
454 bool parseDirectiveRept(SMLoc DirectiveLoc); // ".rept"
455 bool parseDirectiveIrp(SMLoc DirectiveLoc); // ".irp"
456 bool parseDirectiveIrpc(SMLoc DirectiveLoc); // ".irpc"
457 bool parseDirectiveEndr(SMLoc DirectiveLoc); // ".endr"
458
459 // "_emit" or "__emit"
460 bool parseDirectiveMSEmit(SMLoc DirectiveLoc, ParseStatementInfo &Info,
461 size_t Len);
462
463 // "align"
464 bool parseDirectiveMSAlign(SMLoc DirectiveLoc, ParseStatementInfo &Info);
465
466 void initializeDirectiveKindMap();
467 };
468 }
469
470 namespace llvm {
471
472 extern MCAsmParserExtension *createDarwinAsmParser();
473 extern MCAsmParserExtension *createELFAsmParser();
474 extern MCAsmParserExtension *createCOFFAsmParser();
475
476 }
477
478 enum { DEFAULT_ADDRSPACE = 0 };
479
AsmParser(SourceMgr & _SM,MCContext & _Ctx,MCStreamer & _Out,const MCAsmInfo & _MAI)480 AsmParser::AsmParser(SourceMgr &_SM, MCContext &_Ctx, MCStreamer &_Out,
481 const MCAsmInfo &_MAI)
482 : Lexer(_MAI), Ctx(_Ctx), Out(_Out), MAI(_MAI), SrcMgr(_SM),
483 PlatformParser(0), CurBuffer(0), MacrosEnabledFlag(true),
484 CppHashLineNumber(0), AssemblerDialect(~0U), IsDarwin(false),
485 ParsingInlineAsm(false) {
486 // Save the old handler.
487 SavedDiagHandler = SrcMgr.getDiagHandler();
488 SavedDiagContext = SrcMgr.getDiagContext();
489 // Set our own handler which calls the saved handler.
490 SrcMgr.setDiagHandler(DiagHandler, this);
491 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer));
492
493 // Initialize the platform / file format parser.
494 //
495 // FIXME: This is a hack, we need to (majorly) cleanup how these objects are
496 // created.
497 if (_MAI.hasMicrosoftFastStdCallMangling()) {
498 PlatformParser = createCOFFAsmParser();
499 PlatformParser->Initialize(*this);
500 } else if (_MAI.hasSubsectionsViaSymbols()) {
501 PlatformParser = createDarwinAsmParser();
502 PlatformParser->Initialize(*this);
503 IsDarwin = true;
504 } else {
505 PlatformParser = createELFAsmParser();
506 PlatformParser->Initialize(*this);
507 }
508
509 initializeDirectiveKindMap();
510 }
511
~AsmParser()512 AsmParser::~AsmParser() {
513 assert(ActiveMacros.empty() && "Unexpected active macro instantiation!");
514
515 // Destroy any macros.
516 for (StringMap<MCAsmMacro *>::iterator it = MacroMap.begin(),
517 ie = MacroMap.end();
518 it != ie; ++it)
519 delete it->getValue();
520
521 delete PlatformParser;
522 }
523
printMacroInstantiations()524 void AsmParser::printMacroInstantiations() {
525 // Print the active macro instantiation stack.
526 for (std::vector<MacroInstantiation *>::const_reverse_iterator
527 it = ActiveMacros.rbegin(),
528 ie = ActiveMacros.rend();
529 it != ie; ++it)
530 printMessage((*it)->InstantiationLoc, SourceMgr::DK_Note,
531 "while in macro instantiation");
532 }
533
Warning(SMLoc L,const Twine & Msg,ArrayRef<SMRange> Ranges)534 bool AsmParser::Warning(SMLoc L, const Twine &Msg, ArrayRef<SMRange> Ranges) {
535 if (FatalAssemblerWarnings)
536 return Error(L, Msg, Ranges);
537 printMessage(L, SourceMgr::DK_Warning, Msg, Ranges);
538 printMacroInstantiations();
539 return false;
540 }
541
Error(SMLoc L,const Twine & Msg,ArrayRef<SMRange> Ranges)542 bool AsmParser::Error(SMLoc L, const Twine &Msg, ArrayRef<SMRange> Ranges) {
543 HadError = true;
544 printMessage(L, SourceMgr::DK_Error, Msg, Ranges);
545 printMacroInstantiations();
546 return true;
547 }
548
enterIncludeFile(const std::string & Filename)549 bool AsmParser::enterIncludeFile(const std::string &Filename) {
550 std::string IncludedFile;
551 int NewBuf = SrcMgr.AddIncludeFile(Filename, Lexer.getLoc(), IncludedFile);
552 if (NewBuf == -1)
553 return true;
554
555 CurBuffer = NewBuf;
556
557 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer));
558
559 return false;
560 }
561
562 /// Process the specified .incbin file by searching for it in the include paths
563 /// then just emitting the byte contents of the file to the streamer. This
564 /// returns true on failure.
processIncbinFile(const std::string & Filename)565 bool AsmParser::processIncbinFile(const std::string &Filename) {
566 std::string IncludedFile;
567 int NewBuf = SrcMgr.AddIncludeFile(Filename, Lexer.getLoc(), IncludedFile);
568 if (NewBuf == -1)
569 return true;
570
571 // Pick up the bytes from the file and emit them.
572 getStreamer().EmitBytes(SrcMgr.getMemoryBuffer(NewBuf)->getBuffer());
573 return false;
574 }
575
jumpToLoc(SMLoc Loc,int InBuffer)576 void AsmParser::jumpToLoc(SMLoc Loc, int InBuffer) {
577 if (InBuffer != -1) {
578 CurBuffer = InBuffer;
579 } else {
580 CurBuffer = SrcMgr.FindBufferContainingLoc(Loc);
581 }
582 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer), Loc.getPointer());
583 }
584
Lex()585 const AsmToken &AsmParser::Lex() {
586 const AsmToken *tok = &Lexer.Lex();
587
588 if (tok->is(AsmToken::Eof)) {
589 // If this is the end of an included file, pop the parent file off the
590 // include stack.
591 SMLoc ParentIncludeLoc = SrcMgr.getParentIncludeLoc(CurBuffer);
592 if (ParentIncludeLoc != SMLoc()) {
593 jumpToLoc(ParentIncludeLoc);
594 tok = &Lexer.Lex();
595 }
596 }
597
598 if (tok->is(AsmToken::Error))
599 Error(Lexer.getErrLoc(), Lexer.getErr());
600
601 return *tok;
602 }
603
Run(bool NoInitialTextSection,bool NoFinalize)604 bool AsmParser::Run(bool NoInitialTextSection, bool NoFinalize) {
605 // Create the initial section, if requested.
606 if (!NoInitialTextSection)
607 Out.InitSections();
608
609 // Prime the lexer.
610 Lex();
611
612 HadError = false;
613 AsmCond StartingCondState = TheCondState;
614
615 // If we are generating dwarf for assembly source files save the initial text
616 // section and generate a .file directive.
617 if (getContext().getGenDwarfForAssembly()) {
618 getContext().setGenDwarfSection(getStreamer().getCurrentSection().first);
619 MCSymbol *SectionStartSym = getContext().CreateTempSymbol();
620 getStreamer().EmitLabel(SectionStartSym);
621 getContext().setGenDwarfSectionStartSym(SectionStartSym);
622 getStreamer().EmitDwarfFileDirective(getContext().nextGenDwarfFileNumber(),
623 StringRef(),
624 getContext().getMainFileName());
625 }
626
627 // While we have input, parse each statement.
628 while (Lexer.isNot(AsmToken::Eof)) {
629 ParseStatementInfo Info;
630 if (!parseStatement(Info))
631 continue;
632
633 // We had an error, validate that one was emitted and recover by skipping to
634 // the next line.
635 assert(HadError && "Parse statement returned an error, but none emitted!");
636 eatToEndOfStatement();
637 }
638
639 if (TheCondState.TheCond != StartingCondState.TheCond ||
640 TheCondState.Ignore != StartingCondState.Ignore)
641 return TokError("unmatched .ifs or .elses");
642
643 // Check to see there are no empty DwarfFile slots.
644 const SmallVectorImpl<MCDwarfFile *> &MCDwarfFiles =
645 getContext().getMCDwarfFiles();
646 for (unsigned i = 1; i < MCDwarfFiles.size(); i++) {
647 if (!MCDwarfFiles[i])
648 TokError("unassigned file number: " + Twine(i) + " for .file directives");
649 }
650
651 // Check to see that all assembler local symbols were actually defined.
652 // Targets that don't do subsections via symbols may not want this, though,
653 // so conservatively exclude them. Only do this if we're finalizing, though,
654 // as otherwise we won't necessarilly have seen everything yet.
655 if (!NoFinalize && MAI.hasSubsectionsViaSymbols()) {
656 const MCContext::SymbolTable &Symbols = getContext().getSymbols();
657 for (MCContext::SymbolTable::const_iterator i = Symbols.begin(),
658 e = Symbols.end();
659 i != e; ++i) {
660 MCSymbol *Sym = i->getValue();
661 // Variable symbols may not be marked as defined, so check those
662 // explicitly. If we know it's a variable, we have a definition for
663 // the purposes of this check.
664 if (Sym->isTemporary() && !Sym->isVariable() && !Sym->isDefined())
665 // FIXME: We would really like to refer back to where the symbol was
666 // first referenced for a source location. We need to add something
667 // to track that. Currently, we just point to the end of the file.
668 printMessage(
669 getLexer().getLoc(), SourceMgr::DK_Error,
670 "assembler local symbol '" + Sym->getName() + "' not defined");
671 }
672 }
673
674 // Finalize the output stream if there are no errors and if the client wants
675 // us to.
676 if (!HadError && !NoFinalize)
677 Out.Finish();
678
679 return HadError;
680 }
681
checkForValidSection()682 void AsmParser::checkForValidSection() {
683 if (!ParsingInlineAsm && !getStreamer().getCurrentSection().first) {
684 TokError("expected section directive before assembly directive");
685 Out.InitToTextSection();
686 }
687 }
688
689 /// \brief Throw away the rest of the line for testing purposes.
eatToEndOfStatement()690 void AsmParser::eatToEndOfStatement() {
691 while (Lexer.isNot(AsmToken::EndOfStatement) && Lexer.isNot(AsmToken::Eof))
692 Lex();
693
694 // Eat EOL.
695 if (Lexer.is(AsmToken::EndOfStatement))
696 Lex();
697 }
698
parseStringToEndOfStatement()699 StringRef AsmParser::parseStringToEndOfStatement() {
700 const char *Start = getTok().getLoc().getPointer();
701
702 while (Lexer.isNot(AsmToken::EndOfStatement) && Lexer.isNot(AsmToken::Eof))
703 Lex();
704
705 const char *End = getTok().getLoc().getPointer();
706 return StringRef(Start, End - Start);
707 }
708
parseStringToComma()709 StringRef AsmParser::parseStringToComma() {
710 const char *Start = getTok().getLoc().getPointer();
711
712 while (Lexer.isNot(AsmToken::EndOfStatement) &&
713 Lexer.isNot(AsmToken::Comma) && Lexer.isNot(AsmToken::Eof))
714 Lex();
715
716 const char *End = getTok().getLoc().getPointer();
717 return StringRef(Start, End - Start);
718 }
719
720 /// \brief Parse a paren expression and return it.
721 /// NOTE: This assumes the leading '(' has already been consumed.
722 ///
723 /// parenexpr ::= expr)
724 ///
parseParenExpr(const MCExpr * & Res,SMLoc & EndLoc)725 bool AsmParser::parseParenExpr(const MCExpr *&Res, SMLoc &EndLoc) {
726 if (parseExpression(Res))
727 return true;
728 if (Lexer.isNot(AsmToken::RParen))
729 return TokError("expected ')' in parentheses expression");
730 EndLoc = Lexer.getTok().getEndLoc();
731 Lex();
732 return false;
733 }
734
735 /// \brief Parse a bracket expression and return it.
736 /// NOTE: This assumes the leading '[' has already been consumed.
737 ///
738 /// bracketexpr ::= expr]
739 ///
parseBracketExpr(const MCExpr * & Res,SMLoc & EndLoc)740 bool AsmParser::parseBracketExpr(const MCExpr *&Res, SMLoc &EndLoc) {
741 if (parseExpression(Res))
742 return true;
743 if (Lexer.isNot(AsmToken::RBrac))
744 return TokError("expected ']' in brackets expression");
745 EndLoc = Lexer.getTok().getEndLoc();
746 Lex();
747 return false;
748 }
749
750 /// \brief Parse a primary expression and return it.
751 /// primaryexpr ::= (parenexpr
752 /// primaryexpr ::= symbol
753 /// primaryexpr ::= number
754 /// primaryexpr ::= '.'
755 /// primaryexpr ::= ~,+,- primaryexpr
parsePrimaryExpr(const MCExpr * & Res,SMLoc & EndLoc)756 bool AsmParser::parsePrimaryExpr(const MCExpr *&Res, SMLoc &EndLoc) {
757 SMLoc FirstTokenLoc = getLexer().getLoc();
758 AsmToken::TokenKind FirstTokenKind = Lexer.getKind();
759 switch (FirstTokenKind) {
760 default:
761 return TokError("unknown token in expression");
762 // If we have an error assume that we've already handled it.
763 case AsmToken::Error:
764 return true;
765 case AsmToken::Exclaim:
766 Lex(); // Eat the operator.
767 if (parsePrimaryExpr(Res, EndLoc))
768 return true;
769 Res = MCUnaryExpr::CreateLNot(Res, getContext());
770 return false;
771 case AsmToken::Dollar:
772 case AsmToken::At:
773 case AsmToken::String:
774 case AsmToken::Identifier: {
775 StringRef Identifier;
776 if (parseIdentifier(Identifier)) {
777 if (FirstTokenKind == AsmToken::Dollar) {
778 if (Lexer.getMAI().getDollarIsPC()) {
779 // This is a '$' reference, which references the current PC. Emit a
780 // temporary label to the streamer and refer to it.
781 MCSymbol *Sym = Ctx.CreateTempSymbol();
782 Out.EmitLabel(Sym);
783 Res = MCSymbolRefExpr::Create(Sym, MCSymbolRefExpr::VK_None,
784 getContext());
785 EndLoc = FirstTokenLoc;
786 return false;
787 } else
788 return Error(FirstTokenLoc, "invalid token in expression");
789 return true;
790 }
791 }
792
793 EndLoc = SMLoc::getFromPointer(Identifier.end());
794
795 // This is a symbol reference.
796 StringRef SymbolName = Identifier;
797 MCSymbolRefExpr::VariantKind Variant = MCSymbolRefExpr::VK_None;
798 std::pair<StringRef, StringRef> Split = Identifier.split('@');
799
800 // Lookup the symbol variant if used.
801 if (Split.first.size() != Identifier.size()) {
802 Variant = MCSymbolRefExpr::getVariantKindForName(Split.second);
803 if (Variant != MCSymbolRefExpr::VK_Invalid) {
804 SymbolName = Split.first;
805 } else if (MAI.doesAllowAtInName()) {
806 Variant = MCSymbolRefExpr::VK_None;
807 } else {
808 Variant = MCSymbolRefExpr::VK_None;
809 return TokError("invalid variant '" + Split.second + "'");
810 }
811 }
812
813 MCSymbol *Sym = getContext().GetOrCreateSymbol(SymbolName);
814
815 // If this is an absolute variable reference, substitute it now to preserve
816 // semantics in the face of reassignment.
817 if (Sym->isVariable() && isa<MCConstantExpr>(Sym->getVariableValue())) {
818 if (Variant)
819 return Error(EndLoc, "unexpected modifier on variable reference");
820
821 Res = Sym->getVariableValue();
822 return false;
823 }
824
825 // Otherwise create a symbol ref.
826 Res = MCSymbolRefExpr::Create(Sym, Variant, getContext());
827 return false;
828 }
829 case AsmToken::Integer: {
830 SMLoc Loc = getTok().getLoc();
831 int64_t IntVal = getTok().getIntVal();
832 Res = MCConstantExpr::Create(IntVal, getContext());
833 EndLoc = Lexer.getTok().getEndLoc();
834 Lex(); // Eat token.
835 // Look for 'b' or 'f' following an Integer as a directional label
836 if (Lexer.getKind() == AsmToken::Identifier) {
837 StringRef IDVal = getTok().getString();
838 // Lookup the symbol variant if used.
839 std::pair<StringRef, StringRef> Split = IDVal.split('@');
840 MCSymbolRefExpr::VariantKind Variant = MCSymbolRefExpr::VK_None;
841 if (Split.first.size() != IDVal.size()) {
842 Variant = MCSymbolRefExpr::getVariantKindForName(Split.second);
843 if (Variant == MCSymbolRefExpr::VK_Invalid) {
844 Variant = MCSymbolRefExpr::VK_None;
845 return TokError("invalid variant '" + Split.second + "'");
846 }
847 IDVal = Split.first;
848 }
849 if (IDVal == "f" || IDVal == "b") {
850 MCSymbol *Sym =
851 Ctx.GetDirectionalLocalSymbol(IntVal, IDVal == "f" ? 1 : 0);
852 Res = MCSymbolRefExpr::Create(Sym, Variant, getContext());
853 if (IDVal == "b" && Sym->isUndefined())
854 return Error(Loc, "invalid reference to undefined symbol");
855 EndLoc = Lexer.getTok().getEndLoc();
856 Lex(); // Eat identifier.
857 }
858 }
859 return false;
860 }
861 case AsmToken::Real: {
862 APFloat RealVal(APFloat::IEEEdouble, getTok().getString());
863 uint64_t IntVal = RealVal.bitcastToAPInt().getZExtValue();
864 Res = MCConstantExpr::Create(IntVal, getContext());
865 EndLoc = Lexer.getTok().getEndLoc();
866 Lex(); // Eat token.
867 return false;
868 }
869 case AsmToken::Dot: {
870 // This is a '.' reference, which references the current PC. Emit a
871 // temporary label to the streamer and refer to it.
872 MCSymbol *Sym = Ctx.CreateTempSymbol();
873 Out.EmitLabel(Sym);
874 Res = MCSymbolRefExpr::Create(Sym, MCSymbolRefExpr::VK_None, getContext());
875 EndLoc = Lexer.getTok().getEndLoc();
876 Lex(); // Eat identifier.
877 return false;
878 }
879 case AsmToken::LParen:
880 Lex(); // Eat the '('.
881 return parseParenExpr(Res, EndLoc);
882 case AsmToken::LBrac:
883 if (!PlatformParser->HasBracketExpressions())
884 return TokError("brackets expression not supported on this target");
885 Lex(); // Eat the '['.
886 return parseBracketExpr(Res, EndLoc);
887 case AsmToken::Minus:
888 Lex(); // Eat the operator.
889 if (parsePrimaryExpr(Res, EndLoc))
890 return true;
891 Res = MCUnaryExpr::CreateMinus(Res, getContext());
892 return false;
893 case AsmToken::Plus:
894 Lex(); // Eat the operator.
895 if (parsePrimaryExpr(Res, EndLoc))
896 return true;
897 Res = MCUnaryExpr::CreatePlus(Res, getContext());
898 return false;
899 case AsmToken::Tilde:
900 Lex(); // Eat the operator.
901 if (parsePrimaryExpr(Res, EndLoc))
902 return true;
903 Res = MCUnaryExpr::CreateNot(Res, getContext());
904 return false;
905 }
906 }
907
parseExpression(const MCExpr * & Res)908 bool AsmParser::parseExpression(const MCExpr *&Res) {
909 SMLoc EndLoc;
910 return parseExpression(Res, EndLoc);
911 }
912
913 const MCExpr *
applyModifierToExpr(const MCExpr * E,MCSymbolRefExpr::VariantKind Variant)914 AsmParser::applyModifierToExpr(const MCExpr *E,
915 MCSymbolRefExpr::VariantKind Variant) {
916 // Ask the target implementation about this expression first.
917 const MCExpr *NewE = getTargetParser().applyModifierToExpr(E, Variant, Ctx);
918 if (NewE)
919 return NewE;
920 // Recurse over the given expression, rebuilding it to apply the given variant
921 // if there is exactly one symbol.
922 switch (E->getKind()) {
923 case MCExpr::Target:
924 case MCExpr::Constant:
925 return 0;
926
927 case MCExpr::SymbolRef: {
928 const MCSymbolRefExpr *SRE = cast<MCSymbolRefExpr>(E);
929
930 if (SRE->getKind() != MCSymbolRefExpr::VK_None) {
931 TokError("invalid variant on expression '" + getTok().getIdentifier() +
932 "' (already modified)");
933 return E;
934 }
935
936 return MCSymbolRefExpr::Create(&SRE->getSymbol(), Variant, getContext());
937 }
938
939 case MCExpr::Unary: {
940 const MCUnaryExpr *UE = cast<MCUnaryExpr>(E);
941 const MCExpr *Sub = applyModifierToExpr(UE->getSubExpr(), Variant);
942 if (!Sub)
943 return 0;
944 return MCUnaryExpr::Create(UE->getOpcode(), Sub, getContext());
945 }
946
947 case MCExpr::Binary: {
948 const MCBinaryExpr *BE = cast<MCBinaryExpr>(E);
949 const MCExpr *LHS = applyModifierToExpr(BE->getLHS(), Variant);
950 const MCExpr *RHS = applyModifierToExpr(BE->getRHS(), Variant);
951
952 if (!LHS && !RHS)
953 return 0;
954
955 if (!LHS)
956 LHS = BE->getLHS();
957 if (!RHS)
958 RHS = BE->getRHS();
959
960 return MCBinaryExpr::Create(BE->getOpcode(), LHS, RHS, getContext());
961 }
962 }
963
964 llvm_unreachable("Invalid expression kind!");
965 }
966
967 /// \brief Parse an expression and return it.
968 ///
969 /// expr ::= expr &&,|| expr -> lowest.
970 /// expr ::= expr |,^,&,! expr
971 /// expr ::= expr ==,!=,<>,<,<=,>,>= expr
972 /// expr ::= expr <<,>> expr
973 /// expr ::= expr +,- expr
974 /// expr ::= expr *,/,% expr -> highest.
975 /// expr ::= primaryexpr
976 ///
parseExpression(const MCExpr * & Res,SMLoc & EndLoc)977 bool AsmParser::parseExpression(const MCExpr *&Res, SMLoc &EndLoc) {
978 // Parse the expression.
979 Res = 0;
980 if (parsePrimaryExpr(Res, EndLoc) || parseBinOpRHS(1, Res, EndLoc))
981 return true;
982
983 // As a special case, we support 'a op b @ modifier' by rewriting the
984 // expression to include the modifier. This is inefficient, but in general we
985 // expect users to use 'a@modifier op b'.
986 if (Lexer.getKind() == AsmToken::At) {
987 Lex();
988
989 if (Lexer.isNot(AsmToken::Identifier))
990 return TokError("unexpected symbol modifier following '@'");
991
992 MCSymbolRefExpr::VariantKind Variant =
993 MCSymbolRefExpr::getVariantKindForName(getTok().getIdentifier());
994 if (Variant == MCSymbolRefExpr::VK_Invalid)
995 return TokError("invalid variant '" + getTok().getIdentifier() + "'");
996
997 const MCExpr *ModifiedRes = applyModifierToExpr(Res, Variant);
998 if (!ModifiedRes) {
999 return TokError("invalid modifier '" + getTok().getIdentifier() +
1000 "' (no symbols present)");
1001 }
1002
1003 Res = ModifiedRes;
1004 Lex();
1005 }
1006
1007 // Try to constant fold it up front, if possible.
1008 int64_t Value;
1009 if (Res->EvaluateAsAbsolute(Value))
1010 Res = MCConstantExpr::Create(Value, getContext());
1011
1012 return false;
1013 }
1014
parseParenExpression(const MCExpr * & Res,SMLoc & EndLoc)1015 bool AsmParser::parseParenExpression(const MCExpr *&Res, SMLoc &EndLoc) {
1016 Res = 0;
1017 return parseParenExpr(Res, EndLoc) || parseBinOpRHS(1, Res, EndLoc);
1018 }
1019
parseAbsoluteExpression(int64_t & Res)1020 bool AsmParser::parseAbsoluteExpression(int64_t &Res) {
1021 const MCExpr *Expr;
1022
1023 SMLoc StartLoc = Lexer.getLoc();
1024 if (parseExpression(Expr))
1025 return true;
1026
1027 if (!Expr->EvaluateAsAbsolute(Res))
1028 return Error(StartLoc, "expected absolute expression");
1029
1030 return false;
1031 }
1032
getBinOpPrecedence(AsmToken::TokenKind K,MCBinaryExpr::Opcode & Kind)1033 static unsigned getBinOpPrecedence(AsmToken::TokenKind K,
1034 MCBinaryExpr::Opcode &Kind) {
1035 switch (K) {
1036 default:
1037 return 0; // not a binop.
1038
1039 // Lowest Precedence: &&, ||
1040 case AsmToken::AmpAmp:
1041 Kind = MCBinaryExpr::LAnd;
1042 return 1;
1043 case AsmToken::PipePipe:
1044 Kind = MCBinaryExpr::LOr;
1045 return 1;
1046
1047 // Low Precedence: |, &, ^
1048 //
1049 // FIXME: gas seems to support '!' as an infix operator?
1050 case AsmToken::Pipe:
1051 Kind = MCBinaryExpr::Or;
1052 return 2;
1053 case AsmToken::Caret:
1054 Kind = MCBinaryExpr::Xor;
1055 return 2;
1056 case AsmToken::Amp:
1057 Kind = MCBinaryExpr::And;
1058 return 2;
1059
1060 // Low Intermediate Precedence: ==, !=, <>, <, <=, >, >=
1061 case AsmToken::EqualEqual:
1062 Kind = MCBinaryExpr::EQ;
1063 return 3;
1064 case AsmToken::ExclaimEqual:
1065 case AsmToken::LessGreater:
1066 Kind = MCBinaryExpr::NE;
1067 return 3;
1068 case AsmToken::Less:
1069 Kind = MCBinaryExpr::LT;
1070 return 3;
1071 case AsmToken::LessEqual:
1072 Kind = MCBinaryExpr::LTE;
1073 return 3;
1074 case AsmToken::Greater:
1075 Kind = MCBinaryExpr::GT;
1076 return 3;
1077 case AsmToken::GreaterEqual:
1078 Kind = MCBinaryExpr::GTE;
1079 return 3;
1080
1081 // Intermediate Precedence: <<, >>
1082 case AsmToken::LessLess:
1083 Kind = MCBinaryExpr::Shl;
1084 return 4;
1085 case AsmToken::GreaterGreater:
1086 Kind = MCBinaryExpr::Shr;
1087 return 4;
1088
1089 // High Intermediate Precedence: +, -
1090 case AsmToken::Plus:
1091 Kind = MCBinaryExpr::Add;
1092 return 5;
1093 case AsmToken::Minus:
1094 Kind = MCBinaryExpr::Sub;
1095 return 5;
1096
1097 // Highest Precedence: *, /, %
1098 case AsmToken::Star:
1099 Kind = MCBinaryExpr::Mul;
1100 return 6;
1101 case AsmToken::Slash:
1102 Kind = MCBinaryExpr::Div;
1103 return 6;
1104 case AsmToken::Percent:
1105 Kind = MCBinaryExpr::Mod;
1106 return 6;
1107 }
1108 }
1109
1110 /// \brief Parse all binary operators with precedence >= 'Precedence'.
1111 /// Res contains the LHS of the expression on input.
parseBinOpRHS(unsigned Precedence,const MCExpr * & Res,SMLoc & EndLoc)1112 bool AsmParser::parseBinOpRHS(unsigned Precedence, const MCExpr *&Res,
1113 SMLoc &EndLoc) {
1114 while (1) {
1115 MCBinaryExpr::Opcode Kind = MCBinaryExpr::Add;
1116 unsigned TokPrec = getBinOpPrecedence(Lexer.getKind(), Kind);
1117
1118 // If the next token is lower precedence than we are allowed to eat, return
1119 // successfully with what we ate already.
1120 if (TokPrec < Precedence)
1121 return false;
1122
1123 Lex();
1124
1125 // Eat the next primary expression.
1126 const MCExpr *RHS;
1127 if (parsePrimaryExpr(RHS, EndLoc))
1128 return true;
1129
1130 // If BinOp binds less tightly with RHS than the operator after RHS, let
1131 // the pending operator take RHS as its LHS.
1132 MCBinaryExpr::Opcode Dummy;
1133 unsigned NextTokPrec = getBinOpPrecedence(Lexer.getKind(), Dummy);
1134 if (TokPrec < NextTokPrec && parseBinOpRHS(TokPrec + 1, RHS, EndLoc))
1135 return true;
1136
1137 // Merge LHS and RHS according to operator.
1138 Res = MCBinaryExpr::Create(Kind, Res, RHS, getContext());
1139 }
1140 }
1141
1142 /// ParseStatement:
1143 /// ::= EndOfStatement
1144 /// ::= Label* Directive ...Operands... EndOfStatement
1145 /// ::= Label* Identifier OperandList* EndOfStatement
parseStatement(ParseStatementInfo & Info)1146 bool AsmParser::parseStatement(ParseStatementInfo &Info) {
1147 if (Lexer.is(AsmToken::EndOfStatement)) {
1148 Out.AddBlankLine();
1149 Lex();
1150 return false;
1151 }
1152
1153 // Statements always start with an identifier or are a full line comment.
1154 AsmToken ID = getTok();
1155 SMLoc IDLoc = ID.getLoc();
1156 StringRef IDVal;
1157 int64_t LocalLabelVal = -1;
1158 // A full line comment is a '#' as the first token.
1159 if (Lexer.is(AsmToken::Hash))
1160 return parseCppHashLineFilenameComment(IDLoc);
1161
1162 // Allow an integer followed by a ':' as a directional local label.
1163 if (Lexer.is(AsmToken::Integer)) {
1164 LocalLabelVal = getTok().getIntVal();
1165 if (LocalLabelVal < 0) {
1166 if (!TheCondState.Ignore)
1167 return TokError("unexpected token at start of statement");
1168 IDVal = "";
1169 } else {
1170 IDVal = getTok().getString();
1171 Lex(); // Consume the integer token to be used as an identifier token.
1172 if (Lexer.getKind() != AsmToken::Colon) {
1173 if (!TheCondState.Ignore)
1174 return TokError("unexpected token at start of statement");
1175 }
1176 }
1177 } else if (Lexer.is(AsmToken::Dot)) {
1178 // Treat '.' as a valid identifier in this context.
1179 Lex();
1180 IDVal = ".";
1181 } else if (parseIdentifier(IDVal)) {
1182 if (!TheCondState.Ignore)
1183 return TokError("unexpected token at start of statement");
1184 IDVal = "";
1185 }
1186
1187 // Handle conditional assembly here before checking for skipping. We
1188 // have to do this so that .endif isn't skipped in a ".if 0" block for
1189 // example.
1190 StringMap<DirectiveKind>::const_iterator DirKindIt =
1191 DirectiveKindMap.find(IDVal);
1192 DirectiveKind DirKind = (DirKindIt == DirectiveKindMap.end())
1193 ? DK_NO_DIRECTIVE
1194 : DirKindIt->getValue();
1195 switch (DirKind) {
1196 default:
1197 break;
1198 case DK_IF:
1199 return parseDirectiveIf(IDLoc);
1200 case DK_IFB:
1201 return parseDirectiveIfb(IDLoc, true);
1202 case DK_IFNB:
1203 return parseDirectiveIfb(IDLoc, false);
1204 case DK_IFC:
1205 return parseDirectiveIfc(IDLoc, true);
1206 case DK_IFNC:
1207 return parseDirectiveIfc(IDLoc, false);
1208 case DK_IFDEF:
1209 return parseDirectiveIfdef(IDLoc, true);
1210 case DK_IFNDEF:
1211 case DK_IFNOTDEF:
1212 return parseDirectiveIfdef(IDLoc, false);
1213 case DK_ELSEIF:
1214 return parseDirectiveElseIf(IDLoc);
1215 case DK_ELSE:
1216 return parseDirectiveElse(IDLoc);
1217 case DK_ENDIF:
1218 return parseDirectiveEndIf(IDLoc);
1219 }
1220
1221 // Ignore the statement if in the middle of inactive conditional
1222 // (e.g. ".if 0").
1223 if (TheCondState.Ignore) {
1224 eatToEndOfStatement();
1225 return false;
1226 }
1227
1228 // FIXME: Recurse on local labels?
1229
1230 // See what kind of statement we have.
1231 switch (Lexer.getKind()) {
1232 case AsmToken::Colon: {
1233 checkForValidSection();
1234
1235 // identifier ':' -> Label.
1236 Lex();
1237
1238 // Diagnose attempt to use '.' as a label.
1239 if (IDVal == ".")
1240 return Error(IDLoc, "invalid use of pseudo-symbol '.' as a label");
1241
1242 // Diagnose attempt to use a variable as a label.
1243 //
1244 // FIXME: Diagnostics. Note the location of the definition as a label.
1245 // FIXME: This doesn't diagnose assignment to a symbol which has been
1246 // implicitly marked as external.
1247 MCSymbol *Sym;
1248 if (LocalLabelVal == -1)
1249 Sym = getContext().GetOrCreateSymbol(IDVal);
1250 else
1251 Sym = Ctx.CreateDirectionalLocalSymbol(LocalLabelVal);
1252 if (!Sym->isUndefined() || Sym->isVariable())
1253 return Error(IDLoc, "invalid symbol redefinition");
1254
1255 // Emit the label.
1256 if (!ParsingInlineAsm)
1257 Out.EmitLabel(Sym);
1258
1259 // If we are generating dwarf for assembly source files then gather the
1260 // info to make a dwarf label entry for this label if needed.
1261 if (getContext().getGenDwarfForAssembly())
1262 MCGenDwarfLabelEntry::Make(Sym, &getStreamer(), getSourceManager(),
1263 IDLoc);
1264
1265 getTargetParser().onLabelParsed(Sym);
1266
1267 // Consume any end of statement token, if present, to avoid spurious
1268 // AddBlankLine calls().
1269 if (Lexer.is(AsmToken::EndOfStatement)) {
1270 Lex();
1271 if (Lexer.is(AsmToken::Eof))
1272 return false;
1273 }
1274
1275 return false;
1276 }
1277
1278 case AsmToken::Equal:
1279 // identifier '=' ... -> assignment statement
1280 Lex();
1281
1282 return parseAssignment(IDVal, true);
1283
1284 default: // Normal instruction or directive.
1285 break;
1286 }
1287
1288 // If macros are enabled, check to see if this is a macro instantiation.
1289 if (areMacrosEnabled())
1290 if (const MCAsmMacro *M = lookupMacro(IDVal)) {
1291 return handleMacroEntry(M, IDLoc);
1292 }
1293
1294 // Otherwise, we have a normal instruction or directive.
1295
1296 // Directives start with "."
1297 if (IDVal[0] == '.' && IDVal != ".") {
1298 // There are several entities interested in parsing directives:
1299 //
1300 // 1. The target-specific assembly parser. Some directives are target
1301 // specific or may potentially behave differently on certain targets.
1302 // 2. Asm parser extensions. For example, platform-specific parsers
1303 // (like the ELF parser) register themselves as extensions.
1304 // 3. The generic directive parser implemented by this class. These are
1305 // all the directives that behave in a target and platform independent
1306 // manner, or at least have a default behavior that's shared between
1307 // all targets and platforms.
1308
1309 // First query the target-specific parser. It will return 'true' if it
1310 // isn't interested in this directive.
1311 if (!getTargetParser().ParseDirective(ID))
1312 return false;
1313
1314 // Next, check the extention directive map to see if any extension has
1315 // registered itself to parse this directive.
1316 std::pair<MCAsmParserExtension *, DirectiveHandler> Handler =
1317 ExtensionDirectiveMap.lookup(IDVal);
1318 if (Handler.first)
1319 return (*Handler.second)(Handler.first, IDVal, IDLoc);
1320
1321 // Finally, if no one else is interested in this directive, it must be
1322 // generic and familiar to this class.
1323 switch (DirKind) {
1324 default:
1325 break;
1326 case DK_SET:
1327 case DK_EQU:
1328 return parseDirectiveSet(IDVal, true);
1329 case DK_EQUIV:
1330 return parseDirectiveSet(IDVal, false);
1331 case DK_ASCII:
1332 return parseDirectiveAscii(IDVal, false);
1333 case DK_ASCIZ:
1334 case DK_STRING:
1335 return parseDirectiveAscii(IDVal, true);
1336 case DK_BYTE:
1337 return parseDirectiveValue(1);
1338 case DK_SHORT:
1339 case DK_VALUE:
1340 case DK_2BYTE:
1341 return parseDirectiveValue(2);
1342 case DK_LONG:
1343 case DK_INT:
1344 case DK_4BYTE:
1345 return parseDirectiveValue(4);
1346 case DK_QUAD:
1347 case DK_8BYTE:
1348 return parseDirectiveValue(8);
1349 case DK_SINGLE:
1350 case DK_FLOAT:
1351 return parseDirectiveRealValue(APFloat::IEEEsingle);
1352 case DK_DOUBLE:
1353 return parseDirectiveRealValue(APFloat::IEEEdouble);
1354 case DK_ALIGN: {
1355 bool IsPow2 = !getContext().getAsmInfo()->getAlignmentIsInBytes();
1356 return parseDirectiveAlign(IsPow2, /*ExprSize=*/1);
1357 }
1358 case DK_ALIGN32: {
1359 bool IsPow2 = !getContext().getAsmInfo()->getAlignmentIsInBytes();
1360 return parseDirectiveAlign(IsPow2, /*ExprSize=*/4);
1361 }
1362 case DK_BALIGN:
1363 return parseDirectiveAlign(/*IsPow2=*/false, /*ExprSize=*/1);
1364 case DK_BALIGNW:
1365 return parseDirectiveAlign(/*IsPow2=*/false, /*ExprSize=*/2);
1366 case DK_BALIGNL:
1367 return parseDirectiveAlign(/*IsPow2=*/false, /*ExprSize=*/4);
1368 case DK_P2ALIGN:
1369 return parseDirectiveAlign(/*IsPow2=*/true, /*ExprSize=*/1);
1370 case DK_P2ALIGNW:
1371 return parseDirectiveAlign(/*IsPow2=*/true, /*ExprSize=*/2);
1372 case DK_P2ALIGNL:
1373 return parseDirectiveAlign(/*IsPow2=*/true, /*ExprSize=*/4);
1374 case DK_ORG:
1375 return parseDirectiveOrg();
1376 case DK_FILL:
1377 return parseDirectiveFill();
1378 case DK_ZERO:
1379 return parseDirectiveZero();
1380 case DK_EXTERN:
1381 eatToEndOfStatement(); // .extern is the default, ignore it.
1382 return false;
1383 case DK_GLOBL:
1384 case DK_GLOBAL:
1385 return parseDirectiveSymbolAttribute(MCSA_Global);
1386 case DK_LAZY_REFERENCE:
1387 return parseDirectiveSymbolAttribute(MCSA_LazyReference);
1388 case DK_NO_DEAD_STRIP:
1389 return parseDirectiveSymbolAttribute(MCSA_NoDeadStrip);
1390 case DK_SYMBOL_RESOLVER:
1391 return parseDirectiveSymbolAttribute(MCSA_SymbolResolver);
1392 case DK_PRIVATE_EXTERN:
1393 return parseDirectiveSymbolAttribute(MCSA_PrivateExtern);
1394 case DK_REFERENCE:
1395 return parseDirectiveSymbolAttribute(MCSA_Reference);
1396 case DK_WEAK_DEFINITION:
1397 return parseDirectiveSymbolAttribute(MCSA_WeakDefinition);
1398 case DK_WEAK_REFERENCE:
1399 return parseDirectiveSymbolAttribute(MCSA_WeakReference);
1400 case DK_WEAK_DEF_CAN_BE_HIDDEN:
1401 return parseDirectiveSymbolAttribute(MCSA_WeakDefAutoPrivate);
1402 case DK_COMM:
1403 case DK_COMMON:
1404 return parseDirectiveComm(/*IsLocal=*/false);
1405 case DK_LCOMM:
1406 return parseDirectiveComm(/*IsLocal=*/true);
1407 case DK_ABORT:
1408 return parseDirectiveAbort();
1409 case DK_INCLUDE:
1410 return parseDirectiveInclude();
1411 case DK_INCBIN:
1412 return parseDirectiveIncbin();
1413 case DK_CODE16:
1414 case DK_CODE16GCC:
1415 return TokError(Twine(IDVal) + " not supported yet");
1416 case DK_REPT:
1417 return parseDirectiveRept(IDLoc);
1418 case DK_IRP:
1419 return parseDirectiveIrp(IDLoc);
1420 case DK_IRPC:
1421 return parseDirectiveIrpc(IDLoc);
1422 case DK_ENDR:
1423 return parseDirectiveEndr(IDLoc);
1424 case DK_BUNDLE_ALIGN_MODE:
1425 return parseDirectiveBundleAlignMode();
1426 case DK_BUNDLE_LOCK:
1427 return parseDirectiveBundleLock();
1428 case DK_BUNDLE_UNLOCK:
1429 return parseDirectiveBundleUnlock();
1430 case DK_SLEB128:
1431 return parseDirectiveLEB128(true);
1432 case DK_ULEB128:
1433 return parseDirectiveLEB128(false);
1434 case DK_SPACE:
1435 case DK_SKIP:
1436 return parseDirectiveSpace(IDVal);
1437 case DK_FILE:
1438 return parseDirectiveFile(IDLoc);
1439 case DK_LINE:
1440 return parseDirectiveLine();
1441 case DK_LOC:
1442 return parseDirectiveLoc();
1443 case DK_STABS:
1444 return parseDirectiveStabs();
1445 case DK_CFI_SECTIONS:
1446 return parseDirectiveCFISections();
1447 case DK_CFI_STARTPROC:
1448 return parseDirectiveCFIStartProc();
1449 case DK_CFI_ENDPROC:
1450 return parseDirectiveCFIEndProc();
1451 case DK_CFI_DEF_CFA:
1452 return parseDirectiveCFIDefCfa(IDLoc);
1453 case DK_CFI_DEF_CFA_OFFSET:
1454 return parseDirectiveCFIDefCfaOffset();
1455 case DK_CFI_ADJUST_CFA_OFFSET:
1456 return parseDirectiveCFIAdjustCfaOffset();
1457 case DK_CFI_DEF_CFA_REGISTER:
1458 return parseDirectiveCFIDefCfaRegister(IDLoc);
1459 case DK_CFI_OFFSET:
1460 return parseDirectiveCFIOffset(IDLoc);
1461 case DK_CFI_REL_OFFSET:
1462 return parseDirectiveCFIRelOffset(IDLoc);
1463 case DK_CFI_PERSONALITY:
1464 return parseDirectiveCFIPersonalityOrLsda(true);
1465 case DK_CFI_LSDA:
1466 return parseDirectiveCFIPersonalityOrLsda(false);
1467 case DK_CFI_REMEMBER_STATE:
1468 return parseDirectiveCFIRememberState();
1469 case DK_CFI_RESTORE_STATE:
1470 return parseDirectiveCFIRestoreState();
1471 case DK_CFI_SAME_VALUE:
1472 return parseDirectiveCFISameValue(IDLoc);
1473 case DK_CFI_RESTORE:
1474 return parseDirectiveCFIRestore(IDLoc);
1475 case DK_CFI_ESCAPE:
1476 return parseDirectiveCFIEscape();
1477 case DK_CFI_SIGNAL_FRAME:
1478 return parseDirectiveCFISignalFrame();
1479 case DK_CFI_UNDEFINED:
1480 return parseDirectiveCFIUndefined(IDLoc);
1481 case DK_CFI_REGISTER:
1482 return parseDirectiveCFIRegister(IDLoc);
1483 case DK_CFI_WINDOW_SAVE:
1484 return parseDirectiveCFIWindowSave();
1485 case DK_MACROS_ON:
1486 case DK_MACROS_OFF:
1487 return parseDirectiveMacrosOnOff(IDVal);
1488 case DK_MACRO:
1489 return parseDirectiveMacro(IDLoc);
1490 case DK_ENDM:
1491 case DK_ENDMACRO:
1492 return parseDirectiveEndMacro(IDVal);
1493 case DK_PURGEM:
1494 return parseDirectivePurgeMacro(IDLoc);
1495 }
1496
1497 return Error(IDLoc, "unknown directive");
1498 }
1499
1500 // __asm _emit or __asm __emit
1501 if (ParsingInlineAsm && (IDVal == "_emit" || IDVal == "__emit" ||
1502 IDVal == "_EMIT" || IDVal == "__EMIT"))
1503 return parseDirectiveMSEmit(IDLoc, Info, IDVal.size());
1504
1505 // __asm align
1506 if (ParsingInlineAsm && (IDVal == "align" || IDVal == "ALIGN"))
1507 return parseDirectiveMSAlign(IDLoc, Info);
1508
1509 checkForValidSection();
1510
1511 // Canonicalize the opcode to lower case.
1512 std::string OpcodeStr = IDVal.lower();
1513 ParseInstructionInfo IInfo(Info.AsmRewrites);
1514 bool HadError = getTargetParser().ParseInstruction(IInfo, OpcodeStr, IDLoc,
1515 Info.ParsedOperands);
1516 Info.ParseError = HadError;
1517
1518 // Dump the parsed representation, if requested.
1519 if (getShowParsedOperands()) {
1520 SmallString<256> Str;
1521 raw_svector_ostream OS(Str);
1522 OS << "parsed instruction: [";
1523 for (unsigned i = 0; i != Info.ParsedOperands.size(); ++i) {
1524 if (i != 0)
1525 OS << ", ";
1526 Info.ParsedOperands[i]->print(OS);
1527 }
1528 OS << "]";
1529
1530 printMessage(IDLoc, SourceMgr::DK_Note, OS.str());
1531 }
1532
1533 // If we are generating dwarf for assembly source files and the current
1534 // section is the initial text section then generate a .loc directive for
1535 // the instruction.
1536 if (!HadError && getContext().getGenDwarfForAssembly() &&
1537 getContext().getGenDwarfSection() ==
1538 getStreamer().getCurrentSection().first) {
1539
1540 unsigned Line = SrcMgr.FindLineNumber(IDLoc, CurBuffer);
1541
1542 // If we previously parsed a cpp hash file line comment then make sure the
1543 // current Dwarf File is for the CppHashFilename if not then emit the
1544 // Dwarf File table for it and adjust the line number for the .loc.
1545 const SmallVectorImpl<MCDwarfFile *> &MCDwarfFiles =
1546 getContext().getMCDwarfFiles();
1547 if (CppHashFilename.size() != 0) {
1548 if (MCDwarfFiles[getContext().getGenDwarfFileNumber()]->getName() !=
1549 CppHashFilename)
1550 getStreamer().EmitDwarfFileDirective(
1551 getContext().nextGenDwarfFileNumber(), StringRef(),
1552 CppHashFilename);
1553
1554 // Since SrcMgr.FindLineNumber() is slow and messes up the SourceMgr's
1555 // cache with the different Loc from the call above we save the last
1556 // info we queried here with SrcMgr.FindLineNumber().
1557 unsigned CppHashLocLineNo;
1558 if (LastQueryIDLoc == CppHashLoc && LastQueryBuffer == CppHashBuf)
1559 CppHashLocLineNo = LastQueryLine;
1560 else {
1561 CppHashLocLineNo = SrcMgr.FindLineNumber(CppHashLoc, CppHashBuf);
1562 LastQueryLine = CppHashLocLineNo;
1563 LastQueryIDLoc = CppHashLoc;
1564 LastQueryBuffer = CppHashBuf;
1565 }
1566 Line = CppHashLineNumber - 1 + (Line - CppHashLocLineNo);
1567 }
1568
1569 getStreamer().EmitDwarfLocDirective(
1570 getContext().getGenDwarfFileNumber(), Line, 0,
1571 DWARF2_LINE_DEFAULT_IS_STMT ? DWARF2_FLAG_IS_STMT : 0, 0, 0,
1572 StringRef());
1573 }
1574
1575 // If parsing succeeded, match the instruction.
1576 if (!HadError) {
1577 unsigned ErrorInfo;
1578 HadError = getTargetParser().MatchAndEmitInstruction(
1579 IDLoc, Info.Opcode, Info.ParsedOperands, Out, ErrorInfo,
1580 ParsingInlineAsm);
1581 }
1582
1583 // Don't skip the rest of the line, the instruction parser is responsible for
1584 // that.
1585 return false;
1586 }
1587
1588 /// eatToEndOfLine uses the Lexer to eat the characters to the end of the line
1589 /// since they may not be able to be tokenized to get to the end of line token.
eatToEndOfLine()1590 void AsmParser::eatToEndOfLine() {
1591 if (!Lexer.is(AsmToken::EndOfStatement))
1592 Lexer.LexUntilEndOfLine();
1593 // Eat EOL.
1594 Lex();
1595 }
1596
1597 /// parseCppHashLineFilenameComment as this:
1598 /// ::= # number "filename"
1599 /// or just as a full line comment if it doesn't have a number and a string.
parseCppHashLineFilenameComment(const SMLoc & L)1600 bool AsmParser::parseCppHashLineFilenameComment(const SMLoc &L) {
1601 Lex(); // Eat the hash token.
1602
1603 if (getLexer().isNot(AsmToken::Integer)) {
1604 // Consume the line since in cases it is not a well-formed line directive,
1605 // as if were simply a full line comment.
1606 eatToEndOfLine();
1607 return false;
1608 }
1609
1610 int64_t LineNumber = getTok().getIntVal();
1611 Lex();
1612
1613 if (getLexer().isNot(AsmToken::String)) {
1614 eatToEndOfLine();
1615 return false;
1616 }
1617
1618 StringRef Filename = getTok().getString();
1619 // Get rid of the enclosing quotes.
1620 Filename = Filename.substr(1, Filename.size() - 2);
1621
1622 // Save the SMLoc, Filename and LineNumber for later use by diagnostics.
1623 CppHashLoc = L;
1624 CppHashFilename = Filename;
1625 CppHashLineNumber = LineNumber;
1626 CppHashBuf = CurBuffer;
1627
1628 // Ignore any trailing characters, they're just comment.
1629 eatToEndOfLine();
1630 return false;
1631 }
1632
1633 /// \brief will use the last parsed cpp hash line filename comment
1634 /// for the Filename and LineNo if any in the diagnostic.
DiagHandler(const SMDiagnostic & Diag,void * Context)1635 void AsmParser::DiagHandler(const SMDiagnostic &Diag, void *Context) {
1636 const AsmParser *Parser = static_cast<const AsmParser *>(Context);
1637 raw_ostream &OS = errs();
1638
1639 const SourceMgr &DiagSrcMgr = *Diag.getSourceMgr();
1640 const SMLoc &DiagLoc = Diag.getLoc();
1641 int DiagBuf = DiagSrcMgr.FindBufferContainingLoc(DiagLoc);
1642 int CppHashBuf = Parser->SrcMgr.FindBufferContainingLoc(Parser->CppHashLoc);
1643
1644 // Like SourceMgr::printMessage() we need to print the include stack if any
1645 // before printing the message.
1646 int DiagCurBuffer = DiagSrcMgr.FindBufferContainingLoc(DiagLoc);
1647 if (!Parser->SavedDiagHandler && DiagCurBuffer > 0) {
1648 SMLoc ParentIncludeLoc = DiagSrcMgr.getParentIncludeLoc(DiagCurBuffer);
1649 DiagSrcMgr.PrintIncludeStack(ParentIncludeLoc, OS);
1650 }
1651
1652 // If we have not parsed a cpp hash line filename comment or the source
1653 // manager changed or buffer changed (like in a nested include) then just
1654 // print the normal diagnostic using its Filename and LineNo.
1655 if (!Parser->CppHashLineNumber || &DiagSrcMgr != &Parser->SrcMgr ||
1656 DiagBuf != CppHashBuf) {
1657 if (Parser->SavedDiagHandler)
1658 Parser->SavedDiagHandler(Diag, Parser->SavedDiagContext);
1659 else
1660 Diag.print(0, OS);
1661 return;
1662 }
1663
1664 // Use the CppHashFilename and calculate a line number based on the
1665 // CppHashLoc and CppHashLineNumber relative to this Diag's SMLoc for
1666 // the diagnostic.
1667 const std::string &Filename = Parser->CppHashFilename;
1668
1669 int DiagLocLineNo = DiagSrcMgr.FindLineNumber(DiagLoc, DiagBuf);
1670 int CppHashLocLineNo =
1671 Parser->SrcMgr.FindLineNumber(Parser->CppHashLoc, CppHashBuf);
1672 int LineNo =
1673 Parser->CppHashLineNumber - 1 + (DiagLocLineNo - CppHashLocLineNo);
1674
1675 SMDiagnostic NewDiag(*Diag.getSourceMgr(), Diag.getLoc(), Filename, LineNo,
1676 Diag.getColumnNo(), Diag.getKind(), Diag.getMessage(),
1677 Diag.getLineContents(), Diag.getRanges());
1678
1679 if (Parser->SavedDiagHandler)
1680 Parser->SavedDiagHandler(NewDiag, Parser->SavedDiagContext);
1681 else
1682 NewDiag.print(0, OS);
1683 }
1684
1685 // FIXME: This is mostly duplicated from the function in AsmLexer.cpp. The
1686 // difference being that that function accepts '@' as part of identifiers and
1687 // we can't do that. AsmLexer.cpp should probably be changed to handle
1688 // '@' as a special case when needed.
isIdentifierChar(char c)1689 static bool isIdentifierChar(char c) {
1690 return isalnum(static_cast<unsigned char>(c)) || c == '_' || c == '$' ||
1691 c == '.';
1692 }
1693
expandMacro(raw_svector_ostream & OS,StringRef Body,const MCAsmMacroParameters & Parameters,const MCAsmMacroArguments & A,const SMLoc & L)1694 bool AsmParser::expandMacro(raw_svector_ostream &OS, StringRef Body,
1695 const MCAsmMacroParameters &Parameters,
1696 const MCAsmMacroArguments &A, const SMLoc &L) {
1697 unsigned NParameters = Parameters.size();
1698 if ((!IsDarwin || NParameters != 0) && NParameters != A.size())
1699 return Error(L, "Wrong number of arguments");
1700
1701 // A macro without parameters is handled differently on Darwin:
1702 // gas accepts no arguments and does no substitutions
1703 while (!Body.empty()) {
1704 // Scan for the next substitution.
1705 std::size_t End = Body.size(), Pos = 0;
1706 for (; Pos != End; ++Pos) {
1707 // Check for a substitution or escape.
1708 if (IsDarwin && !NParameters) {
1709 // This macro has no parameters, look for $0, $1, etc.
1710 if (Body[Pos] != '$' || Pos + 1 == End)
1711 continue;
1712
1713 char Next = Body[Pos + 1];
1714 if (Next == '$' || Next == 'n' ||
1715 isdigit(static_cast<unsigned char>(Next)))
1716 break;
1717 } else {
1718 // This macro has parameters, look for \foo, \bar, etc.
1719 if (Body[Pos] == '\\' && Pos + 1 != End)
1720 break;
1721 }
1722 }
1723
1724 // Add the prefix.
1725 OS << Body.slice(0, Pos);
1726
1727 // Check if we reached the end.
1728 if (Pos == End)
1729 break;
1730
1731 if (IsDarwin && !NParameters) {
1732 switch (Body[Pos + 1]) {
1733 // $$ => $
1734 case '$':
1735 OS << '$';
1736 break;
1737
1738 // $n => number of arguments
1739 case 'n':
1740 OS << A.size();
1741 break;
1742
1743 // $[0-9] => argument
1744 default: {
1745 // Missing arguments are ignored.
1746 unsigned Index = Body[Pos + 1] - '0';
1747 if (Index >= A.size())
1748 break;
1749
1750 // Otherwise substitute with the token values, with spaces eliminated.
1751 for (MCAsmMacroArgument::const_iterator it = A[Index].begin(),
1752 ie = A[Index].end();
1753 it != ie; ++it)
1754 OS << it->getString();
1755 break;
1756 }
1757 }
1758 Pos += 2;
1759 } else {
1760 unsigned I = Pos + 1;
1761 while (isIdentifierChar(Body[I]) && I + 1 != End)
1762 ++I;
1763
1764 const char *Begin = Body.data() + Pos + 1;
1765 StringRef Argument(Begin, I - (Pos + 1));
1766 unsigned Index = 0;
1767 for (; Index < NParameters; ++Index)
1768 if (Parameters[Index].first == Argument)
1769 break;
1770
1771 if (Index == NParameters) {
1772 if (Body[Pos + 1] == '(' && Body[Pos + 2] == ')')
1773 Pos += 3;
1774 else {
1775 OS << '\\' << Argument;
1776 Pos = I;
1777 }
1778 } else {
1779 for (MCAsmMacroArgument::const_iterator it = A[Index].begin(),
1780 ie = A[Index].end();
1781 it != ie; ++it)
1782 if (it->getKind() == AsmToken::String)
1783 OS << it->getStringContents();
1784 else
1785 OS << it->getString();
1786
1787 Pos += 1 + Argument.size();
1788 }
1789 }
1790 // Update the scan point.
1791 Body = Body.substr(Pos);
1792 }
1793
1794 return false;
1795 }
1796
MacroInstantiation(const MCAsmMacro * M,SMLoc IL,int EB,SMLoc EL,MemoryBuffer * I)1797 MacroInstantiation::MacroInstantiation(const MCAsmMacro *M, SMLoc IL, int EB,
1798 SMLoc EL, MemoryBuffer *I)
1799 : TheMacro(M), Instantiation(I), InstantiationLoc(IL), ExitBuffer(EB),
1800 ExitLoc(EL) {}
1801
isOperator(AsmToken::TokenKind kind)1802 static bool isOperator(AsmToken::TokenKind kind) {
1803 switch (kind) {
1804 default:
1805 return false;
1806 case AsmToken::Plus:
1807 case AsmToken::Minus:
1808 case AsmToken::Tilde:
1809 case AsmToken::Slash:
1810 case AsmToken::Star:
1811 case AsmToken::Dot:
1812 case AsmToken::Equal:
1813 case AsmToken::EqualEqual:
1814 case AsmToken::Pipe:
1815 case AsmToken::PipePipe:
1816 case AsmToken::Caret:
1817 case AsmToken::Amp:
1818 case AsmToken::AmpAmp:
1819 case AsmToken::Exclaim:
1820 case AsmToken::ExclaimEqual:
1821 case AsmToken::Percent:
1822 case AsmToken::Less:
1823 case AsmToken::LessEqual:
1824 case AsmToken::LessLess:
1825 case AsmToken::LessGreater:
1826 case AsmToken::Greater:
1827 case AsmToken::GreaterEqual:
1828 case AsmToken::GreaterGreater:
1829 return true;
1830 }
1831 }
1832
parseMacroArgument(MCAsmMacroArgument & MA,AsmToken::TokenKind & ArgumentDelimiter)1833 bool AsmParser::parseMacroArgument(MCAsmMacroArgument &MA,
1834 AsmToken::TokenKind &ArgumentDelimiter) {
1835 unsigned ParenLevel = 0;
1836 unsigned AddTokens = 0;
1837
1838 // gas accepts arguments separated by whitespace, except on Darwin
1839 if (!IsDarwin)
1840 Lexer.setSkipSpace(false);
1841
1842 for (;;) {
1843 if (Lexer.is(AsmToken::Eof) || Lexer.is(AsmToken::Equal)) {
1844 Lexer.setSkipSpace(true);
1845 return TokError("unexpected token in macro instantiation");
1846 }
1847
1848 if (ParenLevel == 0 && Lexer.is(AsmToken::Comma)) {
1849 // Spaces and commas cannot be mixed to delimit parameters
1850 if (ArgumentDelimiter == AsmToken::Eof)
1851 ArgumentDelimiter = AsmToken::Comma;
1852 else if (ArgumentDelimiter != AsmToken::Comma) {
1853 Lexer.setSkipSpace(true);
1854 return TokError("expected ' ' for macro argument separator");
1855 }
1856 break;
1857 }
1858
1859 if (Lexer.is(AsmToken::Space)) {
1860 Lex(); // Eat spaces
1861
1862 // Spaces can delimit parameters, but could also be part an expression.
1863 // If the token after a space is an operator, add the token and the next
1864 // one into this argument
1865 if (ArgumentDelimiter == AsmToken::Space ||
1866 ArgumentDelimiter == AsmToken::Eof) {
1867 if (isOperator(Lexer.getKind())) {
1868 // Check to see whether the token is used as an operator,
1869 // or part of an identifier
1870 const char *NextChar = getTok().getEndLoc().getPointer();
1871 if (*NextChar == ' ')
1872 AddTokens = 2;
1873 }
1874
1875 if (!AddTokens && ParenLevel == 0) {
1876 if (ArgumentDelimiter == AsmToken::Eof &&
1877 !isOperator(Lexer.getKind()))
1878 ArgumentDelimiter = AsmToken::Space;
1879 break;
1880 }
1881 }
1882 }
1883
1884 // handleMacroEntry relies on not advancing the lexer here
1885 // to be able to fill in the remaining default parameter values
1886 if (Lexer.is(AsmToken::EndOfStatement))
1887 break;
1888
1889 // Adjust the current parentheses level.
1890 if (Lexer.is(AsmToken::LParen))
1891 ++ParenLevel;
1892 else if (Lexer.is(AsmToken::RParen) && ParenLevel)
1893 --ParenLevel;
1894
1895 // Append the token to the current argument list.
1896 MA.push_back(getTok());
1897 if (AddTokens)
1898 AddTokens--;
1899 Lex();
1900 }
1901
1902 Lexer.setSkipSpace(true);
1903 if (ParenLevel != 0)
1904 return TokError("unbalanced parentheses in macro argument");
1905 return false;
1906 }
1907
1908 // Parse the macro instantiation arguments.
parseMacroArguments(const MCAsmMacro * M,MCAsmMacroArguments & A)1909 bool AsmParser::parseMacroArguments(const MCAsmMacro *M,
1910 MCAsmMacroArguments &A) {
1911 const unsigned NParameters = M ? M->Parameters.size() : 0;
1912 // Argument delimiter is initially unknown. It will be set by
1913 // parseMacroArgument()
1914 AsmToken::TokenKind ArgumentDelimiter = AsmToken::Eof;
1915
1916 // Parse two kinds of macro invocations:
1917 // - macros defined without any parameters accept an arbitrary number of them
1918 // - macros defined with parameters accept at most that many of them
1919 for (unsigned Parameter = 0; !NParameters || Parameter < NParameters;
1920 ++Parameter) {
1921 MCAsmMacroArgument MA;
1922
1923 if (parseMacroArgument(MA, ArgumentDelimiter))
1924 return true;
1925
1926 if (!MA.empty() || !NParameters)
1927 A.push_back(MA);
1928 else if (NParameters) {
1929 if (!M->Parameters[Parameter].second.empty())
1930 A.push_back(M->Parameters[Parameter].second);
1931 }
1932
1933 // At the end of the statement, fill in remaining arguments that have
1934 // default values. If there aren't any, then the next argument is
1935 // required but missing
1936 if (Lexer.is(AsmToken::EndOfStatement)) {
1937 if (NParameters && Parameter < NParameters - 1) {
1938 if (M->Parameters[Parameter + 1].second.empty())
1939 return TokError("macro argument '" +
1940 Twine(M->Parameters[Parameter + 1].first) +
1941 "' is missing");
1942 else
1943 continue;
1944 }
1945 return false;
1946 }
1947
1948 if (Lexer.is(AsmToken::Comma))
1949 Lex();
1950 }
1951 return TokError("Too many arguments");
1952 }
1953
lookupMacro(StringRef Name)1954 const MCAsmMacro *AsmParser::lookupMacro(StringRef Name) {
1955 StringMap<MCAsmMacro *>::iterator I = MacroMap.find(Name);
1956 return (I == MacroMap.end()) ? NULL : I->getValue();
1957 }
1958
defineMacro(StringRef Name,const MCAsmMacro & Macro)1959 void AsmParser::defineMacro(StringRef Name, const MCAsmMacro &Macro) {
1960 MacroMap[Name] = new MCAsmMacro(Macro);
1961 }
1962
undefineMacro(StringRef Name)1963 void AsmParser::undefineMacro(StringRef Name) {
1964 StringMap<MCAsmMacro *>::iterator I = MacroMap.find(Name);
1965 if (I != MacroMap.end()) {
1966 delete I->getValue();
1967 MacroMap.erase(I);
1968 }
1969 }
1970
handleMacroEntry(const MCAsmMacro * M,SMLoc NameLoc)1971 bool AsmParser::handleMacroEntry(const MCAsmMacro *M, SMLoc NameLoc) {
1972 // Arbitrarily limit macro nesting depth, to match 'as'. We can eliminate
1973 // this, although we should protect against infinite loops.
1974 if (ActiveMacros.size() == 20)
1975 return TokError("macros cannot be nested more than 20 levels deep");
1976
1977 MCAsmMacroArguments A;
1978 if (parseMacroArguments(M, A))
1979 return true;
1980
1981 // Remove any trailing empty arguments. Do this after-the-fact as we have
1982 // to keep empty arguments in the middle of the list or positionality
1983 // gets off. e.g., "foo 1, , 2" vs. "foo 1, 2,"
1984 while (!A.empty() && A.back().empty())
1985 A.pop_back();
1986
1987 // Macro instantiation is lexical, unfortunately. We construct a new buffer
1988 // to hold the macro body with substitutions.
1989 SmallString<256> Buf;
1990 StringRef Body = M->Body;
1991 raw_svector_ostream OS(Buf);
1992
1993 if (expandMacro(OS, Body, M->Parameters, A, getTok().getLoc()))
1994 return true;
1995
1996 // We include the .endmacro in the buffer as our cue to exit the macro
1997 // instantiation.
1998 OS << ".endmacro\n";
1999
2000 MemoryBuffer *Instantiation =
2001 MemoryBuffer::getMemBufferCopy(OS.str(), "<instantiation>");
2002
2003 // Create the macro instantiation object and add to the current macro
2004 // instantiation stack.
2005 MacroInstantiation *MI = new MacroInstantiation(
2006 M, NameLoc, CurBuffer, getTok().getLoc(), Instantiation);
2007 ActiveMacros.push_back(MI);
2008
2009 // Jump to the macro instantiation and prime the lexer.
2010 CurBuffer = SrcMgr.AddNewSourceBuffer(MI->Instantiation, SMLoc());
2011 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer));
2012 Lex();
2013
2014 return false;
2015 }
2016
handleMacroExit()2017 void AsmParser::handleMacroExit() {
2018 // Jump to the EndOfStatement we should return to, and consume it.
2019 jumpToLoc(ActiveMacros.back()->ExitLoc, ActiveMacros.back()->ExitBuffer);
2020 Lex();
2021
2022 // Pop the instantiation entry.
2023 delete ActiveMacros.back();
2024 ActiveMacros.pop_back();
2025 }
2026
isUsedIn(const MCSymbol * Sym,const MCExpr * Value)2027 static bool isUsedIn(const MCSymbol *Sym, const MCExpr *Value) {
2028 switch (Value->getKind()) {
2029 case MCExpr::Binary: {
2030 const MCBinaryExpr *BE = static_cast<const MCBinaryExpr *>(Value);
2031 return isUsedIn(Sym, BE->getLHS()) || isUsedIn(Sym, BE->getRHS());
2032 }
2033 case MCExpr::Target:
2034 case MCExpr::Constant:
2035 return false;
2036 case MCExpr::SymbolRef: {
2037 const MCSymbol &S =
2038 static_cast<const MCSymbolRefExpr *>(Value)->getSymbol();
2039 if (S.isVariable())
2040 return isUsedIn(Sym, S.getVariableValue());
2041 return &S == Sym;
2042 }
2043 case MCExpr::Unary:
2044 return isUsedIn(Sym, static_cast<const MCUnaryExpr *>(Value)->getSubExpr());
2045 }
2046
2047 llvm_unreachable("Unknown expr kind!");
2048 }
2049
parseAssignment(StringRef Name,bool allow_redef,bool NoDeadStrip)2050 bool AsmParser::parseAssignment(StringRef Name, bool allow_redef,
2051 bool NoDeadStrip) {
2052 // FIXME: Use better location, we should use proper tokens.
2053 SMLoc EqualLoc = Lexer.getLoc();
2054
2055 const MCExpr *Value;
2056 if (parseExpression(Value))
2057 return true;
2058
2059 // Note: we don't count b as used in "a = b". This is to allow
2060 // a = b
2061 // b = c
2062
2063 if (Lexer.isNot(AsmToken::EndOfStatement))
2064 return TokError("unexpected token in assignment");
2065
2066 // Error on assignment to '.'.
2067 if (Name == ".") {
2068 return Error(EqualLoc, ("assignment to pseudo-symbol '.' is unsupported "
2069 "(use '.space' or '.org').)"));
2070 }
2071
2072 // Eat the end of statement marker.
2073 Lex();
2074
2075 // Validate that the LHS is allowed to be a variable (either it has not been
2076 // used as a symbol, or it is an absolute symbol).
2077 MCSymbol *Sym = getContext().LookupSymbol(Name);
2078 if (Sym) {
2079 // Diagnose assignment to a label.
2080 //
2081 // FIXME: Diagnostics. Note the location of the definition as a label.
2082 // FIXME: Diagnose assignment to protected identifier (e.g., register name).
2083 if (isUsedIn(Sym, Value))
2084 return Error(EqualLoc, "Recursive use of '" + Name + "'");
2085 else if (Sym->isUndefined() && !Sym->isUsed() && !Sym->isVariable())
2086 ; // Allow redefinitions of undefined symbols only used in directives.
2087 else if (Sym->isVariable() && !Sym->isUsed() && allow_redef)
2088 ; // Allow redefinitions of variables that haven't yet been used.
2089 else if (!Sym->isUndefined() && (!Sym->isVariable() || !allow_redef))
2090 return Error(EqualLoc, "redefinition of '" + Name + "'");
2091 else if (!Sym->isVariable())
2092 return Error(EqualLoc, "invalid assignment to '" + Name + "'");
2093 else if (!isa<MCConstantExpr>(Sym->getVariableValue()))
2094 return Error(EqualLoc, "invalid reassignment of non-absolute variable '" +
2095 Name + "'");
2096
2097 // Don't count these checks as uses.
2098 Sym->setUsed(false);
2099 } else
2100 Sym = getContext().GetOrCreateSymbol(Name);
2101
2102 // FIXME: Handle '.'.
2103
2104 // Do the assignment.
2105 Out.EmitAssignment(Sym, Value);
2106 if (NoDeadStrip)
2107 Out.EmitSymbolAttribute(Sym, MCSA_NoDeadStrip);
2108
2109 return false;
2110 }
2111
2112 /// parseIdentifier:
2113 /// ::= identifier
2114 /// ::= string
parseIdentifier(StringRef & Res)2115 bool AsmParser::parseIdentifier(StringRef &Res) {
2116 // The assembler has relaxed rules for accepting identifiers, in particular we
2117 // allow things like '.globl $foo' and '.def @feat.00', which would normally be
2118 // separate tokens. At this level, we have already lexed so we cannot (currently)
2119 // handle this as a context dependent token, instead we detect adjacent tokens
2120 // and return the combined identifier.
2121 if (Lexer.is(AsmToken::Dollar) || Lexer.is(AsmToken::At)) {
2122 SMLoc PrefixLoc = getLexer().getLoc();
2123
2124 // Consume the prefix character, and check for a following identifier.
2125 Lex();
2126 if (Lexer.isNot(AsmToken::Identifier))
2127 return true;
2128
2129 // We have a '$' or '@' followed by an identifier, make sure they are adjacent.
2130 if (PrefixLoc.getPointer() + 1 != getTok().getLoc().getPointer())
2131 return true;
2132
2133 // Construct the joined identifier and consume the token.
2134 Res =
2135 StringRef(PrefixLoc.getPointer(), getTok().getIdentifier().size() + 1);
2136 Lex();
2137 return false;
2138 }
2139
2140 if (Lexer.isNot(AsmToken::Identifier) && Lexer.isNot(AsmToken::String))
2141 return true;
2142
2143 Res = getTok().getIdentifier();
2144
2145 Lex(); // Consume the identifier token.
2146
2147 return false;
2148 }
2149
2150 /// parseDirectiveSet:
2151 /// ::= .equ identifier ',' expression
2152 /// ::= .equiv identifier ',' expression
2153 /// ::= .set identifier ',' expression
parseDirectiveSet(StringRef IDVal,bool allow_redef)2154 bool AsmParser::parseDirectiveSet(StringRef IDVal, bool allow_redef) {
2155 StringRef Name;
2156
2157 if (parseIdentifier(Name))
2158 return TokError("expected identifier after '" + Twine(IDVal) + "'");
2159
2160 if (getLexer().isNot(AsmToken::Comma))
2161 return TokError("unexpected token in '" + Twine(IDVal) + "'");
2162 Lex();
2163
2164 return parseAssignment(Name, allow_redef, true);
2165 }
2166
parseEscapedString(std::string & Data)2167 bool AsmParser::parseEscapedString(std::string &Data) {
2168 assert(getLexer().is(AsmToken::String) && "Unexpected current token!");
2169
2170 Data = "";
2171 StringRef Str = getTok().getStringContents();
2172 for (unsigned i = 0, e = Str.size(); i != e; ++i) {
2173 if (Str[i] != '\\') {
2174 Data += Str[i];
2175 continue;
2176 }
2177
2178 // Recognize escaped characters. Note that this escape semantics currently
2179 // loosely follows Darwin 'as'. Notably, it doesn't support hex escapes.
2180 ++i;
2181 if (i == e)
2182 return TokError("unexpected backslash at end of string");
2183
2184 // Recognize octal sequences.
2185 if ((unsigned)(Str[i] - '0') <= 7) {
2186 // Consume up to three octal characters.
2187 unsigned Value = Str[i] - '0';
2188
2189 if (i + 1 != e && ((unsigned)(Str[i + 1] - '0')) <= 7) {
2190 ++i;
2191 Value = Value * 8 + (Str[i] - '0');
2192
2193 if (i + 1 != e && ((unsigned)(Str[i + 1] - '0')) <= 7) {
2194 ++i;
2195 Value = Value * 8 + (Str[i] - '0');
2196 }
2197 }
2198
2199 if (Value > 255)
2200 return TokError("invalid octal escape sequence (out of range)");
2201
2202 Data += (unsigned char)Value;
2203 continue;
2204 }
2205
2206 // Otherwise recognize individual escapes.
2207 switch (Str[i]) {
2208 default:
2209 // Just reject invalid escape sequences for now.
2210 return TokError("invalid escape sequence (unrecognized character)");
2211
2212 case 'b': Data += '\b'; break;
2213 case 'f': Data += '\f'; break;
2214 case 'n': Data += '\n'; break;
2215 case 'r': Data += '\r'; break;
2216 case 't': Data += '\t'; break;
2217 case '"': Data += '"'; break;
2218 case '\\': Data += '\\'; break;
2219 }
2220 }
2221
2222 return false;
2223 }
2224
2225 /// parseDirectiveAscii:
2226 /// ::= ( .ascii | .asciz | .string ) [ "string" ( , "string" )* ]
parseDirectiveAscii(StringRef IDVal,bool ZeroTerminated)2227 bool AsmParser::parseDirectiveAscii(StringRef IDVal, bool ZeroTerminated) {
2228 if (getLexer().isNot(AsmToken::EndOfStatement)) {
2229 checkForValidSection();
2230
2231 for (;;) {
2232 if (getLexer().isNot(AsmToken::String))
2233 return TokError("expected string in '" + Twine(IDVal) + "' directive");
2234
2235 std::string Data;
2236 if (parseEscapedString(Data))
2237 return true;
2238
2239 getStreamer().EmitBytes(Data);
2240 if (ZeroTerminated)
2241 getStreamer().EmitBytes(StringRef("\0", 1));
2242
2243 Lex();
2244
2245 if (getLexer().is(AsmToken::EndOfStatement))
2246 break;
2247
2248 if (getLexer().isNot(AsmToken::Comma))
2249 return TokError("unexpected token in '" + Twine(IDVal) + "' directive");
2250 Lex();
2251 }
2252 }
2253
2254 Lex();
2255 return false;
2256 }
2257
2258 /// parseDirectiveValue
2259 /// ::= (.byte | .short | ... ) [ expression (, expression)* ]
parseDirectiveValue(unsigned Size)2260 bool AsmParser::parseDirectiveValue(unsigned Size) {
2261 if (getLexer().isNot(AsmToken::EndOfStatement)) {
2262 checkForValidSection();
2263
2264 for (;;) {
2265 const MCExpr *Value;
2266 SMLoc ExprLoc = getLexer().getLoc();
2267 if (parseExpression(Value))
2268 return true;
2269
2270 // Special case constant expressions to match code generator.
2271 if (const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value)) {
2272 assert(Size <= 8 && "Invalid size");
2273 uint64_t IntValue = MCE->getValue();
2274 if (!isUIntN(8 * Size, IntValue) && !isIntN(8 * Size, IntValue))
2275 return Error(ExprLoc, "literal value out of range for directive");
2276 getStreamer().EmitIntValue(IntValue, Size);
2277 } else
2278 getStreamer().EmitValue(Value, Size);
2279
2280 if (getLexer().is(AsmToken::EndOfStatement))
2281 break;
2282
2283 // FIXME: Improve diagnostic.
2284 if (getLexer().isNot(AsmToken::Comma))
2285 return TokError("unexpected token in directive");
2286 Lex();
2287 }
2288 }
2289
2290 Lex();
2291 return false;
2292 }
2293
2294 /// parseDirectiveRealValue
2295 /// ::= (.single | .double) [ expression (, expression)* ]
parseDirectiveRealValue(const fltSemantics & Semantics)2296 bool AsmParser::parseDirectiveRealValue(const fltSemantics &Semantics) {
2297 if (getLexer().isNot(AsmToken::EndOfStatement)) {
2298 checkForValidSection();
2299
2300 for (;;) {
2301 // We don't truly support arithmetic on floating point expressions, so we
2302 // have to manually parse unary prefixes.
2303 bool IsNeg = false;
2304 if (getLexer().is(AsmToken::Minus)) {
2305 Lex();
2306 IsNeg = true;
2307 } else if (getLexer().is(AsmToken::Plus))
2308 Lex();
2309
2310 if (getLexer().isNot(AsmToken::Integer) &&
2311 getLexer().isNot(AsmToken::Real) &&
2312 getLexer().isNot(AsmToken::Identifier))
2313 return TokError("unexpected token in directive");
2314
2315 // Convert to an APFloat.
2316 APFloat Value(Semantics);
2317 StringRef IDVal = getTok().getString();
2318 if (getLexer().is(AsmToken::Identifier)) {
2319 if (!IDVal.compare_lower("infinity") || !IDVal.compare_lower("inf"))
2320 Value = APFloat::getInf(Semantics);
2321 else if (!IDVal.compare_lower("nan"))
2322 Value = APFloat::getNaN(Semantics, false, ~0);
2323 else
2324 return TokError("invalid floating point literal");
2325 } else if (Value.convertFromString(IDVal, APFloat::rmNearestTiesToEven) ==
2326 APFloat::opInvalidOp)
2327 return TokError("invalid floating point literal");
2328 if (IsNeg)
2329 Value.changeSign();
2330
2331 // Consume the numeric token.
2332 Lex();
2333
2334 // Emit the value as an integer.
2335 APInt AsInt = Value.bitcastToAPInt();
2336 getStreamer().EmitIntValue(AsInt.getLimitedValue(),
2337 AsInt.getBitWidth() / 8);
2338
2339 if (getLexer().is(AsmToken::EndOfStatement))
2340 break;
2341
2342 if (getLexer().isNot(AsmToken::Comma))
2343 return TokError("unexpected token in directive");
2344 Lex();
2345 }
2346 }
2347
2348 Lex();
2349 return false;
2350 }
2351
2352 /// parseDirectiveZero
2353 /// ::= .zero expression
parseDirectiveZero()2354 bool AsmParser::parseDirectiveZero() {
2355 checkForValidSection();
2356
2357 int64_t NumBytes;
2358 if (parseAbsoluteExpression(NumBytes))
2359 return true;
2360
2361 int64_t Val = 0;
2362 if (getLexer().is(AsmToken::Comma)) {
2363 Lex();
2364 if (parseAbsoluteExpression(Val))
2365 return true;
2366 }
2367
2368 if (getLexer().isNot(AsmToken::EndOfStatement))
2369 return TokError("unexpected token in '.zero' directive");
2370
2371 Lex();
2372
2373 getStreamer().EmitFill(NumBytes, Val);
2374
2375 return false;
2376 }
2377
2378 /// parseDirectiveFill
2379 /// ::= .fill expression [ , expression [ , expression ] ]
parseDirectiveFill()2380 bool AsmParser::parseDirectiveFill() {
2381 checkForValidSection();
2382
2383 int64_t NumValues;
2384 if (parseAbsoluteExpression(NumValues))
2385 return true;
2386
2387 int64_t FillSize = 1;
2388 int64_t FillExpr = 0;
2389
2390 if (getLexer().isNot(AsmToken::EndOfStatement)) {
2391 if (getLexer().isNot(AsmToken::Comma))
2392 return TokError("unexpected token in '.fill' directive");
2393 Lex();
2394
2395 if (parseAbsoluteExpression(FillSize))
2396 return true;
2397
2398 if (getLexer().isNot(AsmToken::EndOfStatement)) {
2399 if (getLexer().isNot(AsmToken::Comma))
2400 return TokError("unexpected token in '.fill' directive");
2401 Lex();
2402
2403 if (parseAbsoluteExpression(FillExpr))
2404 return true;
2405
2406 if (getLexer().isNot(AsmToken::EndOfStatement))
2407 return TokError("unexpected token in '.fill' directive");
2408
2409 Lex();
2410 }
2411 }
2412
2413 if (FillSize != 1 && FillSize != 2 && FillSize != 4 && FillSize != 8)
2414 return TokError("invalid '.fill' size, expected 1, 2, 4, or 8");
2415
2416 for (uint64_t i = 0, e = NumValues; i != e; ++i)
2417 getStreamer().EmitIntValue(FillExpr, FillSize);
2418
2419 return false;
2420 }
2421
2422 /// parseDirectiveOrg
2423 /// ::= .org expression [ , expression ]
parseDirectiveOrg()2424 bool AsmParser::parseDirectiveOrg() {
2425 checkForValidSection();
2426
2427 const MCExpr *Offset;
2428 SMLoc Loc = getTok().getLoc();
2429 if (parseExpression(Offset))
2430 return true;
2431
2432 // Parse optional fill expression.
2433 int64_t FillExpr = 0;
2434 if (getLexer().isNot(AsmToken::EndOfStatement)) {
2435 if (getLexer().isNot(AsmToken::Comma))
2436 return TokError("unexpected token in '.org' directive");
2437 Lex();
2438
2439 if (parseAbsoluteExpression(FillExpr))
2440 return true;
2441
2442 if (getLexer().isNot(AsmToken::EndOfStatement))
2443 return TokError("unexpected token in '.org' directive");
2444 }
2445
2446 Lex();
2447
2448 // Only limited forms of relocatable expressions are accepted here, it
2449 // has to be relative to the current section. The streamer will return
2450 // 'true' if the expression wasn't evaluatable.
2451 if (getStreamer().EmitValueToOffset(Offset, FillExpr))
2452 return Error(Loc, "expected assembly-time absolute expression");
2453
2454 return false;
2455 }
2456
2457 /// parseDirectiveAlign
2458 /// ::= {.align, ...} expression [ , expression [ , expression ]]
parseDirectiveAlign(bool IsPow2,unsigned ValueSize)2459 bool AsmParser::parseDirectiveAlign(bool IsPow2, unsigned ValueSize) {
2460 checkForValidSection();
2461
2462 SMLoc AlignmentLoc = getLexer().getLoc();
2463 int64_t Alignment;
2464 if (parseAbsoluteExpression(Alignment))
2465 return true;
2466
2467 SMLoc MaxBytesLoc;
2468 bool HasFillExpr = false;
2469 int64_t FillExpr = 0;
2470 int64_t MaxBytesToFill = 0;
2471 if (getLexer().isNot(AsmToken::EndOfStatement)) {
2472 if (getLexer().isNot(AsmToken::Comma))
2473 return TokError("unexpected token in directive");
2474 Lex();
2475
2476 // The fill expression can be omitted while specifying a maximum number of
2477 // alignment bytes, e.g:
2478 // .align 3,,4
2479 if (getLexer().isNot(AsmToken::Comma)) {
2480 HasFillExpr = true;
2481 if (parseAbsoluteExpression(FillExpr))
2482 return true;
2483 }
2484
2485 if (getLexer().isNot(AsmToken::EndOfStatement)) {
2486 if (getLexer().isNot(AsmToken::Comma))
2487 return TokError("unexpected token in directive");
2488 Lex();
2489
2490 MaxBytesLoc = getLexer().getLoc();
2491 if (parseAbsoluteExpression(MaxBytesToFill))
2492 return true;
2493
2494 if (getLexer().isNot(AsmToken::EndOfStatement))
2495 return TokError("unexpected token in directive");
2496 }
2497 }
2498
2499 Lex();
2500
2501 if (!HasFillExpr)
2502 FillExpr = 0;
2503
2504 // Compute alignment in bytes.
2505 if (IsPow2) {
2506 // FIXME: Diagnose overflow.
2507 if (Alignment >= 32) {
2508 Error(AlignmentLoc, "invalid alignment value");
2509 Alignment = 31;
2510 }
2511
2512 Alignment = 1ULL << Alignment;
2513 } else {
2514 // Reject alignments that aren't a power of two, for gas compatibility.
2515 if (!isPowerOf2_64(Alignment))
2516 Error(AlignmentLoc, "alignment must be a power of 2");
2517 }
2518
2519 // Diagnose non-sensical max bytes to align.
2520 if (MaxBytesLoc.isValid()) {
2521 if (MaxBytesToFill < 1) {
2522 Error(MaxBytesLoc, "alignment directive can never be satisfied in this "
2523 "many bytes, ignoring maximum bytes expression");
2524 MaxBytesToFill = 0;
2525 }
2526
2527 if (MaxBytesToFill >= Alignment) {
2528 Warning(MaxBytesLoc, "maximum bytes expression exceeds alignment and "
2529 "has no effect");
2530 MaxBytesToFill = 0;
2531 }
2532 }
2533
2534 // Check whether we should use optimal code alignment for this .align
2535 // directive.
2536 bool UseCodeAlign = getStreamer().getCurrentSection().first->UseCodeAlign();
2537 if ((!HasFillExpr || Lexer.getMAI().getTextAlignFillValue() == FillExpr) &&
2538 ValueSize == 1 && UseCodeAlign) {
2539 getStreamer().EmitCodeAlignment(Alignment, MaxBytesToFill);
2540 } else {
2541 // FIXME: Target specific behavior about how the "extra" bytes are filled.
2542 getStreamer().EmitValueToAlignment(Alignment, FillExpr, ValueSize,
2543 MaxBytesToFill);
2544 }
2545
2546 return false;
2547 }
2548
2549 /// parseDirectiveFile
2550 /// ::= .file [number] filename
2551 /// ::= .file number directory filename
parseDirectiveFile(SMLoc DirectiveLoc)2552 bool AsmParser::parseDirectiveFile(SMLoc DirectiveLoc) {
2553 // FIXME: I'm not sure what this is.
2554 int64_t FileNumber = -1;
2555 SMLoc FileNumberLoc = getLexer().getLoc();
2556 if (getLexer().is(AsmToken::Integer)) {
2557 FileNumber = getTok().getIntVal();
2558 Lex();
2559
2560 if (FileNumber < 1)
2561 return TokError("file number less than one");
2562 }
2563
2564 if (getLexer().isNot(AsmToken::String))
2565 return TokError("unexpected token in '.file' directive");
2566
2567 // Usually the directory and filename together, otherwise just the directory.
2568 // Allow the strings to have escaped octal character sequence.
2569 std::string Path = getTok().getString();
2570 if (parseEscapedString(Path))
2571 return true;
2572 Lex();
2573
2574 StringRef Directory;
2575 StringRef Filename;
2576 std::string FilenameData;
2577 if (getLexer().is(AsmToken::String)) {
2578 if (FileNumber == -1)
2579 return TokError("explicit path specified, but no file number");
2580 if (parseEscapedString(FilenameData))
2581 return true;
2582 Filename = FilenameData;
2583 Directory = Path;
2584 Lex();
2585 } else {
2586 Filename = Path;
2587 }
2588
2589 if (getLexer().isNot(AsmToken::EndOfStatement))
2590 return TokError("unexpected token in '.file' directive");
2591
2592 if (FileNumber == -1)
2593 getStreamer().EmitFileDirective(Filename);
2594 else {
2595 if (getContext().getGenDwarfForAssembly() == true)
2596 Error(DirectiveLoc,
2597 "input can't have .file dwarf directives when -g is "
2598 "used to generate dwarf debug info for assembly code");
2599
2600 if (getStreamer().EmitDwarfFileDirective(FileNumber, Directory, Filename))
2601 Error(FileNumberLoc, "file number already allocated");
2602 }
2603
2604 return false;
2605 }
2606
2607 /// parseDirectiveLine
2608 /// ::= .line [number]
parseDirectiveLine()2609 bool AsmParser::parseDirectiveLine() {
2610 if (getLexer().isNot(AsmToken::EndOfStatement)) {
2611 if (getLexer().isNot(AsmToken::Integer))
2612 return TokError("unexpected token in '.line' directive");
2613
2614 int64_t LineNumber = getTok().getIntVal();
2615 (void)LineNumber;
2616 Lex();
2617
2618 // FIXME: Do something with the .line.
2619 }
2620
2621 if (getLexer().isNot(AsmToken::EndOfStatement))
2622 return TokError("unexpected token in '.line' directive");
2623
2624 return false;
2625 }
2626
2627 /// parseDirectiveLoc
2628 /// ::= .loc FileNumber [LineNumber] [ColumnPos] [basic_block] [prologue_end]
2629 /// [epilogue_begin] [is_stmt VALUE] [isa VALUE]
2630 /// The first number is a file number, must have been previously assigned with
2631 /// a .file directive, the second number is the line number and optionally the
2632 /// third number is a column position (zero if not specified). The remaining
2633 /// optional items are .loc sub-directives.
parseDirectiveLoc()2634 bool AsmParser::parseDirectiveLoc() {
2635 if (getLexer().isNot(AsmToken::Integer))
2636 return TokError("unexpected token in '.loc' directive");
2637 int64_t FileNumber = getTok().getIntVal();
2638 if (FileNumber < 1)
2639 return TokError("file number less than one in '.loc' directive");
2640 if (!getContext().isValidDwarfFileNumber(FileNumber))
2641 return TokError("unassigned file number in '.loc' directive");
2642 Lex();
2643
2644 int64_t LineNumber = 0;
2645 if (getLexer().is(AsmToken::Integer)) {
2646 LineNumber = getTok().getIntVal();
2647 if (LineNumber < 0)
2648 return TokError("line number less than zero in '.loc' directive");
2649 Lex();
2650 }
2651
2652 int64_t ColumnPos = 0;
2653 if (getLexer().is(AsmToken::Integer)) {
2654 ColumnPos = getTok().getIntVal();
2655 if (ColumnPos < 0)
2656 return TokError("column position less than zero in '.loc' directive");
2657 Lex();
2658 }
2659
2660 unsigned Flags = DWARF2_LINE_DEFAULT_IS_STMT ? DWARF2_FLAG_IS_STMT : 0;
2661 unsigned Isa = 0;
2662 int64_t Discriminator = 0;
2663 if (getLexer().isNot(AsmToken::EndOfStatement)) {
2664 for (;;) {
2665 if (getLexer().is(AsmToken::EndOfStatement))
2666 break;
2667
2668 StringRef Name;
2669 SMLoc Loc = getTok().getLoc();
2670 if (parseIdentifier(Name))
2671 return TokError("unexpected token in '.loc' directive");
2672
2673 if (Name == "basic_block")
2674 Flags |= DWARF2_FLAG_BASIC_BLOCK;
2675 else if (Name == "prologue_end")
2676 Flags |= DWARF2_FLAG_PROLOGUE_END;
2677 else if (Name == "epilogue_begin")
2678 Flags |= DWARF2_FLAG_EPILOGUE_BEGIN;
2679 else if (Name == "is_stmt") {
2680 Loc = getTok().getLoc();
2681 const MCExpr *Value;
2682 if (parseExpression(Value))
2683 return true;
2684 // The expression must be the constant 0 or 1.
2685 if (const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value)) {
2686 int Value = MCE->getValue();
2687 if (Value == 0)
2688 Flags &= ~DWARF2_FLAG_IS_STMT;
2689 else if (Value == 1)
2690 Flags |= DWARF2_FLAG_IS_STMT;
2691 else
2692 return Error(Loc, "is_stmt value not 0 or 1");
2693 } else {
2694 return Error(Loc, "is_stmt value not the constant value of 0 or 1");
2695 }
2696 } else if (Name == "isa") {
2697 Loc = getTok().getLoc();
2698 const MCExpr *Value;
2699 if (parseExpression(Value))
2700 return true;
2701 // The expression must be a constant greater or equal to 0.
2702 if (const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value)) {
2703 int Value = MCE->getValue();
2704 if (Value < 0)
2705 return Error(Loc, "isa number less than zero");
2706 Isa = Value;
2707 } else {
2708 return Error(Loc, "isa number not a constant value");
2709 }
2710 } else if (Name == "discriminator") {
2711 if (parseAbsoluteExpression(Discriminator))
2712 return true;
2713 } else {
2714 return Error(Loc, "unknown sub-directive in '.loc' directive");
2715 }
2716
2717 if (getLexer().is(AsmToken::EndOfStatement))
2718 break;
2719 }
2720 }
2721
2722 getStreamer().EmitDwarfLocDirective(FileNumber, LineNumber, ColumnPos, Flags,
2723 Isa, Discriminator, StringRef());
2724
2725 return false;
2726 }
2727
2728 /// parseDirectiveStabs
2729 /// ::= .stabs string, number, number, number
parseDirectiveStabs()2730 bool AsmParser::parseDirectiveStabs() {
2731 return TokError("unsupported directive '.stabs'");
2732 }
2733
2734 /// parseDirectiveCFISections
2735 /// ::= .cfi_sections section [, section]
parseDirectiveCFISections()2736 bool AsmParser::parseDirectiveCFISections() {
2737 StringRef Name;
2738 bool EH = false;
2739 bool Debug = false;
2740
2741 if (parseIdentifier(Name))
2742 return TokError("Expected an identifier");
2743
2744 if (Name == ".eh_frame")
2745 EH = true;
2746 else if (Name == ".debug_frame")
2747 Debug = true;
2748
2749 if (getLexer().is(AsmToken::Comma)) {
2750 Lex();
2751
2752 if (parseIdentifier(Name))
2753 return TokError("Expected an identifier");
2754
2755 if (Name == ".eh_frame")
2756 EH = true;
2757 else if (Name == ".debug_frame")
2758 Debug = true;
2759 }
2760
2761 getStreamer().EmitCFISections(EH, Debug);
2762 return false;
2763 }
2764
2765 /// parseDirectiveCFIStartProc
2766 /// ::= .cfi_startproc
parseDirectiveCFIStartProc()2767 bool AsmParser::parseDirectiveCFIStartProc() {
2768 getStreamer().EmitCFIStartProc();
2769 return false;
2770 }
2771
2772 /// parseDirectiveCFIEndProc
2773 /// ::= .cfi_endproc
parseDirectiveCFIEndProc()2774 bool AsmParser::parseDirectiveCFIEndProc() {
2775 getStreamer().EmitCFIEndProc();
2776 return false;
2777 }
2778
2779 /// \brief parse register name or number.
parseRegisterOrRegisterNumber(int64_t & Register,SMLoc DirectiveLoc)2780 bool AsmParser::parseRegisterOrRegisterNumber(int64_t &Register,
2781 SMLoc DirectiveLoc) {
2782 unsigned RegNo;
2783
2784 if (getLexer().isNot(AsmToken::Integer)) {
2785 if (getTargetParser().ParseRegister(RegNo, DirectiveLoc, DirectiveLoc))
2786 return true;
2787 Register = getContext().getRegisterInfo()->getDwarfRegNum(RegNo, true);
2788 } else
2789 return parseAbsoluteExpression(Register);
2790
2791 return false;
2792 }
2793
2794 /// parseDirectiveCFIDefCfa
2795 /// ::= .cfi_def_cfa register, offset
parseDirectiveCFIDefCfa(SMLoc DirectiveLoc)2796 bool AsmParser::parseDirectiveCFIDefCfa(SMLoc DirectiveLoc) {
2797 int64_t Register = 0;
2798 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc))
2799 return true;
2800
2801 if (getLexer().isNot(AsmToken::Comma))
2802 return TokError("unexpected token in directive");
2803 Lex();
2804
2805 int64_t Offset = 0;
2806 if (parseAbsoluteExpression(Offset))
2807 return true;
2808
2809 getStreamer().EmitCFIDefCfa(Register, Offset);
2810 return false;
2811 }
2812
2813 /// parseDirectiveCFIDefCfaOffset
2814 /// ::= .cfi_def_cfa_offset offset
parseDirectiveCFIDefCfaOffset()2815 bool AsmParser::parseDirectiveCFIDefCfaOffset() {
2816 int64_t Offset = 0;
2817 if (parseAbsoluteExpression(Offset))
2818 return true;
2819
2820 getStreamer().EmitCFIDefCfaOffset(Offset);
2821 return false;
2822 }
2823
2824 /// parseDirectiveCFIRegister
2825 /// ::= .cfi_register register, register
parseDirectiveCFIRegister(SMLoc DirectiveLoc)2826 bool AsmParser::parseDirectiveCFIRegister(SMLoc DirectiveLoc) {
2827 int64_t Register1 = 0;
2828 if (parseRegisterOrRegisterNumber(Register1, DirectiveLoc))
2829 return true;
2830
2831 if (getLexer().isNot(AsmToken::Comma))
2832 return TokError("unexpected token in directive");
2833 Lex();
2834
2835 int64_t Register2 = 0;
2836 if (parseRegisterOrRegisterNumber(Register2, DirectiveLoc))
2837 return true;
2838
2839 getStreamer().EmitCFIRegister(Register1, Register2);
2840 return false;
2841 }
2842
2843 /// parseDirectiveCFIWindowSave
2844 /// ::= .cfi_window_save
parseDirectiveCFIWindowSave()2845 bool AsmParser::parseDirectiveCFIWindowSave() {
2846 getStreamer().EmitCFIWindowSave();
2847 return false;
2848 }
2849
2850 /// parseDirectiveCFIAdjustCfaOffset
2851 /// ::= .cfi_adjust_cfa_offset adjustment
parseDirectiveCFIAdjustCfaOffset()2852 bool AsmParser::parseDirectiveCFIAdjustCfaOffset() {
2853 int64_t Adjustment = 0;
2854 if (parseAbsoluteExpression(Adjustment))
2855 return true;
2856
2857 getStreamer().EmitCFIAdjustCfaOffset(Adjustment);
2858 return false;
2859 }
2860
2861 /// parseDirectiveCFIDefCfaRegister
2862 /// ::= .cfi_def_cfa_register register
parseDirectiveCFIDefCfaRegister(SMLoc DirectiveLoc)2863 bool AsmParser::parseDirectiveCFIDefCfaRegister(SMLoc DirectiveLoc) {
2864 int64_t Register = 0;
2865 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc))
2866 return true;
2867
2868 getStreamer().EmitCFIDefCfaRegister(Register);
2869 return false;
2870 }
2871
2872 /// parseDirectiveCFIOffset
2873 /// ::= .cfi_offset register, offset
parseDirectiveCFIOffset(SMLoc DirectiveLoc)2874 bool AsmParser::parseDirectiveCFIOffset(SMLoc DirectiveLoc) {
2875 int64_t Register = 0;
2876 int64_t Offset = 0;
2877
2878 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc))
2879 return true;
2880
2881 if (getLexer().isNot(AsmToken::Comma))
2882 return TokError("unexpected token in directive");
2883 Lex();
2884
2885 if (parseAbsoluteExpression(Offset))
2886 return true;
2887
2888 getStreamer().EmitCFIOffset(Register, Offset);
2889 return false;
2890 }
2891
2892 /// parseDirectiveCFIRelOffset
2893 /// ::= .cfi_rel_offset register, offset
parseDirectiveCFIRelOffset(SMLoc DirectiveLoc)2894 bool AsmParser::parseDirectiveCFIRelOffset(SMLoc DirectiveLoc) {
2895 int64_t Register = 0;
2896
2897 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc))
2898 return true;
2899
2900 if (getLexer().isNot(AsmToken::Comma))
2901 return TokError("unexpected token in directive");
2902 Lex();
2903
2904 int64_t Offset = 0;
2905 if (parseAbsoluteExpression(Offset))
2906 return true;
2907
2908 getStreamer().EmitCFIRelOffset(Register, Offset);
2909 return false;
2910 }
2911
isValidEncoding(int64_t Encoding)2912 static bool isValidEncoding(int64_t Encoding) {
2913 if (Encoding & ~0xff)
2914 return false;
2915
2916 if (Encoding == dwarf::DW_EH_PE_omit)
2917 return true;
2918
2919 const unsigned Format = Encoding & 0xf;
2920 if (Format != dwarf::DW_EH_PE_absptr && Format != dwarf::DW_EH_PE_udata2 &&
2921 Format != dwarf::DW_EH_PE_udata4 && Format != dwarf::DW_EH_PE_udata8 &&
2922 Format != dwarf::DW_EH_PE_sdata2 && Format != dwarf::DW_EH_PE_sdata4 &&
2923 Format != dwarf::DW_EH_PE_sdata8 && Format != dwarf::DW_EH_PE_signed)
2924 return false;
2925
2926 const unsigned Application = Encoding & 0x70;
2927 if (Application != dwarf::DW_EH_PE_absptr &&
2928 Application != dwarf::DW_EH_PE_pcrel)
2929 return false;
2930
2931 return true;
2932 }
2933
2934 /// parseDirectiveCFIPersonalityOrLsda
2935 /// IsPersonality true for cfi_personality, false for cfi_lsda
2936 /// ::= .cfi_personality encoding, [symbol_name]
2937 /// ::= .cfi_lsda encoding, [symbol_name]
parseDirectiveCFIPersonalityOrLsda(bool IsPersonality)2938 bool AsmParser::parseDirectiveCFIPersonalityOrLsda(bool IsPersonality) {
2939 int64_t Encoding = 0;
2940 if (parseAbsoluteExpression(Encoding))
2941 return true;
2942 if (Encoding == dwarf::DW_EH_PE_omit)
2943 return false;
2944
2945 if (!isValidEncoding(Encoding))
2946 return TokError("unsupported encoding.");
2947
2948 if (getLexer().isNot(AsmToken::Comma))
2949 return TokError("unexpected token in directive");
2950 Lex();
2951
2952 StringRef Name;
2953 if (parseIdentifier(Name))
2954 return TokError("expected identifier in directive");
2955
2956 MCSymbol *Sym = getContext().GetOrCreateSymbol(Name);
2957
2958 if (IsPersonality)
2959 getStreamer().EmitCFIPersonality(Sym, Encoding);
2960 else
2961 getStreamer().EmitCFILsda(Sym, Encoding);
2962 return false;
2963 }
2964
2965 /// parseDirectiveCFIRememberState
2966 /// ::= .cfi_remember_state
parseDirectiveCFIRememberState()2967 bool AsmParser::parseDirectiveCFIRememberState() {
2968 getStreamer().EmitCFIRememberState();
2969 return false;
2970 }
2971
2972 /// parseDirectiveCFIRestoreState
2973 /// ::= .cfi_remember_state
parseDirectiveCFIRestoreState()2974 bool AsmParser::parseDirectiveCFIRestoreState() {
2975 getStreamer().EmitCFIRestoreState();
2976 return false;
2977 }
2978
2979 /// parseDirectiveCFISameValue
2980 /// ::= .cfi_same_value register
parseDirectiveCFISameValue(SMLoc DirectiveLoc)2981 bool AsmParser::parseDirectiveCFISameValue(SMLoc DirectiveLoc) {
2982 int64_t Register = 0;
2983
2984 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc))
2985 return true;
2986
2987 getStreamer().EmitCFISameValue(Register);
2988 return false;
2989 }
2990
2991 /// parseDirectiveCFIRestore
2992 /// ::= .cfi_restore register
parseDirectiveCFIRestore(SMLoc DirectiveLoc)2993 bool AsmParser::parseDirectiveCFIRestore(SMLoc DirectiveLoc) {
2994 int64_t Register = 0;
2995 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc))
2996 return true;
2997
2998 getStreamer().EmitCFIRestore(Register);
2999 return false;
3000 }
3001
3002 /// parseDirectiveCFIEscape
3003 /// ::= .cfi_escape expression[,...]
parseDirectiveCFIEscape()3004 bool AsmParser::parseDirectiveCFIEscape() {
3005 std::string Values;
3006 int64_t CurrValue;
3007 if (parseAbsoluteExpression(CurrValue))
3008 return true;
3009
3010 Values.push_back((uint8_t)CurrValue);
3011
3012 while (getLexer().is(AsmToken::Comma)) {
3013 Lex();
3014
3015 if (parseAbsoluteExpression(CurrValue))
3016 return true;
3017
3018 Values.push_back((uint8_t)CurrValue);
3019 }
3020
3021 getStreamer().EmitCFIEscape(Values);
3022 return false;
3023 }
3024
3025 /// parseDirectiveCFISignalFrame
3026 /// ::= .cfi_signal_frame
parseDirectiveCFISignalFrame()3027 bool AsmParser::parseDirectiveCFISignalFrame() {
3028 if (getLexer().isNot(AsmToken::EndOfStatement))
3029 return Error(getLexer().getLoc(),
3030 "unexpected token in '.cfi_signal_frame'");
3031
3032 getStreamer().EmitCFISignalFrame();
3033 return false;
3034 }
3035
3036 /// parseDirectiveCFIUndefined
3037 /// ::= .cfi_undefined register
parseDirectiveCFIUndefined(SMLoc DirectiveLoc)3038 bool AsmParser::parseDirectiveCFIUndefined(SMLoc DirectiveLoc) {
3039 int64_t Register = 0;
3040
3041 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc))
3042 return true;
3043
3044 getStreamer().EmitCFIUndefined(Register);
3045 return false;
3046 }
3047
3048 /// parseDirectiveMacrosOnOff
3049 /// ::= .macros_on
3050 /// ::= .macros_off
parseDirectiveMacrosOnOff(StringRef Directive)3051 bool AsmParser::parseDirectiveMacrosOnOff(StringRef Directive) {
3052 if (getLexer().isNot(AsmToken::EndOfStatement))
3053 return Error(getLexer().getLoc(),
3054 "unexpected token in '" + Directive + "' directive");
3055
3056 setMacrosEnabled(Directive == ".macros_on");
3057 return false;
3058 }
3059
3060 /// parseDirectiveMacro
3061 /// ::= .macro name [parameters]
parseDirectiveMacro(SMLoc DirectiveLoc)3062 bool AsmParser::parseDirectiveMacro(SMLoc DirectiveLoc) {
3063 StringRef Name;
3064 if (parseIdentifier(Name))
3065 return TokError("expected identifier in '.macro' directive");
3066
3067 MCAsmMacroParameters Parameters;
3068 // Argument delimiter is initially unknown. It will be set by
3069 // parseMacroArgument()
3070 AsmToken::TokenKind ArgumentDelimiter = AsmToken::Eof;
3071 if (getLexer().isNot(AsmToken::EndOfStatement)) {
3072 for (;;) {
3073 MCAsmMacroParameter Parameter;
3074 if (parseIdentifier(Parameter.first))
3075 return TokError("expected identifier in '.macro' directive");
3076
3077 if (getLexer().is(AsmToken::Equal)) {
3078 Lex();
3079 if (parseMacroArgument(Parameter.second, ArgumentDelimiter))
3080 return true;
3081 }
3082
3083 Parameters.push_back(Parameter);
3084
3085 if (getLexer().is(AsmToken::Comma))
3086 Lex();
3087 else if (getLexer().is(AsmToken::EndOfStatement))
3088 break;
3089 }
3090 }
3091
3092 // Eat the end of statement.
3093 Lex();
3094
3095 AsmToken EndToken, StartToken = getTok();
3096
3097 // Lex the macro definition.
3098 for (;;) {
3099 // Check whether we have reached the end of the file.
3100 if (getLexer().is(AsmToken::Eof))
3101 return Error(DirectiveLoc, "no matching '.endmacro' in definition");
3102
3103 // Otherwise, check whether we have reach the .endmacro.
3104 if (getLexer().is(AsmToken::Identifier) &&
3105 (getTok().getIdentifier() == ".endm" ||
3106 getTok().getIdentifier() == ".endmacro")) {
3107 EndToken = getTok();
3108 Lex();
3109 if (getLexer().isNot(AsmToken::EndOfStatement))
3110 return TokError("unexpected token in '" + EndToken.getIdentifier() +
3111 "' directive");
3112 break;
3113 }
3114
3115 // Otherwise, scan til the end of the statement.
3116 eatToEndOfStatement();
3117 }
3118
3119 if (lookupMacro(Name)) {
3120 return Error(DirectiveLoc, "macro '" + Name + "' is already defined");
3121 }
3122
3123 const char *BodyStart = StartToken.getLoc().getPointer();
3124 const char *BodyEnd = EndToken.getLoc().getPointer();
3125 StringRef Body = StringRef(BodyStart, BodyEnd - BodyStart);
3126 checkForBadMacro(DirectiveLoc, Name, Body, Parameters);
3127 defineMacro(Name, MCAsmMacro(Name, Body, Parameters));
3128 return false;
3129 }
3130
3131 /// checkForBadMacro
3132 ///
3133 /// With the support added for named parameters there may be code out there that
3134 /// is transitioning from positional parameters. In versions of gas that did
3135 /// not support named parameters they would be ignored on the macro defintion.
3136 /// But to support both styles of parameters this is not possible so if a macro
3137 /// defintion has named parameters but does not use them and has what appears
3138 /// to be positional parameters, strings like $1, $2, ... and $n, then issue a
3139 /// warning that the positional parameter found in body which have no effect.
3140 /// Hoping the developer will either remove the named parameters from the macro
3141 /// definiton so the positional parameters get used if that was what was
3142 /// intended or change the macro to use the named parameters. It is possible
3143 /// this warning will trigger when the none of the named parameters are used
3144 /// and the strings like $1 are infact to simply to be passed trough unchanged.
checkForBadMacro(SMLoc DirectiveLoc,StringRef Name,StringRef Body,MCAsmMacroParameters Parameters)3145 void AsmParser::checkForBadMacro(SMLoc DirectiveLoc, StringRef Name,
3146 StringRef Body,
3147 MCAsmMacroParameters Parameters) {
3148 // If this macro is not defined with named parameters the warning we are
3149 // checking for here doesn't apply.
3150 unsigned NParameters = Parameters.size();
3151 if (NParameters == 0)
3152 return;
3153
3154 bool NamedParametersFound = false;
3155 bool PositionalParametersFound = false;
3156
3157 // Look at the body of the macro for use of both the named parameters and what
3158 // are likely to be positional parameters. This is what expandMacro() is
3159 // doing when it finds the parameters in the body.
3160 while (!Body.empty()) {
3161 // Scan for the next possible parameter.
3162 std::size_t End = Body.size(), Pos = 0;
3163 for (; Pos != End; ++Pos) {
3164 // Check for a substitution or escape.
3165 // This macro is defined with parameters, look for \foo, \bar, etc.
3166 if (Body[Pos] == '\\' && Pos + 1 != End)
3167 break;
3168
3169 // This macro should have parameters, but look for $0, $1, ..., $n too.
3170 if (Body[Pos] != '$' || Pos + 1 == End)
3171 continue;
3172 char Next = Body[Pos + 1];
3173 if (Next == '$' || Next == 'n' ||
3174 isdigit(static_cast<unsigned char>(Next)))
3175 break;
3176 }
3177
3178 // Check if we reached the end.
3179 if (Pos == End)
3180 break;
3181
3182 if (Body[Pos] == '$') {
3183 switch (Body[Pos + 1]) {
3184 // $$ => $
3185 case '$':
3186 break;
3187
3188 // $n => number of arguments
3189 case 'n':
3190 PositionalParametersFound = true;
3191 break;
3192
3193 // $[0-9] => argument
3194 default: {
3195 PositionalParametersFound = true;
3196 break;
3197 }
3198 }
3199 Pos += 2;
3200 } else {
3201 unsigned I = Pos + 1;
3202 while (isIdentifierChar(Body[I]) && I + 1 != End)
3203 ++I;
3204
3205 const char *Begin = Body.data() + Pos + 1;
3206 StringRef Argument(Begin, I - (Pos + 1));
3207 unsigned Index = 0;
3208 for (; Index < NParameters; ++Index)
3209 if (Parameters[Index].first == Argument)
3210 break;
3211
3212 if (Index == NParameters) {
3213 if (Body[Pos + 1] == '(' && Body[Pos + 2] == ')')
3214 Pos += 3;
3215 else {
3216 Pos = I;
3217 }
3218 } else {
3219 NamedParametersFound = true;
3220 Pos += 1 + Argument.size();
3221 }
3222 }
3223 // Update the scan point.
3224 Body = Body.substr(Pos);
3225 }
3226
3227 if (!NamedParametersFound && PositionalParametersFound)
3228 Warning(DirectiveLoc, "macro defined with named parameters which are not "
3229 "used in macro body, possible positional parameter "
3230 "found in body which will have no effect");
3231 }
3232
3233 /// parseDirectiveEndMacro
3234 /// ::= .endm
3235 /// ::= .endmacro
parseDirectiveEndMacro(StringRef Directive)3236 bool AsmParser::parseDirectiveEndMacro(StringRef Directive) {
3237 if (getLexer().isNot(AsmToken::EndOfStatement))
3238 return TokError("unexpected token in '" + Directive + "' directive");
3239
3240 // If we are inside a macro instantiation, terminate the current
3241 // instantiation.
3242 if (isInsideMacroInstantiation()) {
3243 handleMacroExit();
3244 return false;
3245 }
3246
3247 // Otherwise, this .endmacro is a stray entry in the file; well formed
3248 // .endmacro directives are handled during the macro definition parsing.
3249 return TokError("unexpected '" + Directive + "' in file, "
3250 "no current macro definition");
3251 }
3252
3253 /// parseDirectivePurgeMacro
3254 /// ::= .purgem
parseDirectivePurgeMacro(SMLoc DirectiveLoc)3255 bool AsmParser::parseDirectivePurgeMacro(SMLoc DirectiveLoc) {
3256 StringRef Name;
3257 if (parseIdentifier(Name))
3258 return TokError("expected identifier in '.purgem' directive");
3259
3260 if (getLexer().isNot(AsmToken::EndOfStatement))
3261 return TokError("unexpected token in '.purgem' directive");
3262
3263 if (!lookupMacro(Name))
3264 return Error(DirectiveLoc, "macro '" + Name + "' is not defined");
3265
3266 undefineMacro(Name);
3267 return false;
3268 }
3269
3270 /// parseDirectiveBundleAlignMode
3271 /// ::= {.bundle_align_mode} expression
parseDirectiveBundleAlignMode()3272 bool AsmParser::parseDirectiveBundleAlignMode() {
3273 checkForValidSection();
3274
3275 // Expect a single argument: an expression that evaluates to a constant
3276 // in the inclusive range 0-30.
3277 SMLoc ExprLoc = getLexer().getLoc();
3278 int64_t AlignSizePow2;
3279 if (parseAbsoluteExpression(AlignSizePow2))
3280 return true;
3281 else if (getLexer().isNot(AsmToken::EndOfStatement))
3282 return TokError("unexpected token after expression in"
3283 " '.bundle_align_mode' directive");
3284 else if (AlignSizePow2 < 0 || AlignSizePow2 > 30)
3285 return Error(ExprLoc,
3286 "invalid bundle alignment size (expected between 0 and 30)");
3287
3288 Lex();
3289
3290 // Because of AlignSizePow2's verified range we can safely truncate it to
3291 // unsigned.
3292 getStreamer().EmitBundleAlignMode(static_cast<unsigned>(AlignSizePow2));
3293 return false;
3294 }
3295
3296 /// parseDirectiveBundleLock
3297 /// ::= {.bundle_lock} [align_to_end]
parseDirectiveBundleLock()3298 bool AsmParser::parseDirectiveBundleLock() {
3299 checkForValidSection();
3300 bool AlignToEnd = false;
3301
3302 if (getLexer().isNot(AsmToken::EndOfStatement)) {
3303 StringRef Option;
3304 SMLoc Loc = getTok().getLoc();
3305 const char *kInvalidOptionError =
3306 "invalid option for '.bundle_lock' directive";
3307
3308 if (parseIdentifier(Option))
3309 return Error(Loc, kInvalidOptionError);
3310
3311 if (Option != "align_to_end")
3312 return Error(Loc, kInvalidOptionError);
3313 else if (getLexer().isNot(AsmToken::EndOfStatement))
3314 return Error(Loc,
3315 "unexpected token after '.bundle_lock' directive option");
3316 AlignToEnd = true;
3317 }
3318
3319 Lex();
3320
3321 getStreamer().EmitBundleLock(AlignToEnd);
3322 return false;
3323 }
3324
3325 /// parseDirectiveBundleLock
3326 /// ::= {.bundle_lock}
parseDirectiveBundleUnlock()3327 bool AsmParser::parseDirectiveBundleUnlock() {
3328 checkForValidSection();
3329
3330 if (getLexer().isNot(AsmToken::EndOfStatement))
3331 return TokError("unexpected token in '.bundle_unlock' directive");
3332 Lex();
3333
3334 getStreamer().EmitBundleUnlock();
3335 return false;
3336 }
3337
3338 /// parseDirectiveSpace
3339 /// ::= (.skip | .space) expression [ , expression ]
parseDirectiveSpace(StringRef IDVal)3340 bool AsmParser::parseDirectiveSpace(StringRef IDVal) {
3341 checkForValidSection();
3342
3343 int64_t NumBytes;
3344 if (parseAbsoluteExpression(NumBytes))
3345 return true;
3346
3347 int64_t FillExpr = 0;
3348 if (getLexer().isNot(AsmToken::EndOfStatement)) {
3349 if (getLexer().isNot(AsmToken::Comma))
3350 return TokError("unexpected token in '" + Twine(IDVal) + "' directive");
3351 Lex();
3352
3353 if (parseAbsoluteExpression(FillExpr))
3354 return true;
3355
3356 if (getLexer().isNot(AsmToken::EndOfStatement))
3357 return TokError("unexpected token in '" + Twine(IDVal) + "' directive");
3358 }
3359
3360 Lex();
3361
3362 if (NumBytes <= 0)
3363 return TokError("invalid number of bytes in '" + Twine(IDVal) +
3364 "' directive");
3365
3366 // FIXME: Sometimes the fill expr is 'nop' if it isn't supplied, instead of 0.
3367 getStreamer().EmitFill(NumBytes, FillExpr);
3368
3369 return false;
3370 }
3371
3372 /// parseDirectiveLEB128
3373 /// ::= (.sleb128 | .uleb128) [ expression (, expression)* ]
parseDirectiveLEB128(bool Signed)3374 bool AsmParser::parseDirectiveLEB128(bool Signed) {
3375 checkForValidSection();
3376 const MCExpr *Value;
3377
3378 for (;;) {
3379 if (parseExpression(Value))
3380 return true;
3381
3382 if (Signed)
3383 getStreamer().EmitSLEB128Value(Value);
3384 else
3385 getStreamer().EmitULEB128Value(Value);
3386
3387 if (getLexer().is(AsmToken::EndOfStatement))
3388 break;
3389
3390 if (getLexer().isNot(AsmToken::Comma))
3391 return TokError("unexpected token in directive");
3392 Lex();
3393 }
3394
3395 return false;
3396 }
3397
3398 /// parseDirectiveSymbolAttribute
3399 /// ::= { ".globl", ".weak", ... } [ identifier ( , identifier )* ]
parseDirectiveSymbolAttribute(MCSymbolAttr Attr)3400 bool AsmParser::parseDirectiveSymbolAttribute(MCSymbolAttr Attr) {
3401 if (getLexer().isNot(AsmToken::EndOfStatement)) {
3402 for (;;) {
3403 StringRef Name;
3404 SMLoc Loc = getTok().getLoc();
3405
3406 if (parseIdentifier(Name))
3407 return Error(Loc, "expected identifier in directive");
3408
3409 MCSymbol *Sym = getContext().GetOrCreateSymbol(Name);
3410
3411 // Assembler local symbols don't make any sense here. Complain loudly.
3412 if (Sym->isTemporary())
3413 return Error(Loc, "non-local symbol required in directive");
3414
3415 if (!getStreamer().EmitSymbolAttribute(Sym, Attr))
3416 return Error(Loc, "unable to emit symbol attribute");
3417
3418 if (getLexer().is(AsmToken::EndOfStatement))
3419 break;
3420
3421 if (getLexer().isNot(AsmToken::Comma))
3422 return TokError("unexpected token in directive");
3423 Lex();
3424 }
3425 }
3426
3427 Lex();
3428 return false;
3429 }
3430
3431 /// parseDirectiveComm
3432 /// ::= ( .comm | .lcomm ) identifier , size_expression [ , align_expression ]
parseDirectiveComm(bool IsLocal)3433 bool AsmParser::parseDirectiveComm(bool IsLocal) {
3434 checkForValidSection();
3435
3436 SMLoc IDLoc = getLexer().getLoc();
3437 StringRef Name;
3438 if (parseIdentifier(Name))
3439 return TokError("expected identifier in directive");
3440
3441 // Handle the identifier as the key symbol.
3442 MCSymbol *Sym = getContext().GetOrCreateSymbol(Name);
3443
3444 if (getLexer().isNot(AsmToken::Comma))
3445 return TokError("unexpected token in directive");
3446 Lex();
3447
3448 int64_t Size;
3449 SMLoc SizeLoc = getLexer().getLoc();
3450 if (parseAbsoluteExpression(Size))
3451 return true;
3452
3453 int64_t Pow2Alignment = 0;
3454 SMLoc Pow2AlignmentLoc;
3455 if (getLexer().is(AsmToken::Comma)) {
3456 Lex();
3457 Pow2AlignmentLoc = getLexer().getLoc();
3458 if (parseAbsoluteExpression(Pow2Alignment))
3459 return true;
3460
3461 LCOMM::LCOMMType LCOMM = Lexer.getMAI().getLCOMMDirectiveAlignmentType();
3462 if (IsLocal && LCOMM == LCOMM::NoAlignment)
3463 return Error(Pow2AlignmentLoc, "alignment not supported on this target");
3464
3465 // If this target takes alignments in bytes (not log) validate and convert.
3466 if ((!IsLocal && Lexer.getMAI().getCOMMDirectiveAlignmentIsInBytes()) ||
3467 (IsLocal && LCOMM == LCOMM::ByteAlignment)) {
3468 if (!isPowerOf2_64(Pow2Alignment))
3469 return Error(Pow2AlignmentLoc, "alignment must be a power of 2");
3470 Pow2Alignment = Log2_64(Pow2Alignment);
3471 }
3472 }
3473
3474 if (getLexer().isNot(AsmToken::EndOfStatement))
3475 return TokError("unexpected token in '.comm' or '.lcomm' directive");
3476
3477 Lex();
3478
3479 // NOTE: a size of zero for a .comm should create a undefined symbol
3480 // but a size of .lcomm creates a bss symbol of size zero.
3481 if (Size < 0)
3482 return Error(SizeLoc, "invalid '.comm' or '.lcomm' directive size, can't "
3483 "be less than zero");
3484
3485 // NOTE: The alignment in the directive is a power of 2 value, the assembler
3486 // may internally end up wanting an alignment in bytes.
3487 // FIXME: Diagnose overflow.
3488 if (Pow2Alignment < 0)
3489 return Error(Pow2AlignmentLoc, "invalid '.comm' or '.lcomm' directive "
3490 "alignment, can't be less than zero");
3491
3492 if (!Sym->isUndefined())
3493 return Error(IDLoc, "invalid symbol redefinition");
3494
3495 // Create the Symbol as a common or local common with Size and Pow2Alignment
3496 if (IsLocal) {
3497 getStreamer().EmitLocalCommonSymbol(Sym, Size, 1 << Pow2Alignment);
3498 return false;
3499 }
3500
3501 getStreamer().EmitCommonSymbol(Sym, Size, 1 << Pow2Alignment);
3502 return false;
3503 }
3504
3505 /// parseDirectiveAbort
3506 /// ::= .abort [... message ...]
parseDirectiveAbort()3507 bool AsmParser::parseDirectiveAbort() {
3508 // FIXME: Use loc from directive.
3509 SMLoc Loc = getLexer().getLoc();
3510
3511 StringRef Str = parseStringToEndOfStatement();
3512 if (getLexer().isNot(AsmToken::EndOfStatement))
3513 return TokError("unexpected token in '.abort' directive");
3514
3515 Lex();
3516
3517 if (Str.empty())
3518 Error(Loc, ".abort detected. Assembly stopping.");
3519 else
3520 Error(Loc, ".abort '" + Str + "' detected. Assembly stopping.");
3521 // FIXME: Actually abort assembly here.
3522
3523 return false;
3524 }
3525
3526 /// parseDirectiveInclude
3527 /// ::= .include "filename"
parseDirectiveInclude()3528 bool AsmParser::parseDirectiveInclude() {
3529 if (getLexer().isNot(AsmToken::String))
3530 return TokError("expected string in '.include' directive");
3531
3532 // Allow the strings to have escaped octal character sequence.
3533 std::string Filename;
3534 if (parseEscapedString(Filename))
3535 return true;
3536 SMLoc IncludeLoc = getLexer().getLoc();
3537 Lex();
3538
3539 if (getLexer().isNot(AsmToken::EndOfStatement))
3540 return TokError("unexpected token in '.include' directive");
3541
3542 // Attempt to switch the lexer to the included file before consuming the end
3543 // of statement to avoid losing it when we switch.
3544 if (enterIncludeFile(Filename)) {
3545 Error(IncludeLoc, "Could not find include file '" + Filename + "'");
3546 return true;
3547 }
3548
3549 return false;
3550 }
3551
3552 /// parseDirectiveIncbin
3553 /// ::= .incbin "filename"
parseDirectiveIncbin()3554 bool AsmParser::parseDirectiveIncbin() {
3555 if (getLexer().isNot(AsmToken::String))
3556 return TokError("expected string in '.incbin' directive");
3557
3558 // Allow the strings to have escaped octal character sequence.
3559 std::string Filename;
3560 if (parseEscapedString(Filename))
3561 return true;
3562 SMLoc IncbinLoc = getLexer().getLoc();
3563 Lex();
3564
3565 if (getLexer().isNot(AsmToken::EndOfStatement))
3566 return TokError("unexpected token in '.incbin' directive");
3567
3568 // Attempt to process the included file.
3569 if (processIncbinFile(Filename)) {
3570 Error(IncbinLoc, "Could not find incbin file '" + Filename + "'");
3571 return true;
3572 }
3573
3574 return false;
3575 }
3576
3577 /// parseDirectiveIf
3578 /// ::= .if expression
parseDirectiveIf(SMLoc DirectiveLoc)3579 bool AsmParser::parseDirectiveIf(SMLoc DirectiveLoc) {
3580 TheCondStack.push_back(TheCondState);
3581 TheCondState.TheCond = AsmCond::IfCond;
3582 if (TheCondState.Ignore) {
3583 eatToEndOfStatement();
3584 } else {
3585 int64_t ExprValue;
3586 if (parseAbsoluteExpression(ExprValue))
3587 return true;
3588
3589 if (getLexer().isNot(AsmToken::EndOfStatement))
3590 return TokError("unexpected token in '.if' directive");
3591
3592 Lex();
3593
3594 TheCondState.CondMet = ExprValue;
3595 TheCondState.Ignore = !TheCondState.CondMet;
3596 }
3597
3598 return false;
3599 }
3600
3601 /// parseDirectiveIfb
3602 /// ::= .ifb string
parseDirectiveIfb(SMLoc DirectiveLoc,bool ExpectBlank)3603 bool AsmParser::parseDirectiveIfb(SMLoc DirectiveLoc, bool ExpectBlank) {
3604 TheCondStack.push_back(TheCondState);
3605 TheCondState.TheCond = AsmCond::IfCond;
3606
3607 if (TheCondState.Ignore) {
3608 eatToEndOfStatement();
3609 } else {
3610 StringRef Str = parseStringToEndOfStatement();
3611
3612 if (getLexer().isNot(AsmToken::EndOfStatement))
3613 return TokError("unexpected token in '.ifb' directive");
3614
3615 Lex();
3616
3617 TheCondState.CondMet = ExpectBlank == Str.empty();
3618 TheCondState.Ignore = !TheCondState.CondMet;
3619 }
3620
3621 return false;
3622 }
3623
3624 /// parseDirectiveIfc
3625 /// ::= .ifc string1, string2
parseDirectiveIfc(SMLoc DirectiveLoc,bool ExpectEqual)3626 bool AsmParser::parseDirectiveIfc(SMLoc DirectiveLoc, bool ExpectEqual) {
3627 TheCondStack.push_back(TheCondState);
3628 TheCondState.TheCond = AsmCond::IfCond;
3629
3630 if (TheCondState.Ignore) {
3631 eatToEndOfStatement();
3632 } else {
3633 StringRef Str1 = parseStringToComma();
3634
3635 if (getLexer().isNot(AsmToken::Comma))
3636 return TokError("unexpected token in '.ifc' directive");
3637
3638 Lex();
3639
3640 StringRef Str2 = parseStringToEndOfStatement();
3641
3642 if (getLexer().isNot(AsmToken::EndOfStatement))
3643 return TokError("unexpected token in '.ifc' directive");
3644
3645 Lex();
3646
3647 TheCondState.CondMet = ExpectEqual == (Str1 == Str2);
3648 TheCondState.Ignore = !TheCondState.CondMet;
3649 }
3650
3651 return false;
3652 }
3653
3654 /// parseDirectiveIfdef
3655 /// ::= .ifdef symbol
parseDirectiveIfdef(SMLoc DirectiveLoc,bool expect_defined)3656 bool AsmParser::parseDirectiveIfdef(SMLoc DirectiveLoc, bool expect_defined) {
3657 StringRef Name;
3658 TheCondStack.push_back(TheCondState);
3659 TheCondState.TheCond = AsmCond::IfCond;
3660
3661 if (TheCondState.Ignore) {
3662 eatToEndOfStatement();
3663 } else {
3664 if (parseIdentifier(Name))
3665 return TokError("expected identifier after '.ifdef'");
3666
3667 Lex();
3668
3669 MCSymbol *Sym = getContext().LookupSymbol(Name);
3670
3671 if (expect_defined)
3672 TheCondState.CondMet = (Sym != NULL && !Sym->isUndefined());
3673 else
3674 TheCondState.CondMet = (Sym == NULL || Sym->isUndefined());
3675 TheCondState.Ignore = !TheCondState.CondMet;
3676 }
3677
3678 return false;
3679 }
3680
3681 /// parseDirectiveElseIf
3682 /// ::= .elseif expression
parseDirectiveElseIf(SMLoc DirectiveLoc)3683 bool AsmParser::parseDirectiveElseIf(SMLoc DirectiveLoc) {
3684 if (TheCondState.TheCond != AsmCond::IfCond &&
3685 TheCondState.TheCond != AsmCond::ElseIfCond)
3686 Error(DirectiveLoc, "Encountered a .elseif that doesn't follow a .if or "
3687 " an .elseif");
3688 TheCondState.TheCond = AsmCond::ElseIfCond;
3689
3690 bool LastIgnoreState = false;
3691 if (!TheCondStack.empty())
3692 LastIgnoreState = TheCondStack.back().Ignore;
3693 if (LastIgnoreState || TheCondState.CondMet) {
3694 TheCondState.Ignore = true;
3695 eatToEndOfStatement();
3696 } else {
3697 int64_t ExprValue;
3698 if (parseAbsoluteExpression(ExprValue))
3699 return true;
3700
3701 if (getLexer().isNot(AsmToken::EndOfStatement))
3702 return TokError("unexpected token in '.elseif' directive");
3703
3704 Lex();
3705 TheCondState.CondMet = ExprValue;
3706 TheCondState.Ignore = !TheCondState.CondMet;
3707 }
3708
3709 return false;
3710 }
3711
3712 /// parseDirectiveElse
3713 /// ::= .else
parseDirectiveElse(SMLoc DirectiveLoc)3714 bool AsmParser::parseDirectiveElse(SMLoc DirectiveLoc) {
3715 if (getLexer().isNot(AsmToken::EndOfStatement))
3716 return TokError("unexpected token in '.else' directive");
3717
3718 Lex();
3719
3720 if (TheCondState.TheCond != AsmCond::IfCond &&
3721 TheCondState.TheCond != AsmCond::ElseIfCond)
3722 Error(DirectiveLoc, "Encountered a .else that doesn't follow a .if or an "
3723 ".elseif");
3724 TheCondState.TheCond = AsmCond::ElseCond;
3725 bool LastIgnoreState = false;
3726 if (!TheCondStack.empty())
3727 LastIgnoreState = TheCondStack.back().Ignore;
3728 if (LastIgnoreState || TheCondState.CondMet)
3729 TheCondState.Ignore = true;
3730 else
3731 TheCondState.Ignore = false;
3732
3733 return false;
3734 }
3735
3736 /// parseDirectiveEndIf
3737 /// ::= .endif
parseDirectiveEndIf(SMLoc DirectiveLoc)3738 bool AsmParser::parseDirectiveEndIf(SMLoc DirectiveLoc) {
3739 if (getLexer().isNot(AsmToken::EndOfStatement))
3740 return TokError("unexpected token in '.endif' directive");
3741
3742 Lex();
3743
3744 if ((TheCondState.TheCond == AsmCond::NoCond) || TheCondStack.empty())
3745 Error(DirectiveLoc, "Encountered a .endif that doesn't follow a .if or "
3746 ".else");
3747 if (!TheCondStack.empty()) {
3748 TheCondState = TheCondStack.back();
3749 TheCondStack.pop_back();
3750 }
3751
3752 return false;
3753 }
3754
initializeDirectiveKindMap()3755 void AsmParser::initializeDirectiveKindMap() {
3756 DirectiveKindMap[".set"] = DK_SET;
3757 DirectiveKindMap[".equ"] = DK_EQU;
3758 DirectiveKindMap[".equiv"] = DK_EQUIV;
3759 DirectiveKindMap[".ascii"] = DK_ASCII;
3760 DirectiveKindMap[".asciz"] = DK_ASCIZ;
3761 DirectiveKindMap[".string"] = DK_STRING;
3762 DirectiveKindMap[".byte"] = DK_BYTE;
3763 DirectiveKindMap[".short"] = DK_SHORT;
3764 DirectiveKindMap[".value"] = DK_VALUE;
3765 DirectiveKindMap[".2byte"] = DK_2BYTE;
3766 DirectiveKindMap[".long"] = DK_LONG;
3767 DirectiveKindMap[".int"] = DK_INT;
3768 DirectiveKindMap[".4byte"] = DK_4BYTE;
3769 DirectiveKindMap[".quad"] = DK_QUAD;
3770 DirectiveKindMap[".8byte"] = DK_8BYTE;
3771 DirectiveKindMap[".single"] = DK_SINGLE;
3772 DirectiveKindMap[".float"] = DK_FLOAT;
3773 DirectiveKindMap[".double"] = DK_DOUBLE;
3774 DirectiveKindMap[".align"] = DK_ALIGN;
3775 DirectiveKindMap[".align32"] = DK_ALIGN32;
3776 DirectiveKindMap[".balign"] = DK_BALIGN;
3777 DirectiveKindMap[".balignw"] = DK_BALIGNW;
3778 DirectiveKindMap[".balignl"] = DK_BALIGNL;
3779 DirectiveKindMap[".p2align"] = DK_P2ALIGN;
3780 DirectiveKindMap[".p2alignw"] = DK_P2ALIGNW;
3781 DirectiveKindMap[".p2alignl"] = DK_P2ALIGNL;
3782 DirectiveKindMap[".org"] = DK_ORG;
3783 DirectiveKindMap[".fill"] = DK_FILL;
3784 DirectiveKindMap[".zero"] = DK_ZERO;
3785 DirectiveKindMap[".extern"] = DK_EXTERN;
3786 DirectiveKindMap[".globl"] = DK_GLOBL;
3787 DirectiveKindMap[".global"] = DK_GLOBAL;
3788 DirectiveKindMap[".lazy_reference"] = DK_LAZY_REFERENCE;
3789 DirectiveKindMap[".no_dead_strip"] = DK_NO_DEAD_STRIP;
3790 DirectiveKindMap[".symbol_resolver"] = DK_SYMBOL_RESOLVER;
3791 DirectiveKindMap[".private_extern"] = DK_PRIVATE_EXTERN;
3792 DirectiveKindMap[".reference"] = DK_REFERENCE;
3793 DirectiveKindMap[".weak_definition"] = DK_WEAK_DEFINITION;
3794 DirectiveKindMap[".weak_reference"] = DK_WEAK_REFERENCE;
3795 DirectiveKindMap[".weak_def_can_be_hidden"] = DK_WEAK_DEF_CAN_BE_HIDDEN;
3796 DirectiveKindMap[".comm"] = DK_COMM;
3797 DirectiveKindMap[".common"] = DK_COMMON;
3798 DirectiveKindMap[".lcomm"] = DK_LCOMM;
3799 DirectiveKindMap[".abort"] = DK_ABORT;
3800 DirectiveKindMap[".include"] = DK_INCLUDE;
3801 DirectiveKindMap[".incbin"] = DK_INCBIN;
3802 DirectiveKindMap[".code16"] = DK_CODE16;
3803 DirectiveKindMap[".code16gcc"] = DK_CODE16GCC;
3804 DirectiveKindMap[".rept"] = DK_REPT;
3805 DirectiveKindMap[".irp"] = DK_IRP;
3806 DirectiveKindMap[".irpc"] = DK_IRPC;
3807 DirectiveKindMap[".endr"] = DK_ENDR;
3808 DirectiveKindMap[".bundle_align_mode"] = DK_BUNDLE_ALIGN_MODE;
3809 DirectiveKindMap[".bundle_lock"] = DK_BUNDLE_LOCK;
3810 DirectiveKindMap[".bundle_unlock"] = DK_BUNDLE_UNLOCK;
3811 DirectiveKindMap[".if"] = DK_IF;
3812 DirectiveKindMap[".ifb"] = DK_IFB;
3813 DirectiveKindMap[".ifnb"] = DK_IFNB;
3814 DirectiveKindMap[".ifc"] = DK_IFC;
3815 DirectiveKindMap[".ifnc"] = DK_IFNC;
3816 DirectiveKindMap[".ifdef"] = DK_IFDEF;
3817 DirectiveKindMap[".ifndef"] = DK_IFNDEF;
3818 DirectiveKindMap[".ifnotdef"] = DK_IFNOTDEF;
3819 DirectiveKindMap[".elseif"] = DK_ELSEIF;
3820 DirectiveKindMap[".else"] = DK_ELSE;
3821 DirectiveKindMap[".endif"] = DK_ENDIF;
3822 DirectiveKindMap[".skip"] = DK_SKIP;
3823 DirectiveKindMap[".space"] = DK_SPACE;
3824 DirectiveKindMap[".file"] = DK_FILE;
3825 DirectiveKindMap[".line"] = DK_LINE;
3826 DirectiveKindMap[".loc"] = DK_LOC;
3827 DirectiveKindMap[".stabs"] = DK_STABS;
3828 DirectiveKindMap[".sleb128"] = DK_SLEB128;
3829 DirectiveKindMap[".uleb128"] = DK_ULEB128;
3830 DirectiveKindMap[".cfi_sections"] = DK_CFI_SECTIONS;
3831 DirectiveKindMap[".cfi_startproc"] = DK_CFI_STARTPROC;
3832 DirectiveKindMap[".cfi_endproc"] = DK_CFI_ENDPROC;
3833 DirectiveKindMap[".cfi_def_cfa"] = DK_CFI_DEF_CFA;
3834 DirectiveKindMap[".cfi_def_cfa_offset"] = DK_CFI_DEF_CFA_OFFSET;
3835 DirectiveKindMap[".cfi_adjust_cfa_offset"] = DK_CFI_ADJUST_CFA_OFFSET;
3836 DirectiveKindMap[".cfi_def_cfa_register"] = DK_CFI_DEF_CFA_REGISTER;
3837 DirectiveKindMap[".cfi_offset"] = DK_CFI_OFFSET;
3838 DirectiveKindMap[".cfi_rel_offset"] = DK_CFI_REL_OFFSET;
3839 DirectiveKindMap[".cfi_personality"] = DK_CFI_PERSONALITY;
3840 DirectiveKindMap[".cfi_lsda"] = DK_CFI_LSDA;
3841 DirectiveKindMap[".cfi_remember_state"] = DK_CFI_REMEMBER_STATE;
3842 DirectiveKindMap[".cfi_restore_state"] = DK_CFI_RESTORE_STATE;
3843 DirectiveKindMap[".cfi_same_value"] = DK_CFI_SAME_VALUE;
3844 DirectiveKindMap[".cfi_restore"] = DK_CFI_RESTORE;
3845 DirectiveKindMap[".cfi_escape"] = DK_CFI_ESCAPE;
3846 DirectiveKindMap[".cfi_signal_frame"] = DK_CFI_SIGNAL_FRAME;
3847 DirectiveKindMap[".cfi_undefined"] = DK_CFI_UNDEFINED;
3848 DirectiveKindMap[".cfi_register"] = DK_CFI_REGISTER;
3849 DirectiveKindMap[".cfi_window_save"] = DK_CFI_WINDOW_SAVE;
3850 DirectiveKindMap[".macros_on"] = DK_MACROS_ON;
3851 DirectiveKindMap[".macros_off"] = DK_MACROS_OFF;
3852 DirectiveKindMap[".macro"] = DK_MACRO;
3853 DirectiveKindMap[".endm"] = DK_ENDM;
3854 DirectiveKindMap[".endmacro"] = DK_ENDMACRO;
3855 DirectiveKindMap[".purgem"] = DK_PURGEM;
3856 }
3857
parseMacroLikeBody(SMLoc DirectiveLoc)3858 MCAsmMacro *AsmParser::parseMacroLikeBody(SMLoc DirectiveLoc) {
3859 AsmToken EndToken, StartToken = getTok();
3860
3861 unsigned NestLevel = 0;
3862 for (;;) {
3863 // Check whether we have reached the end of the file.
3864 if (getLexer().is(AsmToken::Eof)) {
3865 Error(DirectiveLoc, "no matching '.endr' in definition");
3866 return 0;
3867 }
3868
3869 if (Lexer.is(AsmToken::Identifier) &&
3870 (getTok().getIdentifier() == ".rept")) {
3871 ++NestLevel;
3872 }
3873
3874 // Otherwise, check whether we have reached the .endr.
3875 if (Lexer.is(AsmToken::Identifier) && getTok().getIdentifier() == ".endr") {
3876 if (NestLevel == 0) {
3877 EndToken = getTok();
3878 Lex();
3879 if (Lexer.isNot(AsmToken::EndOfStatement)) {
3880 TokError("unexpected token in '.endr' directive");
3881 return 0;
3882 }
3883 break;
3884 }
3885 --NestLevel;
3886 }
3887
3888 // Otherwise, scan till the end of the statement.
3889 eatToEndOfStatement();
3890 }
3891
3892 const char *BodyStart = StartToken.getLoc().getPointer();
3893 const char *BodyEnd = EndToken.getLoc().getPointer();
3894 StringRef Body = StringRef(BodyStart, BodyEnd - BodyStart);
3895
3896 // We Are Anonymous.
3897 StringRef Name;
3898 MCAsmMacroParameters Parameters;
3899 MacroLikeBodies.push_back(MCAsmMacro(Name, Body, Parameters));
3900 return &MacroLikeBodies.back();
3901 }
3902
instantiateMacroLikeBody(MCAsmMacro * M,SMLoc DirectiveLoc,raw_svector_ostream & OS)3903 void AsmParser::instantiateMacroLikeBody(MCAsmMacro *M, SMLoc DirectiveLoc,
3904 raw_svector_ostream &OS) {
3905 OS << ".endr\n";
3906
3907 MemoryBuffer *Instantiation =
3908 MemoryBuffer::getMemBufferCopy(OS.str(), "<instantiation>");
3909
3910 // Create the macro instantiation object and add to the current macro
3911 // instantiation stack.
3912 MacroInstantiation *MI = new MacroInstantiation(
3913 M, DirectiveLoc, CurBuffer, getTok().getLoc(), Instantiation);
3914 ActiveMacros.push_back(MI);
3915
3916 // Jump to the macro instantiation and prime the lexer.
3917 CurBuffer = SrcMgr.AddNewSourceBuffer(MI->Instantiation, SMLoc());
3918 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer));
3919 Lex();
3920 }
3921
parseDirectiveRept(SMLoc DirectiveLoc)3922 bool AsmParser::parseDirectiveRept(SMLoc DirectiveLoc) {
3923 int64_t Count;
3924 if (parseAbsoluteExpression(Count))
3925 return TokError("unexpected token in '.rept' directive");
3926
3927 if (Count < 0)
3928 return TokError("Count is negative");
3929
3930 if (Lexer.isNot(AsmToken::EndOfStatement))
3931 return TokError("unexpected token in '.rept' directive");
3932
3933 // Eat the end of statement.
3934 Lex();
3935
3936 // Lex the rept definition.
3937 MCAsmMacro *M = parseMacroLikeBody(DirectiveLoc);
3938 if (!M)
3939 return true;
3940
3941 // Macro instantiation is lexical, unfortunately. We construct a new buffer
3942 // to hold the macro body with substitutions.
3943 SmallString<256> Buf;
3944 MCAsmMacroParameters Parameters;
3945 MCAsmMacroArguments A;
3946 raw_svector_ostream OS(Buf);
3947 while (Count--) {
3948 if (expandMacro(OS, M->Body, Parameters, A, getTok().getLoc()))
3949 return true;
3950 }
3951 instantiateMacroLikeBody(M, DirectiveLoc, OS);
3952
3953 return false;
3954 }
3955
3956 /// parseDirectiveIrp
3957 /// ::= .irp symbol,values
parseDirectiveIrp(SMLoc DirectiveLoc)3958 bool AsmParser::parseDirectiveIrp(SMLoc DirectiveLoc) {
3959 MCAsmMacroParameters Parameters;
3960 MCAsmMacroParameter Parameter;
3961
3962 if (parseIdentifier(Parameter.first))
3963 return TokError("expected identifier in '.irp' directive");
3964
3965 Parameters.push_back(Parameter);
3966
3967 if (Lexer.isNot(AsmToken::Comma))
3968 return TokError("expected comma in '.irp' directive");
3969
3970 Lex();
3971
3972 MCAsmMacroArguments A;
3973 if (parseMacroArguments(0, A))
3974 return true;
3975
3976 // Eat the end of statement.
3977 Lex();
3978
3979 // Lex the irp definition.
3980 MCAsmMacro *M = parseMacroLikeBody(DirectiveLoc);
3981 if (!M)
3982 return true;
3983
3984 // Macro instantiation is lexical, unfortunately. We construct a new buffer
3985 // to hold the macro body with substitutions.
3986 SmallString<256> Buf;
3987 raw_svector_ostream OS(Buf);
3988
3989 for (MCAsmMacroArguments::iterator i = A.begin(), e = A.end(); i != e; ++i) {
3990 MCAsmMacroArguments Args;
3991 Args.push_back(*i);
3992
3993 if (expandMacro(OS, M->Body, Parameters, Args, getTok().getLoc()))
3994 return true;
3995 }
3996
3997 instantiateMacroLikeBody(M, DirectiveLoc, OS);
3998
3999 return false;
4000 }
4001
4002 /// parseDirectiveIrpc
4003 /// ::= .irpc symbol,values
parseDirectiveIrpc(SMLoc DirectiveLoc)4004 bool AsmParser::parseDirectiveIrpc(SMLoc DirectiveLoc) {
4005 MCAsmMacroParameters Parameters;
4006 MCAsmMacroParameter Parameter;
4007
4008 if (parseIdentifier(Parameter.first))
4009 return TokError("expected identifier in '.irpc' directive");
4010
4011 Parameters.push_back(Parameter);
4012
4013 if (Lexer.isNot(AsmToken::Comma))
4014 return TokError("expected comma in '.irpc' directive");
4015
4016 Lex();
4017
4018 MCAsmMacroArguments A;
4019 if (parseMacroArguments(0, A))
4020 return true;
4021
4022 if (A.size() != 1 || A.front().size() != 1)
4023 return TokError("unexpected token in '.irpc' directive");
4024
4025 // Eat the end of statement.
4026 Lex();
4027
4028 // Lex the irpc definition.
4029 MCAsmMacro *M = parseMacroLikeBody(DirectiveLoc);
4030 if (!M)
4031 return true;
4032
4033 // Macro instantiation is lexical, unfortunately. We construct a new buffer
4034 // to hold the macro body with substitutions.
4035 SmallString<256> Buf;
4036 raw_svector_ostream OS(Buf);
4037
4038 StringRef Values = A.front().front().getString();
4039 std::size_t I, End = Values.size();
4040 for (I = 0; I < End; ++I) {
4041 MCAsmMacroArgument Arg;
4042 Arg.push_back(AsmToken(AsmToken::Identifier, Values.slice(I, I + 1)));
4043
4044 MCAsmMacroArguments Args;
4045 Args.push_back(Arg);
4046
4047 if (expandMacro(OS, M->Body, Parameters, Args, getTok().getLoc()))
4048 return true;
4049 }
4050
4051 instantiateMacroLikeBody(M, DirectiveLoc, OS);
4052
4053 return false;
4054 }
4055
parseDirectiveEndr(SMLoc DirectiveLoc)4056 bool AsmParser::parseDirectiveEndr(SMLoc DirectiveLoc) {
4057 if (ActiveMacros.empty())
4058 return TokError("unmatched '.endr' directive");
4059
4060 // The only .repl that should get here are the ones created by
4061 // instantiateMacroLikeBody.
4062 assert(getLexer().is(AsmToken::EndOfStatement));
4063
4064 handleMacroExit();
4065 return false;
4066 }
4067
parseDirectiveMSEmit(SMLoc IDLoc,ParseStatementInfo & Info,size_t Len)4068 bool AsmParser::parseDirectiveMSEmit(SMLoc IDLoc, ParseStatementInfo &Info,
4069 size_t Len) {
4070 const MCExpr *Value;
4071 SMLoc ExprLoc = getLexer().getLoc();
4072 if (parseExpression(Value))
4073 return true;
4074 const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value);
4075 if (!MCE)
4076 return Error(ExprLoc, "unexpected expression in _emit");
4077 uint64_t IntValue = MCE->getValue();
4078 if (!isUIntN(8, IntValue) && !isIntN(8, IntValue))
4079 return Error(ExprLoc, "literal value out of range for directive");
4080
4081 Info.AsmRewrites->push_back(AsmRewrite(AOK_Emit, IDLoc, Len));
4082 return false;
4083 }
4084
parseDirectiveMSAlign(SMLoc IDLoc,ParseStatementInfo & Info)4085 bool AsmParser::parseDirectiveMSAlign(SMLoc IDLoc, ParseStatementInfo &Info) {
4086 const MCExpr *Value;
4087 SMLoc ExprLoc = getLexer().getLoc();
4088 if (parseExpression(Value))
4089 return true;
4090 const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value);
4091 if (!MCE)
4092 return Error(ExprLoc, "unexpected expression in align");
4093 uint64_t IntValue = MCE->getValue();
4094 if (!isPowerOf2_64(IntValue))
4095 return Error(ExprLoc, "literal value not a power of two greater then zero");
4096
4097 Info.AsmRewrites->push_back(
4098 AsmRewrite(AOK_Align, IDLoc, 5, Log2_64(IntValue)));
4099 return false;
4100 }
4101
4102 // We are comparing pointers, but the pointers are relative to a single string.
4103 // Thus, this should always be deterministic.
rewritesSort(const AsmRewrite * AsmRewriteA,const AsmRewrite * AsmRewriteB)4104 static int rewritesSort(const AsmRewrite *AsmRewriteA,
4105 const AsmRewrite *AsmRewriteB) {
4106 if (AsmRewriteA->Loc.getPointer() < AsmRewriteB->Loc.getPointer())
4107 return -1;
4108 if (AsmRewriteB->Loc.getPointer() < AsmRewriteA->Loc.getPointer())
4109 return 1;
4110
4111 // It's possible to have a SizeDirective, Imm/ImmPrefix and an Input/Output
4112 // rewrite to the same location. Make sure the SizeDirective rewrite is
4113 // performed first, then the Imm/ImmPrefix and finally the Input/Output. This
4114 // ensures the sort algorithm is stable.
4115 if (AsmRewritePrecedence[AsmRewriteA->Kind] >
4116 AsmRewritePrecedence[AsmRewriteB->Kind])
4117 return -1;
4118
4119 if (AsmRewritePrecedence[AsmRewriteA->Kind] <
4120 AsmRewritePrecedence[AsmRewriteB->Kind])
4121 return 1;
4122 llvm_unreachable("Unstable rewrite sort.");
4123 }
4124
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)4125 bool AsmParser::parseMSInlineAsm(
4126 void *AsmLoc, std::string &AsmString, unsigned &NumOutputs,
4127 unsigned &NumInputs, SmallVectorImpl<std::pair<void *, bool> > &OpDecls,
4128 SmallVectorImpl<std::string> &Constraints,
4129 SmallVectorImpl<std::string> &Clobbers, const MCInstrInfo *MII,
4130 const MCInstPrinter *IP, MCAsmParserSemaCallback &SI) {
4131 SmallVector<void *, 4> InputDecls;
4132 SmallVector<void *, 4> OutputDecls;
4133 SmallVector<bool, 4> InputDeclsAddressOf;
4134 SmallVector<bool, 4> OutputDeclsAddressOf;
4135 SmallVector<std::string, 4> InputConstraints;
4136 SmallVector<std::string, 4> OutputConstraints;
4137 SmallVector<unsigned, 4> ClobberRegs;
4138
4139 SmallVector<AsmRewrite, 4> AsmStrRewrites;
4140
4141 // Prime the lexer.
4142 Lex();
4143
4144 // While we have input, parse each statement.
4145 unsigned InputIdx = 0;
4146 unsigned OutputIdx = 0;
4147 while (getLexer().isNot(AsmToken::Eof)) {
4148 ParseStatementInfo Info(&AsmStrRewrites);
4149 if (parseStatement(Info))
4150 return true;
4151
4152 if (Info.ParseError)
4153 return true;
4154
4155 if (Info.Opcode == ~0U)
4156 continue;
4157
4158 const MCInstrDesc &Desc = MII->get(Info.Opcode);
4159
4160 // Build the list of clobbers, outputs and inputs.
4161 for (unsigned i = 1, e = Info.ParsedOperands.size(); i != e; ++i) {
4162 MCParsedAsmOperand *Operand = Info.ParsedOperands[i];
4163
4164 // Immediate.
4165 if (Operand->isImm())
4166 continue;
4167
4168 // Register operand.
4169 if (Operand->isReg() && !Operand->needAddressOf()) {
4170 unsigned NumDefs = Desc.getNumDefs();
4171 // Clobber.
4172 if (NumDefs && Operand->getMCOperandNum() < NumDefs)
4173 ClobberRegs.push_back(Operand->getReg());
4174 continue;
4175 }
4176
4177 // Expr/Input or Output.
4178 StringRef SymName = Operand->getSymName();
4179 if (SymName.empty())
4180 continue;
4181
4182 void *OpDecl = Operand->getOpDecl();
4183 if (!OpDecl)
4184 continue;
4185
4186 bool isOutput = (i == 1) && Desc.mayStore();
4187 SMLoc Start = SMLoc::getFromPointer(SymName.data());
4188 if (isOutput) {
4189 ++InputIdx;
4190 OutputDecls.push_back(OpDecl);
4191 OutputDeclsAddressOf.push_back(Operand->needAddressOf());
4192 OutputConstraints.push_back('=' + Operand->getConstraint().str());
4193 AsmStrRewrites.push_back(AsmRewrite(AOK_Output, Start, SymName.size()));
4194 } else {
4195 InputDecls.push_back(OpDecl);
4196 InputDeclsAddressOf.push_back(Operand->needAddressOf());
4197 InputConstraints.push_back(Operand->getConstraint().str());
4198 AsmStrRewrites.push_back(AsmRewrite(AOK_Input, Start, SymName.size()));
4199 }
4200 }
4201
4202 // Consider implicit defs to be clobbers. Think of cpuid and push.
4203 const uint16_t *ImpDefs = Desc.getImplicitDefs();
4204 for (unsigned I = 0, E = Desc.getNumImplicitDefs(); I != E; ++I)
4205 ClobberRegs.push_back(ImpDefs[I]);
4206 }
4207
4208 // Set the number of Outputs and Inputs.
4209 NumOutputs = OutputDecls.size();
4210 NumInputs = InputDecls.size();
4211
4212 // Set the unique clobbers.
4213 array_pod_sort(ClobberRegs.begin(), ClobberRegs.end());
4214 ClobberRegs.erase(std::unique(ClobberRegs.begin(), ClobberRegs.end()),
4215 ClobberRegs.end());
4216 Clobbers.assign(ClobberRegs.size(), std::string());
4217 for (unsigned I = 0, E = ClobberRegs.size(); I != E; ++I) {
4218 raw_string_ostream OS(Clobbers[I]);
4219 IP->printRegName(OS, ClobberRegs[I]);
4220 }
4221
4222 // Merge the various outputs and inputs. Output are expected first.
4223 if (NumOutputs || NumInputs) {
4224 unsigned NumExprs = NumOutputs + NumInputs;
4225 OpDecls.resize(NumExprs);
4226 Constraints.resize(NumExprs);
4227 for (unsigned i = 0; i < NumOutputs; ++i) {
4228 OpDecls[i] = std::make_pair(OutputDecls[i], OutputDeclsAddressOf[i]);
4229 Constraints[i] = OutputConstraints[i];
4230 }
4231 for (unsigned i = 0, j = NumOutputs; i < NumInputs; ++i, ++j) {
4232 OpDecls[j] = std::make_pair(InputDecls[i], InputDeclsAddressOf[i]);
4233 Constraints[j] = InputConstraints[i];
4234 }
4235 }
4236
4237 // Build the IR assembly string.
4238 std::string AsmStringIR;
4239 raw_string_ostream OS(AsmStringIR);
4240 const char *AsmStart = SrcMgr.getMemoryBuffer(0)->getBufferStart();
4241 const char *AsmEnd = SrcMgr.getMemoryBuffer(0)->getBufferEnd();
4242 array_pod_sort(AsmStrRewrites.begin(), AsmStrRewrites.end(), rewritesSort);
4243 for (SmallVectorImpl<AsmRewrite>::iterator I = AsmStrRewrites.begin(),
4244 E = AsmStrRewrites.end();
4245 I != E; ++I) {
4246 AsmRewriteKind Kind = (*I).Kind;
4247 if (Kind == AOK_Delete)
4248 continue;
4249
4250 const char *Loc = (*I).Loc.getPointer();
4251 assert(Loc >= AsmStart && "Expected Loc to be at or after Start!");
4252
4253 // Emit everything up to the immediate/expression.
4254 unsigned Len = Loc - AsmStart;
4255 if (Len)
4256 OS << StringRef(AsmStart, Len);
4257
4258 // Skip the original expression.
4259 if (Kind == AOK_Skip) {
4260 AsmStart = Loc + (*I).Len;
4261 continue;
4262 }
4263
4264 unsigned AdditionalSkip = 0;
4265 // Rewrite expressions in $N notation.
4266 switch (Kind) {
4267 default:
4268 break;
4269 case AOK_Imm:
4270 OS << "$$" << (*I).Val;
4271 break;
4272 case AOK_ImmPrefix:
4273 OS << "$$";
4274 break;
4275 case AOK_Input:
4276 OS << '$' << InputIdx++;
4277 break;
4278 case AOK_Output:
4279 OS << '$' << OutputIdx++;
4280 break;
4281 case AOK_SizeDirective:
4282 switch ((*I).Val) {
4283 default: break;
4284 case 8: OS << "byte ptr "; break;
4285 case 16: OS << "word ptr "; break;
4286 case 32: OS << "dword ptr "; break;
4287 case 64: OS << "qword ptr "; break;
4288 case 80: OS << "xword ptr "; break;
4289 case 128: OS << "xmmword ptr "; break;
4290 case 256: OS << "ymmword ptr "; break;
4291 }
4292 break;
4293 case AOK_Emit:
4294 OS << ".byte";
4295 break;
4296 case AOK_Align: {
4297 unsigned Val = (*I).Val;
4298 OS << ".align " << Val;
4299
4300 // Skip the original immediate.
4301 assert(Val < 10 && "Expected alignment less then 2^10.");
4302 AdditionalSkip = (Val < 4) ? 2 : Val < 7 ? 3 : 4;
4303 break;
4304 }
4305 case AOK_DotOperator:
4306 // Insert the dot if the user omitted it.
4307 OS.flush();
4308 if (AsmStringIR.at(AsmStringIR.size() - 1) != '.')
4309 OS << '.';
4310 OS << (*I).Val;
4311 break;
4312 }
4313
4314 // Skip the original expression.
4315 AsmStart = Loc + (*I).Len + AdditionalSkip;
4316 }
4317
4318 // Emit the remainder of the asm string.
4319 if (AsmStart != AsmEnd)
4320 OS << StringRef(AsmStart, AsmEnd - AsmStart);
4321
4322 AsmString = OS.str();
4323 return false;
4324 }
4325
4326 /// \brief Create an MCAsmParser instance.
createMCAsmParser(SourceMgr & SM,MCContext & C,MCStreamer & Out,const MCAsmInfo & MAI)4327 MCAsmParser *llvm::createMCAsmParser(SourceMgr &SM, MCContext &C,
4328 MCStreamer &Out, const MCAsmInfo &MAI) {
4329 return new AsmParser(SM, C, Out, MAI);
4330 }
4331