1 //===- AsmLexer.cpp - Lexer for Assembly Files ----------------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This class implements the lexer for assembly files.
12 //===----------------------------------------------------------------------===//
14 #include "llvm/MC/MCParser/AsmLexer.h"
15 #include "llvm/MC/MCAsmInfo.h"
16 #include "llvm/Support/MemoryBuffer.h"
17 #include "llvm/Support/SMLoc.h"
24 AsmLexer::AsmLexer(const MCAsmInfo &MAI) : MAI(MAI) {
26 IsAtStartOfLine = true;
27 IsAtStartOfStatement = true;
28 AllowAtInIdentifier = !StringRef(MAI.getCommentString()).startswith("@");
31 AsmLexer::~AsmLexer() {
34 void AsmLexer::setBuffer(StringRef Buf, const char *ptr) {
40 CurPtr = CurBuf.begin();
45 /// ReturnError - Set the error to the specified string at the specified
46 /// location. This is defined to always return AsmToken::Error.
47 AsmToken AsmLexer::ReturnError(const char *Loc, const std::string &Msg) {
48 SetError(SMLoc::getFromPointer(Loc), Msg);
50 return AsmToken(AsmToken::Error, StringRef(Loc, CurPtr - Loc));
53 int AsmLexer::getNextChar() {
54 if (CurPtr == CurBuf.end())
56 return (unsigned char)*CurPtr++;
59 /// LexFloatLiteral: [0-9]*[.][0-9]*([eE][+-]?[0-9]*)?
61 /// The leading integral digit sequence and dot should have already been
62 /// consumed, some or all of the fractional digit sequence *can* have been
64 AsmToken AsmLexer::LexFloatLiteral() {
65 // Skip the fractional digit sequence.
66 while (isdigit(*CurPtr))
69 // Check for exponent; we intentionally accept a slighlty wider set of
70 // literals here and rely on the upstream client to reject invalid ones (e.g.,
72 if (*CurPtr == 'e' || *CurPtr == 'E') {
74 if (*CurPtr == '-' || *CurPtr == '+')
76 while (isdigit(*CurPtr))
80 return AsmToken(AsmToken::Real,
81 StringRef(TokStart, CurPtr - TokStart));
84 /// LexHexFloatLiteral matches essentially (.[0-9a-fA-F]*)?[pP][+-]?[0-9a-fA-F]+
85 /// while making sure there are enough actual digits around for the constant to
88 /// The leading "0x[0-9a-fA-F]*" (i.e. integer part) has already been consumed
89 /// before we get here.
90 AsmToken AsmLexer::LexHexFloatLiteral(bool NoIntDigits) {
91 assert((*CurPtr == 'p' || *CurPtr == 'P' || *CurPtr == '.') &&
92 "unexpected parse state in floating hex");
93 bool NoFracDigits = true;
95 // Skip the fractional part if there is one
99 const char *FracStart = CurPtr;
100 while (isxdigit(*CurPtr))
103 NoFracDigits = CurPtr == FracStart;
106 if (NoIntDigits && NoFracDigits)
107 return ReturnError(TokStart, "invalid hexadecimal floating-point constant: "
108 "expected at least one significand digit");
110 // Make sure we do have some kind of proper exponent part
111 if (*CurPtr != 'p' && *CurPtr != 'P')
112 return ReturnError(TokStart, "invalid hexadecimal floating-point constant: "
113 "expected exponent part 'p'");
116 if (*CurPtr == '+' || *CurPtr == '-')
119 // N.b. exponent digits are *not* hex
120 const char *ExpStart = CurPtr;
121 while (isdigit(*CurPtr))
124 if (CurPtr == ExpStart)
125 return ReturnError(TokStart, "invalid hexadecimal floating-point constant: "
126 "expected at least one exponent digit");
128 return AsmToken(AsmToken::Real, StringRef(TokStart, CurPtr - TokStart));
131 /// LexIdentifier: [a-zA-Z_.][a-zA-Z0-9_$.@?]*
132 static bool IsIdentifierChar(char c, bool AllowAt) {
133 return isalnum(c) || c == '_' || c == '$' || c == '.' ||
134 (c == '@' && AllowAt) || c == '?';
136 AsmToken AsmLexer::LexIdentifier() {
137 // Check for floating point literals.
138 if (CurPtr[-1] == '.' && isdigit(*CurPtr)) {
139 // Disambiguate a .1243foo identifier from a floating literal.
140 while (isdigit(*CurPtr))
142 if (*CurPtr == 'e' || *CurPtr == 'E' ||
143 !IsIdentifierChar(*CurPtr, AllowAtInIdentifier))
144 return LexFloatLiteral();
147 while (IsIdentifierChar(*CurPtr, AllowAtInIdentifier))
150 // Handle . as a special case.
151 if (CurPtr == TokStart+1 && TokStart[0] == '.')
152 return AsmToken(AsmToken::Dot, StringRef(TokStart, 1));
154 return AsmToken(AsmToken::Identifier, StringRef(TokStart, CurPtr - TokStart));
157 /// LexSlash: Slash: /
158 /// C-Style Comment: /* ... */
159 AsmToken AsmLexer::LexSlash() {
162 IsAtStartOfStatement = false;
163 break; // C style comment.
166 return LexLineComment();
168 IsAtStartOfStatement = false;
169 return AsmToken(AsmToken::Slash, StringRef(TokStart, 1));
173 ++CurPtr; // skip the star.
174 while (CurPtr != CurBuf.end()) {
177 // End of the comment?
180 ++CurPtr; // End the */.
181 return AsmToken(AsmToken::Comment,
182 StringRef(TokStart, CurPtr - TokStart));
185 return ReturnError(TokStart, "unterminated comment");
188 /// LexLineComment: Comment: #[^\n]*
190 AsmToken AsmLexer::LexLineComment() {
191 // Mark This as an end of statement with a body of the
192 // comment. While it would be nicer to leave this two tokens,
193 // backwards compatability with TargetParsers makes keeping this in this form
195 int CurChar = getNextChar();
196 while (CurChar != '\n' && CurChar != '\r' && CurChar != EOF)
197 CurChar = getNextChar();
199 IsAtStartOfLine = true;
200 // Whis is a whole line comment. leave newline
201 if (IsAtStartOfStatement)
202 return AsmToken(AsmToken::EndOfStatement,
203 StringRef(TokStart, CurPtr - TokStart));
204 IsAtStartOfStatement = true;
206 return AsmToken(AsmToken::EndOfStatement,
207 StringRef(TokStart, CurPtr - 1 - TokStart));
210 static void SkipIgnoredIntegerSuffix(const char *&CurPtr) {
211 // Skip ULL, UL, U, L and LL suffices.
212 if (CurPtr[0] == 'U')
214 if (CurPtr[0] == 'L')
216 if (CurPtr[0] == 'L')
220 // Look ahead to search for first non-hex digit, if it's [hH], then we treat the
221 // integer as a hexadecimal, possibly with leading zeroes.
222 static unsigned doLookAhead(const char *&CurPtr, unsigned DefaultRadix) {
223 const char *FirstHex = nullptr;
224 const char *LookAhead = CurPtr;
226 if (isdigit(*LookAhead)) {
228 } else if (isxdigit(*LookAhead)) {
230 FirstHex = LookAhead;
236 bool isHex = *LookAhead == 'h' || *LookAhead == 'H';
237 CurPtr = isHex || !FirstHex ? LookAhead : FirstHex;
243 static AsmToken intToken(StringRef Ref, APInt &Value)
245 if (Value.isIntN(64))
246 return AsmToken(AsmToken::Integer, Ref, Value);
247 return AsmToken(AsmToken::BigNum, Ref, Value);
250 /// LexDigit: First character is [0-9].
251 /// Local Label: [0-9][:]
252 /// Forward/Backward Label: [0-9][fb]
253 /// Binary integer: 0b[01]+
254 /// Octal integer: 0[0-7]+
255 /// Hex integer: 0x[0-9a-fA-F]+ or [0x]?[0-9][0-9a-fA-F]*[hH]
256 /// Decimal integer: [1-9][0-9]*
257 AsmToken AsmLexer::LexDigit() {
258 // Decimal integer: [1-9][0-9]*
259 if (CurPtr[-1] != '0' || CurPtr[0] == '.') {
260 unsigned Radix = doLookAhead(CurPtr, 10);
261 bool isHex = Radix == 16;
262 // Check for floating point literals.
263 if (!isHex && (*CurPtr == '.' || *CurPtr == 'e')) {
265 return LexFloatLiteral();
268 StringRef Result(TokStart, CurPtr - TokStart);
270 APInt Value(128, 0, true);
271 if (Result.getAsInteger(Radix, Value))
272 return ReturnError(TokStart, !isHex ? "invalid decimal number" :
273 "invalid hexdecimal number");
275 // Consume the [bB][hH].
276 if (Radix == 2 || Radix == 16)
279 // The darwin/x86 (and x86-64) assembler accepts and ignores type
280 // suffices on integer literals.
281 SkipIgnoredIntegerSuffix(CurPtr);
283 return intToken(Result, Value);
286 if ((*CurPtr == 'b') || (*CurPtr == 'B')) {
288 // See if we actually have "0b" as part of something like "jmp 0b\n"
289 if (!isdigit(CurPtr[0])) {
291 StringRef Result(TokStart, CurPtr - TokStart);
292 return AsmToken(AsmToken::Integer, Result, 0);
294 const char *NumStart = CurPtr;
295 while (CurPtr[0] == '0' || CurPtr[0] == '1')
298 // Requires at least one binary digit.
299 if (CurPtr == NumStart)
300 return ReturnError(TokStart, "invalid binary number");
302 StringRef Result(TokStart, CurPtr - TokStart);
304 APInt Value(128, 0, true);
305 if (Result.substr(2).getAsInteger(2, Value))
306 return ReturnError(TokStart, "invalid binary number");
308 // The darwin/x86 (and x86-64) assembler accepts and ignores ULL and LL
309 // suffixes on integer literals.
310 SkipIgnoredIntegerSuffix(CurPtr);
312 return intToken(Result, Value);
315 if ((*CurPtr == 'x') || (*CurPtr == 'X')) {
317 const char *NumStart = CurPtr;
318 while (isxdigit(CurPtr[0]))
321 // "0x.0p0" is valid, and "0x0p0" (but not "0xp0" for example, which will be
322 // diagnosed by LexHexFloatLiteral).
323 if (CurPtr[0] == '.' || CurPtr[0] == 'p' || CurPtr[0] == 'P')
324 return LexHexFloatLiteral(NumStart == CurPtr);
326 // Otherwise requires at least one hex digit.
327 if (CurPtr == NumStart)
328 return ReturnError(CurPtr-2, "invalid hexadecimal number");
330 APInt Result(128, 0);
331 if (StringRef(TokStart, CurPtr - TokStart).getAsInteger(0, Result))
332 return ReturnError(TokStart, "invalid hexadecimal number");
334 // Consume the optional [hH].
335 if (*CurPtr == 'h' || *CurPtr == 'H')
338 // The darwin/x86 (and x86-64) assembler accepts and ignores ULL and LL
339 // suffixes on integer literals.
340 SkipIgnoredIntegerSuffix(CurPtr);
342 return intToken(StringRef(TokStart, CurPtr - TokStart), Result);
345 // Either octal or hexadecimal.
346 APInt Value(128, 0, true);
347 unsigned Radix = doLookAhead(CurPtr, 8);
348 bool isHex = Radix == 16;
349 StringRef Result(TokStart, CurPtr - TokStart);
350 if (Result.getAsInteger(Radix, Value))
351 return ReturnError(TokStart, !isHex ? "invalid octal number" :
352 "invalid hexdecimal number");
358 // The darwin/x86 (and x86-64) assembler accepts and ignores ULL and LL
359 // suffixes on integer literals.
360 SkipIgnoredIntegerSuffix(CurPtr);
362 return intToken(Result, Value);
365 /// LexSingleQuote: Integer: 'b'
366 AsmToken AsmLexer::LexSingleQuote() {
367 int CurChar = getNextChar();
370 CurChar = getNextChar();
373 return ReturnError(TokStart, "unterminated single quote");
375 CurChar = getNextChar();
378 return ReturnError(TokStart, "single quote way too long");
380 // The idea here being that 'c' is basically just an integral
382 StringRef Res = StringRef(TokStart,CurPtr - TokStart);
385 if (Res.startswith("\'\\")) {
386 char theChar = Res[2];
388 default: Value = theChar; break;
389 case '\'': Value = '\''; break;
390 case 't': Value = '\t'; break;
391 case 'n': Value = '\n'; break;
392 case 'b': Value = '\b'; break;
397 return AsmToken(AsmToken::Integer, Res, Value);
401 /// LexQuote: String: "..."
402 AsmToken AsmLexer::LexQuote() {
403 int CurChar = getNextChar();
404 // TODO: does gas allow multiline string constants?
405 while (CurChar != '"') {
406 if (CurChar == '\\') {
408 CurChar = getNextChar();
412 return ReturnError(TokStart, "unterminated string constant");
414 CurChar = getNextChar();
417 return AsmToken(AsmToken::String, StringRef(TokStart, CurPtr - TokStart));
420 StringRef AsmLexer::LexUntilEndOfStatement() {
423 while (!isAtStartOfComment(CurPtr) && // Start of line comment.
424 !isAtStatementSeparator(CurPtr) && // End of statement marker.
425 *CurPtr != '\n' && *CurPtr != '\r' && CurPtr != CurBuf.end()) {
428 return StringRef(TokStart, CurPtr-TokStart);
431 StringRef AsmLexer::LexUntilEndOfLine() {
434 while (*CurPtr != '\n' && *CurPtr != '\r' && CurPtr != CurBuf.end()) {
437 return StringRef(TokStart, CurPtr-TokStart);
440 size_t AsmLexer::peekTokens(MutableArrayRef<AsmToken> Buf,
441 bool ShouldSkipSpace) {
442 const char *SavedTokStart = TokStart;
443 const char *SavedCurPtr = CurPtr;
444 bool SavedAtStartOfLine = IsAtStartOfLine;
445 bool SavedAtStartOfStatement = IsAtStartOfStatement;
446 bool SavedSkipSpace = SkipSpace;
448 std::string SavedErr = getErr();
449 SMLoc SavedErrLoc = getErrLoc();
451 SkipSpace = ShouldSkipSpace;
454 for (ReadCount = 0; ReadCount < Buf.size(); ++ReadCount) {
455 AsmToken Token = LexToken();
457 Buf[ReadCount] = Token;
459 if (Token.is(AsmToken::Eof))
463 SetError(SavedErrLoc, SavedErr);
465 SkipSpace = SavedSkipSpace;
466 IsAtStartOfLine = SavedAtStartOfLine;
467 IsAtStartOfStatement = SavedAtStartOfStatement;
468 CurPtr = SavedCurPtr;
469 TokStart = SavedTokStart;
474 bool AsmLexer::isAtStartOfComment(const char *Ptr) {
475 const char *CommentString = MAI.getCommentString();
477 if (CommentString[1] == '\0')
478 return CommentString[0] == Ptr[0];
480 // FIXME: special case for the bogus "##" comment string in X86MCAsmInfoDarwin
481 if (CommentString[1] == '#')
482 return CommentString[0] == Ptr[0];
484 return strncmp(Ptr, CommentString, strlen(CommentString)) == 0;
487 bool AsmLexer::isAtStatementSeparator(const char *Ptr) {
488 return strncmp(Ptr, MAI.getSeparatorString(),
489 strlen(MAI.getSeparatorString())) == 0;
492 AsmToken AsmLexer::LexToken() {
494 // This always consumes at least one character.
495 int CurChar = getNextChar();
497 if (CurChar == '#' && IsAtStartOfStatement) {
498 // If this starts with a '#', this may be a cpp
499 // hash directive and otherwise a line comment.
500 AsmToken TokenBuf[2];
501 MutableArrayRef<AsmToken> Buf(TokenBuf, 2);
502 size_t num = peekTokens(Buf, true);
503 // There cannot be a space preceeding this
504 if (IsAtStartOfLine && num == 2 && TokenBuf[0].is(AsmToken::Integer) &&
505 TokenBuf[1].is(AsmToken::String)) {
506 CurPtr = TokStart; // reset curPtr;
507 StringRef s = LexUntilEndOfLine();
510 return AsmToken(AsmToken::HashDirective, s);
512 return LexLineComment();
515 if (isAtStartOfComment(TokStart))
516 return LexLineComment();
518 if (isAtStatementSeparator(TokStart)) {
519 CurPtr += strlen(MAI.getSeparatorString()) - 1;
520 IsAtStartOfLine = true;
521 IsAtStartOfStatement = true;
522 return AsmToken(AsmToken::EndOfStatement,
523 StringRef(TokStart, strlen(MAI.getSeparatorString())));
526 // If we're missing a newline at EOF, make sure we still get an
527 // EndOfStatement token before the Eof token.
528 if (CurChar == EOF && !IsAtStartOfStatement) {
529 IsAtStartOfLine = true;
530 IsAtStartOfStatement = true;
531 return AsmToken(AsmToken::EndOfStatement, StringRef(TokStart, 1));
533 IsAtStartOfLine = false;
534 bool OldIsAtStartOfStatement = IsAtStartOfStatement;
535 IsAtStartOfStatement = false;
538 // Handle identifier: [a-zA-Z_.][a-zA-Z0-9_$.@]*
539 if (isalpha(CurChar) || CurChar == '_' || CurChar == '.')
540 return LexIdentifier();
542 // Unknown character, emit an error.
543 return ReturnError(TokStart, "invalid character in input");
545 IsAtStartOfLine = true;
546 IsAtStartOfStatement = true;
547 return AsmToken(AsmToken::Eof, StringRef(TokStart, 0));
551 IsAtStartOfStatement = OldIsAtStartOfStatement;
552 while (*CurPtr == ' ' || *CurPtr == '\t')
555 return LexToken(); // Ignore whitespace.
557 return AsmToken(AsmToken::Space, StringRef(TokStart, CurPtr - TokStart));
560 IsAtStartOfLine = true;
561 IsAtStartOfStatement = true;
562 return AsmToken(AsmToken::EndOfStatement, StringRef(TokStart, 1));
563 case ':': return AsmToken(AsmToken::Colon, StringRef(TokStart, 1));
564 case '+': return AsmToken(AsmToken::Plus, StringRef(TokStart, 1));
565 case '-': return AsmToken(AsmToken::Minus, StringRef(TokStart, 1));
566 case '~': return AsmToken(AsmToken::Tilde, StringRef(TokStart, 1));
567 case '(': return AsmToken(AsmToken::LParen, StringRef(TokStart, 1));
568 case ')': return AsmToken(AsmToken::RParen, StringRef(TokStart, 1));
569 case '[': return AsmToken(AsmToken::LBrac, StringRef(TokStart, 1));
570 case ']': return AsmToken(AsmToken::RBrac, StringRef(TokStart, 1));
571 case '{': return AsmToken(AsmToken::LCurly, StringRef(TokStart, 1));
572 case '}': return AsmToken(AsmToken::RCurly, StringRef(TokStart, 1));
573 case '*': return AsmToken(AsmToken::Star, StringRef(TokStart, 1));
574 case ',': return AsmToken(AsmToken::Comma, StringRef(TokStart, 1));
575 case '$': return AsmToken(AsmToken::Dollar, StringRef(TokStart, 1));
576 case '@': return AsmToken(AsmToken::At, StringRef(TokStart, 1));
577 case '\\': return AsmToken(AsmToken::BackSlash, StringRef(TokStart, 1));
579 if (*CurPtr == '=') {
581 return AsmToken(AsmToken::EqualEqual, StringRef(TokStart, 2));
583 return AsmToken(AsmToken::Equal, StringRef(TokStart, 1));
585 if (*CurPtr == '|') {
587 return AsmToken(AsmToken::PipePipe, StringRef(TokStart, 2));
589 return AsmToken(AsmToken::Pipe, StringRef(TokStart, 1));
590 case '^': return AsmToken(AsmToken::Caret, StringRef(TokStart, 1));
592 if (*CurPtr == '&') {
594 return AsmToken(AsmToken::AmpAmp, StringRef(TokStart, 2));
596 return AsmToken(AsmToken::Amp, StringRef(TokStart, 1));
598 if (*CurPtr == '=') {
600 return AsmToken(AsmToken::ExclaimEqual, StringRef(TokStart, 2));
602 return AsmToken(AsmToken::Exclaim, StringRef(TokStart, 1));
603 case '%': return AsmToken(AsmToken::Percent, StringRef(TokStart, 1));
605 IsAtStartOfStatement = OldIsAtStartOfStatement;
607 case '#': return AsmToken(AsmToken::Hash, StringRef(TokStart, 1));
608 case '\'': return LexSingleQuote();
609 case '"': return LexQuote();
610 case '0': case '1': case '2': case '3': case '4':
611 case '5': case '6': case '7': case '8': case '9':
617 return AsmToken(AsmToken::LessLess, StringRef(TokStart, 2));
620 return AsmToken(AsmToken::LessEqual, StringRef(TokStart, 2));
623 return AsmToken(AsmToken::LessGreater, StringRef(TokStart, 2));
625 return AsmToken(AsmToken::Less, StringRef(TokStart, 1));
631 return AsmToken(AsmToken::GreaterGreater, StringRef(TokStart, 2));
634 return AsmToken(AsmToken::GreaterEqual, StringRef(TokStart, 2));
636 return AsmToken(AsmToken::Greater, StringRef(TokStart, 1));
639 // TODO: Quoted identifiers (objc methods etc)
640 // local labels: [0-9][:]
641 // Forward/backward labels: [0-9][fb]
642 // Integers, fp constants, character constants.