1 //===- AsmLexer.cpp - Lexer 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 lexer for assembly files.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #include "llvm/MC/MCParser/AsmLexer.h"
15 #include "llvm/ADT/APInt.h"
16 #include "llvm/ADT/ArrayRef.h"
17 #include "llvm/ADT/StringExtras.h"
18 #include "llvm/ADT/StringRef.h"
19 #include "llvm/ADT/StringSwitch.h"
20 #include "llvm/MC/MCAsmInfo.h"
21 #include "llvm/MC/MCParser/MCAsmLexer.h"
22 #include "llvm/Support/SMLoc.h"
23 #include "llvm/Support/SaveAndRestore.h"
24 #include <cassert>
25 #include <cctype>
26 #include <cstdio>
27 #include <cstring>
28 #include <string>
29 #include <tuple>
30 #include <utility>
31 
32 using namespace llvm;
33 
AsmLexer(const MCAsmInfo & MAI)34 AsmLexer::AsmLexer(const MCAsmInfo &MAI) : MAI(MAI) {
35   AllowAtInIdentifier = !StringRef(MAI.getCommentString()).startswith("@");
36 }
37 
38 AsmLexer::~AsmLexer() = default;
39 
setBuffer(StringRef Buf,const char * ptr)40 void AsmLexer::setBuffer(StringRef Buf, const char *ptr) {
41   CurBuf = Buf;
42 
43   if (ptr)
44     CurPtr = ptr;
45   else
46     CurPtr = CurBuf.begin();
47 
48   TokStart = nullptr;
49 }
50 
51 /// ReturnError - Set the error to the specified string at the specified
52 /// location.  This is defined to always return AsmToken::Error.
ReturnError(const char * Loc,const std::string & Msg)53 AsmToken AsmLexer::ReturnError(const char *Loc, const std::string &Msg) {
54   SetError(SMLoc::getFromPointer(Loc), Msg);
55 
56   return AsmToken(AsmToken::Error, StringRef(Loc, CurPtr - Loc));
57 }
58 
getNextChar()59 int AsmLexer::getNextChar() {
60   if (CurPtr == CurBuf.end())
61     return EOF;
62   return (unsigned char)*CurPtr++;
63 }
64 
65 /// LexFloatLiteral: [0-9]*[.][0-9]*([eE][+-]?[0-9]*)?
66 ///
67 /// The leading integral digit sequence and dot should have already been
68 /// consumed, some or all of the fractional digit sequence *can* have been
69 /// consumed.
LexFloatLiteral()70 AsmToken AsmLexer::LexFloatLiteral() {
71   // Skip the fractional digit sequence.
72   while (isDigit(*CurPtr))
73     ++CurPtr;
74 
75   // Check for exponent; we intentionally accept a slighlty wider set of
76   // literals here and rely on the upstream client to reject invalid ones (e.g.,
77   // "1e+").
78   if (*CurPtr == 'e' || *CurPtr == 'E') {
79     ++CurPtr;
80     if (*CurPtr == '-' || *CurPtr == '+')
81       ++CurPtr;
82     while (isDigit(*CurPtr))
83       ++CurPtr;
84   }
85 
86   return AsmToken(AsmToken::Real,
87                   StringRef(TokStart, CurPtr - TokStart));
88 }
89 
90 /// LexHexFloatLiteral matches essentially (.[0-9a-fA-F]*)?[pP][+-]?[0-9a-fA-F]+
91 /// while making sure there are enough actual digits around for the constant to
92 /// be valid.
93 ///
94 /// The leading "0x[0-9a-fA-F]*" (i.e. integer part) has already been consumed
95 /// before we get here.
LexHexFloatLiteral(bool NoIntDigits)96 AsmToken AsmLexer::LexHexFloatLiteral(bool NoIntDigits) {
97   assert((*CurPtr == 'p' || *CurPtr == 'P' || *CurPtr == '.') &&
98          "unexpected parse state in floating hex");
99   bool NoFracDigits = true;
100 
101   // Skip the fractional part if there is one
102   if (*CurPtr == '.') {
103     ++CurPtr;
104 
105     const char *FracStart = CurPtr;
106     while (isHexDigit(*CurPtr))
107       ++CurPtr;
108 
109     NoFracDigits = CurPtr == FracStart;
110   }
111 
112   if (NoIntDigits && NoFracDigits)
113     return ReturnError(TokStart, "invalid hexadecimal floating-point constant: "
114                                  "expected at least one significand digit");
115 
116   // Make sure we do have some kind of proper exponent part
117   if (*CurPtr != 'p' && *CurPtr != 'P')
118     return ReturnError(TokStart, "invalid hexadecimal floating-point constant: "
119                                  "expected exponent part 'p'");
120   ++CurPtr;
121 
122   if (*CurPtr == '+' || *CurPtr == '-')
123     ++CurPtr;
124 
125   // N.b. exponent digits are *not* hex
126   const char *ExpStart = CurPtr;
127   while (isDigit(*CurPtr))
128     ++CurPtr;
129 
130   if (CurPtr == ExpStart)
131     return ReturnError(TokStart, "invalid hexadecimal floating-point constant: "
132                                  "expected at least one exponent digit");
133 
134   return AsmToken(AsmToken::Real, StringRef(TokStart, CurPtr - TokStart));
135 }
136 
137 /// LexIdentifier: [a-zA-Z_.][a-zA-Z0-9_$.@?]*
IsIdentifierChar(char c,bool AllowAt)138 static bool IsIdentifierChar(char c, bool AllowAt) {
139   return isAlnum(c) || c == '_' || c == '$' || c == '.' ||
140          (c == '@' && AllowAt) || c == '?';
141 }
142 
LexIdentifier()143 AsmToken AsmLexer::LexIdentifier() {
144   // Check for floating point literals.
145   if (CurPtr[-1] == '.' && isDigit(*CurPtr)) {
146     // Disambiguate a .1243foo identifier from a floating literal.
147     while (isDigit(*CurPtr))
148       ++CurPtr;
149     if (*CurPtr == 'e' || *CurPtr == 'E' ||
150         !IsIdentifierChar(*CurPtr, AllowAtInIdentifier))
151       return LexFloatLiteral();
152   }
153 
154   while (IsIdentifierChar(*CurPtr, AllowAtInIdentifier))
155     ++CurPtr;
156 
157   // Handle . as a special case.
158   if (CurPtr == TokStart+1 && TokStart[0] == '.')
159     return AsmToken(AsmToken::Dot, StringRef(TokStart, 1));
160 
161   return AsmToken(AsmToken::Identifier, StringRef(TokStart, CurPtr - TokStart));
162 }
163 
164 /// LexSlash: Slash: /
165 ///           C-Style Comment: /* ... */
LexSlash()166 AsmToken AsmLexer::LexSlash() {
167   switch (*CurPtr) {
168   case '*':
169     IsAtStartOfStatement = false;
170     break; // C style comment.
171   case '/':
172     ++CurPtr;
173     return LexLineComment();
174   default:
175     IsAtStartOfStatement = false;
176     return AsmToken(AsmToken::Slash, StringRef(TokStart, 1));
177   }
178 
179   // C Style comment.
180   ++CurPtr;  // skip the star.
181   const char *CommentTextStart = CurPtr;
182   while (CurPtr != CurBuf.end()) {
183     switch (*CurPtr++) {
184     case '*':
185       // End of the comment?
186       if (*CurPtr != '/')
187         break;
188       // If we have a CommentConsumer, notify it about the comment.
189       if (CommentConsumer) {
190         CommentConsumer->HandleComment(
191             SMLoc::getFromPointer(CommentTextStart),
192             StringRef(CommentTextStart, CurPtr - 1 - CommentTextStart));
193       }
194       ++CurPtr;   // End the */.
195       return AsmToken(AsmToken::Comment,
196                       StringRef(TokStart, CurPtr - TokStart));
197     }
198   }
199   return ReturnError(TokStart, "unterminated comment");
200 }
201 
202 /// LexLineComment: Comment: #[^\n]*
203 ///                        : //[^\n]*
LexLineComment()204 AsmToken AsmLexer::LexLineComment() {
205   // Mark This as an end of statement with a body of the
206   // comment. While it would be nicer to leave this two tokens,
207   // backwards compatability with TargetParsers makes keeping this in this form
208   // better.
209   const char *CommentTextStart = CurPtr;
210   int CurChar = getNextChar();
211   while (CurChar != '\n' && CurChar != '\r' && CurChar != EOF)
212     CurChar = getNextChar();
213   if (CurChar == '\r' && CurPtr != CurBuf.end() && *CurPtr == '\n')
214     ++CurPtr;
215 
216   // If we have a CommentConsumer, notify it about the comment.
217   if (CommentConsumer) {
218     CommentConsumer->HandleComment(
219         SMLoc::getFromPointer(CommentTextStart),
220         StringRef(CommentTextStart, CurPtr - 1 - CommentTextStart));
221   }
222 
223   IsAtStartOfLine = true;
224   // This is a whole line comment. leave newline
225   if (IsAtStartOfStatement)
226     return AsmToken(AsmToken::EndOfStatement,
227                     StringRef(TokStart, CurPtr - TokStart));
228   IsAtStartOfStatement = true;
229 
230   return AsmToken(AsmToken::EndOfStatement,
231                   StringRef(TokStart, CurPtr - 1 - TokStart));
232 }
233 
SkipIgnoredIntegerSuffix(const char * & CurPtr)234 static void SkipIgnoredIntegerSuffix(const char *&CurPtr) {
235   // Skip ULL, UL, U, L and LL suffices.
236   if (CurPtr[0] == 'U')
237     ++CurPtr;
238   if (CurPtr[0] == 'L')
239     ++CurPtr;
240   if (CurPtr[0] == 'L')
241     ++CurPtr;
242 }
243 
244 // Look ahead to search for first non-hex digit, if it's [hH], then we treat the
245 // integer as a hexadecimal, possibly with leading zeroes.
doHexLookAhead(const char * & CurPtr,unsigned DefaultRadix,bool LexHex)246 static unsigned doHexLookAhead(const char *&CurPtr, unsigned DefaultRadix,
247                                bool LexHex) {
248   const char *FirstNonDec = nullptr;
249   const char *LookAhead = CurPtr;
250   while (true) {
251     if (isDigit(*LookAhead)) {
252       ++LookAhead;
253     } else {
254       if (!FirstNonDec)
255         FirstNonDec = LookAhead;
256 
257       // Keep going if we are looking for a 'h' suffix.
258       if (LexHex && isHexDigit(*LookAhead))
259         ++LookAhead;
260       else
261         break;
262     }
263   }
264   bool isHex = LexHex && (*LookAhead == 'h' || *LookAhead == 'H');
265   CurPtr = isHex || !FirstNonDec ? LookAhead : FirstNonDec;
266   if (isHex)
267     return 16;
268   return DefaultRadix;
269 }
270 
intToken(StringRef Ref,APInt & Value)271 static AsmToken intToken(StringRef Ref, APInt &Value)
272 {
273   if (Value.isIntN(64))
274     return AsmToken(AsmToken::Integer, Ref, Value);
275   return AsmToken(AsmToken::BigNum, Ref, Value);
276 }
277 
278 /// LexDigit: First character is [0-9].
279 ///   Local Label: [0-9][:]
280 ///   Forward/Backward Label: [0-9][fb]
281 ///   Binary integer: 0b[01]+
282 ///   Octal integer: 0[0-7]+
283 ///   Hex integer: 0x[0-9a-fA-F]+ or [0x]?[0-9][0-9a-fA-F]*[hH]
284 ///   Decimal integer: [1-9][0-9]*
LexDigit()285 AsmToken AsmLexer::LexDigit() {
286   // MASM-flavor binary integer: [01]+[bB]
287   // MASM-flavor hexadecimal integer: [0-9][0-9a-fA-F]*[hH]
288   if (LexMasmIntegers && isdigit(CurPtr[-1])) {
289     const char *FirstNonBinary = (CurPtr[-1] != '0' && CurPtr[-1] != '1') ?
290                                    CurPtr - 1 : nullptr;
291     const char *OldCurPtr = CurPtr;
292     while (isHexDigit(*CurPtr)) {
293       if (*CurPtr != '0' && *CurPtr != '1' && !FirstNonBinary)
294         FirstNonBinary = CurPtr;
295       ++CurPtr;
296     }
297 
298     unsigned Radix = 0;
299     if (*CurPtr == 'h' || *CurPtr == 'H') {
300       // hexadecimal number
301       ++CurPtr;
302       Radix = 16;
303     } else if (FirstNonBinary && FirstNonBinary + 1 == CurPtr &&
304                (*FirstNonBinary == 'b' || *FirstNonBinary == 'B'))
305       Radix = 2;
306 
307     if (Radix == 2 || Radix == 16) {
308       StringRef Result(TokStart, CurPtr - TokStart);
309       APInt Value(128, 0, true);
310 
311       if (Result.drop_back().getAsInteger(Radix, Value))
312         return ReturnError(TokStart, Radix == 2 ? "invalid binary number" :
313                              "invalid hexdecimal number");
314 
315       // MSVC accepts and ignores type suffices on integer literals.
316       SkipIgnoredIntegerSuffix(CurPtr);
317 
318       return intToken(Result, Value);
319    }
320 
321     // octal/decimal integers, or floating point numbers, fall through
322     CurPtr = OldCurPtr;
323   }
324 
325   // Decimal integer: [1-9][0-9]*
326   if (CurPtr[-1] != '0' || CurPtr[0] == '.') {
327     unsigned Radix = doHexLookAhead(CurPtr, 10, LexMasmIntegers);
328     bool isHex = Radix == 16;
329     // Check for floating point literals.
330     if (!isHex && (*CurPtr == '.' || *CurPtr == 'e')) {
331       ++CurPtr;
332       return LexFloatLiteral();
333     }
334 
335     StringRef Result(TokStart, CurPtr - TokStart);
336 
337     APInt Value(128, 0, true);
338     if (Result.getAsInteger(Radix, Value))
339       return ReturnError(TokStart, !isHex ? "invalid decimal number" :
340                            "invalid hexdecimal number");
341 
342     // Consume the [hH].
343     if (LexMasmIntegers && Radix == 16)
344       ++CurPtr;
345 
346     // The darwin/x86 (and x86-64) assembler accepts and ignores type
347     // suffices on integer literals.
348     SkipIgnoredIntegerSuffix(CurPtr);
349 
350     return intToken(Result, Value);
351   }
352 
353   if (!LexMasmIntegers && ((*CurPtr == 'b') || (*CurPtr == 'B'))) {
354     ++CurPtr;
355     // See if we actually have "0b" as part of something like "jmp 0b\n"
356     if (!isDigit(CurPtr[0])) {
357       --CurPtr;
358       StringRef Result(TokStart, CurPtr - TokStart);
359       return AsmToken(AsmToken::Integer, Result, 0);
360     }
361     const char *NumStart = CurPtr;
362     while (CurPtr[0] == '0' || CurPtr[0] == '1')
363       ++CurPtr;
364 
365     // Requires at least one binary digit.
366     if (CurPtr == NumStart)
367       return ReturnError(TokStart, "invalid binary number");
368 
369     StringRef Result(TokStart, CurPtr - TokStart);
370 
371     APInt Value(128, 0, true);
372     if (Result.substr(2).getAsInteger(2, Value))
373       return ReturnError(TokStart, "invalid binary number");
374 
375     // The darwin/x86 (and x86-64) assembler accepts and ignores ULL and LL
376     // suffixes on integer literals.
377     SkipIgnoredIntegerSuffix(CurPtr);
378 
379     return intToken(Result, Value);
380   }
381 
382   if ((*CurPtr == 'x') || (*CurPtr == 'X')) {
383     ++CurPtr;
384     const char *NumStart = CurPtr;
385     while (isHexDigit(CurPtr[0]))
386       ++CurPtr;
387 
388     // "0x.0p0" is valid, and "0x0p0" (but not "0xp0" for example, which will be
389     // diagnosed by LexHexFloatLiteral).
390     if (CurPtr[0] == '.' || CurPtr[0] == 'p' || CurPtr[0] == 'P')
391       return LexHexFloatLiteral(NumStart == CurPtr);
392 
393     // Otherwise requires at least one hex digit.
394     if (CurPtr == NumStart)
395       return ReturnError(CurPtr-2, "invalid hexadecimal number");
396 
397     APInt Result(128, 0);
398     if (StringRef(TokStart, CurPtr - TokStart).getAsInteger(0, Result))
399       return ReturnError(TokStart, "invalid hexadecimal number");
400 
401     // Consume the optional [hH].
402     if (LexMasmIntegers && (*CurPtr == 'h' || *CurPtr == 'H'))
403       ++CurPtr;
404 
405     // The darwin/x86 (and x86-64) assembler accepts and ignores ULL and LL
406     // suffixes on integer literals.
407     SkipIgnoredIntegerSuffix(CurPtr);
408 
409     return intToken(StringRef(TokStart, CurPtr - TokStart), Result);
410   }
411 
412   // Either octal or hexadecimal.
413   APInt Value(128, 0, true);
414   unsigned Radix = doHexLookAhead(CurPtr, 8, LexMasmIntegers);
415   bool isHex = Radix == 16;
416   StringRef Result(TokStart, CurPtr - TokStart);
417   if (Result.getAsInteger(Radix, Value))
418     return ReturnError(TokStart, !isHex ? "invalid octal number" :
419                        "invalid hexdecimal number");
420 
421   // Consume the [hH].
422   if (Radix == 16)
423     ++CurPtr;
424 
425   // The darwin/x86 (and x86-64) assembler accepts and ignores ULL and LL
426   // suffixes on integer literals.
427   SkipIgnoredIntegerSuffix(CurPtr);
428 
429   return intToken(Result, Value);
430 }
431 
432 /// LexSingleQuote: Integer: 'b'
LexSingleQuote()433 AsmToken AsmLexer::LexSingleQuote() {
434   int CurChar = getNextChar();
435 
436   if (CurChar == '\\')
437     CurChar = getNextChar();
438 
439   if (CurChar == EOF)
440     return ReturnError(TokStart, "unterminated single quote");
441 
442   CurChar = getNextChar();
443 
444   if (CurChar != '\'')
445     return ReturnError(TokStart, "single quote way too long");
446 
447   // The idea here being that 'c' is basically just an integral
448   // constant.
449   StringRef Res = StringRef(TokStart,CurPtr - TokStart);
450   long long Value;
451 
452   if (Res.startswith("\'\\")) {
453     char theChar = Res[2];
454     switch (theChar) {
455       default: Value = theChar; break;
456       case '\'': Value = '\''; break;
457       case 't': Value = '\t'; break;
458       case 'n': Value = '\n'; break;
459       case 'b': Value = '\b'; break;
460     }
461   } else
462     Value = TokStart[1];
463 
464   return AsmToken(AsmToken::Integer, Res, Value);
465 }
466 
467 /// LexQuote: String: "..."
LexQuote()468 AsmToken AsmLexer::LexQuote() {
469   int CurChar = getNextChar();
470   // TODO: does gas allow multiline string constants?
471   while (CurChar != '"') {
472     if (CurChar == '\\') {
473       // Allow \", etc.
474       CurChar = getNextChar();
475     }
476 
477     if (CurChar == EOF)
478       return ReturnError(TokStart, "unterminated string constant");
479 
480     CurChar = getNextChar();
481   }
482 
483   return AsmToken(AsmToken::String, StringRef(TokStart, CurPtr - TokStart));
484 }
485 
LexUntilEndOfStatement()486 StringRef AsmLexer::LexUntilEndOfStatement() {
487   TokStart = CurPtr;
488 
489   while (!isAtStartOfComment(CurPtr) &&     // Start of line comment.
490          !isAtStatementSeparator(CurPtr) && // End of statement marker.
491          *CurPtr != '\n' && *CurPtr != '\r' && CurPtr != CurBuf.end()) {
492     ++CurPtr;
493   }
494   return StringRef(TokStart, CurPtr-TokStart);
495 }
496 
LexUntilEndOfLine()497 StringRef AsmLexer::LexUntilEndOfLine() {
498   TokStart = CurPtr;
499 
500   while (*CurPtr != '\n' && *CurPtr != '\r' && CurPtr != CurBuf.end()) {
501     ++CurPtr;
502   }
503   return StringRef(TokStart, CurPtr-TokStart);
504 }
505 
peekTokens(MutableArrayRef<AsmToken> Buf,bool ShouldSkipSpace)506 size_t AsmLexer::peekTokens(MutableArrayRef<AsmToken> Buf,
507                             bool ShouldSkipSpace) {
508   SaveAndRestore<const char *> SavedTokenStart(TokStart);
509   SaveAndRestore<const char *> SavedCurPtr(CurPtr);
510   SaveAndRestore<bool> SavedAtStartOfLine(IsAtStartOfLine);
511   SaveAndRestore<bool> SavedAtStartOfStatement(IsAtStartOfStatement);
512   SaveAndRestore<bool> SavedSkipSpace(SkipSpace, ShouldSkipSpace);
513   SaveAndRestore<bool> SavedIsPeeking(IsPeeking, true);
514   std::string SavedErr = getErr();
515   SMLoc SavedErrLoc = getErrLoc();
516 
517   size_t ReadCount;
518   for (ReadCount = 0; ReadCount < Buf.size(); ++ReadCount) {
519     AsmToken Token = LexToken();
520 
521     Buf[ReadCount] = Token;
522 
523     if (Token.is(AsmToken::Eof))
524       break;
525   }
526 
527   SetError(SavedErrLoc, SavedErr);
528   return ReadCount;
529 }
530 
isAtStartOfComment(const char * Ptr)531 bool AsmLexer::isAtStartOfComment(const char *Ptr) {
532   StringRef CommentString = MAI.getCommentString();
533 
534   if (CommentString.size() == 1)
535     return CommentString[0] == Ptr[0];
536 
537   // Allow # preprocessor commments also be counted as comments for "##" cases
538   if (CommentString[1] == '#')
539     return CommentString[0] == Ptr[0];
540 
541   return strncmp(Ptr, CommentString.data(), CommentString.size()) == 0;
542 }
543 
isAtStatementSeparator(const char * Ptr)544 bool AsmLexer::isAtStatementSeparator(const char *Ptr) {
545   return strncmp(Ptr, MAI.getSeparatorString(),
546                  strlen(MAI.getSeparatorString())) == 0;
547 }
548 
LexToken()549 AsmToken AsmLexer::LexToken() {
550   TokStart = CurPtr;
551   // This always consumes at least one character.
552   int CurChar = getNextChar();
553 
554   if (!IsPeeking && CurChar == '#' && IsAtStartOfStatement) {
555     // If this starts with a '#', this may be a cpp
556     // hash directive and otherwise a line comment.
557     AsmToken TokenBuf[2];
558     MutableArrayRef<AsmToken> Buf(TokenBuf, 2);
559     size_t num = peekTokens(Buf, true);
560     // There cannot be a space preceeding this
561     if (IsAtStartOfLine && num == 2 && TokenBuf[0].is(AsmToken::Integer) &&
562         TokenBuf[1].is(AsmToken::String)) {
563       CurPtr = TokStart; // reset curPtr;
564       StringRef s = LexUntilEndOfLine();
565       UnLex(TokenBuf[1]);
566       UnLex(TokenBuf[0]);
567       return AsmToken(AsmToken::HashDirective, s);
568     }
569     return LexLineComment();
570   }
571 
572   if (isAtStartOfComment(TokStart))
573     return LexLineComment();
574 
575   if (isAtStatementSeparator(TokStart)) {
576     CurPtr += strlen(MAI.getSeparatorString()) - 1;
577     IsAtStartOfLine = true;
578     IsAtStartOfStatement = true;
579     return AsmToken(AsmToken::EndOfStatement,
580                     StringRef(TokStart, strlen(MAI.getSeparatorString())));
581   }
582 
583   // If we're missing a newline at EOF, make sure we still get an
584   // EndOfStatement token before the Eof token.
585   if (CurChar == EOF && !IsAtStartOfStatement) {
586     IsAtStartOfLine = true;
587     IsAtStartOfStatement = true;
588     return AsmToken(AsmToken::EndOfStatement, StringRef(TokStart, 1));
589   }
590   IsAtStartOfLine = false;
591   bool OldIsAtStartOfStatement = IsAtStartOfStatement;
592   IsAtStartOfStatement = false;
593   switch (CurChar) {
594   default:
595     // Handle identifier: [a-zA-Z_.][a-zA-Z0-9_$.@]*
596     if (isalpha(CurChar) || CurChar == '_' || CurChar == '.')
597       return LexIdentifier();
598 
599     // Unknown character, emit an error.
600     return ReturnError(TokStart, "invalid character in input");
601   case EOF:
602     IsAtStartOfLine = true;
603     IsAtStartOfStatement = true;
604     return AsmToken(AsmToken::Eof, StringRef(TokStart, 0));
605   case 0:
606   case ' ':
607   case '\t':
608     IsAtStartOfStatement = OldIsAtStartOfStatement;
609     while (*CurPtr == ' ' || *CurPtr == '\t')
610       CurPtr++;
611     if (SkipSpace)
612       return LexToken(); // Ignore whitespace.
613     else
614       return AsmToken(AsmToken::Space, StringRef(TokStart, CurPtr - TokStart));
615   case '\r': {
616     IsAtStartOfLine = true;
617     IsAtStartOfStatement = true;
618     // If this is a CR followed by LF, treat that as one token.
619     if (CurPtr != CurBuf.end() && *CurPtr == '\n')
620       ++CurPtr;
621     return AsmToken(AsmToken::EndOfStatement,
622                     StringRef(TokStart, CurPtr - TokStart));
623   }
624   case '\n':
625     IsAtStartOfLine = true;
626     IsAtStartOfStatement = true;
627     return AsmToken(AsmToken::EndOfStatement, StringRef(TokStart, 1));
628   case ':': return AsmToken(AsmToken::Colon, StringRef(TokStart, 1));
629   case '+': return AsmToken(AsmToken::Plus, StringRef(TokStart, 1));
630   case '~': return AsmToken(AsmToken::Tilde, StringRef(TokStart, 1));
631   case '(': return AsmToken(AsmToken::LParen, StringRef(TokStart, 1));
632   case ')': return AsmToken(AsmToken::RParen, StringRef(TokStart, 1));
633   case '[': return AsmToken(AsmToken::LBrac, StringRef(TokStart, 1));
634   case ']': return AsmToken(AsmToken::RBrac, StringRef(TokStart, 1));
635   case '{': return AsmToken(AsmToken::LCurly, StringRef(TokStart, 1));
636   case '}': return AsmToken(AsmToken::RCurly, StringRef(TokStart, 1));
637   case '*': return AsmToken(AsmToken::Star, StringRef(TokStart, 1));
638   case ',': return AsmToken(AsmToken::Comma, StringRef(TokStart, 1));
639   case '$': return AsmToken(AsmToken::Dollar, StringRef(TokStart, 1));
640   case '@': return AsmToken(AsmToken::At, StringRef(TokStart, 1));
641   case '\\': return AsmToken(AsmToken::BackSlash, StringRef(TokStart, 1));
642   case '=':
643     if (*CurPtr == '=') {
644       ++CurPtr;
645       return AsmToken(AsmToken::EqualEqual, StringRef(TokStart, 2));
646     }
647     return AsmToken(AsmToken::Equal, StringRef(TokStart, 1));
648   case '-':
649     if (*CurPtr == '>') {
650       ++CurPtr;
651       return AsmToken(AsmToken::MinusGreater, StringRef(TokStart, 2));
652     }
653     return AsmToken(AsmToken::Minus, StringRef(TokStart, 1));
654   case '|':
655     if (*CurPtr == '|') {
656       ++CurPtr;
657       return AsmToken(AsmToken::PipePipe, StringRef(TokStart, 2));
658     }
659     return AsmToken(AsmToken::Pipe, StringRef(TokStart, 1));
660   case '^': return AsmToken(AsmToken::Caret, StringRef(TokStart, 1));
661   case '&':
662     if (*CurPtr == '&') {
663       ++CurPtr;
664       return AsmToken(AsmToken::AmpAmp, StringRef(TokStart, 2));
665     }
666     return AsmToken(AsmToken::Amp, StringRef(TokStart, 1));
667   case '!':
668     if (*CurPtr == '=') {
669       ++CurPtr;
670       return AsmToken(AsmToken::ExclaimEqual, StringRef(TokStart, 2));
671     }
672     return AsmToken(AsmToken::Exclaim, StringRef(TokStart, 1));
673   case '%':
674     if (MAI.hasMipsExpressions()) {
675       AsmToken::TokenKind Operator;
676       unsigned OperatorLength;
677 
678       std::tie(Operator, OperatorLength) =
679           StringSwitch<std::pair<AsmToken::TokenKind, unsigned>>(
680               StringRef(CurPtr))
681               .StartsWith("call16", {AsmToken::PercentCall16, 7})
682               .StartsWith("call_hi", {AsmToken::PercentCall_Hi, 8})
683               .StartsWith("call_lo", {AsmToken::PercentCall_Lo, 8})
684               .StartsWith("dtprel_hi", {AsmToken::PercentDtprel_Hi, 10})
685               .StartsWith("dtprel_lo", {AsmToken::PercentDtprel_Lo, 10})
686               .StartsWith("got_disp", {AsmToken::PercentGot_Disp, 9})
687               .StartsWith("got_hi", {AsmToken::PercentGot_Hi, 7})
688               .StartsWith("got_lo", {AsmToken::PercentGot_Lo, 7})
689               .StartsWith("got_ofst", {AsmToken::PercentGot_Ofst, 9})
690               .StartsWith("got_page", {AsmToken::PercentGot_Page, 9})
691               .StartsWith("gottprel", {AsmToken::PercentGottprel, 9})
692               .StartsWith("got", {AsmToken::PercentGot, 4})
693               .StartsWith("gp_rel", {AsmToken::PercentGp_Rel, 7})
694               .StartsWith("higher", {AsmToken::PercentHigher, 7})
695               .StartsWith("highest", {AsmToken::PercentHighest, 8})
696               .StartsWith("hi", {AsmToken::PercentHi, 3})
697               .StartsWith("lo", {AsmToken::PercentLo, 3})
698               .StartsWith("neg", {AsmToken::PercentNeg, 4})
699               .StartsWith("pcrel_hi", {AsmToken::PercentPcrel_Hi, 9})
700               .StartsWith("pcrel_lo", {AsmToken::PercentPcrel_Lo, 9})
701               .StartsWith("tlsgd", {AsmToken::PercentTlsgd, 6})
702               .StartsWith("tlsldm", {AsmToken::PercentTlsldm, 7})
703               .StartsWith("tprel_hi", {AsmToken::PercentTprel_Hi, 9})
704               .StartsWith("tprel_lo", {AsmToken::PercentTprel_Lo, 9})
705               .Default({AsmToken::Percent, 1});
706 
707       if (Operator != AsmToken::Percent) {
708         CurPtr += OperatorLength - 1;
709         return AsmToken(Operator, StringRef(TokStart, OperatorLength));
710       }
711     }
712     return AsmToken(AsmToken::Percent, StringRef(TokStart, 1));
713   case '/':
714     IsAtStartOfStatement = OldIsAtStartOfStatement;
715     return LexSlash();
716   case '#': return AsmToken(AsmToken::Hash, StringRef(TokStart, 1));
717   case '\'': return LexSingleQuote();
718   case '"': return LexQuote();
719   case '0': case '1': case '2': case '3': case '4':
720   case '5': case '6': case '7': case '8': case '9':
721     return LexDigit();
722   case '<':
723     switch (*CurPtr) {
724     case '<':
725       ++CurPtr;
726       return AsmToken(AsmToken::LessLess, StringRef(TokStart, 2));
727     case '=':
728       ++CurPtr;
729       return AsmToken(AsmToken::LessEqual, StringRef(TokStart, 2));
730     case '>':
731       ++CurPtr;
732       return AsmToken(AsmToken::LessGreater, StringRef(TokStart, 2));
733     default:
734       return AsmToken(AsmToken::Less, StringRef(TokStart, 1));
735     }
736   case '>':
737     switch (*CurPtr) {
738     case '>':
739       ++CurPtr;
740       return AsmToken(AsmToken::GreaterGreater, StringRef(TokStart, 2));
741     case '=':
742       ++CurPtr;
743       return AsmToken(AsmToken::GreaterEqual, StringRef(TokStart, 2));
744     default:
745       return AsmToken(AsmToken::Greater, StringRef(TokStart, 1));
746     }
747 
748   // TODO: Quoted identifiers (objc methods etc)
749   // local labels: [0-9][:]
750   // Forward/backward labels: [0-9][fb]
751   // Integers, fp constants, character constants.
752   }
753 }
754