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 
34 AsmLexer::AsmLexer(const MCAsmInfo &MAI) : MAI(MAI) {
35   AllowAtInIdentifier = !StringRef(MAI.getCommentString()).startswith("@");
36 }
37 
38 AsmLexer::~AsmLexer() = default;
39 
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.
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 
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.
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.
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_$.@?]*
138 static bool IsIdentifierChar(char c, bool AllowAt) {
139   return isAlnum(c) || c == '_' || c == '$' || c == '.' ||
140          (c == '@' && AllowAt) || c == '?';
141 }
142 
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: /* ... */
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]*
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 
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.
246 static unsigned doLookAhead(const char *&CurPtr, unsigned DefaultRadix) {
247   const char *FirstHex = nullptr;
248   const char *LookAhead = CurPtr;
249   while (true) {
250     if (isDigit(*LookAhead)) {
251       ++LookAhead;
252     } else if (isHexDigit(*LookAhead)) {
253       if (!FirstHex)
254         FirstHex = LookAhead;
255       ++LookAhead;
256     } else {
257       break;
258     }
259   }
260   bool isHex = *LookAhead == 'h' || *LookAhead == 'H';
261   CurPtr = isHex || !FirstHex ? LookAhead : FirstHex;
262   if (isHex)
263     return 16;
264   return DefaultRadix;
265 }
266 
267 static AsmToken intToken(StringRef Ref, APInt &Value)
268 {
269   if (Value.isIntN(64))
270     return AsmToken(AsmToken::Integer, Ref, Value);
271   return AsmToken(AsmToken::BigNum, Ref, Value);
272 }
273 
274 /// LexDigit: First character is [0-9].
275 ///   Local Label: [0-9][:]
276 ///   Forward/Backward Label: [0-9][fb]
277 ///   Binary integer: 0b[01]+
278 ///   Octal integer: 0[0-7]+
279 ///   Hex integer: 0x[0-9a-fA-F]+ or [0x]?[0-9][0-9a-fA-F]*[hH]
280 ///   Decimal integer: [1-9][0-9]*
281 AsmToken AsmLexer::LexDigit() {
282   // MASM-flavor binary integer: [01]+[bB]
283   // MASM-flavor hexadecimal integer: [0-9][0-9a-fA-F]*[hH]
284   if (IsParsingMSInlineAsm && isdigit(CurPtr[-1])) {
285     const char *FirstNonBinary = (CurPtr[-1] != '0' && CurPtr[-1] != '1') ?
286                                    CurPtr - 1 : nullptr;
287     const char *OldCurPtr = CurPtr;
288     while (isHexDigit(*CurPtr)) {
289       if (*CurPtr != '0' && *CurPtr != '1' && !FirstNonBinary)
290         FirstNonBinary = CurPtr;
291       ++CurPtr;
292     }
293 
294     unsigned Radix = 0;
295     if (*CurPtr == 'h' || *CurPtr == 'H') {
296       // hexadecimal number
297       ++CurPtr;
298       Radix = 16;
299     } else if (FirstNonBinary && FirstNonBinary + 1 == CurPtr &&
300                (*FirstNonBinary == 'b' || *FirstNonBinary == 'B'))
301       Radix = 2;
302 
303     if (Radix == 2 || Radix == 16) {
304       StringRef Result(TokStart, CurPtr - TokStart);
305       APInt Value(128, 0, true);
306 
307       if (Result.drop_back().getAsInteger(Radix, Value))
308         return ReturnError(TokStart, Radix == 2 ? "invalid binary number" :
309                              "invalid hexdecimal number");
310 
311       // MSVC accepts and ignores type suffices on integer literals.
312       SkipIgnoredIntegerSuffix(CurPtr);
313 
314       return intToken(Result, Value);
315    }
316 
317     // octal/decimal integers, or floating point numbers, fall through
318     CurPtr = OldCurPtr;
319   }
320 
321   // Decimal integer: [1-9][0-9]*
322   if (CurPtr[-1] != '0' || CurPtr[0] == '.') {
323     unsigned Radix = doLookAhead(CurPtr, 10);
324     bool isHex = Radix == 16;
325     // Check for floating point literals.
326     if (!isHex && (*CurPtr == '.' || *CurPtr == 'e')) {
327       ++CurPtr;
328       return LexFloatLiteral();
329     }
330 
331     StringRef Result(TokStart, CurPtr - TokStart);
332 
333     APInt Value(128, 0, true);
334     if (Result.getAsInteger(Radix, Value))
335       return ReturnError(TokStart, !isHex ? "invalid decimal number" :
336                            "invalid hexdecimal number");
337 
338     // Consume the [bB][hH].
339     if (Radix == 2 || Radix == 16)
340       ++CurPtr;
341 
342     // The darwin/x86 (and x86-64) assembler accepts and ignores type
343     // suffices on integer literals.
344     SkipIgnoredIntegerSuffix(CurPtr);
345 
346     return intToken(Result, Value);
347   }
348 
349   if (!IsParsingMSInlineAsm && ((*CurPtr == 'b') || (*CurPtr == 'B'))) {
350     ++CurPtr;
351     // See if we actually have "0b" as part of something like "jmp 0b\n"
352     if (!isDigit(CurPtr[0])) {
353       --CurPtr;
354       StringRef Result(TokStart, CurPtr - TokStart);
355       return AsmToken(AsmToken::Integer, Result, 0);
356     }
357     const char *NumStart = CurPtr;
358     while (CurPtr[0] == '0' || CurPtr[0] == '1')
359       ++CurPtr;
360 
361     // Requires at least one binary digit.
362     if (CurPtr == NumStart)
363       return ReturnError(TokStart, "invalid binary number");
364 
365     StringRef Result(TokStart, CurPtr - TokStart);
366 
367     APInt Value(128, 0, true);
368     if (Result.substr(2).getAsInteger(2, Value))
369       return ReturnError(TokStart, "invalid binary number");
370 
371     // The darwin/x86 (and x86-64) assembler accepts and ignores ULL and LL
372     // suffixes on integer literals.
373     SkipIgnoredIntegerSuffix(CurPtr);
374 
375     return intToken(Result, Value);
376   }
377 
378   if ((*CurPtr == 'x') || (*CurPtr == 'X')) {
379     ++CurPtr;
380     const char *NumStart = CurPtr;
381     while (isHexDigit(CurPtr[0]))
382       ++CurPtr;
383 
384     // "0x.0p0" is valid, and "0x0p0" (but not "0xp0" for example, which will be
385     // diagnosed by LexHexFloatLiteral).
386     if (CurPtr[0] == '.' || CurPtr[0] == 'p' || CurPtr[0] == 'P')
387       return LexHexFloatLiteral(NumStart == CurPtr);
388 
389     // Otherwise requires at least one hex digit.
390     if (CurPtr == NumStart)
391       return ReturnError(CurPtr-2, "invalid hexadecimal number");
392 
393     APInt Result(128, 0);
394     if (StringRef(TokStart, CurPtr - TokStart).getAsInteger(0, Result))
395       return ReturnError(TokStart, "invalid hexadecimal number");
396 
397     // Consume the optional [hH].
398     if (!IsParsingMSInlineAsm && (*CurPtr == 'h' || *CurPtr == 'H'))
399       ++CurPtr;
400 
401     // The darwin/x86 (and x86-64) assembler accepts and ignores ULL and LL
402     // suffixes on integer literals.
403     SkipIgnoredIntegerSuffix(CurPtr);
404 
405     return intToken(StringRef(TokStart, CurPtr - TokStart), Result);
406   }
407 
408   // Either octal or hexadecimal.
409   APInt Value(128, 0, true);
410   unsigned Radix = doLookAhead(CurPtr, 8);
411   bool isHex = Radix == 16;
412   StringRef Result(TokStart, CurPtr - TokStart);
413   if (Result.getAsInteger(Radix, Value))
414     return ReturnError(TokStart, !isHex ? "invalid octal number" :
415                        "invalid hexdecimal number");
416 
417   // Consume the [hH].
418   if (Radix == 16)
419     ++CurPtr;
420 
421   // The darwin/x86 (and x86-64) assembler accepts and ignores ULL and LL
422   // suffixes on integer literals.
423   SkipIgnoredIntegerSuffix(CurPtr);
424 
425   return intToken(Result, Value);
426 }
427 
428 /// LexSingleQuote: Integer: 'b'
429 AsmToken AsmLexer::LexSingleQuote() {
430   int CurChar = getNextChar();
431 
432   if (CurChar == '\\')
433     CurChar = getNextChar();
434 
435   if (CurChar == EOF)
436     return ReturnError(TokStart, "unterminated single quote");
437 
438   CurChar = getNextChar();
439 
440   if (CurChar != '\'')
441     return ReturnError(TokStart, "single quote way too long");
442 
443   // The idea here being that 'c' is basically just an integral
444   // constant.
445   StringRef Res = StringRef(TokStart,CurPtr - TokStart);
446   long long Value;
447 
448   if (Res.startswith("\'\\")) {
449     char theChar = Res[2];
450     switch (theChar) {
451       default: Value = theChar; break;
452       case '\'': Value = '\''; break;
453       case 't': Value = '\t'; break;
454       case 'n': Value = '\n'; break;
455       case 'b': Value = '\b'; break;
456     }
457   } else
458     Value = TokStart[1];
459 
460   return AsmToken(AsmToken::Integer, Res, Value);
461 }
462 
463 /// LexQuote: String: "..."
464 AsmToken AsmLexer::LexQuote() {
465   int CurChar = getNextChar();
466   // TODO: does gas allow multiline string constants?
467   while (CurChar != '"') {
468     if (CurChar == '\\') {
469       // Allow \", etc.
470       CurChar = getNextChar();
471     }
472 
473     if (CurChar == EOF)
474       return ReturnError(TokStart, "unterminated string constant");
475 
476     CurChar = getNextChar();
477   }
478 
479   return AsmToken(AsmToken::String, StringRef(TokStart, CurPtr - TokStart));
480 }
481 
482 StringRef AsmLexer::LexUntilEndOfStatement() {
483   TokStart = CurPtr;
484 
485   while (!isAtStartOfComment(CurPtr) &&     // Start of line comment.
486          !isAtStatementSeparator(CurPtr) && // End of statement marker.
487          *CurPtr != '\n' && *CurPtr != '\r' && CurPtr != CurBuf.end()) {
488     ++CurPtr;
489   }
490   return StringRef(TokStart, CurPtr-TokStart);
491 }
492 
493 StringRef AsmLexer::LexUntilEndOfLine() {
494   TokStart = CurPtr;
495 
496   while (*CurPtr != '\n' && *CurPtr != '\r' && CurPtr != CurBuf.end()) {
497     ++CurPtr;
498   }
499   return StringRef(TokStart, CurPtr-TokStart);
500 }
501 
502 size_t AsmLexer::peekTokens(MutableArrayRef<AsmToken> Buf,
503                             bool ShouldSkipSpace) {
504   SaveAndRestore<const char *> SavedTokenStart(TokStart);
505   SaveAndRestore<const char *> SavedCurPtr(CurPtr);
506   SaveAndRestore<bool> SavedAtStartOfLine(IsAtStartOfLine);
507   SaveAndRestore<bool> SavedAtStartOfStatement(IsAtStartOfStatement);
508   SaveAndRestore<bool> SavedSkipSpace(SkipSpace, ShouldSkipSpace);
509   SaveAndRestore<bool> SavedIsPeeking(IsPeeking, true);
510   std::string SavedErr = getErr();
511   SMLoc SavedErrLoc = getErrLoc();
512 
513   size_t ReadCount;
514   for (ReadCount = 0; ReadCount < Buf.size(); ++ReadCount) {
515     AsmToken Token = LexToken();
516 
517     Buf[ReadCount] = Token;
518 
519     if (Token.is(AsmToken::Eof))
520       break;
521   }
522 
523   SetError(SavedErrLoc, SavedErr);
524   return ReadCount;
525 }
526 
527 bool AsmLexer::isAtStartOfComment(const char *Ptr) {
528   StringRef CommentString = MAI.getCommentString();
529 
530   if (CommentString.size() == 1)
531     return CommentString[0] == Ptr[0];
532 
533   // Allow # preprocessor commments also be counted as comments for "##" cases
534   if (CommentString[1] == '#')
535     return CommentString[0] == Ptr[0];
536 
537   return strncmp(Ptr, CommentString.data(), CommentString.size()) == 0;
538 }
539 
540 bool AsmLexer::isAtStatementSeparator(const char *Ptr) {
541   return strncmp(Ptr, MAI.getSeparatorString(),
542                  strlen(MAI.getSeparatorString())) == 0;
543 }
544 
545 AsmToken AsmLexer::LexToken() {
546   TokStart = CurPtr;
547   // This always consumes at least one character.
548   int CurChar = getNextChar();
549 
550   if (!IsPeeking && CurChar == '#' && IsAtStartOfStatement) {
551     // If this starts with a '#', this may be a cpp
552     // hash directive and otherwise a line comment.
553     AsmToken TokenBuf[2];
554     MutableArrayRef<AsmToken> Buf(TokenBuf, 2);
555     size_t num = peekTokens(Buf, true);
556     // There cannot be a space preceeding this
557     if (IsAtStartOfLine && num == 2 && TokenBuf[0].is(AsmToken::Integer) &&
558         TokenBuf[1].is(AsmToken::String)) {
559       CurPtr = TokStart; // reset curPtr;
560       StringRef s = LexUntilEndOfLine();
561       UnLex(TokenBuf[1]);
562       UnLex(TokenBuf[0]);
563       return AsmToken(AsmToken::HashDirective, s);
564     }
565     return LexLineComment();
566   }
567 
568   if (isAtStartOfComment(TokStart))
569     return LexLineComment();
570 
571   if (isAtStatementSeparator(TokStart)) {
572     CurPtr += strlen(MAI.getSeparatorString()) - 1;
573     IsAtStartOfLine = true;
574     IsAtStartOfStatement = true;
575     return AsmToken(AsmToken::EndOfStatement,
576                     StringRef(TokStart, strlen(MAI.getSeparatorString())));
577   }
578 
579   // If we're missing a newline at EOF, make sure we still get an
580   // EndOfStatement token before the Eof token.
581   if (CurChar == EOF && !IsAtStartOfStatement) {
582     IsAtStartOfLine = true;
583     IsAtStartOfStatement = true;
584     return AsmToken(AsmToken::EndOfStatement, StringRef(TokStart, 1));
585   }
586   IsAtStartOfLine = false;
587   bool OldIsAtStartOfStatement = IsAtStartOfStatement;
588   IsAtStartOfStatement = false;
589   switch (CurChar) {
590   default:
591     // Handle identifier: [a-zA-Z_.][a-zA-Z0-9_$.@]*
592     if (isalpha(CurChar) || CurChar == '_' || CurChar == '.')
593       return LexIdentifier();
594 
595     // Unknown character, emit an error.
596     return ReturnError(TokStart, "invalid character in input");
597   case EOF:
598     IsAtStartOfLine = true;
599     IsAtStartOfStatement = true;
600     return AsmToken(AsmToken::Eof, StringRef(TokStart, 0));
601   case 0:
602   case ' ':
603   case '\t':
604     IsAtStartOfStatement = OldIsAtStartOfStatement;
605     while (*CurPtr == ' ' || *CurPtr == '\t')
606       CurPtr++;
607     if (SkipSpace)
608       return LexToken(); // Ignore whitespace.
609     else
610       return AsmToken(AsmToken::Space, StringRef(TokStart, CurPtr - TokStart));
611   case '\r': {
612     IsAtStartOfLine = true;
613     IsAtStartOfStatement = true;
614     // If this is a CR followed by LF, treat that as one token.
615     if (CurPtr != CurBuf.end() && *CurPtr == '\n')
616       ++CurPtr;
617     return AsmToken(AsmToken::EndOfStatement,
618                     StringRef(TokStart, CurPtr - TokStart));
619   }
620   case '\n':
621     IsAtStartOfLine = true;
622     IsAtStartOfStatement = true;
623     return AsmToken(AsmToken::EndOfStatement, StringRef(TokStart, 1));
624   case ':': return AsmToken(AsmToken::Colon, StringRef(TokStart, 1));
625   case '+': return AsmToken(AsmToken::Plus, StringRef(TokStart, 1));
626   case '-': return AsmToken(AsmToken::Minus, StringRef(TokStart, 1));
627   case '~': return AsmToken(AsmToken::Tilde, StringRef(TokStart, 1));
628   case '(': return AsmToken(AsmToken::LParen, StringRef(TokStart, 1));
629   case ')': return AsmToken(AsmToken::RParen, StringRef(TokStart, 1));
630   case '[': return AsmToken(AsmToken::LBrac, StringRef(TokStart, 1));
631   case ']': return AsmToken(AsmToken::RBrac, StringRef(TokStart, 1));
632   case '{': return AsmToken(AsmToken::LCurly, StringRef(TokStart, 1));
633   case '}': return AsmToken(AsmToken::RCurly, StringRef(TokStart, 1));
634   case '*': return AsmToken(AsmToken::Star, StringRef(TokStart, 1));
635   case ',': return AsmToken(AsmToken::Comma, StringRef(TokStart, 1));
636   case '$': return AsmToken(AsmToken::Dollar, StringRef(TokStart, 1));
637   case '@': return AsmToken(AsmToken::At, StringRef(TokStart, 1));
638   case '\\': return AsmToken(AsmToken::BackSlash, StringRef(TokStart, 1));
639   case '=':
640     if (*CurPtr == '=') {
641       ++CurPtr;
642       return AsmToken(AsmToken::EqualEqual, StringRef(TokStart, 2));
643     }
644     return AsmToken(AsmToken::Equal, StringRef(TokStart, 1));
645   case '|':
646     if (*CurPtr == '|') {
647       ++CurPtr;
648       return AsmToken(AsmToken::PipePipe, StringRef(TokStart, 2));
649     }
650     return AsmToken(AsmToken::Pipe, StringRef(TokStart, 1));
651   case '^': return AsmToken(AsmToken::Caret, StringRef(TokStart, 1));
652   case '&':
653     if (*CurPtr == '&') {
654       ++CurPtr;
655       return AsmToken(AsmToken::AmpAmp, StringRef(TokStart, 2));
656     }
657     return AsmToken(AsmToken::Amp, StringRef(TokStart, 1));
658   case '!':
659     if (*CurPtr == '=') {
660       ++CurPtr;
661       return AsmToken(AsmToken::ExclaimEqual, StringRef(TokStart, 2));
662     }
663     return AsmToken(AsmToken::Exclaim, StringRef(TokStart, 1));
664   case '%':
665     if (MAI.hasMipsExpressions()) {
666       AsmToken::TokenKind Operator;
667       unsigned OperatorLength;
668 
669       std::tie(Operator, OperatorLength) =
670           StringSwitch<std::pair<AsmToken::TokenKind, unsigned>>(
671               StringRef(CurPtr))
672               .StartsWith("call16", {AsmToken::PercentCall16, 7})
673               .StartsWith("call_hi", {AsmToken::PercentCall_Hi, 8})
674               .StartsWith("call_lo", {AsmToken::PercentCall_Lo, 8})
675               .StartsWith("dtprel_hi", {AsmToken::PercentDtprel_Hi, 10})
676               .StartsWith("dtprel_lo", {AsmToken::PercentDtprel_Lo, 10})
677               .StartsWith("got_disp", {AsmToken::PercentGot_Disp, 9})
678               .StartsWith("got_hi", {AsmToken::PercentGot_Hi, 7})
679               .StartsWith("got_lo", {AsmToken::PercentGot_Lo, 7})
680               .StartsWith("got_ofst", {AsmToken::PercentGot_Ofst, 9})
681               .StartsWith("got_page", {AsmToken::PercentGot_Page, 9})
682               .StartsWith("gottprel", {AsmToken::PercentGottprel, 9})
683               .StartsWith("got", {AsmToken::PercentGot, 4})
684               .StartsWith("gp_rel", {AsmToken::PercentGp_Rel, 7})
685               .StartsWith("higher", {AsmToken::PercentHigher, 7})
686               .StartsWith("highest", {AsmToken::PercentHighest, 8})
687               .StartsWith("hi", {AsmToken::PercentHi, 3})
688               .StartsWith("lo", {AsmToken::PercentLo, 3})
689               .StartsWith("neg", {AsmToken::PercentNeg, 4})
690               .StartsWith("pcrel_hi", {AsmToken::PercentPcrel_Hi, 9})
691               .StartsWith("pcrel_lo", {AsmToken::PercentPcrel_Lo, 9})
692               .StartsWith("tlsgd", {AsmToken::PercentTlsgd, 6})
693               .StartsWith("tlsldm", {AsmToken::PercentTlsldm, 7})
694               .StartsWith("tprel_hi", {AsmToken::PercentTprel_Hi, 9})
695               .StartsWith("tprel_lo", {AsmToken::PercentTprel_Lo, 9})
696               .Default({AsmToken::Percent, 1});
697 
698       if (Operator != AsmToken::Percent) {
699         CurPtr += OperatorLength - 1;
700         return AsmToken(Operator, StringRef(TokStart, OperatorLength));
701       }
702     }
703     return AsmToken(AsmToken::Percent, StringRef(TokStart, 1));
704   case '/':
705     IsAtStartOfStatement = OldIsAtStartOfStatement;
706     return LexSlash();
707   case '#': return AsmToken(AsmToken::Hash, StringRef(TokStart, 1));
708   case '\'': return LexSingleQuote();
709   case '"': return LexQuote();
710   case '0': case '1': case '2': case '3': case '4':
711   case '5': case '6': case '7': case '8': case '9':
712     return LexDigit();
713   case '<':
714     switch (*CurPtr) {
715     case '<':
716       ++CurPtr;
717       return AsmToken(AsmToken::LessLess, StringRef(TokStart, 2));
718     case '=':
719       ++CurPtr;
720       return AsmToken(AsmToken::LessEqual, StringRef(TokStart, 2));
721     case '>':
722       ++CurPtr;
723       return AsmToken(AsmToken::LessGreater, StringRef(TokStart, 2));
724     default:
725       return AsmToken(AsmToken::Less, StringRef(TokStart, 1));
726     }
727   case '>':
728     switch (*CurPtr) {
729     case '>':
730       ++CurPtr;
731       return AsmToken(AsmToken::GreaterGreater, StringRef(TokStart, 2));
732     case '=':
733       ++CurPtr;
734       return AsmToken(AsmToken::GreaterEqual, StringRef(TokStart, 2));
735     default:
736       return AsmToken(AsmToken::Greater, StringRef(TokStart, 1));
737     }
738 
739   // TODO: Quoted identifiers (objc methods etc)
740   // local labels: [0-9][:]
741   // Forward/backward labels: [0-9][fb]
742   // Integers, fp constants, character constants.
743   }
744 }
745