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