1 //===- AsmLexer.cpp - Lexer for Assembly Files ----------------------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // This class implements the lexer for assembly files.
10 //
11 //===----------------------------------------------------------------------===//
12 
13 #include "llvm/MC/MCParser/AsmLexer.h"
14 #include "llvm/ADT/APInt.h"
15 #include "llvm/ADT/ArrayRef.h"
16 #include "llvm/ADT/StringExtras.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 /// The leading integral digit sequence and dot should have already been
65 /// consumed, some or all of the fractional digit sequence *can* have been
66 /// consumed.
67 AsmToken AsmLexer::LexFloatLiteral() {
68   // Skip the fractional digit sequence.
69   while (isDigit(*CurPtr))
70     ++CurPtr;
71 
72   if (*CurPtr == '-' || *CurPtr == '+')
73     return ReturnError(CurPtr, "Invalid sign in float literal");
74 
75   // Check for exponent
76   if ((*CurPtr == 'e' || *CurPtr == 'E')) {
77     ++CurPtr;
78 
79     if (*CurPtr == '-' || *CurPtr == '+')
80       ++CurPtr;
81 
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 
150     if (!IsIdentifierChar(*CurPtr, AllowAtInIdentifier) ||
151         *CurPtr == 'e' || *CurPtr == 'E')
152       return LexFloatLiteral();
153   }
154 
155   while (IsIdentifierChar(*CurPtr, AllowAtInIdentifier))
156     ++CurPtr;
157 
158   // Handle . as a special case.
159   if (CurPtr == TokStart+1 && TokStart[0] == '.')
160     return AsmToken(AsmToken::Dot, StringRef(TokStart, 1));
161 
162   return AsmToken(AsmToken::Identifier, StringRef(TokStart, CurPtr - TokStart));
163 }
164 
165 /// LexSlash: Slash: /
166 ///           C-Style Comment: /* ... */
167 AsmToken AsmLexer::LexSlash() {
168   switch (*CurPtr) {
169   case '*':
170     IsAtStartOfStatement = false;
171     break; // C style comment.
172   case '/':
173     ++CurPtr;
174     return LexLineComment();
175   default:
176     IsAtStartOfStatement = false;
177     return AsmToken(AsmToken::Slash, StringRef(TokStart, 1));
178   }
179 
180   // C Style comment.
181   ++CurPtr;  // skip the star.
182   const char *CommentTextStart = CurPtr;
183   while (CurPtr != CurBuf.end()) {
184     switch (*CurPtr++) {
185     case '*':
186       // End of the comment?
187       if (*CurPtr != '/')
188         break;
189       // If we have a CommentConsumer, notify it about the comment.
190       if (CommentConsumer) {
191         CommentConsumer->HandleComment(
192             SMLoc::getFromPointer(CommentTextStart),
193             StringRef(CommentTextStart, CurPtr - 1 - CommentTextStart));
194       }
195       ++CurPtr;   // End the */.
196       return AsmToken(AsmToken::Comment,
197                       StringRef(TokStart, CurPtr - TokStart));
198     }
199   }
200   return ReturnError(TokStart, "unterminated comment");
201 }
202 
203 /// LexLineComment: Comment: #[^\n]*
204 ///                        : //[^\n]*
205 AsmToken AsmLexer::LexLineComment() {
206   // Mark This as an end of statement with a body of the
207   // comment. While it would be nicer to leave this two tokens,
208   // backwards compatability with TargetParsers makes keeping this in this form
209   // better.
210   const char *CommentTextStart = CurPtr;
211   int CurChar = getNextChar();
212   while (CurChar != '\n' && CurChar != '\r' && CurChar != EOF)
213     CurChar = getNextChar();
214   if (CurChar == '\r' && CurPtr != CurBuf.end() && *CurPtr == '\n')
215     ++CurPtr;
216 
217   // If we have a CommentConsumer, notify it about the comment.
218   if (CommentConsumer) {
219     CommentConsumer->HandleComment(
220         SMLoc::getFromPointer(CommentTextStart),
221         StringRef(CommentTextStart, CurPtr - 1 - CommentTextStart));
222   }
223 
224   IsAtStartOfLine = true;
225   // This is a whole line comment. leave newline
226   if (IsAtStartOfStatement)
227     return AsmToken(AsmToken::EndOfStatement,
228                     StringRef(TokStart, CurPtr - TokStart));
229   IsAtStartOfStatement = true;
230 
231   return AsmToken(AsmToken::EndOfStatement,
232                   StringRef(TokStart, CurPtr - 1 - TokStart));
233 }
234 
235 static void SkipIgnoredIntegerSuffix(const char *&CurPtr) {
236   // Skip ULL, UL, U, L and LL suffices.
237   if (CurPtr[0] == 'U')
238     ++CurPtr;
239   if (CurPtr[0] == 'L')
240     ++CurPtr;
241   if (CurPtr[0] == 'L')
242     ++CurPtr;
243 }
244 
245 // Look ahead to search for first non-hex digit, if it's [hH], then we treat the
246 // integer as a hexadecimal, possibly with leading zeroes.
247 static unsigned doHexLookAhead(const char *&CurPtr, unsigned DefaultRadix,
248                                bool LexHex) {
249   const char *FirstNonDec = nullptr;
250   const char *LookAhead = CurPtr;
251   while (true) {
252     if (isDigit(*LookAhead)) {
253       ++LookAhead;
254     } else {
255       if (!FirstNonDec)
256         FirstNonDec = LookAhead;
257 
258       // Keep going if we are looking for a 'h' suffix.
259       if (LexHex && isHexDigit(*LookAhead))
260         ++LookAhead;
261       else
262         break;
263     }
264   }
265   bool isHex = LexHex && (*LookAhead == 'h' || *LookAhead == 'H');
266   CurPtr = isHex || !FirstNonDec ? LookAhead : FirstNonDec;
267   if (isHex)
268     return 16;
269   return DefaultRadix;
270 }
271 
272 static AsmToken intToken(StringRef Ref, APInt &Value)
273 {
274   if (Value.isIntN(64))
275     return AsmToken(AsmToken::Integer, Ref, Value);
276   return AsmToken(AsmToken::BigNum, Ref, Value);
277 }
278 
279 /// LexDigit: First character is [0-9].
280 ///   Local Label: [0-9][:]
281 ///   Forward/Backward Label: [0-9][fb]
282 ///   Binary integer: 0b[01]+
283 ///   Octal integer: 0[0-7]+
284 ///   Hex integer: 0x[0-9a-fA-F]+ or [0x]?[0-9][0-9a-fA-F]*[hH]
285 ///   Decimal integer: [1-9][0-9]*
286 AsmToken AsmLexer::LexDigit() {
287   // MASM-flavor binary integer: [01]+[bB]
288   // MASM-flavor hexadecimal integer: [0-9][0-9a-fA-F]*[hH]
289   if (LexMasmIntegers && isdigit(CurPtr[-1])) {
290     const char *FirstNonBinary = (CurPtr[-1] != '0' && CurPtr[-1] != '1') ?
291                                    CurPtr - 1 : nullptr;
292     const char *OldCurPtr = CurPtr;
293     while (isHexDigit(*CurPtr)) {
294       if (*CurPtr != '0' && *CurPtr != '1' && !FirstNonBinary)
295         FirstNonBinary = CurPtr;
296       ++CurPtr;
297     }
298 
299     unsigned Radix = 0;
300     if (*CurPtr == 'h' || *CurPtr == 'H') {
301       // hexadecimal number
302       ++CurPtr;
303       Radix = 16;
304     } else if (FirstNonBinary && FirstNonBinary + 1 == CurPtr &&
305                (*FirstNonBinary == 'b' || *FirstNonBinary == 'B'))
306       Radix = 2;
307 
308     if (Radix == 2 || Radix == 16) {
309       StringRef Result(TokStart, CurPtr - TokStart);
310       APInt Value(128, 0, true);
311 
312       if (Result.drop_back().getAsInteger(Radix, Value))
313         return ReturnError(TokStart, Radix == 2 ? "invalid binary number" :
314                              "invalid hexdecimal number");
315 
316       // MSVC accepts and ignores type suffices on integer literals.
317       SkipIgnoredIntegerSuffix(CurPtr);
318 
319       return intToken(Result, Value);
320    }
321 
322     // octal/decimal integers, or floating point numbers, fall through
323     CurPtr = OldCurPtr;
324   }
325 
326   // Decimal integer: [1-9][0-9]*
327   if (CurPtr[-1] != '0' || CurPtr[0] == '.') {
328     unsigned Radix = doHexLookAhead(CurPtr, 10, LexMasmIntegers);
329     bool isHex = Radix == 16;
330     // Check for floating point literals.
331     if (!isHex && (*CurPtr == '.' || *CurPtr == 'e' || *CurPtr == 'E')) {
332       if (*CurPtr == '.')
333         ++CurPtr;
334       return LexFloatLiteral();
335     }
336 
337     StringRef Result(TokStart, CurPtr - TokStart);
338 
339     APInt Value(128, 0, true);
340     if (Result.getAsInteger(Radix, Value))
341       return ReturnError(TokStart, !isHex ? "invalid decimal number" :
342                            "invalid hexdecimal number");
343 
344     // Consume the [hH].
345     if (LexMasmIntegers && Radix == 16)
346       ++CurPtr;
347 
348     // The darwin/x86 (and x86-64) assembler accepts and ignores type
349     // suffices on integer literals.
350     SkipIgnoredIntegerSuffix(CurPtr);
351 
352     return intToken(Result, Value);
353   }
354 
355   if (!LexMasmIntegers && ((*CurPtr == 'b') || (*CurPtr == 'B'))) {
356     ++CurPtr;
357     // See if we actually have "0b" as part of something like "jmp 0b\n"
358     if (!isDigit(CurPtr[0])) {
359       --CurPtr;
360       StringRef Result(TokStart, CurPtr - TokStart);
361       return AsmToken(AsmToken::Integer, Result, 0);
362     }
363     const char *NumStart = CurPtr;
364     while (CurPtr[0] == '0' || CurPtr[0] == '1')
365       ++CurPtr;
366 
367     // Requires at least one binary digit.
368     if (CurPtr == NumStart)
369       return ReturnError(TokStart, "invalid binary number");
370 
371     StringRef Result(TokStart, CurPtr - TokStart);
372 
373     APInt Value(128, 0, true);
374     if (Result.substr(2).getAsInteger(2, Value))
375       return ReturnError(TokStart, "invalid binary number");
376 
377     // The darwin/x86 (and x86-64) assembler accepts and ignores ULL and LL
378     // suffixes on integer literals.
379     SkipIgnoredIntegerSuffix(CurPtr);
380 
381     return intToken(Result, Value);
382   }
383 
384   if ((*CurPtr == 'x') || (*CurPtr == 'X')) {
385     ++CurPtr;
386     const char *NumStart = CurPtr;
387     while (isHexDigit(CurPtr[0]))
388       ++CurPtr;
389 
390     // "0x.0p0" is valid, and "0x0p0" (but not "0xp0" for example, which will be
391     // diagnosed by LexHexFloatLiteral).
392     if (CurPtr[0] == '.' || CurPtr[0] == 'p' || CurPtr[0] == 'P')
393       return LexHexFloatLiteral(NumStart == CurPtr);
394 
395     // Otherwise requires at least one hex digit.
396     if (CurPtr == NumStart)
397       return ReturnError(CurPtr-2, "invalid hexadecimal number");
398 
399     APInt Result(128, 0);
400     if (StringRef(TokStart, CurPtr - TokStart).getAsInteger(0, Result))
401       return ReturnError(TokStart, "invalid hexadecimal number");
402 
403     // Consume the optional [hH].
404     if (LexMasmIntegers && (*CurPtr == 'h' || *CurPtr == 'H'))
405       ++CurPtr;
406 
407     // The darwin/x86 (and x86-64) assembler accepts and ignores ULL and LL
408     // suffixes on integer literals.
409     SkipIgnoredIntegerSuffix(CurPtr);
410 
411     return intToken(StringRef(TokStart, CurPtr - TokStart), Result);
412   }
413 
414   // Either octal or hexadecimal.
415   APInt Value(128, 0, true);
416   unsigned Radix = doHexLookAhead(CurPtr, 8, LexMasmIntegers);
417   bool isHex = Radix == 16;
418   StringRef Result(TokStart, CurPtr - TokStart);
419   if (Result.getAsInteger(Radix, Value))
420     return ReturnError(TokStart, !isHex ? "invalid octal number" :
421                        "invalid hexdecimal number");
422 
423   // Consume the [hH].
424   if (Radix == 16)
425     ++CurPtr;
426 
427   // The darwin/x86 (and x86-64) assembler accepts and ignores ULL and LL
428   // suffixes on integer literals.
429   SkipIgnoredIntegerSuffix(CurPtr);
430 
431   return intToken(Result, Value);
432 }
433 
434 /// LexSingleQuote: Integer: 'b'
435 AsmToken AsmLexer::LexSingleQuote() {
436   int CurChar = getNextChar();
437 
438   if (CurChar == '\\')
439     CurChar = getNextChar();
440 
441   if (CurChar == EOF)
442     return ReturnError(TokStart, "unterminated single quote");
443 
444   CurChar = getNextChar();
445 
446   if (CurChar != '\'')
447     return ReturnError(TokStart, "single quote way too long");
448 
449   // The idea here being that 'c' is basically just an integral
450   // constant.
451   StringRef Res = StringRef(TokStart,CurPtr - TokStart);
452   long long Value;
453 
454   if (Res.startswith("\'\\")) {
455     char theChar = Res[2];
456     switch (theChar) {
457       default: Value = theChar; break;
458       case '\'': Value = '\''; break;
459       case 't': Value = '\t'; break;
460       case 'n': Value = '\n'; break;
461       case 'b': Value = '\b'; break;
462     }
463   } else
464     Value = TokStart[1];
465 
466   return AsmToken(AsmToken::Integer, Res, Value);
467 }
468 
469 /// LexQuote: String: "..."
470 AsmToken AsmLexer::LexQuote() {
471   int CurChar = getNextChar();
472   // TODO: does gas allow multiline string constants?
473   while (CurChar != '"') {
474     if (CurChar == '\\') {
475       // Allow \", etc.
476       CurChar = getNextChar();
477     }
478 
479     if (CurChar == EOF)
480       return ReturnError(TokStart, "unterminated string constant");
481 
482     CurChar = getNextChar();
483   }
484 
485   return AsmToken(AsmToken::String, StringRef(TokStart, CurPtr - TokStart));
486 }
487 
488 StringRef AsmLexer::LexUntilEndOfStatement() {
489   TokStart = CurPtr;
490 
491   while (!isAtStartOfComment(CurPtr) &&     // Start of line comment.
492          !isAtStatementSeparator(CurPtr) && // End of statement marker.
493          *CurPtr != '\n' && *CurPtr != '\r' && CurPtr != CurBuf.end()) {
494     ++CurPtr;
495   }
496   return StringRef(TokStart, CurPtr-TokStart);
497 }
498 
499 StringRef AsmLexer::LexUntilEndOfLine() {
500   TokStart = CurPtr;
501 
502   while (*CurPtr != '\n' && *CurPtr != '\r' && CurPtr != CurBuf.end()) {
503     ++CurPtr;
504   }
505   return StringRef(TokStart, CurPtr-TokStart);
506 }
507 
508 size_t AsmLexer::peekTokens(MutableArrayRef<AsmToken> Buf,
509                             bool ShouldSkipSpace) {
510   SaveAndRestore<const char *> SavedTokenStart(TokStart);
511   SaveAndRestore<const char *> SavedCurPtr(CurPtr);
512   SaveAndRestore<bool> SavedAtStartOfLine(IsAtStartOfLine);
513   SaveAndRestore<bool> SavedAtStartOfStatement(IsAtStartOfStatement);
514   SaveAndRestore<bool> SavedSkipSpace(SkipSpace, ShouldSkipSpace);
515   SaveAndRestore<bool> SavedIsPeeking(IsPeeking, true);
516   std::string SavedErr = getErr();
517   SMLoc SavedErrLoc = getErrLoc();
518 
519   size_t ReadCount;
520   for (ReadCount = 0; ReadCount < Buf.size(); ++ReadCount) {
521     AsmToken Token = LexToken();
522 
523     Buf[ReadCount] = Token;
524 
525     if (Token.is(AsmToken::Eof))
526       break;
527   }
528 
529   SetError(SavedErrLoc, SavedErr);
530   return ReadCount;
531 }
532 
533 bool AsmLexer::isAtStartOfComment(const char *Ptr) {
534   StringRef CommentString = MAI.getCommentString();
535 
536   if (CommentString.size() == 1)
537     return CommentString[0] == Ptr[0];
538 
539   // Allow # preprocessor commments also be counted as comments for "##" cases
540   if (CommentString[1] == '#')
541     return CommentString[0] == Ptr[0];
542 
543   return strncmp(Ptr, CommentString.data(), CommentString.size()) == 0;
544 }
545 
546 bool AsmLexer::isAtStatementSeparator(const char *Ptr) {
547   return strncmp(Ptr, MAI.getSeparatorString(),
548                  strlen(MAI.getSeparatorString())) == 0;
549 }
550 
551 AsmToken AsmLexer::LexToken() {
552   TokStart = CurPtr;
553   // This always consumes at least one character.
554   int CurChar = getNextChar();
555 
556   if (!IsPeeking && CurChar == '#' && IsAtStartOfStatement) {
557     // If this starts with a '#', this may be a cpp
558     // hash directive and otherwise a line comment.
559     AsmToken TokenBuf[2];
560     MutableArrayRef<AsmToken> Buf(TokenBuf, 2);
561     size_t num = peekTokens(Buf, true);
562     // There cannot be a space preceding this
563     if (IsAtStartOfLine && num == 2 && TokenBuf[0].is(AsmToken::Integer) &&
564         TokenBuf[1].is(AsmToken::String)) {
565       CurPtr = TokStart; // reset curPtr;
566       StringRef s = LexUntilEndOfLine();
567       UnLex(TokenBuf[1]);
568       UnLex(TokenBuf[0]);
569       return AsmToken(AsmToken::HashDirective, s);
570     }
571     return LexLineComment();
572   }
573 
574   if (isAtStartOfComment(TokStart))
575     return LexLineComment();
576 
577   if (isAtStatementSeparator(TokStart)) {
578     CurPtr += strlen(MAI.getSeparatorString()) - 1;
579     IsAtStartOfLine = true;
580     IsAtStartOfStatement = true;
581     return AsmToken(AsmToken::EndOfStatement,
582                     StringRef(TokStart, strlen(MAI.getSeparatorString())));
583   }
584 
585   // If we're missing a newline at EOF, make sure we still get an
586   // EndOfStatement token before the Eof token.
587   if (CurChar == EOF && !IsAtStartOfStatement) {
588     IsAtStartOfLine = true;
589     IsAtStartOfStatement = true;
590     return AsmToken(AsmToken::EndOfStatement, StringRef(TokStart, 1));
591   }
592   IsAtStartOfLine = false;
593   bool OldIsAtStartOfStatement = IsAtStartOfStatement;
594   IsAtStartOfStatement = false;
595   switch (CurChar) {
596   default:
597     // Handle identifier: [a-zA-Z_.][a-zA-Z0-9_$.@]*
598     if (isalpha(CurChar) || CurChar == '_' || CurChar == '.')
599       return LexIdentifier();
600 
601     // Unknown character, emit an error.
602     return ReturnError(TokStart, "invalid character in input");
603   case EOF:
604     IsAtStartOfLine = true;
605     IsAtStartOfStatement = true;
606     return AsmToken(AsmToken::Eof, StringRef(TokStart, 0));
607   case 0:
608   case ' ':
609   case '\t':
610     IsAtStartOfStatement = OldIsAtStartOfStatement;
611     while (*CurPtr == ' ' || *CurPtr == '\t')
612       CurPtr++;
613     if (SkipSpace)
614       return LexToken(); // Ignore whitespace.
615     else
616       return AsmToken(AsmToken::Space, StringRef(TokStart, CurPtr - TokStart));
617   case '\r': {
618     IsAtStartOfLine = true;
619     IsAtStartOfStatement = true;
620     // If this is a CR followed by LF, treat that as one token.
621     if (CurPtr != CurBuf.end() && *CurPtr == '\n')
622       ++CurPtr;
623     return AsmToken(AsmToken::EndOfStatement,
624                     StringRef(TokStart, CurPtr - TokStart));
625   }
626   case '\n':
627     IsAtStartOfLine = true;
628     IsAtStartOfStatement = true;
629     return AsmToken(AsmToken::EndOfStatement, StringRef(TokStart, 1));
630   case ':': return AsmToken(AsmToken::Colon, StringRef(TokStart, 1));
631   case '+': return AsmToken(AsmToken::Plus, StringRef(TokStart, 1));
632   case '~': return AsmToken(AsmToken::Tilde, StringRef(TokStart, 1));
633   case '(': return AsmToken(AsmToken::LParen, StringRef(TokStart, 1));
634   case ')': return AsmToken(AsmToken::RParen, StringRef(TokStart, 1));
635   case '[': return AsmToken(AsmToken::LBrac, StringRef(TokStart, 1));
636   case ']': return AsmToken(AsmToken::RBrac, StringRef(TokStart, 1));
637   case '{': return AsmToken(AsmToken::LCurly, StringRef(TokStart, 1));
638   case '}': return AsmToken(AsmToken::RCurly, StringRef(TokStart, 1));
639   case '*': return AsmToken(AsmToken::Star, StringRef(TokStart, 1));
640   case ',': return AsmToken(AsmToken::Comma, StringRef(TokStart, 1));
641   case '$': return AsmToken(AsmToken::Dollar, StringRef(TokStart, 1));
642   case '@': return AsmToken(AsmToken::At, StringRef(TokStart, 1));
643   case '\\': return AsmToken(AsmToken::BackSlash, StringRef(TokStart, 1));
644   case '=':
645     if (*CurPtr == '=') {
646       ++CurPtr;
647       return AsmToken(AsmToken::EqualEqual, StringRef(TokStart, 2));
648     }
649     return AsmToken(AsmToken::Equal, StringRef(TokStart, 1));
650   case '-':
651     if (*CurPtr == '>') {
652       ++CurPtr;
653       return AsmToken(AsmToken::MinusGreater, StringRef(TokStart, 2));
654     }
655     return AsmToken(AsmToken::Minus, StringRef(TokStart, 1));
656   case '|':
657     if (*CurPtr == '|') {
658       ++CurPtr;
659       return AsmToken(AsmToken::PipePipe, StringRef(TokStart, 2));
660     }
661     return AsmToken(AsmToken::Pipe, StringRef(TokStart, 1));
662   case '^': return AsmToken(AsmToken::Caret, StringRef(TokStart, 1));
663   case '&':
664     if (*CurPtr == '&') {
665       ++CurPtr;
666       return AsmToken(AsmToken::AmpAmp, StringRef(TokStart, 2));
667     }
668     return AsmToken(AsmToken::Amp, StringRef(TokStart, 1));
669   case '!':
670     if (*CurPtr == '=') {
671       ++CurPtr;
672       return AsmToken(AsmToken::ExclaimEqual, StringRef(TokStart, 2));
673     }
674     return AsmToken(AsmToken::Exclaim, StringRef(TokStart, 1));
675   case '%':
676     if (MAI.hasMipsExpressions()) {
677       AsmToken::TokenKind Operator;
678       unsigned OperatorLength;
679 
680       std::tie(Operator, OperatorLength) =
681           StringSwitch<std::pair<AsmToken::TokenKind, unsigned>>(
682               StringRef(CurPtr))
683               .StartsWith("call16", {AsmToken::PercentCall16, 7})
684               .StartsWith("call_hi", {AsmToken::PercentCall_Hi, 8})
685               .StartsWith("call_lo", {AsmToken::PercentCall_Lo, 8})
686               .StartsWith("dtprel_hi", {AsmToken::PercentDtprel_Hi, 10})
687               .StartsWith("dtprel_lo", {AsmToken::PercentDtprel_Lo, 10})
688               .StartsWith("got_disp", {AsmToken::PercentGot_Disp, 9})
689               .StartsWith("got_hi", {AsmToken::PercentGot_Hi, 7})
690               .StartsWith("got_lo", {AsmToken::PercentGot_Lo, 7})
691               .StartsWith("got_ofst", {AsmToken::PercentGot_Ofst, 9})
692               .StartsWith("got_page", {AsmToken::PercentGot_Page, 9})
693               .StartsWith("gottprel", {AsmToken::PercentGottprel, 9})
694               .StartsWith("got", {AsmToken::PercentGot, 4})
695               .StartsWith("gp_rel", {AsmToken::PercentGp_Rel, 7})
696               .StartsWith("higher", {AsmToken::PercentHigher, 7})
697               .StartsWith("highest", {AsmToken::PercentHighest, 8})
698               .StartsWith("hi", {AsmToken::PercentHi, 3})
699               .StartsWith("lo", {AsmToken::PercentLo, 3})
700               .StartsWith("neg", {AsmToken::PercentNeg, 4})
701               .StartsWith("pcrel_hi", {AsmToken::PercentPcrel_Hi, 9})
702               .StartsWith("pcrel_lo", {AsmToken::PercentPcrel_Lo, 9})
703               .StartsWith("tlsgd", {AsmToken::PercentTlsgd, 6})
704               .StartsWith("tlsldm", {AsmToken::PercentTlsldm, 7})
705               .StartsWith("tprel_hi", {AsmToken::PercentTprel_Hi, 9})
706               .StartsWith("tprel_lo", {AsmToken::PercentTprel_Lo, 9})
707               .Default({AsmToken::Percent, 1});
708 
709       if (Operator != AsmToken::Percent) {
710         CurPtr += OperatorLength - 1;
711         return AsmToken(Operator, StringRef(TokStart, OperatorLength));
712       }
713     }
714     return AsmToken(AsmToken::Percent, StringRef(TokStart, 1));
715   case '/':
716     IsAtStartOfStatement = OldIsAtStartOfStatement;
717     return LexSlash();
718   case '#': return AsmToken(AsmToken::Hash, StringRef(TokStart, 1));
719   case '\'': return LexSingleQuote();
720   case '"': return LexQuote();
721   case '0': case '1': case '2': case '3': case '4':
722   case '5': case '6': case '7': case '8': case '9':
723     return LexDigit();
724   case '<':
725     switch (*CurPtr) {
726     case '<':
727       ++CurPtr;
728       return AsmToken(AsmToken::LessLess, StringRef(TokStart, 2));
729     case '=':
730       ++CurPtr;
731       return AsmToken(AsmToken::LessEqual, StringRef(TokStart, 2));
732     case '>':
733       ++CurPtr;
734       return AsmToken(AsmToken::LessGreater, StringRef(TokStart, 2));
735     default:
736       return AsmToken(AsmToken::Less, StringRef(TokStart, 1));
737     }
738   case '>':
739     switch (*CurPtr) {
740     case '>':
741       ++CurPtr;
742       return AsmToken(AsmToken::GreaterGreater, StringRef(TokStart, 2));
743     case '=':
744       ++CurPtr;
745       return AsmToken(AsmToken::GreaterEqual, StringRef(TokStart, 2));
746     default:
747       return AsmToken(AsmToken::Greater, StringRef(TokStart, 1));
748     }
749 
750   // TODO: Quoted identifiers (objc methods etc)
751   // local labels: [0-9][:]
752   // Forward/backward labels: [0-9][fb]
753   // Integers, fp constants, character constants.
754   }
755 }
756