xref: /freebsd/contrib/llvm-project/llvm/lib/MC/MCParser/AsmLexer.cpp (revision a521f2116473fbd8c09db395518f060a27d02334)
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                          bool EndStatementAtEOF) {
41   CurBuf = Buf;
42 
43   if (ptr)
44     CurPtr = ptr;
45   else
46     CurPtr = CurBuf.begin();
47 
48   TokStart = nullptr;
49   this->EndStatementAtEOF = EndStatementAtEOF;
50 }
51 
52 /// ReturnError - Set the error to the specified string at the specified
53 /// location.  This is defined to always return AsmToken::Error.
54 AsmToken AsmLexer::ReturnError(const char *Loc, const std::string &Msg) {
55   SetError(SMLoc::getFromPointer(Loc), Msg);
56 
57   return AsmToken(AsmToken::Error, StringRef(Loc, CurPtr - Loc));
58 }
59 
60 int AsmLexer::getNextChar() {
61   if (CurPtr == CurBuf.end())
62     return EOF;
63   return (unsigned char)*CurPtr++;
64 }
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   if (*CurPtr == '-' || *CurPtr == '+')
75     return ReturnError(CurPtr, "Invalid sign in float literal");
76 
77   // Check for exponent
78   if ((*CurPtr == 'e' || *CurPtr == 'E')) {
79     ++CurPtr;
80 
81     if (*CurPtr == '-' || *CurPtr == '+')
82       ++CurPtr;
83 
84     while (isDigit(*CurPtr))
85       ++CurPtr;
86   }
87 
88   return AsmToken(AsmToken::Real,
89                   StringRef(TokStart, CurPtr - TokStart));
90 }
91 
92 /// LexHexFloatLiteral matches essentially (.[0-9a-fA-F]*)?[pP][+-]?[0-9a-fA-F]+
93 /// while making sure there are enough actual digits around for the constant to
94 /// be valid.
95 ///
96 /// The leading "0x[0-9a-fA-F]*" (i.e. integer part) has already been consumed
97 /// before we get here.
98 AsmToken AsmLexer::LexHexFloatLiteral(bool NoIntDigits) {
99   assert((*CurPtr == 'p' || *CurPtr == 'P' || *CurPtr == '.') &&
100          "unexpected parse state in floating hex");
101   bool NoFracDigits = true;
102 
103   // Skip the fractional part if there is one
104   if (*CurPtr == '.') {
105     ++CurPtr;
106 
107     const char *FracStart = CurPtr;
108     while (isHexDigit(*CurPtr))
109       ++CurPtr;
110 
111     NoFracDigits = CurPtr == FracStart;
112   }
113 
114   if (NoIntDigits && NoFracDigits)
115     return ReturnError(TokStart, "invalid hexadecimal floating-point constant: "
116                                  "expected at least one significand digit");
117 
118   // Make sure we do have some kind of proper exponent part
119   if (*CurPtr != 'p' && *CurPtr != 'P')
120     return ReturnError(TokStart, "invalid hexadecimal floating-point constant: "
121                                  "expected exponent part 'p'");
122   ++CurPtr;
123 
124   if (*CurPtr == '+' || *CurPtr == '-')
125     ++CurPtr;
126 
127   // N.b. exponent digits are *not* hex
128   const char *ExpStart = CurPtr;
129   while (isDigit(*CurPtr))
130     ++CurPtr;
131 
132   if (CurPtr == ExpStart)
133     return ReturnError(TokStart, "invalid hexadecimal floating-point constant: "
134                                  "expected at least one exponent digit");
135 
136   return AsmToken(AsmToken::Real, StringRef(TokStart, CurPtr - TokStart));
137 }
138 
139 /// LexIdentifier: [a-zA-Z_.][a-zA-Z0-9_$.@?]*
140 static bool IsIdentifierChar(char c, bool AllowAt) {
141   return isAlnum(c) || c == '_' || c == '$' || c == '.' ||
142          (c == '@' && AllowAt) || c == '?';
143 }
144 
145 AsmToken AsmLexer::LexIdentifier() {
146   // Check for floating point literals.
147   if (CurPtr[-1] == '.' && isDigit(*CurPtr)) {
148     // Disambiguate a .1243foo identifier from a floating literal.
149     while (isDigit(*CurPtr))
150       ++CurPtr;
151 
152     if (!IsIdentifierChar(*CurPtr, AllowAtInIdentifier) ||
153         *CurPtr == 'e' || *CurPtr == 'E')
154       return LexFloatLiteral();
155   }
156 
157   while (IsIdentifierChar(*CurPtr, AllowAtInIdentifier))
158     ++CurPtr;
159 
160   // Handle . as a special case.
161   if (CurPtr == TokStart+1 && TokStart[0] == '.')
162     return AsmToken(AsmToken::Dot, StringRef(TokStart, 1));
163 
164   return AsmToken(AsmToken::Identifier, StringRef(TokStart, CurPtr - TokStart));
165 }
166 
167 /// LexSlash: Slash: /
168 ///           C-Style Comment: /* ... */
169 AsmToken AsmLexer::LexSlash() {
170   switch (*CurPtr) {
171   case '*':
172     IsAtStartOfStatement = false;
173     break; // C style comment.
174   case '/':
175     ++CurPtr;
176     return LexLineComment();
177   default:
178     IsAtStartOfStatement = false;
179     return AsmToken(AsmToken::Slash, StringRef(TokStart, 1));
180   }
181 
182   // C Style comment.
183   ++CurPtr;  // skip the star.
184   const char *CommentTextStart = CurPtr;
185   while (CurPtr != CurBuf.end()) {
186     switch (*CurPtr++) {
187     case '*':
188       // End of the comment?
189       if (*CurPtr != '/')
190         break;
191       // If we have a CommentConsumer, notify it about the comment.
192       if (CommentConsumer) {
193         CommentConsumer->HandleComment(
194             SMLoc::getFromPointer(CommentTextStart),
195             StringRef(CommentTextStart, CurPtr - 1 - CommentTextStart));
196       }
197       ++CurPtr;   // End the */.
198       return AsmToken(AsmToken::Comment,
199                       StringRef(TokStart, CurPtr - TokStart));
200     }
201   }
202   return ReturnError(TokStart, "unterminated comment");
203 }
204 
205 /// LexLineComment: Comment: #[^\n]*
206 ///                        : //[^\n]*
207 AsmToken AsmLexer::LexLineComment() {
208   // Mark This as an end of statement with a body of the
209   // comment. While it would be nicer to leave this two tokens,
210   // backwards compatability with TargetParsers makes keeping this in this form
211   // better.
212   const char *CommentTextStart = CurPtr;
213   int CurChar = getNextChar();
214   while (CurChar != '\n' && CurChar != '\r' && CurChar != EOF)
215     CurChar = getNextChar();
216   if (CurChar == '\r' && CurPtr != CurBuf.end() && *CurPtr == '\n')
217     ++CurPtr;
218 
219   // If we have a CommentConsumer, notify it about the comment.
220   if (CommentConsumer) {
221     CommentConsumer->HandleComment(
222         SMLoc::getFromPointer(CommentTextStart),
223         StringRef(CommentTextStart, CurPtr - 1 - CommentTextStart));
224   }
225 
226   IsAtStartOfLine = true;
227   // This is a whole line comment. leave newline
228   if (IsAtStartOfStatement)
229     return AsmToken(AsmToken::EndOfStatement,
230                     StringRef(TokStart, CurPtr - TokStart));
231   IsAtStartOfStatement = true;
232 
233   return AsmToken(AsmToken::EndOfStatement,
234                   StringRef(TokStart, CurPtr - 1 - TokStart));
235 }
236 
237 static void SkipIgnoredIntegerSuffix(const char *&CurPtr) {
238   // Skip ULL, UL, U, L and LL suffices.
239   if (CurPtr[0] == 'U')
240     ++CurPtr;
241   if (CurPtr[0] == 'L')
242     ++CurPtr;
243   if (CurPtr[0] == 'L')
244     ++CurPtr;
245 }
246 
247 // Look ahead to search for first non-hex digit, if it's [hH], then we treat the
248 // integer as a hexadecimal, possibly with leading zeroes.
249 static unsigned doHexLookAhead(const char *&CurPtr, unsigned DefaultRadix,
250                                bool LexHex) {
251   const char *FirstNonDec = nullptr;
252   const char *LookAhead = CurPtr;
253   while (true) {
254     if (isDigit(*LookAhead)) {
255       ++LookAhead;
256     } else {
257       if (!FirstNonDec)
258         FirstNonDec = LookAhead;
259 
260       // Keep going if we are looking for a 'h' suffix.
261       if (LexHex && isHexDigit(*LookAhead))
262         ++LookAhead;
263       else
264         break;
265     }
266   }
267   bool isHex = LexHex && (*LookAhead == 'h' || *LookAhead == 'H');
268   CurPtr = isHex || !FirstNonDec ? LookAhead : FirstNonDec;
269   if (isHex)
270     return 16;
271   return DefaultRadix;
272 }
273 
274 static AsmToken intToken(StringRef Ref, APInt &Value)
275 {
276   if (Value.isIntN(64))
277     return AsmToken(AsmToken::Integer, Ref, Value);
278   return AsmToken(AsmToken::BigNum, Ref, Value);
279 }
280 
281 /// LexDigit: First character is [0-9].
282 ///   Local Label: [0-9][:]
283 ///   Forward/Backward Label: [0-9][fb]
284 ///   Binary integer: 0b[01]+
285 ///   Octal integer: 0[0-7]+
286 ///   Hex integer: 0x[0-9a-fA-F]+ or [0x]?[0-9][0-9a-fA-F]*[hH]
287 ///   Decimal integer: [1-9][0-9]*
288 AsmToken AsmLexer::LexDigit() {
289   // MASM-flavor binary integer: [01]+[bB]
290   // MASM-flavor hexadecimal integer: [0-9][0-9a-fA-F]*[hH]
291   if (LexMasmIntegers && isdigit(CurPtr[-1])) {
292     const char *FirstNonBinary = (CurPtr[-1] != '0' && CurPtr[-1] != '1') ?
293                                    CurPtr - 1 : nullptr;
294     const char *OldCurPtr = CurPtr;
295     while (isHexDigit(*CurPtr)) {
296       if (*CurPtr != '0' && *CurPtr != '1' && !FirstNonBinary)
297         FirstNonBinary = CurPtr;
298       ++CurPtr;
299     }
300 
301     unsigned Radix = 0;
302     if (*CurPtr == 'h' || *CurPtr == 'H') {
303       // hexadecimal number
304       ++CurPtr;
305       Radix = 16;
306     } else if (FirstNonBinary && FirstNonBinary + 1 == CurPtr &&
307                (*FirstNonBinary == 'b' || *FirstNonBinary == 'B'))
308       Radix = 2;
309 
310     if (Radix == 2 || Radix == 16) {
311       StringRef Result(TokStart, CurPtr - TokStart);
312       APInt Value(128, 0, true);
313 
314       if (Result.drop_back().getAsInteger(Radix, Value))
315         return ReturnError(TokStart, Radix == 2 ? "invalid binary number" :
316                              "invalid hexdecimal number");
317 
318       // MSVC accepts and ignores type suffices on integer literals.
319       SkipIgnoredIntegerSuffix(CurPtr);
320 
321       return intToken(Result, Value);
322    }
323 
324     // octal/decimal integers, or floating point numbers, fall through
325     CurPtr = OldCurPtr;
326   }
327 
328   // Decimal integer: [1-9][0-9]*
329   if (CurPtr[-1] != '0' || CurPtr[0] == '.') {
330     unsigned Radix = doHexLookAhead(CurPtr, 10, LexMasmIntegers);
331     bool isHex = Radix == 16;
332     // Check for floating point literals.
333     if (!isHex && (*CurPtr == '.' || *CurPtr == 'e' || *CurPtr == 'E')) {
334       if (*CurPtr == '.')
335         ++CurPtr;
336       return LexFloatLiteral();
337     }
338 
339     StringRef Result(TokStart, CurPtr - TokStart);
340 
341     APInt Value(128, 0, true);
342     if (Result.getAsInteger(Radix, Value))
343       return ReturnError(TokStart, !isHex ? "invalid decimal number" :
344                            "invalid hexdecimal number");
345 
346     // Consume the [hH].
347     if (LexMasmIntegers && Radix == 16)
348       ++CurPtr;
349 
350     // The darwin/x86 (and x86-64) assembler accepts and ignores type
351     // suffices on integer literals.
352     SkipIgnoredIntegerSuffix(CurPtr);
353 
354     return intToken(Result, Value);
355   }
356 
357   if (!LexMasmIntegers && ((*CurPtr == 'b') || (*CurPtr == 'B'))) {
358     ++CurPtr;
359     // See if we actually have "0b" as part of something like "jmp 0b\n"
360     if (!isDigit(CurPtr[0])) {
361       --CurPtr;
362       StringRef Result(TokStart, CurPtr - TokStart);
363       return AsmToken(AsmToken::Integer, Result, 0);
364     }
365     const char *NumStart = CurPtr;
366     while (CurPtr[0] == '0' || CurPtr[0] == '1')
367       ++CurPtr;
368 
369     // Requires at least one binary digit.
370     if (CurPtr == NumStart)
371       return ReturnError(TokStart, "invalid binary number");
372 
373     StringRef Result(TokStart, CurPtr - TokStart);
374 
375     APInt Value(128, 0, true);
376     if (Result.substr(2).getAsInteger(2, Value))
377       return ReturnError(TokStart, "invalid binary number");
378 
379     // The darwin/x86 (and x86-64) assembler accepts and ignores ULL and LL
380     // suffixes on integer literals.
381     SkipIgnoredIntegerSuffix(CurPtr);
382 
383     return intToken(Result, Value);
384   }
385 
386   if ((*CurPtr == 'x') || (*CurPtr == 'X')) {
387     ++CurPtr;
388     const char *NumStart = CurPtr;
389     while (isHexDigit(CurPtr[0]))
390       ++CurPtr;
391 
392     // "0x.0p0" is valid, and "0x0p0" (but not "0xp0" for example, which will be
393     // diagnosed by LexHexFloatLiteral).
394     if (CurPtr[0] == '.' || CurPtr[0] == 'p' || CurPtr[0] == 'P')
395       return LexHexFloatLiteral(NumStart == CurPtr);
396 
397     // Otherwise requires at least one hex digit.
398     if (CurPtr == NumStart)
399       return ReturnError(CurPtr-2, "invalid hexadecimal number");
400 
401     APInt Result(128, 0);
402     if (StringRef(TokStart, CurPtr - TokStart).getAsInteger(0, Result))
403       return ReturnError(TokStart, "invalid hexadecimal number");
404 
405     // Consume the optional [hH].
406     if (LexMasmIntegers && (*CurPtr == 'h' || *CurPtr == 'H'))
407       ++CurPtr;
408 
409     // The darwin/x86 (and x86-64) assembler accepts and ignores ULL and LL
410     // suffixes on integer literals.
411     SkipIgnoredIntegerSuffix(CurPtr);
412 
413     return intToken(StringRef(TokStart, CurPtr - TokStart), Result);
414   }
415 
416   // Either octal or hexadecimal.
417   APInt Value(128, 0, true);
418   unsigned Radix = doHexLookAhead(CurPtr, 8, LexMasmIntegers);
419   bool isHex = Radix == 16;
420   StringRef Result(TokStart, CurPtr - TokStart);
421   if (Result.getAsInteger(Radix, Value))
422     return ReturnError(TokStart, !isHex ? "invalid octal number" :
423                        "invalid hexdecimal number");
424 
425   // Consume the [hH].
426   if (Radix == 16)
427     ++CurPtr;
428 
429   // The darwin/x86 (and x86-64) assembler accepts and ignores ULL and LL
430   // suffixes on integer literals.
431   SkipIgnoredIntegerSuffix(CurPtr);
432 
433   return intToken(Result, Value);
434 }
435 
436 /// LexSingleQuote: Integer: 'b'
437 AsmToken AsmLexer::LexSingleQuote() {
438   int CurChar = getNextChar();
439 
440   if (CurChar == '\\')
441     CurChar = getNextChar();
442 
443   if (CurChar == EOF)
444     return ReturnError(TokStart, "unterminated single quote");
445 
446   CurChar = getNextChar();
447 
448   if (CurChar != '\'')
449     return ReturnError(TokStart, "single quote way too long");
450 
451   // The idea here being that 'c' is basically just an integral
452   // constant.
453   StringRef Res = StringRef(TokStart,CurPtr - TokStart);
454   long long Value;
455 
456   if (Res.startswith("\'\\")) {
457     char theChar = Res[2];
458     switch (theChar) {
459       default: Value = theChar; break;
460       case '\'': Value = '\''; break;
461       case 't': Value = '\t'; break;
462       case 'n': Value = '\n'; break;
463       case 'b': Value = '\b'; break;
464     }
465   } else
466     Value = TokStart[1];
467 
468   return AsmToken(AsmToken::Integer, Res, Value);
469 }
470 
471 /// LexQuote: String: "..."
472 AsmToken AsmLexer::LexQuote() {
473   int CurChar = getNextChar();
474   // TODO: does gas allow multiline string constants?
475   while (CurChar != '"') {
476     if (CurChar == '\\') {
477       // Allow \", etc.
478       CurChar = getNextChar();
479     }
480 
481     if (CurChar == EOF)
482       return ReturnError(TokStart, "unterminated string constant");
483 
484     CurChar = getNextChar();
485   }
486 
487   return AsmToken(AsmToken::String, StringRef(TokStart, CurPtr - TokStart));
488 }
489 
490 StringRef AsmLexer::LexUntilEndOfStatement() {
491   TokStart = CurPtr;
492 
493   while (!isAtStartOfComment(CurPtr) &&     // Start of line comment.
494          !isAtStatementSeparator(CurPtr) && // End of statement marker.
495          *CurPtr != '\n' && *CurPtr != '\r' && CurPtr != CurBuf.end()) {
496     ++CurPtr;
497   }
498   return StringRef(TokStart, CurPtr-TokStart);
499 }
500 
501 StringRef AsmLexer::LexUntilEndOfLine() {
502   TokStart = CurPtr;
503 
504   while (*CurPtr != '\n' && *CurPtr != '\r' && CurPtr != CurBuf.end()) {
505     ++CurPtr;
506   }
507   return StringRef(TokStart, CurPtr-TokStart);
508 }
509 
510 size_t AsmLexer::peekTokens(MutableArrayRef<AsmToken> Buf,
511                             bool ShouldSkipSpace) {
512   SaveAndRestore<const char *> SavedTokenStart(TokStart);
513   SaveAndRestore<const char *> SavedCurPtr(CurPtr);
514   SaveAndRestore<bool> SavedAtStartOfLine(IsAtStartOfLine);
515   SaveAndRestore<bool> SavedAtStartOfStatement(IsAtStartOfStatement);
516   SaveAndRestore<bool> SavedSkipSpace(SkipSpace, ShouldSkipSpace);
517   SaveAndRestore<bool> SavedIsPeeking(IsPeeking, true);
518   std::string SavedErr = getErr();
519   SMLoc SavedErrLoc = getErrLoc();
520 
521   size_t ReadCount;
522   for (ReadCount = 0; ReadCount < Buf.size(); ++ReadCount) {
523     AsmToken Token = LexToken();
524 
525     Buf[ReadCount] = Token;
526 
527     if (Token.is(AsmToken::Eof))
528       break;
529   }
530 
531   SetError(SavedErrLoc, SavedErr);
532   return ReadCount;
533 }
534 
535 bool AsmLexer::isAtStartOfComment(const char *Ptr) {
536   StringRef CommentString = MAI.getCommentString();
537 
538   if (CommentString.size() == 1)
539     return CommentString[0] == Ptr[0];
540 
541   // Allow # preprocessor commments also be counted as comments for "##" cases
542   if (CommentString[1] == '#')
543     return CommentString[0] == Ptr[0];
544 
545   return strncmp(Ptr, CommentString.data(), CommentString.size()) == 0;
546 }
547 
548 bool AsmLexer::isAtStatementSeparator(const char *Ptr) {
549   return strncmp(Ptr, MAI.getSeparatorString(),
550                  strlen(MAI.getSeparatorString())) == 0;
551 }
552 
553 AsmToken AsmLexer::LexToken() {
554   TokStart = CurPtr;
555   // This always consumes at least one character.
556   int CurChar = getNextChar();
557 
558   if (!IsPeeking && CurChar == '#' && IsAtStartOfStatement) {
559     // If this starts with a '#', this may be a cpp
560     // hash directive and otherwise a line comment.
561     AsmToken TokenBuf[2];
562     MutableArrayRef<AsmToken> Buf(TokenBuf, 2);
563     size_t num = peekTokens(Buf, true);
564     // There cannot be a space preceding this
565     if (IsAtStartOfLine && num == 2 && TokenBuf[0].is(AsmToken::Integer) &&
566         TokenBuf[1].is(AsmToken::String)) {
567       CurPtr = TokStart; // reset curPtr;
568       StringRef s = LexUntilEndOfLine();
569       UnLex(TokenBuf[1]);
570       UnLex(TokenBuf[0]);
571       return AsmToken(AsmToken::HashDirective, s);
572     }
573     return LexLineComment();
574   }
575 
576   if (isAtStartOfComment(TokStart))
577     return LexLineComment();
578 
579   if (isAtStatementSeparator(TokStart)) {
580     CurPtr += strlen(MAI.getSeparatorString()) - 1;
581     IsAtStartOfLine = true;
582     IsAtStartOfStatement = true;
583     return AsmToken(AsmToken::EndOfStatement,
584                     StringRef(TokStart, strlen(MAI.getSeparatorString())));
585   }
586 
587   // If we're missing a newline at EOF, make sure we still get an
588   // EndOfStatement token before the Eof token.
589   if (CurChar == EOF && !IsAtStartOfStatement && EndStatementAtEOF) {
590     IsAtStartOfLine = true;
591     IsAtStartOfStatement = true;
592     return AsmToken(AsmToken::EndOfStatement, StringRef(TokStart, 1));
593   }
594   IsAtStartOfLine = false;
595   bool OldIsAtStartOfStatement = IsAtStartOfStatement;
596   IsAtStartOfStatement = false;
597   switch (CurChar) {
598   default:
599     if (MAI.doesAllowSymbolAtNameStart()) {
600       // Handle Microsoft-style identifier: [a-zA-Z_$.@?][a-zA-Z0-9_$.@?]*
601       if (!isDigit(CurChar) &&
602           IsIdentifierChar(CurChar, MAI.doesAllowAtInName()))
603         return LexIdentifier();
604     } else {
605       // Handle identifier: [a-zA-Z_.][a-zA-Z0-9_$.@]*
606       if (isalpha(CurChar) || CurChar == '_' || CurChar == '.')
607         return LexIdentifier();
608     }
609 
610     // Unknown character, emit an error.
611     return ReturnError(TokStart, "invalid character in input");
612   case EOF:
613     if (EndStatementAtEOF) {
614       IsAtStartOfLine = true;
615       IsAtStartOfStatement = true;
616     }
617     return AsmToken(AsmToken::Eof, StringRef(TokStart, 0));
618   case 0:
619   case ' ':
620   case '\t':
621     IsAtStartOfStatement = OldIsAtStartOfStatement;
622     while (*CurPtr == ' ' || *CurPtr == '\t')
623       CurPtr++;
624     if (SkipSpace)
625       return LexToken(); // Ignore whitespace.
626     else
627       return AsmToken(AsmToken::Space, StringRef(TokStart, CurPtr - TokStart));
628   case '\r': {
629     IsAtStartOfLine = true;
630     IsAtStartOfStatement = true;
631     // If this is a CR followed by LF, treat that as one token.
632     if (CurPtr != CurBuf.end() && *CurPtr == '\n')
633       ++CurPtr;
634     return AsmToken(AsmToken::EndOfStatement,
635                     StringRef(TokStart, CurPtr - TokStart));
636   }
637   case '\n':
638     IsAtStartOfLine = true;
639     IsAtStartOfStatement = true;
640     return AsmToken(AsmToken::EndOfStatement, StringRef(TokStart, 1));
641   case ':': return AsmToken(AsmToken::Colon, StringRef(TokStart, 1));
642   case '+': return AsmToken(AsmToken::Plus, StringRef(TokStart, 1));
643   case '~': return AsmToken(AsmToken::Tilde, StringRef(TokStart, 1));
644   case '(': return AsmToken(AsmToken::LParen, StringRef(TokStart, 1));
645   case ')': return AsmToken(AsmToken::RParen, StringRef(TokStart, 1));
646   case '[': return AsmToken(AsmToken::LBrac, StringRef(TokStart, 1));
647   case ']': return AsmToken(AsmToken::RBrac, StringRef(TokStart, 1));
648   case '{': return AsmToken(AsmToken::LCurly, StringRef(TokStart, 1));
649   case '}': return AsmToken(AsmToken::RCurly, StringRef(TokStart, 1));
650   case '*': return AsmToken(AsmToken::Star, StringRef(TokStart, 1));
651   case ',': return AsmToken(AsmToken::Comma, StringRef(TokStart, 1));
652   case '$': return AsmToken(AsmToken::Dollar, StringRef(TokStart, 1));
653   case '@': return AsmToken(AsmToken::At, StringRef(TokStart, 1));
654   case '\\': return AsmToken(AsmToken::BackSlash, StringRef(TokStart, 1));
655   case '=':
656     if (*CurPtr == '=') {
657       ++CurPtr;
658       return AsmToken(AsmToken::EqualEqual, StringRef(TokStart, 2));
659     }
660     return AsmToken(AsmToken::Equal, StringRef(TokStart, 1));
661   case '-':
662     if (*CurPtr == '>') {
663       ++CurPtr;
664       return AsmToken(AsmToken::MinusGreater, StringRef(TokStart, 2));
665     }
666     return AsmToken(AsmToken::Minus, StringRef(TokStart, 1));
667   case '|':
668     if (*CurPtr == '|') {
669       ++CurPtr;
670       return AsmToken(AsmToken::PipePipe, StringRef(TokStart, 2));
671     }
672     return AsmToken(AsmToken::Pipe, StringRef(TokStart, 1));
673   case '^': return AsmToken(AsmToken::Caret, StringRef(TokStart, 1));
674   case '&':
675     if (*CurPtr == '&') {
676       ++CurPtr;
677       return AsmToken(AsmToken::AmpAmp, StringRef(TokStart, 2));
678     }
679     return AsmToken(AsmToken::Amp, StringRef(TokStart, 1));
680   case '!':
681     if (*CurPtr == '=') {
682       ++CurPtr;
683       return AsmToken(AsmToken::ExclaimEqual, StringRef(TokStart, 2));
684     }
685     return AsmToken(AsmToken::Exclaim, StringRef(TokStart, 1));
686   case '%':
687     if (MAI.hasMipsExpressions()) {
688       AsmToken::TokenKind Operator;
689       unsigned OperatorLength;
690 
691       std::tie(Operator, OperatorLength) =
692           StringSwitch<std::pair<AsmToken::TokenKind, unsigned>>(
693               StringRef(CurPtr))
694               .StartsWith("call16", {AsmToken::PercentCall16, 7})
695               .StartsWith("call_hi", {AsmToken::PercentCall_Hi, 8})
696               .StartsWith("call_lo", {AsmToken::PercentCall_Lo, 8})
697               .StartsWith("dtprel_hi", {AsmToken::PercentDtprel_Hi, 10})
698               .StartsWith("dtprel_lo", {AsmToken::PercentDtprel_Lo, 10})
699               .StartsWith("got_disp", {AsmToken::PercentGot_Disp, 9})
700               .StartsWith("got_hi", {AsmToken::PercentGot_Hi, 7})
701               .StartsWith("got_lo", {AsmToken::PercentGot_Lo, 7})
702               .StartsWith("got_ofst", {AsmToken::PercentGot_Ofst, 9})
703               .StartsWith("got_page", {AsmToken::PercentGot_Page, 9})
704               .StartsWith("gottprel", {AsmToken::PercentGottprel, 9})
705               .StartsWith("got", {AsmToken::PercentGot, 4})
706               .StartsWith("gp_rel", {AsmToken::PercentGp_Rel, 7})
707               .StartsWith("higher", {AsmToken::PercentHigher, 7})
708               .StartsWith("highest", {AsmToken::PercentHighest, 8})
709               .StartsWith("hi", {AsmToken::PercentHi, 3})
710               .StartsWith("lo", {AsmToken::PercentLo, 3})
711               .StartsWith("neg", {AsmToken::PercentNeg, 4})
712               .StartsWith("pcrel_hi", {AsmToken::PercentPcrel_Hi, 9})
713               .StartsWith("pcrel_lo", {AsmToken::PercentPcrel_Lo, 9})
714               .StartsWith("tlsgd", {AsmToken::PercentTlsgd, 6})
715               .StartsWith("tlsldm", {AsmToken::PercentTlsldm, 7})
716               .StartsWith("tprel_hi", {AsmToken::PercentTprel_Hi, 9})
717               .StartsWith("tprel_lo", {AsmToken::PercentTprel_Lo, 9})
718               .Default({AsmToken::Percent, 1});
719 
720       if (Operator != AsmToken::Percent) {
721         CurPtr += OperatorLength - 1;
722         return AsmToken(Operator, StringRef(TokStart, OperatorLength));
723       }
724     }
725     return AsmToken(AsmToken::Percent, StringRef(TokStart, 1));
726   case '/':
727     IsAtStartOfStatement = OldIsAtStartOfStatement;
728     return LexSlash();
729   case '#': return AsmToken(AsmToken::Hash, StringRef(TokStart, 1));
730   case '\'': return LexSingleQuote();
731   case '"': return LexQuote();
732   case '0': case '1': case '2': case '3': case '4':
733   case '5': case '6': case '7': case '8': case '9':
734     return LexDigit();
735   case '<':
736     switch (*CurPtr) {
737     case '<':
738       ++CurPtr;
739       return AsmToken(AsmToken::LessLess, StringRef(TokStart, 2));
740     case '=':
741       ++CurPtr;
742       return AsmToken(AsmToken::LessEqual, StringRef(TokStart, 2));
743     case '>':
744       ++CurPtr;
745       return AsmToken(AsmToken::LessGreater, StringRef(TokStart, 2));
746     default:
747       return AsmToken(AsmToken::Less, StringRef(TokStart, 1));
748     }
749   case '>':
750     switch (*CurPtr) {
751     case '>':
752       ++CurPtr;
753       return AsmToken(AsmToken::GreaterGreater, StringRef(TokStart, 2));
754     case '=':
755       ++CurPtr;
756       return AsmToken(AsmToken::GreaterEqual, StringRef(TokStart, 2));
757     default:
758       return AsmToken(AsmToken::Greater, StringRef(TokStart, 1));
759     }
760 
761   // TODO: Quoted identifiers (objc methods etc)
762   // local labels: [0-9][:]
763   // Forward/backward labels: [0-9][fb]
764   // Integers, fp constants, character constants.
765   }
766 }
767