lib/AsmParser/LLLexer.cpp

   1 //===- LLLexer.cpp - Lexer for .ll 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 // Implement the Lexer for .ll files.
  11 //
  12 //===----------------------------------------------------------------------===//
  13
  14 #include "LLLexer.h"
  15 #include "llvm/DerivedTypes.h"
  16 #include "llvm/Instruction.h"
  17 #include "llvm/Support/MemoryBuffer.h"
  18 #include "llvm/Support/MathExtras.h"
  19 #include "llvm/Support/raw_ostream.h"
  20 #include "llvm/Assembly/Parser.h"
  21 #include <cstdlib>
  22 #include <cstring>
  23 using namespace llvm;
  24
  25 bool LLLexer::Error(LocTy ErrorLoc, const std::string &Msg) const {
  26   // Scan backward to find the start of the line.
  27   const char *LineStart = ErrorLoc;
  28   while (LineStart != CurBuf->getBufferStart() &&
  29          LineStart[-1] != '\n' && LineStart[-1] != '\r')
  30     --LineStart;
  31   // Get the end of the line.
  32   const char *LineEnd = ErrorLoc;
  33   while (LineEnd != CurBuf->getBufferEnd() &&
  34          LineEnd[0] != '\n' && LineEnd[0] != '\r')
  35     ++LineEnd;
  36
  37   unsigned LineNo = 1;
  38   for (const char *FP = CurBuf->getBufferStart(); FP != ErrorLoc; ++FP)
  39     if (*FP == '\n') ++LineNo;
  40
  41   std::string LineContents(LineStart, LineEnd);
  42   ErrorInfo.setError(Msg, LineNo, ErrorLoc-LineStart, LineContents);
  43   return true;
  44 }
  45
  46 //===----------------------------------------------------------------------===//
  47 // Helper functions.
  48 //===----------------------------------------------------------------------===//
  49
  50 // atoull - Convert an ascii string of decimal digits into the unsigned long
  51 // long representation... this does not have to do input error checking,
  52 // because we know that the input will be matched by a suitable regex...
  53 //
  54 uint64_t LLLexer::atoull(const char *Buffer, const char *End) {
  55   uint64_t Result = 0;
  56   for (; Buffer != End; Buffer++) {
  57     uint64_t OldRes = Result;
  58     Result *= 10;
  59     Result += *Buffer-'0';
  60     if (Result < OldRes) {  // Uh, oh, overflow detected!!!
  61       Error("constant bigger than 64 bits detected!");
  62       return 0;
  63     }
  64   }
  65   return Result;
  66 }
  67
  68 uint64_t LLLexer::HexIntToVal(const char *Buffer, const char *End) {
  69   uint64_t Result = 0;
  70   for (; Buffer != End; ++Buffer) {
  71     uint64_t OldRes = Result;
  72     Result *= 16;
  73     char C = *Buffer;
  74     if (C >= '0' && C <= '9')
  75       Result += C-'0';
  76     else if (C >= 'A' && C <= 'F')
  77       Result += C-'A'+10;
  78     else if (C >= 'a' && C <= 'f')
  79       Result += C-'a'+10;
  80
  81     if (Result < OldRes) {   // Uh, oh, overflow detected!!!
  82       Error("constant bigger than 64 bits detected!");
  83       return 0;
  84     }
  85   }
  86   return Result;
  87 }
  88
  89 void LLLexer::HexToIntPair(const char *Buffer, const char *End,
  90                            uint64_t Pair[2]) {
  91   Pair[0] = 0;
  92   for (int i=0; i<16; i++, Buffer++) {
  93     assert(Buffer != End);
  94     Pair[0] *= 16;
  95     char C = *Buffer;
  96     if (C >= '0' && C <= '9')
  97       Pair[0] += C-'0';
  98     else if (C >= 'A' && C <= 'F')
  99       Pair[0] += C-'A'+10;
 100     else if (C >= 'a' && C <= 'f')
 101       Pair[0] += C-'a'+10;
 102   }
 103   Pair[1] = 0;
 104   for (int i=0; i<16 && Buffer != End; i++, Buffer++) {
 105     Pair[1] *= 16;
 106     char C = *Buffer;
 107     if (C >= '0' && C <= '9')
 108       Pair[1] += C-'0';
 109     else if (C >= 'A' && C <= 'F')
 110       Pair[1] += C-'A'+10;
 111     else if (C >= 'a' && C <= 'f')
 112       Pair[1] += C-'a'+10;
 113   }
 114   if (Buffer != End)
 115     Error("constant bigger than 128 bits detected!");
 116 }
 117
 118 /// FP80HexToIntPair - translate an 80 bit FP80 number (20 hexits) into
 119 /// { low64, high16 } as usual for an APInt.
 120 void LLLexer::FP80HexToIntPair(const char *Buffer, const char *End,
 121                            uint64_t Pair[2]) {
 122   Pair[1] = 0;
 123   for (int i=0; i<4 && Buffer != End; i++, Buffer++) {
 124     assert(Buffer != End);
 125     Pair[1] *= 16;
 126     char C = *Buffer;
 127     if (C >= '0' && C <= '9')
 128       Pair[1] += C-'0';
 129     else if (C >= 'A' && C <= 'F')
 130       Pair[1] += C-'A'+10;
 131     else if (C >= 'a' && C <= 'f')
 132       Pair[1] += C-'a'+10;
 133   }
 134   Pair[0] = 0;
 135   for (int i=0; i<16; i++, Buffer++) {
 136     Pair[0] *= 16;
 137     char C = *Buffer;
 138     if (C >= '0' && C <= '9')
 139       Pair[0] += C-'0';
 140     else if (C >= 'A' && C <= 'F')
 141       Pair[0] += C-'A'+10;
 142     else if (C >= 'a' && C <= 'f')
 143       Pair[0] += C-'a'+10;
 144   }
 145   if (Buffer != End)
 146     Error("constant bigger than 128 bits detected!");
 147 }
 148
 149 // UnEscapeLexed - Run through the specified buffer and change \xx codes to the
 150 // appropriate character.
 151 static void UnEscapeLexed(std::string &Str) {
 152   if (Str.empty()) return;
 153
 154   char *Buffer = &Str[0], *EndBuffer = Buffer+Str.size();
 155   char *BOut = Buffer;
 156   for (char *BIn = Buffer; BIn != EndBuffer; ) {
 157     if (BIn[0] == '\\') {
 158       if (BIn < EndBuffer-1 && BIn[1] == '\\') {
 159         *BOut++ = '\\'; // Two \ becomes one
 160         BIn += 2;
 161       } else if (BIn < EndBuffer-2 && isxdigit(BIn[1]) && isxdigit(BIn[2])) {
 162         char Tmp = BIn[3]; BIn[3] = 0;      // Terminate string
 163         *BOut = (char)strtol(BIn+1, 0, 16); // Convert to number
 164         BIn[3] = Tmp;                       // Restore character
 165         BIn += 3;                           // Skip over handled chars
 166         ++BOut;
 167       } else {
 168         *BOut++ = *BIn++;
 169       }
 170     } else {
 171       *BOut++ = *BIn++;
 172     }
 173   }
 174   Str.resize(BOut-Buffer);
 175 }
 176
 177 /// isLabelChar - Return true for [-a-zA-Z$._0-9].
 178 static bool isLabelChar(char C) {
 179   return isalnum(C) || C == '-' || C == '$' || C == '.' || C == '_';
 180 }
 181
 182
 183 /// isLabelTail - Return true if this pointer points to a valid end of a label.
 184 static const char *isLabelTail(const char *CurPtr) {
 185   while (1) {
 186     if (CurPtr[0] == ':') return CurPtr+1;
 187     if (!isLabelChar(CurPtr[0])) return 0;
 188     ++CurPtr;
 189   }
 190 }
 191
 192
 193
 194 //===----------------------------------------------------------------------===//
 195 // Lexer definition.
 196 //===----------------------------------------------------------------------===//
 197
 198 LLLexer::LLLexer(MemoryBuffer *StartBuf, ParseError &Err)
 199   : CurBuf(StartBuf), ErrorInfo(Err), APFloatVal(0.0) {
 200   CurPtr = CurBuf->getBufferStart();
 201 }
 202
 203 std::string LLLexer::getFilename() const {
 204   return CurBuf->getBufferIdentifier();
 205 }
 206
 207 int LLLexer::getNextChar() {
 208   char CurChar = *CurPtr++;
 209   switch (CurChar) {
 210   default: return (unsigned char)CurChar;
 211   case 0:
 212     // A nul character in the stream is either the end of the current buffer or
 213     // a random nul in the file.  Disambiguate that here.
 214     if (CurPtr-1 != CurBuf->getBufferEnd())
 215       return 0;  // Just whitespace.
 216
 217     // Otherwise, return end of file.
 218     --CurPtr;  // Another call to lex will return EOF again.
 219     return EOF;
 220   }
 221 }
 222
 223
 224 lltok::Kind LLLexer::LexToken() {
 225   TokStart = CurPtr;
 226
 227   int CurChar = getNextChar();
 228   switch (CurChar) {
 229   default:
 230     // Handle letters: [a-zA-Z_]
 231     if (isalpha(CurChar) || CurChar == '_')
 232       return LexIdentifier();
 233
 234     return lltok::Error;
 235   case EOF: return lltok::Eof;
 236   case 0:
 237   case ' ':
 238   case '\t':
 239   case '\n':
 240   case '\r':
 241     // Ignore whitespace.
 242     return LexToken();
 243   case '+': return LexPositive();
 244   case '@': return LexAt();
 245   case '%': return LexPercent();
 246   case '"': return LexQuote();
 247   case '.':
 248     if (const char *Ptr = isLabelTail(CurPtr)) {
 249       CurPtr = Ptr;
 250       StrVal.assign(TokStart, CurPtr-1);
 251       return lltok::LabelStr;
 252     }
 253     if (CurPtr[0] == '.' && CurPtr[1] == '.') {
 254       CurPtr += 2;
 255       return lltok::dotdotdot;
 256     }
 257     return lltok::Error;
 258   case '$':
 259     if (const char *Ptr = isLabelTail(CurPtr)) {
 260       CurPtr = Ptr;
 261       StrVal.assign(TokStart, CurPtr-1);
 262       return lltok::LabelStr;
 263     }
 264     return lltok::Error;
 265   case ';':
 266     SkipLineComment();
 267     return LexToken();
 268   case '!': return lltok::Metadata;
 269   case '0': case '1': case '2': case '3': case '4':
 270   case '5': case '6': case '7': case '8': case '9':
 271   case '-':
 272     return LexDigitOrNegative();
 273   case '=': return lltok::equal;
 274   case '[': return lltok::lsquare;
 275   case ']': return lltok::rsquare;
 276   case '{': return lltok::lbrace;
 277   case '}': return lltok::rbrace;
 278   case '<': return lltok::less;
 279   case '>': return lltok::greater;
 280   case '(': return lltok::lparen;
 281   case ')': return lltok::rparen;
 282   case ',': return lltok::comma;
 283   case '*': return lltok::star;
 284   case '\\': return lltok::backslash;
 285   }
 286 }
 287
 288 void LLLexer::SkipLineComment() {
 289   while (1) {
 290     if (CurPtr[0] == '\n' || CurPtr[0] == '\r' || getNextChar() == EOF)
 291       return;
 292   }
 293 }
 294
 295 /// LexAt - Lex all tokens that start with an @ character:
 296 ///   GlobalVar   @\"[^\"]*\"
 297 ///   GlobalVar   @[-a-zA-Z$._][-a-zA-Z$._0-9]*
 298 ///   GlobalVarID @[0-9]+
 299 lltok::Kind LLLexer::LexAt() {
 300   // Handle AtStringConstant: @\"[^\"]*\"
 301   if (CurPtr[0] == '"') {
 302     ++CurPtr;
 303
 304     while (1) {
 305       int CurChar = getNextChar();
 306
 307       if (CurChar == EOF) {
 308         Error("end of file in global variable name");
 309         return lltok::Error;
 310       }
 311       if (CurChar == '"') {
 312         StrVal.assign(TokStart+2, CurPtr-1);
 313         UnEscapeLexed(StrVal);
 314         return lltok::GlobalVar;
 315       }
 316     }
 317   }
 318
 319   // Handle GlobalVarName: @[-a-zA-Z$._][-a-zA-Z$._0-9]*
 320   if (isalpha(CurPtr[0]) || CurPtr[0] == '-' || CurPtr[0] == '$' ||
 321       CurPtr[0] == '.' || CurPtr[0] == '_') {
 322     ++CurPtr;
 323     while (isalnum(CurPtr[0]) || CurPtr[0] == '-' || CurPtr[0] == '$' ||
 324            CurPtr[0] == '.' || CurPtr[0] == '_')
 325       ++CurPtr;
 326
 327     StrVal.assign(TokStart+1, CurPtr);   // Skip @
 328     return lltok::GlobalVar;
 329   }
 330
 331   // Handle GlobalVarID: @[0-9]+
 332   if (isdigit(CurPtr[0])) {
 333     for (++CurPtr; isdigit(CurPtr[0]); ++CurPtr)
 334       /*empty*/;
 335
 336     uint64_t Val = atoull(TokStart+1, CurPtr);
 337     if ((unsigned)Val != Val)
 338       Error("invalid value number (too large)!");
 339     UIntVal = unsigned(Val);
 340     return lltok::GlobalID;
 341   }
 342
 343   return lltok::Error;
 344 }
 345
 346
 347 /// LexPercent - Lex all tokens that start with a % character:
 348 ///   LocalVar   ::= %\"[^\"]*\"
 349 ///   LocalVar   ::= %[-a-zA-Z$._][-a-zA-Z$._0-9]*
 350 ///   LocalVarID ::= %[0-9]+
 351 lltok::Kind LLLexer::LexPercent() {
 352   // Handle LocalVarName: %\"[^\"]*\"
 353   if (CurPtr[0] == '"') {
 354     ++CurPtr;
 355
 356     while (1) {
 357       int CurChar = getNextChar();
 358
 359       if (CurChar == EOF) {
 360         Error("end of file in string constant");
 361         return lltok::Error;
 362       }
 363       if (CurChar == '"') {
 364         StrVal.assign(TokStart+2, CurPtr-1);
 365         UnEscapeLexed(StrVal);
 366         return lltok::LocalVar;
 367       }
 368     }
 369   }
 370
 371   // Handle LocalVarName: %[-a-zA-Z$._][-a-zA-Z$._0-9]*
 372   if (isalpha(CurPtr[0]) || CurPtr[0] == '-' || CurPtr[0] == '$' ||
 373       CurPtr[0] == '.' || CurPtr[0] == '_') {
 374     ++CurPtr;
 375     while (isalnum(CurPtr[0]) || CurPtr[0] == '-' || CurPtr[0] == '$' ||
 376            CurPtr[0] == '.' || CurPtr[0] == '_')
 377       ++CurPtr;
 378
 379     StrVal.assign(TokStart+1, CurPtr);   // Skip %
 380     return lltok::LocalVar;
 381   }
 382
 383   // Handle LocalVarID: %[0-9]+
 384   if (isdigit(CurPtr[0])) {
 385     for (++CurPtr; isdigit(CurPtr[0]); ++CurPtr)
 386       /*empty*/;
 387
 388     uint64_t Val = atoull(TokStart+1, CurPtr);
 389     if ((unsigned)Val != Val)
 390       Error("invalid value number (too large)!");
 391     UIntVal = unsigned(Val);
 392     return lltok::LocalVarID;
 393   }
 394
 395   return lltok::Error;
 396 }
 397
 398 /// LexQuote - Lex all tokens that start with a " character:
 399 ///   QuoteLabel        "[^"]+":
 400 ///   StringConstant    "[^"]*"
 401 lltok::Kind LLLexer::LexQuote() {
 402   while (1) {
 403     int CurChar = getNextChar();
 404
 405     if (CurChar == EOF) {
 406       Error("end of file in quoted string");
 407       return lltok::Error;
 408     }
 409
 410     if (CurChar != '"') continue;
 411
 412     if (CurPtr[0] != ':') {
 413       StrVal.assign(TokStart+1, CurPtr-1);
 414       UnEscapeLexed(StrVal);
 415       return lltok::StringConstant;
 416     }
 417
 418     ++CurPtr;
 419     StrVal.assign(TokStart+1, CurPtr-2);
 420     UnEscapeLexed(StrVal);
 421     return lltok::LabelStr;
 422   }
 423 }
 424
 425 static bool JustWhitespaceNewLine(const char *&Ptr) {
 426   const char *ThisPtr = Ptr;
 427   while (*ThisPtr == ' ' || *ThisPtr == '\t')
 428     ++ThisPtr;
 429   if (*ThisPtr == '\n' || *ThisPtr == '\r') {
 430     Ptr = ThisPtr;
 431     return true;
 432   }
 433   return false;
 434 }
 435
 436
 437 /// LexIdentifier: Handle several related productions:
 438 ///    Label           [-a-zA-Z$._0-9]+:
 439 ///    IntegerType     i[0-9]+
 440 ///    Keyword         sdiv, float, ...
 441 ///    HexIntConstant  [us]0x[0-9A-Fa-f]+
 442 lltok::Kind LLLexer::LexIdentifier() {
 443   const char *StartChar = CurPtr;
 444   const char *IntEnd = CurPtr[-1] == 'i' ? 0 : StartChar;
 445   const char *KeywordEnd = 0;
 446
 447   for (; isLabelChar(*CurPtr); ++CurPtr) {
 448     // If we decide this is an integer, remember the end of the sequence.
 449     if (!IntEnd && !isdigit(*CurPtr)) IntEnd = CurPtr;
 450     if (!KeywordEnd && !isalnum(*CurPtr) && *CurPtr != '_') KeywordEnd = CurPtr;
 451   }
 452
 453   // If we stopped due to a colon, this really is a label.
 454   if (*CurPtr == ':') {
 455     StrVal.assign(StartChar-1, CurPtr++);
 456     return lltok::LabelStr;
 457   }
 458
 459   // Otherwise, this wasn't a label.  If this was valid as an integer type,
 460   // return it.
 461   if (IntEnd == 0) IntEnd = CurPtr;
 462   if (IntEnd != StartChar) {
 463     CurPtr = IntEnd;
 464     uint64_t NumBits = atoull(StartChar, CurPtr);
 465     if (NumBits < IntegerType::MIN_INT_BITS ||
 466         NumBits > IntegerType::MAX_INT_BITS) {
 467       Error("bitwidth for integer type out of range!");
 468       return lltok::Error;
 469     }
 470     TyVal = IntegerType::get(NumBits);
 471     return lltok::Type;
 472   }
 473
 474   // Otherwise, this was a letter sequence.  See which keyword this is.
 475   if (KeywordEnd == 0) KeywordEnd = CurPtr;
 476   CurPtr = KeywordEnd;
 477   --StartChar;
 478   unsigned Len = CurPtr-StartChar;
 479 #define KEYWORD(STR) \
 480   if (Len == strlen(#STR) && !memcmp(StartChar, #STR, strlen(#STR))) \
 481     return lltok::kw_##STR;
 482
 483   KEYWORD(begin);   KEYWORD(end);
 484   KEYWORD(true);    KEYWORD(false);
 485   KEYWORD(declare); KEYWORD(define);
 486   KEYWORD(global);  KEYWORD(constant);
 487
 488   KEYWORD(private);
 489   KEYWORD(internal);
 490   KEYWORD(available_externally);
 491   KEYWORD(linkonce);
 492   KEYWORD(linkonce_odr);
 493   KEYWORD(weak);
 494   KEYWORD(weak_odr);
 495   KEYWORD(appending);
 496   KEYWORD(dllimport);
 497   KEYWORD(dllexport);
 498   KEYWORD(common);
 499   KEYWORD(default);
 500   KEYWORD(hidden);
 501   KEYWORD(protected);
 502   KEYWORD(extern_weak);
 503   KEYWORD(external);
 504   KEYWORD(thread_local);
 505   KEYWORD(zeroinitializer);
 506   KEYWORD(undef);
 507   KEYWORD(null);
 508   KEYWORD(to);
 509   KEYWORD(tail);
 510   KEYWORD(target);
 511   KEYWORD(triple);
 512   KEYWORD(deplibs);
 513   KEYWORD(datalayout);
 514   KEYWORD(volatile);
 515   KEYWORD(align);
 516   KEYWORD(addrspace);
 517   KEYWORD(section);
 518   KEYWORD(alias);
 519   KEYWORD(module);
 520   KEYWORD(asm);
 521   KEYWORD(sideeffect);
 522   KEYWORD(gc);
 523
 524   KEYWORD(ccc);
 525   KEYWORD(fastcc);
 526   KEYWORD(coldcc);
 527   KEYWORD(x86_stdcallcc);
 528   KEYWORD(x86_fastcallcc);
 529   KEYWORD(cc);
 530   KEYWORD(c);
 531
 532   KEYWORD(signext);
 533   KEYWORD(zeroext);
 534   KEYWORD(inreg);
 535   KEYWORD(sret);
 536   KEYWORD(nounwind);
 537   KEYWORD(noreturn);
 538   KEYWORD(noalias);
 539   KEYWORD(nocapture);
 540   KEYWORD(byval);
 541   KEYWORD(nest);
 542   KEYWORD(readnone);
 543   KEYWORD(readonly);
 544
 545   KEYWORD(noinline);
 546   KEYWORD(alwaysinline);
 547   KEYWORD(optsize);
 548   KEYWORD(ssp);
 549   KEYWORD(sspreq);
 550
 551   KEYWORD(type);
 552   KEYWORD(opaque);
 553
 554   KEYWORD(eq); KEYWORD(ne); KEYWORD(slt); KEYWORD(sgt); KEYWORD(sle);
 555   KEYWORD(sge); KEYWORD(ult); KEYWORD(ugt); KEYWORD(ule); KEYWORD(uge);
 556   KEYWORD(oeq); KEYWORD(one); KEYWORD(olt); KEYWORD(ogt); KEYWORD(ole);
 557   KEYWORD(oge); KEYWORD(ord); KEYWORD(uno); KEYWORD(ueq); KEYWORD(une);
 558
 559   KEYWORD(x);
 560 #undef KEYWORD
 561
 562   // Keywords for types.
 563 #define TYPEKEYWORD(STR, LLVMTY) \
 564   if (Len == strlen(STR) && !memcmp(StartChar, STR, strlen(STR))) { \
 565     TyVal = LLVMTY; return lltok::Type; }
 566   TYPEKEYWORD("void",      Type::VoidTy);
 567   TYPEKEYWORD("float",     Type::FloatTy);
 568   TYPEKEYWORD("double",    Type::DoubleTy);
 569   TYPEKEYWORD("x86_fp80",  Type::X86_FP80Ty);
 570   TYPEKEYWORD("fp128",     Type::FP128Ty);
 571   TYPEKEYWORD("ppc_fp128", Type::PPC_FP128Ty);
 572   TYPEKEYWORD("label",     Type::LabelTy);
 573 #undef TYPEKEYWORD
 574
 575   // Handle special forms for autoupgrading.  Drop these in LLVM 3.0.  This is
 576   // to avoid conflicting with the sext/zext instructions, below.
 577   if (Len == 4 && !memcmp(StartChar, "sext", 4)) {
 578     // Scan CurPtr ahead, seeing if there is just whitespace before the newline.
 579     if (JustWhitespaceNewLine(CurPtr))
 580       return lltok::kw_signext;
 581   } else if (Len == 4 && !memcmp(StartChar, "zext", 4)) {
 582     // Scan CurPtr ahead, seeing if there is just whitespace before the newline.
 583     if (JustWhitespaceNewLine(CurPtr))
 584       return lltok::kw_zeroext;
 585   }
 586
 587   // Keywords for instructions.
 588 #define INSTKEYWORD(STR, Enum) \
 589   if (Len == strlen(#STR) && !memcmp(StartChar, #STR, strlen(#STR))) { \
 590     UIntVal = Instruction::Enum; return lltok::kw_##STR; }
 591
 592   INSTKEYWORD(add,   Add);  INSTKEYWORD(sub,   Sub);  INSTKEYWORD(mul,   Mul);
 593   INSTKEYWORD(udiv,  UDiv); INSTKEYWORD(sdiv,  SDiv); INSTKEYWORD(fdiv,  FDiv);
 594   INSTKEYWORD(urem,  URem); INSTKEYWORD(srem,  SRem); INSTKEYWORD(frem,  FRem);
 595   INSTKEYWORD(shl,   Shl);  INSTKEYWORD(lshr,  LShr); INSTKEYWORD(ashr,  AShr);
 596   INSTKEYWORD(and,   And);  INSTKEYWORD(or,    Or);   INSTKEYWORD(xor,   Xor);
 597   INSTKEYWORD(icmp,  ICmp); INSTKEYWORD(fcmp,  FCmp);
 598   INSTKEYWORD(vicmp, VICmp); INSTKEYWORD(vfcmp, VFCmp);
 599
 600   INSTKEYWORD(phi,         PHI);
 601   INSTKEYWORD(call,        Call);
 602   INSTKEYWORD(trunc,       Trunc);
 603   INSTKEYWORD(zext,        ZExt);
 604   INSTKEYWORD(sext,        SExt);
 605   INSTKEYWORD(fptrunc,     FPTrunc);
 606   INSTKEYWORD(fpext,       FPExt);
 607   INSTKEYWORD(uitofp,      UIToFP);
 608   INSTKEYWORD(sitofp,      SIToFP);
 609   INSTKEYWORD(fptoui,      FPToUI);
 610   INSTKEYWORD(fptosi,      FPToSI);
 611   INSTKEYWORD(inttoptr,    IntToPtr);
 612   INSTKEYWORD(ptrtoint,    PtrToInt);
 613   INSTKEYWORD(bitcast,     BitCast);
 614   INSTKEYWORD(select,      Select);
 615   INSTKEYWORD(va_arg,      VAArg);
 616   INSTKEYWORD(ret,         Ret);
 617   INSTKEYWORD(br,          Br);
 618   INSTKEYWORD(switch,      Switch);
 619   INSTKEYWORD(invoke,      Invoke);
 620   INSTKEYWORD(unwind,      Unwind);
 621   INSTKEYWORD(unreachable, Unreachable);
 622
 623   INSTKEYWORD(malloc,      Malloc);
 624   INSTKEYWORD(alloca,      Alloca);
 625   INSTKEYWORD(free,        Free);
 626   INSTKEYWORD(load,        Load);
 627   INSTKEYWORD(store,       Store);
 628   INSTKEYWORD(getelementptr, GetElementPtr);
 629
 630   INSTKEYWORD(extractelement, ExtractElement);
 631   INSTKEYWORD(insertelement,  InsertElement);
 632   INSTKEYWORD(shufflevector,  ShuffleVector);
 633   INSTKEYWORD(getresult,      ExtractValue);
 634   INSTKEYWORD(extractvalue,   ExtractValue);
 635   INSTKEYWORD(insertvalue,    InsertValue);
 636 #undef INSTKEYWORD
 637
 638   // Check for [us]0x[0-9A-Fa-f]+ which are Hexadecimal constant generated by
 639   // the CFE to avoid forcing it to deal with 64-bit numbers.
 640   if ((TokStart[0] == 'u' || TokStart[0] == 's') &&
 641       TokStart[1] == '0' && TokStart[2] == 'x' && isxdigit(TokStart[3])) {
 642     int len = CurPtr-TokStart-3;
 643     uint32_t bits = len * 4;
 644     APInt Tmp(bits, TokStart+3, len, 16);
 645     uint32_t activeBits = Tmp.getActiveBits();
 646     if (activeBits > 0 && activeBits < bits)
 647       Tmp.trunc(activeBits);
 648     APSIntVal = APSInt(Tmp, TokStart[0] == 'u');
 649     return lltok::APSInt;
 650   }
 651
 652   // If this is "cc1234", return this as just "cc".
 653   if (TokStart[0] == 'c' && TokStart[1] == 'c') {
 654     CurPtr = TokStart+2;
 655     return lltok::kw_cc;
 656   }
 657
 658   // If this starts with "call", return it as CALL.  This is to support old
 659   // broken .ll files.  FIXME: remove this with LLVM 3.0.
 660   if (CurPtr-TokStart > 4 && !memcmp(TokStart, "call", 4)) {
 661     CurPtr = TokStart+4;
 662     UIntVal = Instruction::Call;
 663     return lltok::kw_call;
 664   }
 665
 666   // Finally, if this isn't known, return an error.
 667   CurPtr = TokStart+1;
 668   return lltok::Error;
 669 }
 670
 671
 672 /// Lex0x: Handle productions that start with 0x, knowing that it matches and
 673 /// that this is not a label:
 674 ///    HexFPConstant     0x[0-9A-Fa-f]+
 675 ///    HexFP80Constant   0xK[0-9A-Fa-f]+
 676 ///    HexFP128Constant  0xL[0-9A-Fa-f]+
 677 ///    HexPPC128Constant 0xM[0-9A-Fa-f]+
 678 lltok::Kind LLLexer::Lex0x() {
 679   CurPtr = TokStart + 2;
 680
 681   char Kind;
 682   if (CurPtr[0] >= 'K' && CurPtr[0] <= 'M') {
 683     Kind = *CurPtr++;
 684   } else {
 685     Kind = 'J';
 686   }
 687
 688   if (!isxdigit(CurPtr[0])) {
 689     // Bad token, return it as an error.
 690     CurPtr = TokStart+1;
 691     return lltok::Error;
 692   }
 693
 694   while (isxdigit(CurPtr[0]))
 695     ++CurPtr;
 696
 697   if (Kind == 'J') {
 698     // HexFPConstant - Floating point constant represented in IEEE format as a
 699     // hexadecimal number for when exponential notation is not precise enough.
 700     // Float and double only.
 701     APFloatVal = APFloat(BitsToDouble(HexIntToVal(TokStart+2, CurPtr)));
 702     return lltok::APFloat;
 703   }
 704
 705   uint64_t Pair[2];
 706   switch (Kind) {
 707   default: assert(0 && "Unknown kind!");
 708   case 'K':
 709     // F80HexFPConstant - x87 long double in hexadecimal format (10 bytes)
 710     FP80HexToIntPair(TokStart+3, CurPtr, Pair);
 711     APFloatVal = APFloat(APInt(80, 2, Pair));
 712     return lltok::APFloat;
 713   case 'L':
 714     // F128HexFPConstant - IEEE 128-bit in hexadecimal format (16 bytes)
 715     HexToIntPair(TokStart+3, CurPtr, Pair);
 716     APFloatVal = APFloat(APInt(128, 2, Pair), true);
 717     return lltok::APFloat;
 718   case 'M':
 719     // PPC128HexFPConstant - PowerPC 128-bit in hexadecimal format (16 bytes)
 720     HexToIntPair(TokStart+3, CurPtr, Pair);
 721     APFloatVal = APFloat(APInt(128, 2, Pair));
 722     return lltok::APFloat;
 723   }
 724 }
 725
 726 /// LexIdentifier: Handle several related productions:
 727 ///    Label             [-a-zA-Z$._0-9]+:
 728 ///    NInteger          -[0-9]+
 729 ///    FPConstant        [-+]?[0-9]+[.][0-9]*([eE][-+]?[0-9]+)?
 730 ///    PInteger          [0-9]+
 731 ///    HexFPConstant     0x[0-9A-Fa-f]+
 732 ///    HexFP80Constant   0xK[0-9A-Fa-f]+
 733 ///    HexFP128Constant  0xL[0-9A-Fa-f]+
 734 ///    HexPPC128Constant 0xM[0-9A-Fa-f]+
 735 lltok::Kind LLLexer::LexDigitOrNegative() {
 736   // If the letter after the negative is a number, this is probably a label.
 737   if (!isdigit(TokStart[0]) && !isdigit(CurPtr[0])) {
 738     // Okay, this is not a number after the -, it's probably a label.
 739     if (const char *End = isLabelTail(CurPtr)) {
 740       StrVal.assign(TokStart, End-1);
 741       CurPtr = End;
 742       return lltok::LabelStr;
 743     }
 744
 745     return lltok::Error;
 746   }
 747
 748   // At this point, it is either a label, int or fp constant.
 749
 750   // Skip digits, we have at least one.
 751   for (; isdigit(CurPtr[0]); ++CurPtr)
 752     /*empty*/;
 753
 754   // Check to see if this really is a label afterall, e.g. "-1:".
 755   if (isLabelChar(CurPtr[0]) || CurPtr[0] == ':') {
 756     if (const char *End = isLabelTail(CurPtr)) {
 757       StrVal.assign(TokStart, End-1);
 758       CurPtr = End;
 759       return lltok::LabelStr;
 760     }
 761   }
 762
 763   // If the next character is a '.', then it is a fp value, otherwise its
 764   // integer.
 765   if (CurPtr[0] != '.') {
 766     if (TokStart[0] == '0' && TokStart[1] == 'x')
 767       return Lex0x();
 768     unsigned Len = CurPtr-TokStart;
 769     uint32_t numBits = ((Len * 64) / 19) + 2;
 770     APInt Tmp(numBits, TokStart, Len, 10);
 771     if (TokStart[0] == '-') {
 772       uint32_t minBits = Tmp.getMinSignedBits();
 773       if (minBits > 0 && minBits < numBits)
 774         Tmp.trunc(minBits);
 775       APSIntVal = APSInt(Tmp, false);
 776     } else {
 777       uint32_t activeBits = Tmp.getActiveBits();
 778       if (activeBits > 0 && activeBits < numBits)
 779         Tmp.trunc(activeBits);
 780       APSIntVal = APSInt(Tmp, true);
 781     }
 782     return lltok::APSInt;
 783   }
 784
 785   ++CurPtr;
 786
 787   // Skip over [0-9]*([eE][-+]?[0-9]+)?
 788   while (isdigit(CurPtr[0])) ++CurPtr;
 789
 790   if (CurPtr[0] == 'e' || CurPtr[0] == 'E') {
 791     if (isdigit(CurPtr[1]) ||
 792         ((CurPtr[1] == '-' || CurPtr[1] == '+') && isdigit(CurPtr[2]))) {
 793       CurPtr += 2;
 794       while (isdigit(CurPtr[0])) ++CurPtr;
 795     }
 796   }
 797
 798   APFloatVal = APFloat(atof(TokStart));
 799   return lltok::APFloat;
 800 }
 801
 802 ///    FPConstant  [-+]?[0-9]+[.][0-9]*([eE][-+]?[0-9]+)?
 803 lltok::Kind LLLexer::LexPositive() {
 804   // If the letter after the negative is a number, this is probably not a
 805   // label.
 806   if (!isdigit(CurPtr[0]))
 807     return lltok::Error;
 808
 809   // Skip digits.
 810   for (++CurPtr; isdigit(CurPtr[0]); ++CurPtr)
 811     /*empty*/;
 812
 813   // At this point, we need a '.'.
 814   if (CurPtr[0] != '.') {
 815     CurPtr = TokStart+1;
 816     return lltok::Error;
 817   }
 818
 819   ++CurPtr;
 820
 821   // Skip over [0-9]*([eE][-+]?[0-9]+)?
 822   while (isdigit(CurPtr[0])) ++CurPtr;
 823
 824   if (CurPtr[0] == 'e' || CurPtr[0] == 'E') {
 825     if (isdigit(CurPtr[1]) ||
 826         ((CurPtr[1] == '-' || CurPtr[1] == '+') && isdigit(CurPtr[2]))) {
 827       CurPtr += 2;
 828       while (isdigit(CurPtr[0])) ++CurPtr;
 829     }
 830   }
 831
 832   APFloatVal = APFloat(atof(TokStart));
 833   return lltok::APFloat;
 834 }