1 //===- MILexer.cpp - Machine instructions lexer implementation ------------===//
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 file implements the lexing of machine instructions.
10 //
11 //===----------------------------------------------------------------------===//
12 
13 #include "MILexer.h"
14 #include "llvm/ADT/APSInt.h"
15 #include "llvm/ADT/None.h"
16 #include "llvm/ADT/STLExtras.h"
17 #include "llvm/ADT/StringExtras.h"
18 #include "llvm/ADT/StringSwitch.h"
19 #include "llvm/ADT/StringRef.h"
20 #include "llvm/ADT/Twine.h"
21 #include <algorithm>
22 #include <cassert>
23 #include <cctype>
24 #include <string>
25 
26 using namespace llvm;
27 
28 namespace {
29 
30 using ErrorCallbackType =
31     function_ref<void(StringRef::iterator Loc, const Twine &)>;
32 
33 /// This class provides a way to iterate and get characters from the source
34 /// string.
35 class Cursor {
36   const char *Ptr = nullptr;
37   const char *End = nullptr;
38 
39 public:
40   Cursor(NoneType) {}
41 
42   explicit Cursor(StringRef Str) {
43     Ptr = Str.data();
44     End = Ptr + Str.size();
45   }
46 
47   bool isEOF() const { return Ptr == End; }
48 
49   char peek(int I = 0) const { return End - Ptr <= I ? 0 : Ptr[I]; }
50 
51   void advance(unsigned I = 1) { Ptr += I; }
52 
53   StringRef remaining() const { return StringRef(Ptr, End - Ptr); }
54 
55   StringRef upto(Cursor C) const {
56     assert(C.Ptr >= Ptr && C.Ptr <= End);
57     return StringRef(Ptr, C.Ptr - Ptr);
58   }
59 
60   StringRef::iterator location() const { return Ptr; }
61 
62   operator bool() const { return Ptr != nullptr; }
63 };
64 
65 } // end anonymous namespace
66 
67 MIToken &MIToken::reset(TokenKind Kind, StringRef Range) {
68   this->Kind = Kind;
69   this->Range = Range;
70   return *this;
71 }
72 
73 MIToken &MIToken::setStringValue(StringRef StrVal) {
74   StringValue = StrVal;
75   return *this;
76 }
77 
78 MIToken &MIToken::setOwnedStringValue(std::string StrVal) {
79   StringValueStorage = std::move(StrVal);
80   StringValue = StringValueStorage;
81   return *this;
82 }
83 
84 MIToken &MIToken::setIntegerValue(APSInt IntVal) {
85   this->IntVal = std::move(IntVal);
86   return *this;
87 }
88 
89 /// Skip the leading whitespace characters and return the updated cursor.
90 static Cursor skipWhitespace(Cursor C) {
91   while (isblank(C.peek()))
92     C.advance();
93   return C;
94 }
95 
96 static bool isNewlineChar(char C) { return C == '\n' || C == '\r'; }
97 
98 /// Skip a line comment and return the updated cursor.
99 static Cursor skipComment(Cursor C) {
100   if (C.peek() != ';')
101     return C;
102   while (!isNewlineChar(C.peek()) && !C.isEOF())
103     C.advance();
104   return C;
105 }
106 
107 /// Return true if the given character satisfies the following regular
108 /// expression: [-a-zA-Z$._0-9]
109 static bool isIdentifierChar(char C) {
110   return isalpha(C) || isdigit(C) || C == '_' || C == '-' || C == '.' ||
111          C == '$';
112 }
113 
114 /// Unescapes the given string value.
115 ///
116 /// Expects the string value to be quoted.
117 static std::string unescapeQuotedString(StringRef Value) {
118   assert(Value.front() == '"' && Value.back() == '"');
119   Cursor C = Cursor(Value.substr(1, Value.size() - 2));
120 
121   std::string Str;
122   Str.reserve(C.remaining().size());
123   while (!C.isEOF()) {
124     char Char = C.peek();
125     if (Char == '\\') {
126       if (C.peek(1) == '\\') {
127         // Two '\' become one
128         Str += '\\';
129         C.advance(2);
130         continue;
131       }
132       if (isxdigit(C.peek(1)) && isxdigit(C.peek(2))) {
133         Str += hexDigitValue(C.peek(1)) * 16 + hexDigitValue(C.peek(2));
134         C.advance(3);
135         continue;
136       }
137     }
138     Str += Char;
139     C.advance();
140   }
141   return Str;
142 }
143 
144 /// Lex a string constant using the following regular expression: \"[^\"]*\"
145 static Cursor lexStringConstant(Cursor C, ErrorCallbackType ErrorCallback) {
146   assert(C.peek() == '"');
147   for (C.advance(); C.peek() != '"'; C.advance()) {
148     if (C.isEOF() || isNewlineChar(C.peek())) {
149       ErrorCallback(
150           C.location(),
151           "end of machine instruction reached before the closing '\"'");
152       return None;
153     }
154   }
155   C.advance();
156   return C;
157 }
158 
159 static Cursor lexName(Cursor C, MIToken &Token, MIToken::TokenKind Type,
160                       unsigned PrefixLength, ErrorCallbackType ErrorCallback) {
161   auto Range = C;
162   C.advance(PrefixLength);
163   if (C.peek() == '"') {
164     if (Cursor R = lexStringConstant(C, ErrorCallback)) {
165       StringRef String = Range.upto(R);
166       Token.reset(Type, String)
167           .setOwnedStringValue(
168               unescapeQuotedString(String.drop_front(PrefixLength)));
169       return R;
170     }
171     Token.reset(MIToken::Error, Range.remaining());
172     return Range;
173   }
174   while (isIdentifierChar(C.peek()))
175     C.advance();
176   Token.reset(Type, Range.upto(C))
177       .setStringValue(Range.upto(C).drop_front(PrefixLength));
178   return C;
179 }
180 
181 static MIToken::TokenKind getIdentifierKind(StringRef Identifier) {
182   return StringSwitch<MIToken::TokenKind>(Identifier)
183       .Case("_", MIToken::underscore)
184       .Case("implicit", MIToken::kw_implicit)
185       .Case("implicit-def", MIToken::kw_implicit_define)
186       .Case("def", MIToken::kw_def)
187       .Case("dead", MIToken::kw_dead)
188       .Case("killed", MIToken::kw_killed)
189       .Case("undef", MIToken::kw_undef)
190       .Case("internal", MIToken::kw_internal)
191       .Case("early-clobber", MIToken::kw_early_clobber)
192       .Case("debug-use", MIToken::kw_debug_use)
193       .Case("renamable", MIToken::kw_renamable)
194       .Case("tied-def", MIToken::kw_tied_def)
195       .Case("frame-setup", MIToken::kw_frame_setup)
196       .Case("frame-destroy", MIToken::kw_frame_destroy)
197       .Case("nnan", MIToken::kw_nnan)
198       .Case("ninf", MIToken::kw_ninf)
199       .Case("nsz", MIToken::kw_nsz)
200       .Case("arcp", MIToken::kw_arcp)
201       .Case("contract", MIToken::kw_contract)
202       .Case("afn", MIToken::kw_afn)
203       .Case("reassoc", MIToken::kw_reassoc)
204       .Case("nuw" , MIToken::kw_nuw)
205       .Case("nsw" , MIToken::kw_nsw)
206       .Case("exact" , MIToken::kw_exact)
207       .Case("debug-location", MIToken::kw_debug_location)
208       .Case("same_value", MIToken::kw_cfi_same_value)
209       .Case("offset", MIToken::kw_cfi_offset)
210       .Case("rel_offset", MIToken::kw_cfi_rel_offset)
211       .Case("def_cfa_register", MIToken::kw_cfi_def_cfa_register)
212       .Case("def_cfa_offset", MIToken::kw_cfi_def_cfa_offset)
213       .Case("adjust_cfa_offset", MIToken::kw_cfi_adjust_cfa_offset)
214       .Case("escape", MIToken::kw_cfi_escape)
215       .Case("def_cfa", MIToken::kw_cfi_def_cfa)
216       .Case("remember_state", MIToken::kw_cfi_remember_state)
217       .Case("restore", MIToken::kw_cfi_restore)
218       .Case("restore_state", MIToken::kw_cfi_restore_state)
219       .Case("undefined", MIToken::kw_cfi_undefined)
220       .Case("register", MIToken::kw_cfi_register)
221       .Case("window_save", MIToken::kw_cfi_window_save)
222       .Case("negate_ra_sign_state", MIToken::kw_cfi_aarch64_negate_ra_sign_state)
223       .Case("blockaddress", MIToken::kw_blockaddress)
224       .Case("intrinsic", MIToken::kw_intrinsic)
225       .Case("target-index", MIToken::kw_target_index)
226       .Case("half", MIToken::kw_half)
227       .Case("float", MIToken::kw_float)
228       .Case("double", MIToken::kw_double)
229       .Case("x86_fp80", MIToken::kw_x86_fp80)
230       .Case("fp128", MIToken::kw_fp128)
231       .Case("ppc_fp128", MIToken::kw_ppc_fp128)
232       .Case("target-flags", MIToken::kw_target_flags)
233       .Case("volatile", MIToken::kw_volatile)
234       .Case("non-temporal", MIToken::kw_non_temporal)
235       .Case("dereferenceable", MIToken::kw_dereferenceable)
236       .Case("invariant", MIToken::kw_invariant)
237       .Case("align", MIToken::kw_align)
238       .Case("addrspace", MIToken::kw_addrspace)
239       .Case("stack", MIToken::kw_stack)
240       .Case("got", MIToken::kw_got)
241       .Case("jump-table", MIToken::kw_jump_table)
242       .Case("constant-pool", MIToken::kw_constant_pool)
243       .Case("call-entry", MIToken::kw_call_entry)
244       .Case("liveout", MIToken::kw_liveout)
245       .Case("address-taken", MIToken::kw_address_taken)
246       .Case("landing-pad", MIToken::kw_landing_pad)
247       .Case("liveins", MIToken::kw_liveins)
248       .Case("successors", MIToken::kw_successors)
249       .Case("floatpred", MIToken::kw_floatpred)
250       .Case("intpred", MIToken::kw_intpred)
251       .Case("pre-instr-symbol", MIToken::kw_pre_instr_symbol)
252       .Case("post-instr-symbol", MIToken::kw_post_instr_symbol)
253       .Case("unknown-size", MIToken::kw_unknown_size)
254       .Default(MIToken::Identifier);
255 }
256 
257 static Cursor maybeLexIdentifier(Cursor C, MIToken &Token) {
258   if (!isalpha(C.peek()) && C.peek() != '_')
259     return None;
260   auto Range = C;
261   while (isIdentifierChar(C.peek()))
262     C.advance();
263   auto Identifier = Range.upto(C);
264   Token.reset(getIdentifierKind(Identifier), Identifier)
265       .setStringValue(Identifier);
266   return C;
267 }
268 
269 static Cursor maybeLexMachineBasicBlock(Cursor C, MIToken &Token,
270                                         ErrorCallbackType ErrorCallback) {
271   bool IsReference = C.remaining().startswith("%bb.");
272   if (!IsReference && !C.remaining().startswith("bb."))
273     return None;
274   auto Range = C;
275   unsigned PrefixLength = IsReference ? 4 : 3;
276   C.advance(PrefixLength); // Skip '%bb.' or 'bb.'
277   if (!isdigit(C.peek())) {
278     Token.reset(MIToken::Error, C.remaining());
279     ErrorCallback(C.location(), "expected a number after '%bb.'");
280     return C;
281   }
282   auto NumberRange = C;
283   while (isdigit(C.peek()))
284     C.advance();
285   StringRef Number = NumberRange.upto(C);
286   unsigned StringOffset = PrefixLength + Number.size(); // Drop '%bb.<id>'
287   // TODO: The format bb.<id>.<irname> is supported only when it's not a
288   // reference. Once we deprecate the format where the irname shows up, we
289   // should only lex forward if it is a reference.
290   if (C.peek() == '.') {
291     C.advance(); // Skip '.'
292     ++StringOffset;
293     while (isIdentifierChar(C.peek()))
294       C.advance();
295   }
296   Token.reset(IsReference ? MIToken::MachineBasicBlock
297                           : MIToken::MachineBasicBlockLabel,
298               Range.upto(C))
299       .setIntegerValue(APSInt(Number))
300       .setStringValue(Range.upto(C).drop_front(StringOffset));
301   return C;
302 }
303 
304 static Cursor maybeLexIndex(Cursor C, MIToken &Token, StringRef Rule,
305                             MIToken::TokenKind Kind) {
306   if (!C.remaining().startswith(Rule) || !isdigit(C.peek(Rule.size())))
307     return None;
308   auto Range = C;
309   C.advance(Rule.size());
310   auto NumberRange = C;
311   while (isdigit(C.peek()))
312     C.advance();
313   Token.reset(Kind, Range.upto(C)).setIntegerValue(APSInt(NumberRange.upto(C)));
314   return C;
315 }
316 
317 static Cursor maybeLexIndexAndName(Cursor C, MIToken &Token, StringRef Rule,
318                                    MIToken::TokenKind Kind) {
319   if (!C.remaining().startswith(Rule) || !isdigit(C.peek(Rule.size())))
320     return None;
321   auto Range = C;
322   C.advance(Rule.size());
323   auto NumberRange = C;
324   while (isdigit(C.peek()))
325     C.advance();
326   StringRef Number = NumberRange.upto(C);
327   unsigned StringOffset = Rule.size() + Number.size();
328   if (C.peek() == '.') {
329     C.advance();
330     ++StringOffset;
331     while (isIdentifierChar(C.peek()))
332       C.advance();
333   }
334   Token.reset(Kind, Range.upto(C))
335       .setIntegerValue(APSInt(Number))
336       .setStringValue(Range.upto(C).drop_front(StringOffset));
337   return C;
338 }
339 
340 static Cursor maybeLexJumpTableIndex(Cursor C, MIToken &Token) {
341   return maybeLexIndex(C, Token, "%jump-table.", MIToken::JumpTableIndex);
342 }
343 
344 static Cursor maybeLexStackObject(Cursor C, MIToken &Token) {
345   return maybeLexIndexAndName(C, Token, "%stack.", MIToken::StackObject);
346 }
347 
348 static Cursor maybeLexFixedStackObject(Cursor C, MIToken &Token) {
349   return maybeLexIndex(C, Token, "%fixed-stack.", MIToken::FixedStackObject);
350 }
351 
352 static Cursor maybeLexConstantPoolItem(Cursor C, MIToken &Token) {
353   return maybeLexIndex(C, Token, "%const.", MIToken::ConstantPoolItem);
354 }
355 
356 static Cursor maybeLexSubRegisterIndex(Cursor C, MIToken &Token,
357                                        ErrorCallbackType ErrorCallback) {
358   const StringRef Rule = "%subreg.";
359   if (!C.remaining().startswith(Rule))
360     return None;
361   return lexName(C, Token, MIToken::SubRegisterIndex, Rule.size(),
362                  ErrorCallback);
363 }
364 
365 static Cursor maybeLexIRBlock(Cursor C, MIToken &Token,
366                               ErrorCallbackType ErrorCallback) {
367   const StringRef Rule = "%ir-block.";
368   if (!C.remaining().startswith(Rule))
369     return None;
370   if (isdigit(C.peek(Rule.size())))
371     return maybeLexIndex(C, Token, Rule, MIToken::IRBlock);
372   return lexName(C, Token, MIToken::NamedIRBlock, Rule.size(), ErrorCallback);
373 }
374 
375 static Cursor maybeLexIRValue(Cursor C, MIToken &Token,
376                               ErrorCallbackType ErrorCallback) {
377   const StringRef Rule = "%ir.";
378   if (!C.remaining().startswith(Rule))
379     return None;
380   if (isdigit(C.peek(Rule.size())))
381     return maybeLexIndex(C, Token, Rule, MIToken::IRValue);
382   return lexName(C, Token, MIToken::NamedIRValue, Rule.size(), ErrorCallback);
383 }
384 
385 static Cursor maybeLexStringConstant(Cursor C, MIToken &Token,
386                                      ErrorCallbackType ErrorCallback) {
387   if (C.peek() != '"')
388     return None;
389   return lexName(C, Token, MIToken::StringConstant, /*PrefixLength=*/0,
390                  ErrorCallback);
391 }
392 
393 static Cursor lexVirtualRegister(Cursor C, MIToken &Token) {
394   auto Range = C;
395   C.advance(); // Skip '%'
396   auto NumberRange = C;
397   while (isdigit(C.peek()))
398     C.advance();
399   Token.reset(MIToken::VirtualRegister, Range.upto(C))
400       .setIntegerValue(APSInt(NumberRange.upto(C)));
401   return C;
402 }
403 
404 /// Returns true for a character allowed in a register name.
405 static bool isRegisterChar(char C) {
406   return isIdentifierChar(C) && C != '.';
407 }
408 
409 static Cursor lexNamedVirtualRegister(Cursor C, MIToken &Token) {
410   Cursor Range = C;
411   C.advance(); // Skip '%'
412   while (isRegisterChar(C.peek()))
413     C.advance();
414   Token.reset(MIToken::NamedVirtualRegister, Range.upto(C))
415       .setStringValue(Range.upto(C).drop_front(1)); // Drop the '%'
416   return C;
417 }
418 
419 static Cursor maybeLexRegister(Cursor C, MIToken &Token,
420                                ErrorCallbackType ErrorCallback) {
421   if (C.peek() != '%' && C.peek() != '$')
422     return None;
423 
424   if (C.peek() == '%') {
425     if (isdigit(C.peek(1)))
426       return lexVirtualRegister(C, Token);
427 
428     if (isRegisterChar(C.peek(1)))
429       return lexNamedVirtualRegister(C, Token);
430 
431     return None;
432   }
433 
434   assert(C.peek() == '$');
435   auto Range = C;
436   C.advance(); // Skip '$'
437   while (isRegisterChar(C.peek()))
438     C.advance();
439   Token.reset(MIToken::NamedRegister, Range.upto(C))
440       .setStringValue(Range.upto(C).drop_front(1)); // Drop the '$'
441   return C;
442 }
443 
444 static Cursor maybeLexGlobalValue(Cursor C, MIToken &Token,
445                                   ErrorCallbackType ErrorCallback) {
446   if (C.peek() != '@')
447     return None;
448   if (!isdigit(C.peek(1)))
449     return lexName(C, Token, MIToken::NamedGlobalValue, /*PrefixLength=*/1,
450                    ErrorCallback);
451   auto Range = C;
452   C.advance(1); // Skip the '@'
453   auto NumberRange = C;
454   while (isdigit(C.peek()))
455     C.advance();
456   Token.reset(MIToken::GlobalValue, Range.upto(C))
457       .setIntegerValue(APSInt(NumberRange.upto(C)));
458   return C;
459 }
460 
461 static Cursor maybeLexExternalSymbol(Cursor C, MIToken &Token,
462                                      ErrorCallbackType ErrorCallback) {
463   if (C.peek() != '&')
464     return None;
465   return lexName(C, Token, MIToken::ExternalSymbol, /*PrefixLength=*/1,
466                  ErrorCallback);
467 }
468 
469 static Cursor maybeLexMCSymbol(Cursor C, MIToken &Token,
470                                ErrorCallbackType ErrorCallback) {
471   const StringRef Rule = "<mcsymbol ";
472   if (!C.remaining().startswith(Rule))
473     return None;
474   auto Start = C;
475   C.advance(Rule.size());
476 
477   // Try a simple unquoted name.
478   if (C.peek() != '"') {
479     while (isIdentifierChar(C.peek()))
480       C.advance();
481     StringRef String = Start.upto(C).drop_front(Rule.size());
482     if (C.peek() != '>') {
483       ErrorCallback(C.location(),
484                     "expected the '<mcsymbol ...' to be closed by a '>'");
485       Token.reset(MIToken::Error, Start.remaining());
486       return Start;
487     }
488     C.advance();
489 
490     Token.reset(MIToken::MCSymbol, Start.upto(C)).setStringValue(String);
491     return C;
492   }
493 
494   // Otherwise lex out a quoted name.
495   Cursor R = lexStringConstant(C, ErrorCallback);
496   if (!R) {
497     ErrorCallback(C.location(),
498                   "unable to parse quoted string from opening quote");
499     Token.reset(MIToken::Error, Start.remaining());
500     return Start;
501   }
502   StringRef String = Start.upto(R).drop_front(Rule.size());
503   if (R.peek() != '>') {
504     ErrorCallback(R.location(),
505                   "expected the '<mcsymbol ...' to be closed by a '>'");
506     Token.reset(MIToken::Error, Start.remaining());
507     return Start;
508   }
509   R.advance();
510 
511   Token.reset(MIToken::MCSymbol, Start.upto(R))
512       .setOwnedStringValue(unescapeQuotedString(String));
513   return R;
514 }
515 
516 static bool isValidHexFloatingPointPrefix(char C) {
517   return C == 'H' || C == 'K' || C == 'L' || C == 'M';
518 }
519 
520 static Cursor lexFloatingPointLiteral(Cursor Range, Cursor C, MIToken &Token) {
521   C.advance();
522   // Skip over [0-9]*([eE][-+]?[0-9]+)?
523   while (isdigit(C.peek()))
524     C.advance();
525   if ((C.peek() == 'e' || C.peek() == 'E') &&
526       (isdigit(C.peek(1)) ||
527        ((C.peek(1) == '-' || C.peek(1) == '+') && isdigit(C.peek(2))))) {
528     C.advance(2);
529     while (isdigit(C.peek()))
530       C.advance();
531   }
532   Token.reset(MIToken::FloatingPointLiteral, Range.upto(C));
533   return C;
534 }
535 
536 static Cursor maybeLexHexadecimalLiteral(Cursor C, MIToken &Token) {
537   if (C.peek() != '0' || (C.peek(1) != 'x' && C.peek(1) != 'X'))
538     return None;
539   Cursor Range = C;
540   C.advance(2);
541   unsigned PrefLen = 2;
542   if (isValidHexFloatingPointPrefix(C.peek())) {
543     C.advance();
544     PrefLen++;
545   }
546   while (isxdigit(C.peek()))
547     C.advance();
548   StringRef StrVal = Range.upto(C);
549   if (StrVal.size() <= PrefLen)
550     return None;
551   if (PrefLen == 2)
552     Token.reset(MIToken::HexLiteral, Range.upto(C));
553   else // It must be 3, which means that there was a floating-point prefix.
554     Token.reset(MIToken::FloatingPointLiteral, Range.upto(C));
555   return C;
556 }
557 
558 static Cursor maybeLexNumericalLiteral(Cursor C, MIToken &Token) {
559   if (!isdigit(C.peek()) && (C.peek() != '-' || !isdigit(C.peek(1))))
560     return None;
561   auto Range = C;
562   C.advance();
563   while (isdigit(C.peek()))
564     C.advance();
565   if (C.peek() == '.')
566     return lexFloatingPointLiteral(Range, C, Token);
567   StringRef StrVal = Range.upto(C);
568   Token.reset(MIToken::IntegerLiteral, StrVal).setIntegerValue(APSInt(StrVal));
569   return C;
570 }
571 
572 static MIToken::TokenKind getMetadataKeywordKind(StringRef Identifier) {
573   return StringSwitch<MIToken::TokenKind>(Identifier)
574       .Case("!tbaa", MIToken::md_tbaa)
575       .Case("!alias.scope", MIToken::md_alias_scope)
576       .Case("!noalias", MIToken::md_noalias)
577       .Case("!range", MIToken::md_range)
578       .Case("!DIExpression", MIToken::md_diexpr)
579       .Case("!DILocation", MIToken::md_dilocation)
580       .Default(MIToken::Error);
581 }
582 
583 static Cursor maybeLexExlaim(Cursor C, MIToken &Token,
584                              ErrorCallbackType ErrorCallback) {
585   if (C.peek() != '!')
586     return None;
587   auto Range = C;
588   C.advance(1);
589   if (isdigit(C.peek()) || !isIdentifierChar(C.peek())) {
590     Token.reset(MIToken::exclaim, Range.upto(C));
591     return C;
592   }
593   while (isIdentifierChar(C.peek()))
594     C.advance();
595   StringRef StrVal = Range.upto(C);
596   Token.reset(getMetadataKeywordKind(StrVal), StrVal);
597   if (Token.isError())
598     ErrorCallback(Token.location(),
599                   "use of unknown metadata keyword '" + StrVal + "'");
600   return C;
601 }
602 
603 static MIToken::TokenKind symbolToken(char C) {
604   switch (C) {
605   case ',':
606     return MIToken::comma;
607   case '.':
608     return MIToken::dot;
609   case '=':
610     return MIToken::equal;
611   case ':':
612     return MIToken::colon;
613   case '(':
614     return MIToken::lparen;
615   case ')':
616     return MIToken::rparen;
617   case '{':
618     return MIToken::lbrace;
619   case '}':
620     return MIToken::rbrace;
621   case '+':
622     return MIToken::plus;
623   case '-':
624     return MIToken::minus;
625   case '<':
626     return MIToken::less;
627   case '>':
628     return MIToken::greater;
629   default:
630     return MIToken::Error;
631   }
632 }
633 
634 static Cursor maybeLexSymbol(Cursor C, MIToken &Token) {
635   MIToken::TokenKind Kind;
636   unsigned Length = 1;
637   if (C.peek() == ':' && C.peek(1) == ':') {
638     Kind = MIToken::coloncolon;
639     Length = 2;
640   } else
641     Kind = symbolToken(C.peek());
642   if (Kind == MIToken::Error)
643     return None;
644   auto Range = C;
645   C.advance(Length);
646   Token.reset(Kind, Range.upto(C));
647   return C;
648 }
649 
650 static Cursor maybeLexNewline(Cursor C, MIToken &Token) {
651   if (!isNewlineChar(C.peek()))
652     return None;
653   auto Range = C;
654   C.advance();
655   Token.reset(MIToken::Newline, Range.upto(C));
656   return C;
657 }
658 
659 static Cursor maybeLexEscapedIRValue(Cursor C, MIToken &Token,
660                                      ErrorCallbackType ErrorCallback) {
661   if (C.peek() != '`')
662     return None;
663   auto Range = C;
664   C.advance();
665   auto StrRange = C;
666   while (C.peek() != '`') {
667     if (C.isEOF() || isNewlineChar(C.peek())) {
668       ErrorCallback(
669           C.location(),
670           "end of machine instruction reached before the closing '`'");
671       Token.reset(MIToken::Error, Range.remaining());
672       return C;
673     }
674     C.advance();
675   }
676   StringRef Value = StrRange.upto(C);
677   C.advance();
678   Token.reset(MIToken::QuotedIRValue, Range.upto(C)).setStringValue(Value);
679   return C;
680 }
681 
682 StringRef llvm::lexMIToken(StringRef Source, MIToken &Token,
683                            ErrorCallbackType ErrorCallback) {
684   auto C = skipComment(skipWhitespace(Cursor(Source)));
685   if (C.isEOF()) {
686     Token.reset(MIToken::Eof, C.remaining());
687     return C.remaining();
688   }
689 
690   if (Cursor R = maybeLexMachineBasicBlock(C, Token, ErrorCallback))
691     return R.remaining();
692   if (Cursor R = maybeLexIdentifier(C, Token))
693     return R.remaining();
694   if (Cursor R = maybeLexJumpTableIndex(C, Token))
695     return R.remaining();
696   if (Cursor R = maybeLexStackObject(C, Token))
697     return R.remaining();
698   if (Cursor R = maybeLexFixedStackObject(C, Token))
699     return R.remaining();
700   if (Cursor R = maybeLexConstantPoolItem(C, Token))
701     return R.remaining();
702   if (Cursor R = maybeLexSubRegisterIndex(C, Token, ErrorCallback))
703     return R.remaining();
704   if (Cursor R = maybeLexIRBlock(C, Token, ErrorCallback))
705     return R.remaining();
706   if (Cursor R = maybeLexIRValue(C, Token, ErrorCallback))
707     return R.remaining();
708   if (Cursor R = maybeLexRegister(C, Token, ErrorCallback))
709     return R.remaining();
710   if (Cursor R = maybeLexGlobalValue(C, Token, ErrorCallback))
711     return R.remaining();
712   if (Cursor R = maybeLexExternalSymbol(C, Token, ErrorCallback))
713     return R.remaining();
714   if (Cursor R = maybeLexMCSymbol(C, Token, ErrorCallback))
715     return R.remaining();
716   if (Cursor R = maybeLexHexadecimalLiteral(C, Token))
717     return R.remaining();
718   if (Cursor R = maybeLexNumericalLiteral(C, Token))
719     return R.remaining();
720   if (Cursor R = maybeLexExlaim(C, Token, ErrorCallback))
721     return R.remaining();
722   if (Cursor R = maybeLexSymbol(C, Token))
723     return R.remaining();
724   if (Cursor R = maybeLexNewline(C, Token))
725     return R.remaining();
726   if (Cursor R = maybeLexEscapedIRValue(C, Token, ErrorCallback))
727     return R.remaining();
728   if (Cursor R = maybeLexStringConstant(C, Token, ErrorCallback))
729     return R.remaining();
730 
731   Token.reset(MIToken::Error, C.remaining());
732   ErrorCallback(C.location(),
733                 Twine("unexpected character '") + Twine(C.peek()) + "'");
734   return C.remaining();
735 }
736