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/None.h"
15 #include "llvm/ADT/StringExtras.h"
16 #include "llvm/ADT/StringSwitch.h"
17 #include "llvm/ADT/Twine.h"
18 #include <cassert>
19 #include <cctype>
20 #include <string>
21 
22 using namespace llvm;
23 
24 namespace {
25 
26 using ErrorCallbackType =
27     function_ref<void(StringRef::iterator Loc, const Twine &)>;
28 
29 /// This class provides a way to iterate and get characters from the source
30 /// string.
31 class Cursor {
32   const char *Ptr = nullptr;
33   const char *End = nullptr;
34 
35 public:
36   Cursor(NoneType) {}
37 
38   explicit Cursor(StringRef Str) {
39     Ptr = Str.data();
40     End = Ptr + Str.size();
41   }
42 
43   bool isEOF() const { return Ptr == End; }
44 
45   char peek(int I = 0) const { return End - Ptr <= I ? 0 : Ptr[I]; }
46 
47   void advance(unsigned I = 1) { Ptr += I; }
48 
49   StringRef remaining() const { return StringRef(Ptr, End - Ptr); }
50 
51   StringRef upto(Cursor C) const {
52     assert(C.Ptr >= Ptr && C.Ptr <= End);
53     return StringRef(Ptr, C.Ptr - Ptr);
54   }
55 
56   StringRef::iterator location() const { return Ptr; }
57 
58   operator bool() const { return Ptr != nullptr; }
59 };
60 
61 } // end anonymous namespace
62 
63 MIToken &MIToken::reset(TokenKind Kind, StringRef Range) {
64   this->Kind = Kind;
65   this->Range = Range;
66   return *this;
67 }
68 
69 MIToken &MIToken::setStringValue(StringRef StrVal) {
70   StringValue = StrVal;
71   return *this;
72 }
73 
74 MIToken &MIToken::setOwnedStringValue(std::string StrVal) {
75   StringValueStorage = std::move(StrVal);
76   StringValue = StringValueStorage;
77   return *this;
78 }
79 
80 MIToken &MIToken::setIntegerValue(APSInt IntVal) {
81   this->IntVal = std::move(IntVal);
82   return *this;
83 }
84 
85 /// Skip the leading whitespace characters and return the updated cursor.
86 static Cursor skipWhitespace(Cursor C) {
87   while (isblank(C.peek()))
88     C.advance();
89   return C;
90 }
91 
92 static bool isNewlineChar(char C) { return C == '\n' || C == '\r'; }
93 
94 /// Skip a line comment and return the updated cursor.
95 static Cursor skipComment(Cursor C) {
96   if (C.peek() != ';')
97     return C;
98   while (!isNewlineChar(C.peek()) && !C.isEOF())
99     C.advance();
100   return C;
101 }
102 
103 /// Machine operands can have comments, enclosed between /* and */.
104 /// This eats up all tokens, including /* and */.
105 static Cursor skipMachineOperandComment(Cursor C) {
106   if (C.peek() != '/' || C.peek(1) != '*')
107     return C;
108 
109   while (C.peek() != '*' || C.peek(1) != '/')
110     C.advance();
111 
112   C.advance();
113   C.advance();
114   return C;
115 }
116 
117 /// Return true if the given character satisfies the following regular
118 /// expression: [-a-zA-Z$._0-9]
119 static bool isIdentifierChar(char C) {
120   return isalpha(C) || isdigit(C) || C == '_' || C == '-' || C == '.' ||
121          C == '$';
122 }
123 
124 /// Unescapes the given string value.
125 ///
126 /// Expects the string value to be quoted.
127 static std::string unescapeQuotedString(StringRef Value) {
128   assert(Value.front() == '"' && Value.back() == '"');
129   Cursor C = Cursor(Value.substr(1, Value.size() - 2));
130 
131   std::string Str;
132   Str.reserve(C.remaining().size());
133   while (!C.isEOF()) {
134     char Char = C.peek();
135     if (Char == '\\') {
136       if (C.peek(1) == '\\') {
137         // Two '\' become one
138         Str += '\\';
139         C.advance(2);
140         continue;
141       }
142       if (isxdigit(C.peek(1)) && isxdigit(C.peek(2))) {
143         Str += hexDigitValue(C.peek(1)) * 16 + hexDigitValue(C.peek(2));
144         C.advance(3);
145         continue;
146       }
147     }
148     Str += Char;
149     C.advance();
150   }
151   return Str;
152 }
153 
154 /// Lex a string constant using the following regular expression: \"[^\"]*\"
155 static Cursor lexStringConstant(Cursor C, ErrorCallbackType ErrorCallback) {
156   assert(C.peek() == '"');
157   for (C.advance(); C.peek() != '"'; C.advance()) {
158     if (C.isEOF() || isNewlineChar(C.peek())) {
159       ErrorCallback(
160           C.location(),
161           "end of machine instruction reached before the closing '\"'");
162       return None;
163     }
164   }
165   C.advance();
166   return C;
167 }
168 
169 static Cursor lexName(Cursor C, MIToken &Token, MIToken::TokenKind Type,
170                       unsigned PrefixLength, ErrorCallbackType ErrorCallback) {
171   auto Range = C;
172   C.advance(PrefixLength);
173   if (C.peek() == '"') {
174     if (Cursor R = lexStringConstant(C, ErrorCallback)) {
175       StringRef String = Range.upto(R);
176       Token.reset(Type, String)
177           .setOwnedStringValue(
178               unescapeQuotedString(String.drop_front(PrefixLength)));
179       return R;
180     }
181     Token.reset(MIToken::Error, Range.remaining());
182     return Range;
183   }
184   while (isIdentifierChar(C.peek()))
185     C.advance();
186   Token.reset(Type, Range.upto(C))
187       .setStringValue(Range.upto(C).drop_front(PrefixLength));
188   return C;
189 }
190 
191 static MIToken::TokenKind getIdentifierKind(StringRef Identifier) {
192   return StringSwitch<MIToken::TokenKind>(Identifier)
193       .Case("_", MIToken::underscore)
194       .Case("implicit", MIToken::kw_implicit)
195       .Case("implicit-def", MIToken::kw_implicit_define)
196       .Case("def", MIToken::kw_def)
197       .Case("dead", MIToken::kw_dead)
198       .Case("killed", MIToken::kw_killed)
199       .Case("undef", MIToken::kw_undef)
200       .Case("internal", MIToken::kw_internal)
201       .Case("early-clobber", MIToken::kw_early_clobber)
202       .Case("debug-use", MIToken::kw_debug_use)
203       .Case("renamable", MIToken::kw_renamable)
204       .Case("tied-def", MIToken::kw_tied_def)
205       .Case("frame-setup", MIToken::kw_frame_setup)
206       .Case("frame-destroy", MIToken::kw_frame_destroy)
207       .Case("nnan", MIToken::kw_nnan)
208       .Case("ninf", MIToken::kw_ninf)
209       .Case("nsz", MIToken::kw_nsz)
210       .Case("arcp", MIToken::kw_arcp)
211       .Case("contract", MIToken::kw_contract)
212       .Case("afn", MIToken::kw_afn)
213       .Case("reassoc", MIToken::kw_reassoc)
214       .Case("nuw", MIToken::kw_nuw)
215       .Case("nsw", MIToken::kw_nsw)
216       .Case("exact", MIToken::kw_exact)
217       .Case("nofpexcept", MIToken::kw_nofpexcept)
218       .Case("debug-location", MIToken::kw_debug_location)
219       .Case("debug-instr-number", MIToken::kw_debug_instr_number)
220       .Case("same_value", MIToken::kw_cfi_same_value)
221       .Case("offset", MIToken::kw_cfi_offset)
222       .Case("rel_offset", MIToken::kw_cfi_rel_offset)
223       .Case("def_cfa_register", MIToken::kw_cfi_def_cfa_register)
224       .Case("def_cfa_offset", MIToken::kw_cfi_def_cfa_offset)
225       .Case("adjust_cfa_offset", MIToken::kw_cfi_adjust_cfa_offset)
226       .Case("escape", MIToken::kw_cfi_escape)
227       .Case("def_cfa", MIToken::kw_cfi_def_cfa)
228       .Case("llvm_def_aspace_cfa", MIToken::kw_cfi_llvm_def_aspace_cfa)
229       .Case("remember_state", MIToken::kw_cfi_remember_state)
230       .Case("restore", MIToken::kw_cfi_restore)
231       .Case("restore_state", MIToken::kw_cfi_restore_state)
232       .Case("undefined", MIToken::kw_cfi_undefined)
233       .Case("register", MIToken::kw_cfi_register)
234       .Case("window_save", MIToken::kw_cfi_window_save)
235       .Case("negate_ra_sign_state",
236             MIToken::kw_cfi_aarch64_negate_ra_sign_state)
237       .Case("blockaddress", MIToken::kw_blockaddress)
238       .Case("intrinsic", MIToken::kw_intrinsic)
239       .Case("target-index", MIToken::kw_target_index)
240       .Case("half", MIToken::kw_half)
241       .Case("float", MIToken::kw_float)
242       .Case("double", MIToken::kw_double)
243       .Case("x86_fp80", MIToken::kw_x86_fp80)
244       .Case("fp128", MIToken::kw_fp128)
245       .Case("ppc_fp128", MIToken::kw_ppc_fp128)
246       .Case("target-flags", MIToken::kw_target_flags)
247       .Case("volatile", MIToken::kw_volatile)
248       .Case("non-temporal", MIToken::kw_non_temporal)
249       .Case("dereferenceable", MIToken::kw_dereferenceable)
250       .Case("invariant", MIToken::kw_invariant)
251       .Case("align", MIToken::kw_align)
252       .Case("basealign", MIToken::kw_basealign)
253       .Case("addrspace", MIToken::kw_addrspace)
254       .Case("stack", MIToken::kw_stack)
255       .Case("got", MIToken::kw_got)
256       .Case("jump-table", MIToken::kw_jump_table)
257       .Case("constant-pool", MIToken::kw_constant_pool)
258       .Case("call-entry", MIToken::kw_call_entry)
259       .Case("custom", MIToken::kw_custom)
260       .Case("liveout", MIToken::kw_liveout)
261       .Case("address-taken", MIToken::kw_address_taken)
262       .Case("landing-pad", MIToken::kw_landing_pad)
263       .Case("inlineasm-br-indirect-target",
264             MIToken::kw_inlineasm_br_indirect_target)
265       .Case("ehfunclet-entry", MIToken::kw_ehfunclet_entry)
266       .Case("liveins", MIToken::kw_liveins)
267       .Case("successors", MIToken::kw_successors)
268       .Case("floatpred", MIToken::kw_floatpred)
269       .Case("intpred", MIToken::kw_intpred)
270       .Case("shufflemask", MIToken::kw_shufflemask)
271       .Case("pre-instr-symbol", MIToken::kw_pre_instr_symbol)
272       .Case("post-instr-symbol", MIToken::kw_post_instr_symbol)
273       .Case("heap-alloc-marker", MIToken::kw_heap_alloc_marker)
274       .Case("bbsections", MIToken::kw_bbsections)
275       .Case("unknown-size", MIToken::kw_unknown_size)
276       .Case("unknown-address", MIToken::kw_unknown_address)
277       .Case("distinct", MIToken::kw_distinct)
278       .Default(MIToken::Identifier);
279 }
280 
281 static Cursor maybeLexIdentifier(Cursor C, MIToken &Token) {
282   if (!isalpha(C.peek()) && C.peek() != '_')
283     return None;
284   auto Range = C;
285   while (isIdentifierChar(C.peek()))
286     C.advance();
287   auto Identifier = Range.upto(C);
288   Token.reset(getIdentifierKind(Identifier), Identifier)
289       .setStringValue(Identifier);
290   return C;
291 }
292 
293 static Cursor maybeLexMachineBasicBlock(Cursor C, MIToken &Token,
294                                         ErrorCallbackType ErrorCallback) {
295   bool IsReference = C.remaining().startswith("%bb.");
296   if (!IsReference && !C.remaining().startswith("bb."))
297     return None;
298   auto Range = C;
299   unsigned PrefixLength = IsReference ? 4 : 3;
300   C.advance(PrefixLength); // Skip '%bb.' or 'bb.'
301   if (!isdigit(C.peek())) {
302     Token.reset(MIToken::Error, C.remaining());
303     ErrorCallback(C.location(), "expected a number after '%bb.'");
304     return C;
305   }
306   auto NumberRange = C;
307   while (isdigit(C.peek()))
308     C.advance();
309   StringRef Number = NumberRange.upto(C);
310   unsigned StringOffset = PrefixLength + Number.size(); // Drop '%bb.<id>'
311   // TODO: The format bb.<id>.<irname> is supported only when it's not a
312   // reference. Once we deprecate the format where the irname shows up, we
313   // should only lex forward if it is a reference.
314   if (C.peek() == '.') {
315     C.advance(); // Skip '.'
316     ++StringOffset;
317     while (isIdentifierChar(C.peek()))
318       C.advance();
319   }
320   Token.reset(IsReference ? MIToken::MachineBasicBlock
321                           : MIToken::MachineBasicBlockLabel,
322               Range.upto(C))
323       .setIntegerValue(APSInt(Number))
324       .setStringValue(Range.upto(C).drop_front(StringOffset));
325   return C;
326 }
327 
328 static Cursor maybeLexIndex(Cursor C, MIToken &Token, StringRef Rule,
329                             MIToken::TokenKind Kind) {
330   if (!C.remaining().startswith(Rule) || !isdigit(C.peek(Rule.size())))
331     return None;
332   auto Range = C;
333   C.advance(Rule.size());
334   auto NumberRange = C;
335   while (isdigit(C.peek()))
336     C.advance();
337   Token.reset(Kind, Range.upto(C)).setIntegerValue(APSInt(NumberRange.upto(C)));
338   return C;
339 }
340 
341 static Cursor maybeLexIndexAndName(Cursor C, MIToken &Token, StringRef Rule,
342                                    MIToken::TokenKind Kind) {
343   if (!C.remaining().startswith(Rule) || !isdigit(C.peek(Rule.size())))
344     return None;
345   auto Range = C;
346   C.advance(Rule.size());
347   auto NumberRange = C;
348   while (isdigit(C.peek()))
349     C.advance();
350   StringRef Number = NumberRange.upto(C);
351   unsigned StringOffset = Rule.size() + Number.size();
352   if (C.peek() == '.') {
353     C.advance();
354     ++StringOffset;
355     while (isIdentifierChar(C.peek()))
356       C.advance();
357   }
358   Token.reset(Kind, Range.upto(C))
359       .setIntegerValue(APSInt(Number))
360       .setStringValue(Range.upto(C).drop_front(StringOffset));
361   return C;
362 }
363 
364 static Cursor maybeLexJumpTableIndex(Cursor C, MIToken &Token) {
365   return maybeLexIndex(C, Token, "%jump-table.", MIToken::JumpTableIndex);
366 }
367 
368 static Cursor maybeLexStackObject(Cursor C, MIToken &Token) {
369   return maybeLexIndexAndName(C, Token, "%stack.", MIToken::StackObject);
370 }
371 
372 static Cursor maybeLexFixedStackObject(Cursor C, MIToken &Token) {
373   return maybeLexIndex(C, Token, "%fixed-stack.", MIToken::FixedStackObject);
374 }
375 
376 static Cursor maybeLexConstantPoolItem(Cursor C, MIToken &Token) {
377   return maybeLexIndex(C, Token, "%const.", MIToken::ConstantPoolItem);
378 }
379 
380 static Cursor maybeLexSubRegisterIndex(Cursor C, MIToken &Token,
381                                        ErrorCallbackType ErrorCallback) {
382   const StringRef Rule = "%subreg.";
383   if (!C.remaining().startswith(Rule))
384     return None;
385   return lexName(C, Token, MIToken::SubRegisterIndex, Rule.size(),
386                  ErrorCallback);
387 }
388 
389 static Cursor maybeLexIRBlock(Cursor C, MIToken &Token,
390                               ErrorCallbackType ErrorCallback) {
391   const StringRef Rule = "%ir-block.";
392   if (!C.remaining().startswith(Rule))
393     return None;
394   if (isdigit(C.peek(Rule.size())))
395     return maybeLexIndex(C, Token, Rule, MIToken::IRBlock);
396   return lexName(C, Token, MIToken::NamedIRBlock, Rule.size(), ErrorCallback);
397 }
398 
399 static Cursor maybeLexIRValue(Cursor C, MIToken &Token,
400                               ErrorCallbackType ErrorCallback) {
401   const StringRef Rule = "%ir.";
402   if (!C.remaining().startswith(Rule))
403     return None;
404   if (isdigit(C.peek(Rule.size())))
405     return maybeLexIndex(C, Token, Rule, MIToken::IRValue);
406   return lexName(C, Token, MIToken::NamedIRValue, Rule.size(), ErrorCallback);
407 }
408 
409 static Cursor maybeLexStringConstant(Cursor C, MIToken &Token,
410                                      ErrorCallbackType ErrorCallback) {
411   if (C.peek() != '"')
412     return None;
413   return lexName(C, Token, MIToken::StringConstant, /*PrefixLength=*/0,
414                  ErrorCallback);
415 }
416 
417 static Cursor lexVirtualRegister(Cursor C, MIToken &Token) {
418   auto Range = C;
419   C.advance(); // Skip '%'
420   auto NumberRange = C;
421   while (isdigit(C.peek()))
422     C.advance();
423   Token.reset(MIToken::VirtualRegister, Range.upto(C))
424       .setIntegerValue(APSInt(NumberRange.upto(C)));
425   return C;
426 }
427 
428 /// Returns true for a character allowed in a register name.
429 static bool isRegisterChar(char C) {
430   return isIdentifierChar(C) && C != '.';
431 }
432 
433 static Cursor lexNamedVirtualRegister(Cursor C, MIToken &Token) {
434   Cursor Range = C;
435   C.advance(); // Skip '%'
436   while (isRegisterChar(C.peek()))
437     C.advance();
438   Token.reset(MIToken::NamedVirtualRegister, Range.upto(C))
439       .setStringValue(Range.upto(C).drop_front(1)); // Drop the '%'
440   return C;
441 }
442 
443 static Cursor maybeLexRegister(Cursor C, MIToken &Token,
444                                ErrorCallbackType ErrorCallback) {
445   if (C.peek() != '%' && C.peek() != '$')
446     return None;
447 
448   if (C.peek() == '%') {
449     if (isdigit(C.peek(1)))
450       return lexVirtualRegister(C, Token);
451 
452     if (isRegisterChar(C.peek(1)))
453       return lexNamedVirtualRegister(C, Token);
454 
455     return None;
456   }
457 
458   assert(C.peek() == '$');
459   auto Range = C;
460   C.advance(); // Skip '$'
461   while (isRegisterChar(C.peek()))
462     C.advance();
463   Token.reset(MIToken::NamedRegister, Range.upto(C))
464       .setStringValue(Range.upto(C).drop_front(1)); // Drop the '$'
465   return C;
466 }
467 
468 static Cursor maybeLexGlobalValue(Cursor C, MIToken &Token,
469                                   ErrorCallbackType ErrorCallback) {
470   if (C.peek() != '@')
471     return None;
472   if (!isdigit(C.peek(1)))
473     return lexName(C, Token, MIToken::NamedGlobalValue, /*PrefixLength=*/1,
474                    ErrorCallback);
475   auto Range = C;
476   C.advance(1); // Skip the '@'
477   auto NumberRange = C;
478   while (isdigit(C.peek()))
479     C.advance();
480   Token.reset(MIToken::GlobalValue, Range.upto(C))
481       .setIntegerValue(APSInt(NumberRange.upto(C)));
482   return C;
483 }
484 
485 static Cursor maybeLexExternalSymbol(Cursor C, MIToken &Token,
486                                      ErrorCallbackType ErrorCallback) {
487   if (C.peek() != '&')
488     return None;
489   return lexName(C, Token, MIToken::ExternalSymbol, /*PrefixLength=*/1,
490                  ErrorCallback);
491 }
492 
493 static Cursor maybeLexMCSymbol(Cursor C, MIToken &Token,
494                                ErrorCallbackType ErrorCallback) {
495   const StringRef Rule = "<mcsymbol ";
496   if (!C.remaining().startswith(Rule))
497     return None;
498   auto Start = C;
499   C.advance(Rule.size());
500 
501   // Try a simple unquoted name.
502   if (C.peek() != '"') {
503     while (isIdentifierChar(C.peek()))
504       C.advance();
505     StringRef String = Start.upto(C).drop_front(Rule.size());
506     if (C.peek() != '>') {
507       ErrorCallback(C.location(),
508                     "expected the '<mcsymbol ...' to be closed by a '>'");
509       Token.reset(MIToken::Error, Start.remaining());
510       return Start;
511     }
512     C.advance();
513 
514     Token.reset(MIToken::MCSymbol, Start.upto(C)).setStringValue(String);
515     return C;
516   }
517 
518   // Otherwise lex out a quoted name.
519   Cursor R = lexStringConstant(C, ErrorCallback);
520   if (!R) {
521     ErrorCallback(C.location(),
522                   "unable to parse quoted string from opening quote");
523     Token.reset(MIToken::Error, Start.remaining());
524     return Start;
525   }
526   StringRef String = Start.upto(R).drop_front(Rule.size());
527   if (R.peek() != '>') {
528     ErrorCallback(R.location(),
529                   "expected the '<mcsymbol ...' to be closed by a '>'");
530     Token.reset(MIToken::Error, Start.remaining());
531     return Start;
532   }
533   R.advance();
534 
535   Token.reset(MIToken::MCSymbol, Start.upto(R))
536       .setOwnedStringValue(unescapeQuotedString(String));
537   return R;
538 }
539 
540 static bool isValidHexFloatingPointPrefix(char C) {
541   return C == 'H' || C == 'K' || C == 'L' || C == 'M' || C == 'R';
542 }
543 
544 static Cursor lexFloatingPointLiteral(Cursor Range, Cursor C, MIToken &Token) {
545   C.advance();
546   // Skip over [0-9]*([eE][-+]?[0-9]+)?
547   while (isdigit(C.peek()))
548     C.advance();
549   if ((C.peek() == 'e' || C.peek() == 'E') &&
550       (isdigit(C.peek(1)) ||
551        ((C.peek(1) == '-' || C.peek(1) == '+') && isdigit(C.peek(2))))) {
552     C.advance(2);
553     while (isdigit(C.peek()))
554       C.advance();
555   }
556   Token.reset(MIToken::FloatingPointLiteral, Range.upto(C));
557   return C;
558 }
559 
560 static Cursor maybeLexHexadecimalLiteral(Cursor C, MIToken &Token) {
561   if (C.peek() != '0' || (C.peek(1) != 'x' && C.peek(1) != 'X'))
562     return None;
563   Cursor Range = C;
564   C.advance(2);
565   unsigned PrefLen = 2;
566   if (isValidHexFloatingPointPrefix(C.peek())) {
567     C.advance();
568     PrefLen++;
569   }
570   while (isxdigit(C.peek()))
571     C.advance();
572   StringRef StrVal = Range.upto(C);
573   if (StrVal.size() <= PrefLen)
574     return None;
575   if (PrefLen == 2)
576     Token.reset(MIToken::HexLiteral, Range.upto(C));
577   else // It must be 3, which means that there was a floating-point prefix.
578     Token.reset(MIToken::FloatingPointLiteral, Range.upto(C));
579   return C;
580 }
581 
582 static Cursor maybeLexNumericalLiteral(Cursor C, MIToken &Token) {
583   if (!isdigit(C.peek()) && (C.peek() != '-' || !isdigit(C.peek(1))))
584     return None;
585   auto Range = C;
586   C.advance();
587   while (isdigit(C.peek()))
588     C.advance();
589   if (C.peek() == '.')
590     return lexFloatingPointLiteral(Range, C, Token);
591   StringRef StrVal = Range.upto(C);
592   Token.reset(MIToken::IntegerLiteral, StrVal).setIntegerValue(APSInt(StrVal));
593   return C;
594 }
595 
596 static MIToken::TokenKind getMetadataKeywordKind(StringRef Identifier) {
597   return StringSwitch<MIToken::TokenKind>(Identifier)
598       .Case("!tbaa", MIToken::md_tbaa)
599       .Case("!alias.scope", MIToken::md_alias_scope)
600       .Case("!noalias", MIToken::md_noalias)
601       .Case("!range", MIToken::md_range)
602       .Case("!DIExpression", MIToken::md_diexpr)
603       .Case("!DILocation", MIToken::md_dilocation)
604       .Default(MIToken::Error);
605 }
606 
607 static Cursor maybeLexExclaim(Cursor C, MIToken &Token,
608                               ErrorCallbackType ErrorCallback) {
609   if (C.peek() != '!')
610     return None;
611   auto Range = C;
612   C.advance(1);
613   if (isdigit(C.peek()) || !isIdentifierChar(C.peek())) {
614     Token.reset(MIToken::exclaim, Range.upto(C));
615     return C;
616   }
617   while (isIdentifierChar(C.peek()))
618     C.advance();
619   StringRef StrVal = Range.upto(C);
620   Token.reset(getMetadataKeywordKind(StrVal), StrVal);
621   if (Token.isError())
622     ErrorCallback(Token.location(),
623                   "use of unknown metadata keyword '" + StrVal + "'");
624   return C;
625 }
626 
627 static MIToken::TokenKind symbolToken(char C) {
628   switch (C) {
629   case ',':
630     return MIToken::comma;
631   case '.':
632     return MIToken::dot;
633   case '=':
634     return MIToken::equal;
635   case ':':
636     return MIToken::colon;
637   case '(':
638     return MIToken::lparen;
639   case ')':
640     return MIToken::rparen;
641   case '{':
642     return MIToken::lbrace;
643   case '}':
644     return MIToken::rbrace;
645   case '+':
646     return MIToken::plus;
647   case '-':
648     return MIToken::minus;
649   case '<':
650     return MIToken::less;
651   case '>':
652     return MIToken::greater;
653   default:
654     return MIToken::Error;
655   }
656 }
657 
658 static Cursor maybeLexSymbol(Cursor C, MIToken &Token) {
659   MIToken::TokenKind Kind;
660   unsigned Length = 1;
661   if (C.peek() == ':' && C.peek(1) == ':') {
662     Kind = MIToken::coloncolon;
663     Length = 2;
664   } else
665     Kind = symbolToken(C.peek());
666   if (Kind == MIToken::Error)
667     return None;
668   auto Range = C;
669   C.advance(Length);
670   Token.reset(Kind, Range.upto(C));
671   return C;
672 }
673 
674 static Cursor maybeLexNewline(Cursor C, MIToken &Token) {
675   if (!isNewlineChar(C.peek()))
676     return None;
677   auto Range = C;
678   C.advance();
679   Token.reset(MIToken::Newline, Range.upto(C));
680   return C;
681 }
682 
683 static Cursor maybeLexEscapedIRValue(Cursor C, MIToken &Token,
684                                      ErrorCallbackType ErrorCallback) {
685   if (C.peek() != '`')
686     return None;
687   auto Range = C;
688   C.advance();
689   auto StrRange = C;
690   while (C.peek() != '`') {
691     if (C.isEOF() || isNewlineChar(C.peek())) {
692       ErrorCallback(
693           C.location(),
694           "end of machine instruction reached before the closing '`'");
695       Token.reset(MIToken::Error, Range.remaining());
696       return C;
697     }
698     C.advance();
699   }
700   StringRef Value = StrRange.upto(C);
701   C.advance();
702   Token.reset(MIToken::QuotedIRValue, Range.upto(C)).setStringValue(Value);
703   return C;
704 }
705 
706 StringRef llvm::lexMIToken(StringRef Source, MIToken &Token,
707                            ErrorCallbackType ErrorCallback) {
708   auto C = skipComment(skipWhitespace(Cursor(Source)));
709   if (C.isEOF()) {
710     Token.reset(MIToken::Eof, C.remaining());
711     return C.remaining();
712   }
713 
714   C = skipMachineOperandComment(C);
715 
716   if (Cursor R = maybeLexMachineBasicBlock(C, Token, ErrorCallback))
717     return R.remaining();
718   if (Cursor R = maybeLexIdentifier(C, Token))
719     return R.remaining();
720   if (Cursor R = maybeLexJumpTableIndex(C, Token))
721     return R.remaining();
722   if (Cursor R = maybeLexStackObject(C, Token))
723     return R.remaining();
724   if (Cursor R = maybeLexFixedStackObject(C, Token))
725     return R.remaining();
726   if (Cursor R = maybeLexConstantPoolItem(C, Token))
727     return R.remaining();
728   if (Cursor R = maybeLexSubRegisterIndex(C, Token, ErrorCallback))
729     return R.remaining();
730   if (Cursor R = maybeLexIRBlock(C, Token, ErrorCallback))
731     return R.remaining();
732   if (Cursor R = maybeLexIRValue(C, Token, ErrorCallback))
733     return R.remaining();
734   if (Cursor R = maybeLexRegister(C, Token, ErrorCallback))
735     return R.remaining();
736   if (Cursor R = maybeLexGlobalValue(C, Token, ErrorCallback))
737     return R.remaining();
738   if (Cursor R = maybeLexExternalSymbol(C, Token, ErrorCallback))
739     return R.remaining();
740   if (Cursor R = maybeLexMCSymbol(C, Token, ErrorCallback))
741     return R.remaining();
742   if (Cursor R = maybeLexHexadecimalLiteral(C, Token))
743     return R.remaining();
744   if (Cursor R = maybeLexNumericalLiteral(C, Token))
745     return R.remaining();
746   if (Cursor R = maybeLexExclaim(C, Token, ErrorCallback))
747     return R.remaining();
748   if (Cursor R = maybeLexSymbol(C, Token))
749     return R.remaining();
750   if (Cursor R = maybeLexNewline(C, Token))
751     return R.remaining();
752   if (Cursor R = maybeLexEscapedIRValue(C, Token, ErrorCallback))
753     return R.remaining();
754   if (Cursor R = maybeLexStringConstant(C, Token, ErrorCallback))
755     return R.remaining();
756 
757   Token.reset(MIToken::Error, C.remaining());
758   ErrorCallback(C.location(),
759                 Twine("unexpected character '") + Twine(C.peek()) + "'");
760   return C.remaining();
761 }
762