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/StringExtras.h"
15 #include "llvm/ADT/StringSwitch.h"
16 #include "llvm/ADT/Twine.h"
17 #include <cassert>
18 #include <cctype>
19 #include <string>
20 
21 using namespace llvm;
22 
23 namespace {
24 
25 using ErrorCallbackType =
26     function_ref<void(StringRef::iterator Loc, const Twine &)>;
27 
28 /// This class provides a way to iterate and get characters from the source
29 /// string.
30 class Cursor {
31   const char *Ptr = nullptr;
32   const char *End = nullptr;
33 
34 public:
Cursor(std::nullopt_t)35   Cursor(std::nullopt_t) {}
36 
Cursor(StringRef Str)37   explicit Cursor(StringRef Str) {
38     Ptr = Str.data();
39     End = Ptr + Str.size();
40   }
41 
isEOF() const42   bool isEOF() const { return Ptr == End; }
43 
peek(int I=0) const44   char peek(int I = 0) const { return End - Ptr <= I ? 0 : Ptr[I]; }
45 
advance(unsigned I=1)46   void advance(unsigned I = 1) { Ptr += I; }
47 
remaining() const48   StringRef remaining() const { return StringRef(Ptr, End - Ptr); }
49 
upto(Cursor C) const50   StringRef upto(Cursor C) const {
51     assert(C.Ptr >= Ptr && C.Ptr <= End);
52     return StringRef(Ptr, C.Ptr - Ptr);
53   }
54 
location() const55   StringRef::iterator location() const { return Ptr; }
56 
operator bool() const57   operator bool() const { return Ptr != nullptr; }
58 };
59 
60 } // end anonymous namespace
61 
reset(TokenKind Kind,StringRef Range)62 MIToken &MIToken::reset(TokenKind Kind, StringRef Range) {
63   this->Kind = Kind;
64   this->Range = Range;
65   return *this;
66 }
67 
setStringValue(StringRef StrVal)68 MIToken &MIToken::setStringValue(StringRef StrVal) {
69   StringValue = StrVal;
70   return *this;
71 }
72 
setOwnedStringValue(std::string StrVal)73 MIToken &MIToken::setOwnedStringValue(std::string StrVal) {
74   StringValueStorage = std::move(StrVal);
75   StringValue = StringValueStorage;
76   return *this;
77 }
78 
setIntegerValue(APSInt IntVal)79 MIToken &MIToken::setIntegerValue(APSInt IntVal) {
80   this->IntVal = std::move(IntVal);
81   return *this;
82 }
83 
84 /// Skip the leading whitespace characters and return the updated cursor.
skipWhitespace(Cursor C)85 static Cursor skipWhitespace(Cursor C) {
86   while (isblank(C.peek()))
87     C.advance();
88   return C;
89 }
90 
isNewlineChar(char C)91 static bool isNewlineChar(char C) { return C == '\n' || C == '\r'; }
92 
93 /// Skip a line comment and return the updated cursor.
skipComment(Cursor C)94 static Cursor skipComment(Cursor C) {
95   if (C.peek() != ';')
96     return C;
97   while (!isNewlineChar(C.peek()) && !C.isEOF())
98     C.advance();
99   return C;
100 }
101 
102 /// Machine operands can have comments, enclosed between /* and */.
103 /// This eats up all tokens, including /* and */.
skipMachineOperandComment(Cursor C)104 static Cursor skipMachineOperandComment(Cursor C) {
105   if (C.peek() != '/' || C.peek(1) != '*')
106     return C;
107 
108   while (C.peek() != '*' || C.peek(1) != '/')
109     C.advance();
110 
111   C.advance();
112   C.advance();
113   return C;
114 }
115 
116 /// Return true if the given character satisfies the following regular
117 /// expression: [-a-zA-Z$._0-9]
isIdentifierChar(char C)118 static bool isIdentifierChar(char C) {
119   return isalpha(C) || isdigit(C) || C == '_' || C == '-' || C == '.' ||
120          C == '$';
121 }
122 
123 /// Unescapes the given string value.
124 ///
125 /// Expects the string value to be quoted.
unescapeQuotedString(StringRef Value)126 static std::string unescapeQuotedString(StringRef Value) {
127   assert(Value.front() == '"' && Value.back() == '"');
128   Cursor C = Cursor(Value.substr(1, Value.size() - 2));
129 
130   std::string Str;
131   Str.reserve(C.remaining().size());
132   while (!C.isEOF()) {
133     char Char = C.peek();
134     if (Char == '\\') {
135       if (C.peek(1) == '\\') {
136         // Two '\' become one
137         Str += '\\';
138         C.advance(2);
139         continue;
140       }
141       if (isxdigit(C.peek(1)) && isxdigit(C.peek(2))) {
142         Str += hexDigitValue(C.peek(1)) * 16 + hexDigitValue(C.peek(2));
143         C.advance(3);
144         continue;
145       }
146     }
147     Str += Char;
148     C.advance();
149   }
150   return Str;
151 }
152 
153 /// Lex a string constant using the following regular expression: \"[^\"]*\"
lexStringConstant(Cursor C,ErrorCallbackType ErrorCallback)154 static Cursor lexStringConstant(Cursor C, ErrorCallbackType ErrorCallback) {
155   assert(C.peek() == '"');
156   for (C.advance(); C.peek() != '"'; C.advance()) {
157     if (C.isEOF() || isNewlineChar(C.peek())) {
158       ErrorCallback(
159           C.location(),
160           "end of machine instruction reached before the closing '\"'");
161       return std::nullopt;
162     }
163   }
164   C.advance();
165   return C;
166 }
167 
lexName(Cursor C,MIToken & Token,MIToken::TokenKind Type,unsigned PrefixLength,ErrorCallbackType ErrorCallback)168 static Cursor lexName(Cursor C, MIToken &Token, MIToken::TokenKind Type,
169                       unsigned PrefixLength, ErrorCallbackType ErrorCallback) {
170   auto Range = C;
171   C.advance(PrefixLength);
172   if (C.peek() == '"') {
173     if (Cursor R = lexStringConstant(C, ErrorCallback)) {
174       StringRef String = Range.upto(R);
175       Token.reset(Type, String)
176           .setOwnedStringValue(
177               unescapeQuotedString(String.drop_front(PrefixLength)));
178       return R;
179     }
180     Token.reset(MIToken::Error, Range.remaining());
181     return Range;
182   }
183   while (isIdentifierChar(C.peek()))
184     C.advance();
185   Token.reset(Type, Range.upto(C))
186       .setStringValue(Range.upto(C).drop_front(PrefixLength));
187   return C;
188 }
189 
getIdentifierKind(StringRef Identifier)190 static MIToken::TokenKind getIdentifierKind(StringRef Identifier) {
191   return StringSwitch<MIToken::TokenKind>(Identifier)
192       .Case("_", MIToken::underscore)
193       .Case("implicit", MIToken::kw_implicit)
194       .Case("implicit-def", MIToken::kw_implicit_define)
195       .Case("def", MIToken::kw_def)
196       .Case("dead", MIToken::kw_dead)
197       .Case("killed", MIToken::kw_killed)
198       .Case("undef", MIToken::kw_undef)
199       .Case("internal", MIToken::kw_internal)
200       .Case("early-clobber", MIToken::kw_early_clobber)
201       .Case("debug-use", MIToken::kw_debug_use)
202       .Case("renamable", MIToken::kw_renamable)
203       .Case("tied-def", MIToken::kw_tied_def)
204       .Case("frame-setup", MIToken::kw_frame_setup)
205       .Case("frame-destroy", MIToken::kw_frame_destroy)
206       .Case("nnan", MIToken::kw_nnan)
207       .Case("ninf", MIToken::kw_ninf)
208       .Case("nsz", MIToken::kw_nsz)
209       .Case("arcp", MIToken::kw_arcp)
210       .Case("contract", MIToken::kw_contract)
211       .Case("afn", MIToken::kw_afn)
212       .Case("reassoc", MIToken::kw_reassoc)
213       .Case("nuw", MIToken::kw_nuw)
214       .Case("nsw", MIToken::kw_nsw)
215       .Case("exact", MIToken::kw_exact)
216       .Case("nofpexcept", MIToken::kw_nofpexcept)
217       .Case("unpredictable", MIToken::kw_unpredictable)
218       .Case("debug-location", MIToken::kw_debug_location)
219       .Case("debug-instr-number", MIToken::kw_debug_instr_number)
220       .Case("dbg-instr-ref", MIToken::kw_dbg_instr_ref)
221       .Case("same_value", MIToken::kw_cfi_same_value)
222       .Case("offset", MIToken::kw_cfi_offset)
223       .Case("rel_offset", MIToken::kw_cfi_rel_offset)
224       .Case("def_cfa_register", MIToken::kw_cfi_def_cfa_register)
225       .Case("def_cfa_offset", MIToken::kw_cfi_def_cfa_offset)
226       .Case("adjust_cfa_offset", MIToken::kw_cfi_adjust_cfa_offset)
227       .Case("escape", MIToken::kw_cfi_escape)
228       .Case("def_cfa", MIToken::kw_cfi_def_cfa)
229       .Case("llvm_def_aspace_cfa", MIToken::kw_cfi_llvm_def_aspace_cfa)
230       .Case("remember_state", MIToken::kw_cfi_remember_state)
231       .Case("restore", MIToken::kw_cfi_restore)
232       .Case("restore_state", MIToken::kw_cfi_restore_state)
233       .Case("undefined", MIToken::kw_cfi_undefined)
234       .Case("register", MIToken::kw_cfi_register)
235       .Case("window_save", MIToken::kw_cfi_window_save)
236       .Case("negate_ra_sign_state",
237             MIToken::kw_cfi_aarch64_negate_ra_sign_state)
238       .Case("blockaddress", MIToken::kw_blockaddress)
239       .Case("intrinsic", MIToken::kw_intrinsic)
240       .Case("target-index", MIToken::kw_target_index)
241       .Case("half", MIToken::kw_half)
242       .Case("float", MIToken::kw_float)
243       .Case("double", MIToken::kw_double)
244       .Case("x86_fp80", MIToken::kw_x86_fp80)
245       .Case("fp128", MIToken::kw_fp128)
246       .Case("ppc_fp128", MIToken::kw_ppc_fp128)
247       .Case("target-flags", MIToken::kw_target_flags)
248       .Case("volatile", MIToken::kw_volatile)
249       .Case("non-temporal", MIToken::kw_non_temporal)
250       .Case("dereferenceable", MIToken::kw_dereferenceable)
251       .Case("invariant", MIToken::kw_invariant)
252       .Case("align", MIToken::kw_align)
253       .Case("basealign", MIToken::kw_basealign)
254       .Case("addrspace", MIToken::kw_addrspace)
255       .Case("stack", MIToken::kw_stack)
256       .Case("got", MIToken::kw_got)
257       .Case("jump-table", MIToken::kw_jump_table)
258       .Case("constant-pool", MIToken::kw_constant_pool)
259       .Case("call-entry", MIToken::kw_call_entry)
260       .Case("custom", MIToken::kw_custom)
261       .Case("liveout", MIToken::kw_liveout)
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("pcsections", MIToken::kw_pcsections)
275       .Case("cfi-type", MIToken::kw_cfi_type)
276       .Case("bbsections", MIToken::kw_bbsections)
277       .Case("bb_id", MIToken::kw_bb_id)
278       .Case("unknown-size", MIToken::kw_unknown_size)
279       .Case("unknown-address", MIToken::kw_unknown_address)
280       .Case("distinct", MIToken::kw_distinct)
281       .Case("ir-block-address-taken", MIToken::kw_ir_block_address_taken)
282       .Case("machine-block-address-taken",
283             MIToken::kw_machine_block_address_taken)
284       .Case("call-frame-size", MIToken::kw_call_frame_size)
285       .Case("noconvergent", MIToken::kw_noconvergent)
286       .Default(MIToken::Identifier);
287 }
288 
maybeLexIdentifier(Cursor C,MIToken & Token)289 static Cursor maybeLexIdentifier(Cursor C, MIToken &Token) {
290   if (!isalpha(C.peek()) && C.peek() != '_')
291     return std::nullopt;
292   auto Range = C;
293   while (isIdentifierChar(C.peek()))
294     C.advance();
295   auto Identifier = Range.upto(C);
296   Token.reset(getIdentifierKind(Identifier), Identifier)
297       .setStringValue(Identifier);
298   return C;
299 }
300 
maybeLexMachineBasicBlock(Cursor C,MIToken & Token,ErrorCallbackType ErrorCallback)301 static Cursor maybeLexMachineBasicBlock(Cursor C, MIToken &Token,
302                                         ErrorCallbackType ErrorCallback) {
303   bool IsReference = C.remaining().starts_with("%bb.");
304   if (!IsReference && !C.remaining().starts_with("bb."))
305     return std::nullopt;
306   auto Range = C;
307   unsigned PrefixLength = IsReference ? 4 : 3;
308   C.advance(PrefixLength); // Skip '%bb.' or 'bb.'
309   if (!isdigit(C.peek())) {
310     Token.reset(MIToken::Error, C.remaining());
311     ErrorCallback(C.location(), "expected a number after '%bb.'");
312     return C;
313   }
314   auto NumberRange = C;
315   while (isdigit(C.peek()))
316     C.advance();
317   StringRef Number = NumberRange.upto(C);
318   unsigned StringOffset = PrefixLength + Number.size(); // Drop '%bb.<id>'
319   // TODO: The format bb.<id>.<irname> is supported only when it's not a
320   // reference. Once we deprecate the format where the irname shows up, we
321   // should only lex forward if it is a reference.
322   if (C.peek() == '.') {
323     C.advance(); // Skip '.'
324     ++StringOffset;
325     while (isIdentifierChar(C.peek()))
326       C.advance();
327   }
328   Token.reset(IsReference ? MIToken::MachineBasicBlock
329                           : MIToken::MachineBasicBlockLabel,
330               Range.upto(C))
331       .setIntegerValue(APSInt(Number))
332       .setStringValue(Range.upto(C).drop_front(StringOffset));
333   return C;
334 }
335 
maybeLexIndex(Cursor C,MIToken & Token,StringRef Rule,MIToken::TokenKind Kind)336 static Cursor maybeLexIndex(Cursor C, MIToken &Token, StringRef Rule,
337                             MIToken::TokenKind Kind) {
338   if (!C.remaining().starts_with(Rule) || !isdigit(C.peek(Rule.size())))
339     return std::nullopt;
340   auto Range = C;
341   C.advance(Rule.size());
342   auto NumberRange = C;
343   while (isdigit(C.peek()))
344     C.advance();
345   Token.reset(Kind, Range.upto(C)).setIntegerValue(APSInt(NumberRange.upto(C)));
346   return C;
347 }
348 
maybeLexIndexAndName(Cursor C,MIToken & Token,StringRef Rule,MIToken::TokenKind Kind)349 static Cursor maybeLexIndexAndName(Cursor C, MIToken &Token, StringRef Rule,
350                                    MIToken::TokenKind Kind) {
351   if (!C.remaining().starts_with(Rule) || !isdigit(C.peek(Rule.size())))
352     return std::nullopt;
353   auto Range = C;
354   C.advance(Rule.size());
355   auto NumberRange = C;
356   while (isdigit(C.peek()))
357     C.advance();
358   StringRef Number = NumberRange.upto(C);
359   unsigned StringOffset = Rule.size() + Number.size();
360   if (C.peek() == '.') {
361     C.advance();
362     ++StringOffset;
363     while (isIdentifierChar(C.peek()))
364       C.advance();
365   }
366   Token.reset(Kind, Range.upto(C))
367       .setIntegerValue(APSInt(Number))
368       .setStringValue(Range.upto(C).drop_front(StringOffset));
369   return C;
370 }
371 
maybeLexJumpTableIndex(Cursor C,MIToken & Token)372 static Cursor maybeLexJumpTableIndex(Cursor C, MIToken &Token) {
373   return maybeLexIndex(C, Token, "%jump-table.", MIToken::JumpTableIndex);
374 }
375 
maybeLexStackObject(Cursor C,MIToken & Token)376 static Cursor maybeLexStackObject(Cursor C, MIToken &Token) {
377   return maybeLexIndexAndName(C, Token, "%stack.", MIToken::StackObject);
378 }
379 
maybeLexFixedStackObject(Cursor C,MIToken & Token)380 static Cursor maybeLexFixedStackObject(Cursor C, MIToken &Token) {
381   return maybeLexIndex(C, Token, "%fixed-stack.", MIToken::FixedStackObject);
382 }
383 
maybeLexConstantPoolItem(Cursor C,MIToken & Token)384 static Cursor maybeLexConstantPoolItem(Cursor C, MIToken &Token) {
385   return maybeLexIndex(C, Token, "%const.", MIToken::ConstantPoolItem);
386 }
387 
maybeLexSubRegisterIndex(Cursor C,MIToken & Token,ErrorCallbackType ErrorCallback)388 static Cursor maybeLexSubRegisterIndex(Cursor C, MIToken &Token,
389                                        ErrorCallbackType ErrorCallback) {
390   const StringRef Rule = "%subreg.";
391   if (!C.remaining().starts_with(Rule))
392     return std::nullopt;
393   return lexName(C, Token, MIToken::SubRegisterIndex, Rule.size(),
394                  ErrorCallback);
395 }
396 
maybeLexIRBlock(Cursor C,MIToken & Token,ErrorCallbackType ErrorCallback)397 static Cursor maybeLexIRBlock(Cursor C, MIToken &Token,
398                               ErrorCallbackType ErrorCallback) {
399   const StringRef Rule = "%ir-block.";
400   if (!C.remaining().starts_with(Rule))
401     return std::nullopt;
402   if (isdigit(C.peek(Rule.size())))
403     return maybeLexIndex(C, Token, Rule, MIToken::IRBlock);
404   return lexName(C, Token, MIToken::NamedIRBlock, Rule.size(), ErrorCallback);
405 }
406 
maybeLexIRValue(Cursor C,MIToken & Token,ErrorCallbackType ErrorCallback)407 static Cursor maybeLexIRValue(Cursor C, MIToken &Token,
408                               ErrorCallbackType ErrorCallback) {
409   const StringRef Rule = "%ir.";
410   if (!C.remaining().starts_with(Rule))
411     return std::nullopt;
412   if (isdigit(C.peek(Rule.size())))
413     return maybeLexIndex(C, Token, Rule, MIToken::IRValue);
414   return lexName(C, Token, MIToken::NamedIRValue, Rule.size(), ErrorCallback);
415 }
416 
maybeLexStringConstant(Cursor C,MIToken & Token,ErrorCallbackType ErrorCallback)417 static Cursor maybeLexStringConstant(Cursor C, MIToken &Token,
418                                      ErrorCallbackType ErrorCallback) {
419   if (C.peek() != '"')
420     return std::nullopt;
421   return lexName(C, Token, MIToken::StringConstant, /*PrefixLength=*/0,
422                  ErrorCallback);
423 }
424 
lexVirtualRegister(Cursor C,MIToken & Token)425 static Cursor lexVirtualRegister(Cursor C, MIToken &Token) {
426   auto Range = C;
427   C.advance(); // Skip '%'
428   auto NumberRange = C;
429   while (isdigit(C.peek()))
430     C.advance();
431   Token.reset(MIToken::VirtualRegister, Range.upto(C))
432       .setIntegerValue(APSInt(NumberRange.upto(C)));
433   return C;
434 }
435 
436 /// Returns true for a character allowed in a register name.
isRegisterChar(char C)437 static bool isRegisterChar(char C) {
438   return isIdentifierChar(C) && C != '.';
439 }
440 
lexNamedVirtualRegister(Cursor C,MIToken & Token)441 static Cursor lexNamedVirtualRegister(Cursor C, MIToken &Token) {
442   Cursor Range = C;
443   C.advance(); // Skip '%'
444   while (isRegisterChar(C.peek()))
445     C.advance();
446   Token.reset(MIToken::NamedVirtualRegister, Range.upto(C))
447       .setStringValue(Range.upto(C).drop_front(1)); // Drop the '%'
448   return C;
449 }
450 
maybeLexRegister(Cursor C,MIToken & Token,ErrorCallbackType ErrorCallback)451 static Cursor maybeLexRegister(Cursor C, MIToken &Token,
452                                ErrorCallbackType ErrorCallback) {
453   if (C.peek() != '%' && C.peek() != '$')
454     return std::nullopt;
455 
456   if (C.peek() == '%') {
457     if (isdigit(C.peek(1)))
458       return lexVirtualRegister(C, Token);
459 
460     if (isRegisterChar(C.peek(1)))
461       return lexNamedVirtualRegister(C, Token);
462 
463     return std::nullopt;
464   }
465 
466   assert(C.peek() == '$');
467   auto Range = C;
468   C.advance(); // Skip '$'
469   while (isRegisterChar(C.peek()))
470     C.advance();
471   Token.reset(MIToken::NamedRegister, Range.upto(C))
472       .setStringValue(Range.upto(C).drop_front(1)); // Drop the '$'
473   return C;
474 }
475 
maybeLexGlobalValue(Cursor C,MIToken & Token,ErrorCallbackType ErrorCallback)476 static Cursor maybeLexGlobalValue(Cursor C, MIToken &Token,
477                                   ErrorCallbackType ErrorCallback) {
478   if (C.peek() != '@')
479     return std::nullopt;
480   if (!isdigit(C.peek(1)))
481     return lexName(C, Token, MIToken::NamedGlobalValue, /*PrefixLength=*/1,
482                    ErrorCallback);
483   auto Range = C;
484   C.advance(1); // Skip the '@'
485   auto NumberRange = C;
486   while (isdigit(C.peek()))
487     C.advance();
488   Token.reset(MIToken::GlobalValue, Range.upto(C))
489       .setIntegerValue(APSInt(NumberRange.upto(C)));
490   return C;
491 }
492 
maybeLexExternalSymbol(Cursor C,MIToken & Token,ErrorCallbackType ErrorCallback)493 static Cursor maybeLexExternalSymbol(Cursor C, MIToken &Token,
494                                      ErrorCallbackType ErrorCallback) {
495   if (C.peek() != '&')
496     return std::nullopt;
497   return lexName(C, Token, MIToken::ExternalSymbol, /*PrefixLength=*/1,
498                  ErrorCallback);
499 }
500 
maybeLexMCSymbol(Cursor C,MIToken & Token,ErrorCallbackType ErrorCallback)501 static Cursor maybeLexMCSymbol(Cursor C, MIToken &Token,
502                                ErrorCallbackType ErrorCallback) {
503   const StringRef Rule = "<mcsymbol ";
504   if (!C.remaining().starts_with(Rule))
505     return std::nullopt;
506   auto Start = C;
507   C.advance(Rule.size());
508 
509   // Try a simple unquoted name.
510   if (C.peek() != '"') {
511     while (isIdentifierChar(C.peek()))
512       C.advance();
513     StringRef String = Start.upto(C).drop_front(Rule.size());
514     if (C.peek() != '>') {
515       ErrorCallback(C.location(),
516                     "expected the '<mcsymbol ...' to be closed by a '>'");
517       Token.reset(MIToken::Error, Start.remaining());
518       return Start;
519     }
520     C.advance();
521 
522     Token.reset(MIToken::MCSymbol, Start.upto(C)).setStringValue(String);
523     return C;
524   }
525 
526   // Otherwise lex out a quoted name.
527   Cursor R = lexStringConstant(C, ErrorCallback);
528   if (!R) {
529     ErrorCallback(C.location(),
530                   "unable to parse quoted string from opening quote");
531     Token.reset(MIToken::Error, Start.remaining());
532     return Start;
533   }
534   StringRef String = Start.upto(R).drop_front(Rule.size());
535   if (R.peek() != '>') {
536     ErrorCallback(R.location(),
537                   "expected the '<mcsymbol ...' to be closed by a '>'");
538     Token.reset(MIToken::Error, Start.remaining());
539     return Start;
540   }
541   R.advance();
542 
543   Token.reset(MIToken::MCSymbol, Start.upto(R))
544       .setOwnedStringValue(unescapeQuotedString(String));
545   return R;
546 }
547 
isValidHexFloatingPointPrefix(char C)548 static bool isValidHexFloatingPointPrefix(char C) {
549   return C == 'H' || C == 'K' || C == 'L' || C == 'M' || C == 'R';
550 }
551 
lexFloatingPointLiteral(Cursor Range,Cursor C,MIToken & Token)552 static Cursor lexFloatingPointLiteral(Cursor Range, Cursor C, MIToken &Token) {
553   C.advance();
554   // Skip over [0-9]*([eE][-+]?[0-9]+)?
555   while (isdigit(C.peek()))
556     C.advance();
557   if ((C.peek() == 'e' || C.peek() == 'E') &&
558       (isdigit(C.peek(1)) ||
559        ((C.peek(1) == '-' || C.peek(1) == '+') && isdigit(C.peek(2))))) {
560     C.advance(2);
561     while (isdigit(C.peek()))
562       C.advance();
563   }
564   Token.reset(MIToken::FloatingPointLiteral, Range.upto(C));
565   return C;
566 }
567 
maybeLexHexadecimalLiteral(Cursor C,MIToken & Token)568 static Cursor maybeLexHexadecimalLiteral(Cursor C, MIToken &Token) {
569   if (C.peek() != '0' || (C.peek(1) != 'x' && C.peek(1) != 'X'))
570     return std::nullopt;
571   Cursor Range = C;
572   C.advance(2);
573   unsigned PrefLen = 2;
574   if (isValidHexFloatingPointPrefix(C.peek())) {
575     C.advance();
576     PrefLen++;
577   }
578   while (isxdigit(C.peek()))
579     C.advance();
580   StringRef StrVal = Range.upto(C);
581   if (StrVal.size() <= PrefLen)
582     return std::nullopt;
583   if (PrefLen == 2)
584     Token.reset(MIToken::HexLiteral, Range.upto(C));
585   else // It must be 3, which means that there was a floating-point prefix.
586     Token.reset(MIToken::FloatingPointLiteral, Range.upto(C));
587   return C;
588 }
589 
maybeLexNumericalLiteral(Cursor C,MIToken & Token)590 static Cursor maybeLexNumericalLiteral(Cursor C, MIToken &Token) {
591   if (!isdigit(C.peek()) && (C.peek() != '-' || !isdigit(C.peek(1))))
592     return std::nullopt;
593   auto Range = C;
594   C.advance();
595   while (isdigit(C.peek()))
596     C.advance();
597   if (C.peek() == '.')
598     return lexFloatingPointLiteral(Range, C, Token);
599   StringRef StrVal = Range.upto(C);
600   Token.reset(MIToken::IntegerLiteral, StrVal).setIntegerValue(APSInt(StrVal));
601   return C;
602 }
603 
getMetadataKeywordKind(StringRef Identifier)604 static MIToken::TokenKind getMetadataKeywordKind(StringRef Identifier) {
605   return StringSwitch<MIToken::TokenKind>(Identifier)
606       .Case("!tbaa", MIToken::md_tbaa)
607       .Case("!alias.scope", MIToken::md_alias_scope)
608       .Case("!noalias", MIToken::md_noalias)
609       .Case("!range", MIToken::md_range)
610       .Case("!DIExpression", MIToken::md_diexpr)
611       .Case("!DILocation", MIToken::md_dilocation)
612       .Default(MIToken::Error);
613 }
614 
maybeLexExclaim(Cursor C,MIToken & Token,ErrorCallbackType ErrorCallback)615 static Cursor maybeLexExclaim(Cursor C, MIToken &Token,
616                               ErrorCallbackType ErrorCallback) {
617   if (C.peek() != '!')
618     return std::nullopt;
619   auto Range = C;
620   C.advance(1);
621   if (isdigit(C.peek()) || !isIdentifierChar(C.peek())) {
622     Token.reset(MIToken::exclaim, Range.upto(C));
623     return C;
624   }
625   while (isIdentifierChar(C.peek()))
626     C.advance();
627   StringRef StrVal = Range.upto(C);
628   Token.reset(getMetadataKeywordKind(StrVal), StrVal);
629   if (Token.isError())
630     ErrorCallback(Token.location(),
631                   "use of unknown metadata keyword '" + StrVal + "'");
632   return C;
633 }
634 
symbolToken(char C)635 static MIToken::TokenKind symbolToken(char C) {
636   switch (C) {
637   case ',':
638     return MIToken::comma;
639   case '.':
640     return MIToken::dot;
641   case '=':
642     return MIToken::equal;
643   case ':':
644     return MIToken::colon;
645   case '(':
646     return MIToken::lparen;
647   case ')':
648     return MIToken::rparen;
649   case '{':
650     return MIToken::lbrace;
651   case '}':
652     return MIToken::rbrace;
653   case '+':
654     return MIToken::plus;
655   case '-':
656     return MIToken::minus;
657   case '<':
658     return MIToken::less;
659   case '>':
660     return MIToken::greater;
661   default:
662     return MIToken::Error;
663   }
664 }
665 
maybeLexSymbol(Cursor C,MIToken & Token)666 static Cursor maybeLexSymbol(Cursor C, MIToken &Token) {
667   MIToken::TokenKind Kind;
668   unsigned Length = 1;
669   if (C.peek() == ':' && C.peek(1) == ':') {
670     Kind = MIToken::coloncolon;
671     Length = 2;
672   } else
673     Kind = symbolToken(C.peek());
674   if (Kind == MIToken::Error)
675     return std::nullopt;
676   auto Range = C;
677   C.advance(Length);
678   Token.reset(Kind, Range.upto(C));
679   return C;
680 }
681 
maybeLexNewline(Cursor C,MIToken & Token)682 static Cursor maybeLexNewline(Cursor C, MIToken &Token) {
683   if (!isNewlineChar(C.peek()))
684     return std::nullopt;
685   auto Range = C;
686   C.advance();
687   Token.reset(MIToken::Newline, Range.upto(C));
688   return C;
689 }
690 
maybeLexEscapedIRValue(Cursor C,MIToken & Token,ErrorCallbackType ErrorCallback)691 static Cursor maybeLexEscapedIRValue(Cursor C, MIToken &Token,
692                                      ErrorCallbackType ErrorCallback) {
693   if (C.peek() != '`')
694     return std::nullopt;
695   auto Range = C;
696   C.advance();
697   auto StrRange = C;
698   while (C.peek() != '`') {
699     if (C.isEOF() || isNewlineChar(C.peek())) {
700       ErrorCallback(
701           C.location(),
702           "end of machine instruction reached before the closing '`'");
703       Token.reset(MIToken::Error, Range.remaining());
704       return C;
705     }
706     C.advance();
707   }
708   StringRef Value = StrRange.upto(C);
709   C.advance();
710   Token.reset(MIToken::QuotedIRValue, Range.upto(C)).setStringValue(Value);
711   return C;
712 }
713 
lexMIToken(StringRef Source,MIToken & Token,ErrorCallbackType ErrorCallback)714 StringRef llvm::lexMIToken(StringRef Source, MIToken &Token,
715                            ErrorCallbackType ErrorCallback) {
716   auto C = skipComment(skipWhitespace(Cursor(Source)));
717   if (C.isEOF()) {
718     Token.reset(MIToken::Eof, C.remaining());
719     return C.remaining();
720   }
721 
722   C = skipMachineOperandComment(C);
723 
724   if (Cursor R = maybeLexMachineBasicBlock(C, Token, ErrorCallback))
725     return R.remaining();
726   if (Cursor R = maybeLexIdentifier(C, Token))
727     return R.remaining();
728   if (Cursor R = maybeLexJumpTableIndex(C, Token))
729     return R.remaining();
730   if (Cursor R = maybeLexStackObject(C, Token))
731     return R.remaining();
732   if (Cursor R = maybeLexFixedStackObject(C, Token))
733     return R.remaining();
734   if (Cursor R = maybeLexConstantPoolItem(C, Token))
735     return R.remaining();
736   if (Cursor R = maybeLexSubRegisterIndex(C, Token, ErrorCallback))
737     return R.remaining();
738   if (Cursor R = maybeLexIRBlock(C, Token, ErrorCallback))
739     return R.remaining();
740   if (Cursor R = maybeLexIRValue(C, Token, ErrorCallback))
741     return R.remaining();
742   if (Cursor R = maybeLexRegister(C, Token, ErrorCallback))
743     return R.remaining();
744   if (Cursor R = maybeLexGlobalValue(C, Token, ErrorCallback))
745     return R.remaining();
746   if (Cursor R = maybeLexExternalSymbol(C, Token, ErrorCallback))
747     return R.remaining();
748   if (Cursor R = maybeLexMCSymbol(C, Token, ErrorCallback))
749     return R.remaining();
750   if (Cursor R = maybeLexHexadecimalLiteral(C, Token))
751     return R.remaining();
752   if (Cursor R = maybeLexNumericalLiteral(C, Token))
753     return R.remaining();
754   if (Cursor R = maybeLexExclaim(C, Token, ErrorCallback))
755     return R.remaining();
756   if (Cursor R = maybeLexSymbol(C, Token))
757     return R.remaining();
758   if (Cursor R = maybeLexNewline(C, Token))
759     return R.remaining();
760   if (Cursor R = maybeLexEscapedIRValue(C, Token, ErrorCallback))
761     return R.remaining();
762   if (Cursor R = maybeLexStringConstant(C, Token, ErrorCallback))
763     return R.remaining();
764 
765   Token.reset(MIToken::Error, C.remaining());
766   ErrorCallback(C.location(),
767                 Twine("unexpected character '") + Twine(C.peek()) + "'");
768   return C.remaining();
769 }
770