xref: /freebsd/contrib/llvm-project/llvm/lib/CodeGen/MIRParser/MIParser.cpp (revision c57c26179033f64c2011a2d2a904ee3fa62e826a)
1 //===- MIParser.cpp - Machine instructions parser 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 parsing of machine instructions.
10 //
11 //===----------------------------------------------------------------------===//
12 
13 #include "llvm/CodeGen/MIRParser/MIParser.h"
14 #include "MILexer.h"
15 #include "llvm/ADT/APInt.h"
16 #include "llvm/ADT/APSInt.h"
17 #include "llvm/ADT/ArrayRef.h"
18 #include "llvm/ADT/DenseMap.h"
19 #include "llvm/ADT/SmallVector.h"
20 #include "llvm/ADT/StringMap.h"
21 #include "llvm/ADT/StringRef.h"
22 #include "llvm/ADT/StringSwitch.h"
23 #include "llvm/ADT/Twine.h"
24 #include "llvm/Analysis/MemoryLocation.h"
25 #include "llvm/AsmParser/Parser.h"
26 #include "llvm/AsmParser/SlotMapping.h"
27 #include "llvm/CodeGen/LowLevelType.h"
28 #include "llvm/CodeGen/MIRFormatter.h"
29 #include "llvm/CodeGen/MIRPrinter.h"
30 #include "llvm/CodeGen/MachineBasicBlock.h"
31 #include "llvm/CodeGen/MachineFrameInfo.h"
32 #include "llvm/CodeGen/MachineFunction.h"
33 #include "llvm/CodeGen/MachineInstr.h"
34 #include "llvm/CodeGen/MachineInstrBuilder.h"
35 #include "llvm/CodeGen/MachineMemOperand.h"
36 #include "llvm/CodeGen/MachineOperand.h"
37 #include "llvm/CodeGen/MachineRegisterInfo.h"
38 #include "llvm/CodeGen/PseudoSourceValueManager.h"
39 #include "llvm/CodeGen/RegisterBank.h"
40 #include "llvm/CodeGen/RegisterBankInfo.h"
41 #include "llvm/CodeGen/TargetInstrInfo.h"
42 #include "llvm/CodeGen/TargetRegisterInfo.h"
43 #include "llvm/CodeGen/TargetSubtargetInfo.h"
44 #include "llvm/IR/BasicBlock.h"
45 #include "llvm/IR/Constants.h"
46 #include "llvm/IR/DataLayout.h"
47 #include "llvm/IR/DebugInfoMetadata.h"
48 #include "llvm/IR/DebugLoc.h"
49 #include "llvm/IR/Function.h"
50 #include "llvm/IR/InstrTypes.h"
51 #include "llvm/IR/Instructions.h"
52 #include "llvm/IR/Intrinsics.h"
53 #include "llvm/IR/Metadata.h"
54 #include "llvm/IR/Module.h"
55 #include "llvm/IR/ModuleSlotTracker.h"
56 #include "llvm/IR/Type.h"
57 #include "llvm/IR/Value.h"
58 #include "llvm/IR/ValueSymbolTable.h"
59 #include "llvm/MC/LaneBitmask.h"
60 #include "llvm/MC/MCContext.h"
61 #include "llvm/MC/MCDwarf.h"
62 #include "llvm/MC/MCInstrDesc.h"
63 #include "llvm/Support/AtomicOrdering.h"
64 #include "llvm/Support/BranchProbability.h"
65 #include "llvm/Support/Casting.h"
66 #include "llvm/Support/ErrorHandling.h"
67 #include "llvm/Support/MemoryBuffer.h"
68 #include "llvm/Support/SMLoc.h"
69 #include "llvm/Support/SourceMgr.h"
70 #include "llvm/Target/TargetIntrinsicInfo.h"
71 #include "llvm/Target/TargetMachine.h"
72 #include <cassert>
73 #include <cctype>
74 #include <cstddef>
75 #include <cstdint>
76 #include <limits>
77 #include <string>
78 #include <utility>
79 
80 using namespace llvm;
81 
82 void PerTargetMIParsingState::setTarget(
83   const TargetSubtargetInfo &NewSubtarget) {
84 
85   // If the subtarget changed, over conservatively assume everything is invalid.
86   if (&Subtarget == &NewSubtarget)
87     return;
88 
89   Names2InstrOpCodes.clear();
90   Names2Regs.clear();
91   Names2RegMasks.clear();
92   Names2SubRegIndices.clear();
93   Names2TargetIndices.clear();
94   Names2DirectTargetFlags.clear();
95   Names2BitmaskTargetFlags.clear();
96   Names2MMOTargetFlags.clear();
97 
98   initNames2RegClasses();
99   initNames2RegBanks();
100 }
101 
102 void PerTargetMIParsingState::initNames2Regs() {
103   if (!Names2Regs.empty())
104     return;
105 
106   // The '%noreg' register is the register 0.
107   Names2Regs.insert(std::make_pair("noreg", 0));
108   const auto *TRI = Subtarget.getRegisterInfo();
109   assert(TRI && "Expected target register info");
110 
111   for (unsigned I = 0, E = TRI->getNumRegs(); I < E; ++I) {
112     bool WasInserted =
113         Names2Regs.insert(std::make_pair(StringRef(TRI->getName(I)).lower(), I))
114             .second;
115     (void)WasInserted;
116     assert(WasInserted && "Expected registers to be unique case-insensitively");
117   }
118 }
119 
120 bool PerTargetMIParsingState::getRegisterByName(StringRef RegName,
121                                                 Register &Reg) {
122   initNames2Regs();
123   auto RegInfo = Names2Regs.find(RegName);
124   if (RegInfo == Names2Regs.end())
125     return true;
126   Reg = RegInfo->getValue();
127   return false;
128 }
129 
130 void PerTargetMIParsingState::initNames2InstrOpCodes() {
131   if (!Names2InstrOpCodes.empty())
132     return;
133   const auto *TII = Subtarget.getInstrInfo();
134   assert(TII && "Expected target instruction info");
135   for (unsigned I = 0, E = TII->getNumOpcodes(); I < E; ++I)
136     Names2InstrOpCodes.insert(std::make_pair(StringRef(TII->getName(I)), I));
137 }
138 
139 bool PerTargetMIParsingState::parseInstrName(StringRef InstrName,
140                                              unsigned &OpCode) {
141   initNames2InstrOpCodes();
142   auto InstrInfo = Names2InstrOpCodes.find(InstrName);
143   if (InstrInfo == Names2InstrOpCodes.end())
144     return true;
145   OpCode = InstrInfo->getValue();
146   return false;
147 }
148 
149 void PerTargetMIParsingState::initNames2RegMasks() {
150   if (!Names2RegMasks.empty())
151     return;
152   const auto *TRI = Subtarget.getRegisterInfo();
153   assert(TRI && "Expected target register info");
154   ArrayRef<const uint32_t *> RegMasks = TRI->getRegMasks();
155   ArrayRef<const char *> RegMaskNames = TRI->getRegMaskNames();
156   assert(RegMasks.size() == RegMaskNames.size());
157   for (size_t I = 0, E = RegMasks.size(); I < E; ++I)
158     Names2RegMasks.insert(
159         std::make_pair(StringRef(RegMaskNames[I]).lower(), RegMasks[I]));
160 }
161 
162 const uint32_t *PerTargetMIParsingState::getRegMask(StringRef Identifier) {
163   initNames2RegMasks();
164   auto RegMaskInfo = Names2RegMasks.find(Identifier);
165   if (RegMaskInfo == Names2RegMasks.end())
166     return nullptr;
167   return RegMaskInfo->getValue();
168 }
169 
170 void PerTargetMIParsingState::initNames2SubRegIndices() {
171   if (!Names2SubRegIndices.empty())
172     return;
173   const TargetRegisterInfo *TRI = Subtarget.getRegisterInfo();
174   for (unsigned I = 1, E = TRI->getNumSubRegIndices(); I < E; ++I)
175     Names2SubRegIndices.insert(
176         std::make_pair(TRI->getSubRegIndexName(I), I));
177 }
178 
179 unsigned PerTargetMIParsingState::getSubRegIndex(StringRef Name) {
180   initNames2SubRegIndices();
181   auto SubRegInfo = Names2SubRegIndices.find(Name);
182   if (SubRegInfo == Names2SubRegIndices.end())
183     return 0;
184   return SubRegInfo->getValue();
185 }
186 
187 void PerTargetMIParsingState::initNames2TargetIndices() {
188   if (!Names2TargetIndices.empty())
189     return;
190   const auto *TII = Subtarget.getInstrInfo();
191   assert(TII && "Expected target instruction info");
192   auto Indices = TII->getSerializableTargetIndices();
193   for (const auto &I : Indices)
194     Names2TargetIndices.insert(std::make_pair(StringRef(I.second), I.first));
195 }
196 
197 bool PerTargetMIParsingState::getTargetIndex(StringRef Name, int &Index) {
198   initNames2TargetIndices();
199   auto IndexInfo = Names2TargetIndices.find(Name);
200   if (IndexInfo == Names2TargetIndices.end())
201     return true;
202   Index = IndexInfo->second;
203   return false;
204 }
205 
206 void PerTargetMIParsingState::initNames2DirectTargetFlags() {
207   if (!Names2DirectTargetFlags.empty())
208     return;
209 
210   const auto *TII = Subtarget.getInstrInfo();
211   assert(TII && "Expected target instruction info");
212   auto Flags = TII->getSerializableDirectMachineOperandTargetFlags();
213   for (const auto &I : Flags)
214     Names2DirectTargetFlags.insert(
215         std::make_pair(StringRef(I.second), I.first));
216 }
217 
218 bool PerTargetMIParsingState::getDirectTargetFlag(StringRef Name,
219                                                   unsigned &Flag) {
220   initNames2DirectTargetFlags();
221   auto FlagInfo = Names2DirectTargetFlags.find(Name);
222   if (FlagInfo == Names2DirectTargetFlags.end())
223     return true;
224   Flag = FlagInfo->second;
225   return false;
226 }
227 
228 void PerTargetMIParsingState::initNames2BitmaskTargetFlags() {
229   if (!Names2BitmaskTargetFlags.empty())
230     return;
231 
232   const auto *TII = Subtarget.getInstrInfo();
233   assert(TII && "Expected target instruction info");
234   auto Flags = TII->getSerializableBitmaskMachineOperandTargetFlags();
235   for (const auto &I : Flags)
236     Names2BitmaskTargetFlags.insert(
237         std::make_pair(StringRef(I.second), I.first));
238 }
239 
240 bool PerTargetMIParsingState::getBitmaskTargetFlag(StringRef Name,
241                                                    unsigned &Flag) {
242   initNames2BitmaskTargetFlags();
243   auto FlagInfo = Names2BitmaskTargetFlags.find(Name);
244   if (FlagInfo == Names2BitmaskTargetFlags.end())
245     return true;
246   Flag = FlagInfo->second;
247   return false;
248 }
249 
250 void PerTargetMIParsingState::initNames2MMOTargetFlags() {
251   if (!Names2MMOTargetFlags.empty())
252     return;
253 
254   const auto *TII = Subtarget.getInstrInfo();
255   assert(TII && "Expected target instruction info");
256   auto Flags = TII->getSerializableMachineMemOperandTargetFlags();
257   for (const auto &I : Flags)
258     Names2MMOTargetFlags.insert(std::make_pair(StringRef(I.second), I.first));
259 }
260 
261 bool PerTargetMIParsingState::getMMOTargetFlag(StringRef Name,
262                                                MachineMemOperand::Flags &Flag) {
263   initNames2MMOTargetFlags();
264   auto FlagInfo = Names2MMOTargetFlags.find(Name);
265   if (FlagInfo == Names2MMOTargetFlags.end())
266     return true;
267   Flag = FlagInfo->second;
268   return false;
269 }
270 
271 void PerTargetMIParsingState::initNames2RegClasses() {
272   if (!Names2RegClasses.empty())
273     return;
274 
275   const TargetRegisterInfo *TRI = Subtarget.getRegisterInfo();
276   for (unsigned I = 0, E = TRI->getNumRegClasses(); I < E; ++I) {
277     const auto *RC = TRI->getRegClass(I);
278     Names2RegClasses.insert(
279         std::make_pair(StringRef(TRI->getRegClassName(RC)).lower(), RC));
280   }
281 }
282 
283 void PerTargetMIParsingState::initNames2RegBanks() {
284   if (!Names2RegBanks.empty())
285     return;
286 
287   const RegisterBankInfo *RBI = Subtarget.getRegBankInfo();
288   // If the target does not support GlobalISel, we may not have a
289   // register bank info.
290   if (!RBI)
291     return;
292 
293   for (unsigned I = 0, E = RBI->getNumRegBanks(); I < E; ++I) {
294     const auto &RegBank = RBI->getRegBank(I);
295     Names2RegBanks.insert(
296         std::make_pair(StringRef(RegBank.getName()).lower(), &RegBank));
297   }
298 }
299 
300 const TargetRegisterClass *
301 PerTargetMIParsingState::getRegClass(StringRef Name) {
302   auto RegClassInfo = Names2RegClasses.find(Name);
303   if (RegClassInfo == Names2RegClasses.end())
304     return nullptr;
305   return RegClassInfo->getValue();
306 }
307 
308 const RegisterBank *PerTargetMIParsingState::getRegBank(StringRef Name) {
309   auto RegBankInfo = Names2RegBanks.find(Name);
310   if (RegBankInfo == Names2RegBanks.end())
311     return nullptr;
312   return RegBankInfo->getValue();
313 }
314 
315 PerFunctionMIParsingState::PerFunctionMIParsingState(MachineFunction &MF,
316     SourceMgr &SM, const SlotMapping &IRSlots, PerTargetMIParsingState &T)
317   : MF(MF), SM(&SM), IRSlots(IRSlots), Target(T) {
318 }
319 
320 VRegInfo &PerFunctionMIParsingState::getVRegInfo(Register Num) {
321   auto I = VRegInfos.insert(std::make_pair(Num, nullptr));
322   if (I.second) {
323     MachineRegisterInfo &MRI = MF.getRegInfo();
324     VRegInfo *Info = new (Allocator) VRegInfo;
325     Info->VReg = MRI.createIncompleteVirtualRegister();
326     I.first->second = Info;
327   }
328   return *I.first->second;
329 }
330 
331 VRegInfo &PerFunctionMIParsingState::getVRegInfoNamed(StringRef RegName) {
332   assert(RegName != "" && "Expected named reg.");
333 
334   auto I = VRegInfosNamed.insert(std::make_pair(RegName.str(), nullptr));
335   if (I.second) {
336     VRegInfo *Info = new (Allocator) VRegInfo;
337     Info->VReg = MF.getRegInfo().createIncompleteVirtualRegister(RegName);
338     I.first->second = Info;
339   }
340   return *I.first->second;
341 }
342 
343 static void mapValueToSlot(const Value *V, ModuleSlotTracker &MST,
344                            DenseMap<unsigned, const Value *> &Slots2Values) {
345   int Slot = MST.getLocalSlot(V);
346   if (Slot == -1)
347     return;
348   Slots2Values.insert(std::make_pair(unsigned(Slot), V));
349 }
350 
351 /// Creates the mapping from slot numbers to function's unnamed IR values.
352 static void initSlots2Values(const Function &F,
353                              DenseMap<unsigned, const Value *> &Slots2Values) {
354   ModuleSlotTracker MST(F.getParent(), /*ShouldInitializeAllMetadata=*/false);
355   MST.incorporateFunction(F);
356   for (const auto &Arg : F.args())
357     mapValueToSlot(&Arg, MST, Slots2Values);
358   for (const auto &BB : F) {
359     mapValueToSlot(&BB, MST, Slots2Values);
360     for (const auto &I : BB)
361       mapValueToSlot(&I, MST, Slots2Values);
362   }
363 }
364 
365 const Value* PerFunctionMIParsingState::getIRValue(unsigned Slot) {
366   if (Slots2Values.empty())
367     initSlots2Values(MF.getFunction(), Slots2Values);
368   return Slots2Values.lookup(Slot);
369 }
370 
371 namespace {
372 
373 /// A wrapper struct around the 'MachineOperand' struct that includes a source
374 /// range and other attributes.
375 struct ParsedMachineOperand {
376   MachineOperand Operand;
377   StringRef::iterator Begin;
378   StringRef::iterator End;
379   std::optional<unsigned> TiedDefIdx;
380 
381   ParsedMachineOperand(const MachineOperand &Operand, StringRef::iterator Begin,
382                        StringRef::iterator End,
383                        std::optional<unsigned> &TiedDefIdx)
384       : Operand(Operand), Begin(Begin), End(End), TiedDefIdx(TiedDefIdx) {
385     if (TiedDefIdx)
386       assert(Operand.isReg() && Operand.isUse() &&
387              "Only used register operands can be tied");
388   }
389 };
390 
391 class MIParser {
392   MachineFunction &MF;
393   SMDiagnostic &Error;
394   StringRef Source, CurrentSource;
395   SMRange SourceRange;
396   MIToken Token;
397   PerFunctionMIParsingState &PFS;
398   /// Maps from slot numbers to function's unnamed basic blocks.
399   DenseMap<unsigned, const BasicBlock *> Slots2BasicBlocks;
400 
401 public:
402   MIParser(PerFunctionMIParsingState &PFS, SMDiagnostic &Error,
403            StringRef Source);
404   MIParser(PerFunctionMIParsingState &PFS, SMDiagnostic &Error,
405            StringRef Source, SMRange SourceRange);
406 
407   /// \p SkipChar gives the number of characters to skip before looking
408   /// for the next token.
409   void lex(unsigned SkipChar = 0);
410 
411   /// Report an error at the current location with the given message.
412   ///
413   /// This function always return true.
414   bool error(const Twine &Msg);
415 
416   /// Report an error at the given location with the given message.
417   ///
418   /// This function always return true.
419   bool error(StringRef::iterator Loc, const Twine &Msg);
420 
421   bool
422   parseBasicBlockDefinitions(DenseMap<unsigned, MachineBasicBlock *> &MBBSlots);
423   bool parseBasicBlocks();
424   bool parse(MachineInstr *&MI);
425   bool parseStandaloneMBB(MachineBasicBlock *&MBB);
426   bool parseStandaloneNamedRegister(Register &Reg);
427   bool parseStandaloneVirtualRegister(VRegInfo *&Info);
428   bool parseStandaloneRegister(Register &Reg);
429   bool parseStandaloneStackObject(int &FI);
430   bool parseStandaloneMDNode(MDNode *&Node);
431   bool parseMachineMetadata();
432   bool parseMDTuple(MDNode *&MD, bool IsDistinct);
433   bool parseMDNodeVector(SmallVectorImpl<Metadata *> &Elts);
434   bool parseMetadata(Metadata *&MD);
435 
436   bool
437   parseBasicBlockDefinition(DenseMap<unsigned, MachineBasicBlock *> &MBBSlots);
438   bool parseBasicBlock(MachineBasicBlock &MBB,
439                        MachineBasicBlock *&AddFalthroughFrom);
440   bool parseBasicBlockLiveins(MachineBasicBlock &MBB);
441   bool parseBasicBlockSuccessors(MachineBasicBlock &MBB);
442 
443   bool parseNamedRegister(Register &Reg);
444   bool parseVirtualRegister(VRegInfo *&Info);
445   bool parseNamedVirtualRegister(VRegInfo *&Info);
446   bool parseRegister(Register &Reg, VRegInfo *&VRegInfo);
447   bool parseRegisterFlag(unsigned &Flags);
448   bool parseRegisterClassOrBank(VRegInfo &RegInfo);
449   bool parseSubRegisterIndex(unsigned &SubReg);
450   bool parseRegisterTiedDefIndex(unsigned &TiedDefIdx);
451   bool parseRegisterOperand(MachineOperand &Dest,
452                             std::optional<unsigned> &TiedDefIdx,
453                             bool IsDef = false);
454   bool parseImmediateOperand(MachineOperand &Dest);
455   bool parseIRConstant(StringRef::iterator Loc, StringRef StringValue,
456                        const Constant *&C);
457   bool parseIRConstant(StringRef::iterator Loc, const Constant *&C);
458   bool parseLowLevelType(StringRef::iterator Loc, LLT &Ty);
459   bool parseTypedImmediateOperand(MachineOperand &Dest);
460   bool parseFPImmediateOperand(MachineOperand &Dest);
461   bool parseMBBReference(MachineBasicBlock *&MBB);
462   bool parseMBBOperand(MachineOperand &Dest);
463   bool parseStackFrameIndex(int &FI);
464   bool parseStackObjectOperand(MachineOperand &Dest);
465   bool parseFixedStackFrameIndex(int &FI);
466   bool parseFixedStackObjectOperand(MachineOperand &Dest);
467   bool parseGlobalValue(GlobalValue *&GV);
468   bool parseGlobalAddressOperand(MachineOperand &Dest);
469   bool parseConstantPoolIndexOperand(MachineOperand &Dest);
470   bool parseSubRegisterIndexOperand(MachineOperand &Dest);
471   bool parseJumpTableIndexOperand(MachineOperand &Dest);
472   bool parseExternalSymbolOperand(MachineOperand &Dest);
473   bool parseMCSymbolOperand(MachineOperand &Dest);
474   [[nodiscard]] bool parseMDNode(MDNode *&Node);
475   bool parseDIExpression(MDNode *&Expr);
476   bool parseDILocation(MDNode *&Expr);
477   bool parseMetadataOperand(MachineOperand &Dest);
478   bool parseCFIOffset(int &Offset);
479   bool parseCFIRegister(Register &Reg);
480   bool parseCFIAddressSpace(unsigned &AddressSpace);
481   bool parseCFIEscapeValues(std::string& Values);
482   bool parseCFIOperand(MachineOperand &Dest);
483   bool parseIRBlock(BasicBlock *&BB, const Function &F);
484   bool parseBlockAddressOperand(MachineOperand &Dest);
485   bool parseIntrinsicOperand(MachineOperand &Dest);
486   bool parsePredicateOperand(MachineOperand &Dest);
487   bool parseShuffleMaskOperand(MachineOperand &Dest);
488   bool parseTargetIndexOperand(MachineOperand &Dest);
489   bool parseDbgInstrRefOperand(MachineOperand &Dest);
490   bool parseCustomRegisterMaskOperand(MachineOperand &Dest);
491   bool parseLiveoutRegisterMaskOperand(MachineOperand &Dest);
492   bool parseMachineOperand(const unsigned OpCode, const unsigned OpIdx,
493                            MachineOperand &Dest,
494                            std::optional<unsigned> &TiedDefIdx);
495   bool parseMachineOperandAndTargetFlags(const unsigned OpCode,
496                                          const unsigned OpIdx,
497                                          MachineOperand &Dest,
498                                          std::optional<unsigned> &TiedDefIdx);
499   bool parseOffset(int64_t &Offset);
500   bool parseIRBlockAddressTaken(BasicBlock *&BB);
501   bool parseAlignment(uint64_t &Alignment);
502   bool parseAddrspace(unsigned &Addrspace);
503   bool parseSectionID(std::optional<MBBSectionID> &SID);
504   bool parseBBID(std::optional<UniqueBBID> &BBID);
505   bool parseCallFrameSize(unsigned &CallFrameSize);
506   bool parseOperandsOffset(MachineOperand &Op);
507   bool parseIRValue(const Value *&V);
508   bool parseMemoryOperandFlag(MachineMemOperand::Flags &Flags);
509   bool parseMemoryPseudoSourceValue(const PseudoSourceValue *&PSV);
510   bool parseMachinePointerInfo(MachinePointerInfo &Dest);
511   bool parseOptionalScope(LLVMContext &Context, SyncScope::ID &SSID);
512   bool parseOptionalAtomicOrdering(AtomicOrdering &Order);
513   bool parseMachineMemoryOperand(MachineMemOperand *&Dest);
514   bool parsePreOrPostInstrSymbol(MCSymbol *&Symbol);
515   bool parseHeapAllocMarker(MDNode *&Node);
516   bool parsePCSections(MDNode *&Node);
517 
518   bool parseTargetImmMnemonic(const unsigned OpCode, const unsigned OpIdx,
519                               MachineOperand &Dest, const MIRFormatter &MF);
520 
521 private:
522   /// Convert the integer literal in the current token into an unsigned integer.
523   ///
524   /// Return true if an error occurred.
525   bool getUnsigned(unsigned &Result);
526 
527   /// Convert the integer literal in the current token into an uint64.
528   ///
529   /// Return true if an error occurred.
530   bool getUint64(uint64_t &Result);
531 
532   /// Convert the hexadecimal literal in the current token into an unsigned
533   ///  APInt with a minimum bitwidth required to represent the value.
534   ///
535   /// Return true if the literal does not represent an integer value.
536   bool getHexUint(APInt &Result);
537 
538   /// If the current token is of the given kind, consume it and return false.
539   /// Otherwise report an error and return true.
540   bool expectAndConsume(MIToken::TokenKind TokenKind);
541 
542   /// If the current token is of the given kind, consume it and return true.
543   /// Otherwise return false.
544   bool consumeIfPresent(MIToken::TokenKind TokenKind);
545 
546   bool parseInstruction(unsigned &OpCode, unsigned &Flags);
547 
548   bool assignRegisterTies(MachineInstr &MI,
549                           ArrayRef<ParsedMachineOperand> Operands);
550 
551   bool verifyImplicitOperands(ArrayRef<ParsedMachineOperand> Operands,
552                               const MCInstrDesc &MCID);
553 
554   const BasicBlock *getIRBlock(unsigned Slot);
555   const BasicBlock *getIRBlock(unsigned Slot, const Function &F);
556 
557   /// Get or create an MCSymbol for a given name.
558   MCSymbol *getOrCreateMCSymbol(StringRef Name);
559 
560   /// parseStringConstant
561   ///   ::= StringConstant
562   bool parseStringConstant(std::string &Result);
563 
564   /// Map the location in the MI string to the corresponding location specified
565   /// in `SourceRange`.
566   SMLoc mapSMLoc(StringRef::iterator Loc);
567 };
568 
569 } // end anonymous namespace
570 
571 MIParser::MIParser(PerFunctionMIParsingState &PFS, SMDiagnostic &Error,
572                    StringRef Source)
573     : MF(PFS.MF), Error(Error), Source(Source), CurrentSource(Source), PFS(PFS)
574 {}
575 
576 MIParser::MIParser(PerFunctionMIParsingState &PFS, SMDiagnostic &Error,
577                    StringRef Source, SMRange SourceRange)
578     : MF(PFS.MF), Error(Error), Source(Source), CurrentSource(Source),
579       SourceRange(SourceRange), PFS(PFS) {}
580 
581 void MIParser::lex(unsigned SkipChar) {
582   CurrentSource = lexMIToken(
583       CurrentSource.slice(SkipChar, StringRef::npos), Token,
584       [this](StringRef::iterator Loc, const Twine &Msg) { error(Loc, Msg); });
585 }
586 
587 bool MIParser::error(const Twine &Msg) { return error(Token.location(), Msg); }
588 
589 bool MIParser::error(StringRef::iterator Loc, const Twine &Msg) {
590   const SourceMgr &SM = *PFS.SM;
591   assert(Loc >= Source.data() && Loc <= (Source.data() + Source.size()));
592   const MemoryBuffer &Buffer = *SM.getMemoryBuffer(SM.getMainFileID());
593   if (Loc >= Buffer.getBufferStart() && Loc <= Buffer.getBufferEnd()) {
594     // Create an ordinary diagnostic when the source manager's buffer is the
595     // source string.
596     Error = SM.GetMessage(SMLoc::getFromPointer(Loc), SourceMgr::DK_Error, Msg);
597     return true;
598   }
599   // Create a diagnostic for a YAML string literal.
600   Error = SMDiagnostic(SM, SMLoc(), Buffer.getBufferIdentifier(), 1,
601                        Loc - Source.data(), SourceMgr::DK_Error, Msg.str(),
602                        Source, std::nullopt, std::nullopt);
603   return true;
604 }
605 
606 SMLoc MIParser::mapSMLoc(StringRef::iterator Loc) {
607   assert(SourceRange.isValid() && "Invalid source range");
608   assert(Loc >= Source.data() && Loc <= (Source.data() + Source.size()));
609   return SMLoc::getFromPointer(SourceRange.Start.getPointer() +
610                                (Loc - Source.data()));
611 }
612 
613 typedef function_ref<bool(StringRef::iterator Loc, const Twine &)>
614     ErrorCallbackType;
615 
616 static const char *toString(MIToken::TokenKind TokenKind) {
617   switch (TokenKind) {
618   case MIToken::comma:
619     return "','";
620   case MIToken::equal:
621     return "'='";
622   case MIToken::colon:
623     return "':'";
624   case MIToken::lparen:
625     return "'('";
626   case MIToken::rparen:
627     return "')'";
628   default:
629     return "<unknown token>";
630   }
631 }
632 
633 bool MIParser::expectAndConsume(MIToken::TokenKind TokenKind) {
634   if (Token.isNot(TokenKind))
635     return error(Twine("expected ") + toString(TokenKind));
636   lex();
637   return false;
638 }
639 
640 bool MIParser::consumeIfPresent(MIToken::TokenKind TokenKind) {
641   if (Token.isNot(TokenKind))
642     return false;
643   lex();
644   return true;
645 }
646 
647 // Parse Machine Basic Block Section ID.
648 bool MIParser::parseSectionID(std::optional<MBBSectionID> &SID) {
649   assert(Token.is(MIToken::kw_bbsections));
650   lex();
651   if (Token.is(MIToken::IntegerLiteral)) {
652     unsigned Value = 0;
653     if (getUnsigned(Value))
654       return error("Unknown Section ID");
655     SID = MBBSectionID{Value};
656   } else {
657     const StringRef &S = Token.stringValue();
658     if (S == "Exception")
659       SID = MBBSectionID::ExceptionSectionID;
660     else if (S == "Cold")
661       SID = MBBSectionID::ColdSectionID;
662     else
663       return error("Unknown Section ID");
664   }
665   lex();
666   return false;
667 }
668 
669 // Parse Machine Basic Block ID.
670 bool MIParser::parseBBID(std::optional<UniqueBBID> &BBID) {
671   assert(Token.is(MIToken::kw_bb_id));
672   lex();
673   unsigned BaseID = 0;
674   unsigned CloneID = 0;
675   if (getUnsigned(BaseID))
676     return error("Unknown BB ID");
677   lex();
678   if (Token.is(MIToken::IntegerLiteral)) {
679     if (getUnsigned(CloneID))
680       return error("Unknown Clone ID");
681     lex();
682   }
683   BBID = {BaseID, CloneID};
684   return false;
685 }
686 
687 // Parse basic block call frame size.
688 bool MIParser::parseCallFrameSize(unsigned &CallFrameSize) {
689   assert(Token.is(MIToken::kw_call_frame_size));
690   lex();
691   unsigned Value = 0;
692   if (getUnsigned(Value))
693     return error("Unknown call frame size");
694   CallFrameSize = Value;
695   lex();
696   return false;
697 }
698 
699 bool MIParser::parseBasicBlockDefinition(
700     DenseMap<unsigned, MachineBasicBlock *> &MBBSlots) {
701   assert(Token.is(MIToken::MachineBasicBlockLabel));
702   unsigned ID = 0;
703   if (getUnsigned(ID))
704     return true;
705   auto Loc = Token.location();
706   auto Name = Token.stringValue();
707   lex();
708   bool MachineBlockAddressTaken = false;
709   BasicBlock *AddressTakenIRBlock = nullptr;
710   bool IsLandingPad = false;
711   bool IsInlineAsmBrIndirectTarget = false;
712   bool IsEHFuncletEntry = false;
713   std::optional<MBBSectionID> SectionID;
714   uint64_t Alignment = 0;
715   std::optional<UniqueBBID> BBID;
716   unsigned CallFrameSize = 0;
717   BasicBlock *BB = nullptr;
718   if (consumeIfPresent(MIToken::lparen)) {
719     do {
720       // TODO: Report an error when multiple same attributes are specified.
721       switch (Token.kind()) {
722       case MIToken::kw_machine_block_address_taken:
723         MachineBlockAddressTaken = true;
724         lex();
725         break;
726       case MIToken::kw_ir_block_address_taken:
727         if (parseIRBlockAddressTaken(AddressTakenIRBlock))
728           return true;
729         break;
730       case MIToken::kw_landing_pad:
731         IsLandingPad = true;
732         lex();
733         break;
734       case MIToken::kw_inlineasm_br_indirect_target:
735         IsInlineAsmBrIndirectTarget = true;
736         lex();
737         break;
738       case MIToken::kw_ehfunclet_entry:
739         IsEHFuncletEntry = true;
740         lex();
741         break;
742       case MIToken::kw_align:
743         if (parseAlignment(Alignment))
744           return true;
745         break;
746       case MIToken::IRBlock:
747       case MIToken::NamedIRBlock:
748         // TODO: Report an error when both name and ir block are specified.
749         if (parseIRBlock(BB, MF.getFunction()))
750           return true;
751         lex();
752         break;
753       case MIToken::kw_bbsections:
754         if (parseSectionID(SectionID))
755           return true;
756         break;
757       case MIToken::kw_bb_id:
758         if (parseBBID(BBID))
759           return true;
760         break;
761       case MIToken::kw_call_frame_size:
762         if (parseCallFrameSize(CallFrameSize))
763           return true;
764         break;
765       default:
766         break;
767       }
768     } while (consumeIfPresent(MIToken::comma));
769     if (expectAndConsume(MIToken::rparen))
770       return true;
771   }
772   if (expectAndConsume(MIToken::colon))
773     return true;
774 
775   if (!Name.empty()) {
776     BB = dyn_cast_or_null<BasicBlock>(
777         MF.getFunction().getValueSymbolTable()->lookup(Name));
778     if (!BB)
779       return error(Loc, Twine("basic block '") + Name +
780                             "' is not defined in the function '" +
781                             MF.getName() + "'");
782   }
783   auto *MBB = MF.CreateMachineBasicBlock(BB);
784   MF.insert(MF.end(), MBB);
785   bool WasInserted = MBBSlots.insert(std::make_pair(ID, MBB)).second;
786   if (!WasInserted)
787     return error(Loc, Twine("redefinition of machine basic block with id #") +
788                           Twine(ID));
789   if (Alignment)
790     MBB->setAlignment(Align(Alignment));
791   if (MachineBlockAddressTaken)
792     MBB->setMachineBlockAddressTaken();
793   if (AddressTakenIRBlock)
794     MBB->setAddressTakenIRBlock(AddressTakenIRBlock);
795   MBB->setIsEHPad(IsLandingPad);
796   MBB->setIsInlineAsmBrIndirectTarget(IsInlineAsmBrIndirectTarget);
797   MBB->setIsEHFuncletEntry(IsEHFuncletEntry);
798   if (SectionID) {
799     MBB->setSectionID(*SectionID);
800     MF.setBBSectionsType(BasicBlockSection::List);
801   }
802   if (BBID.has_value()) {
803     // BBSectionsType is set to `List` if any basic blocks has `SectionID`.
804     // Here, we set it to `Labels` if it hasn't been set above.
805     if (!MF.hasBBSections())
806       MF.setBBSectionsType(BasicBlockSection::Labels);
807     MBB->setBBID(BBID.value());
808   }
809   MBB->setCallFrameSize(CallFrameSize);
810   return false;
811 }
812 
813 bool MIParser::parseBasicBlockDefinitions(
814     DenseMap<unsigned, MachineBasicBlock *> &MBBSlots) {
815   lex();
816   // Skip until the first machine basic block.
817   while (Token.is(MIToken::Newline))
818     lex();
819   if (Token.isErrorOrEOF())
820     return Token.isError();
821   if (Token.isNot(MIToken::MachineBasicBlockLabel))
822     return error("expected a basic block definition before instructions");
823   unsigned BraceDepth = 0;
824   do {
825     if (parseBasicBlockDefinition(MBBSlots))
826       return true;
827     bool IsAfterNewline = false;
828     // Skip until the next machine basic block.
829     while (true) {
830       if ((Token.is(MIToken::MachineBasicBlockLabel) && IsAfterNewline) ||
831           Token.isErrorOrEOF())
832         break;
833       else if (Token.is(MIToken::MachineBasicBlockLabel))
834         return error("basic block definition should be located at the start of "
835                      "the line");
836       else if (consumeIfPresent(MIToken::Newline)) {
837         IsAfterNewline = true;
838         continue;
839       }
840       IsAfterNewline = false;
841       if (Token.is(MIToken::lbrace))
842         ++BraceDepth;
843       if (Token.is(MIToken::rbrace)) {
844         if (!BraceDepth)
845           return error("extraneous closing brace ('}')");
846         --BraceDepth;
847       }
848       lex();
849     }
850     // Verify that we closed all of the '{' at the end of a file or a block.
851     if (!Token.isError() && BraceDepth)
852       return error("expected '}'"); // FIXME: Report a note that shows '{'.
853   } while (!Token.isErrorOrEOF());
854   return Token.isError();
855 }
856 
857 bool MIParser::parseBasicBlockLiveins(MachineBasicBlock &MBB) {
858   assert(Token.is(MIToken::kw_liveins));
859   lex();
860   if (expectAndConsume(MIToken::colon))
861     return true;
862   if (Token.isNewlineOrEOF()) // Allow an empty list of liveins.
863     return false;
864   do {
865     if (Token.isNot(MIToken::NamedRegister))
866       return error("expected a named register");
867     Register Reg;
868     if (parseNamedRegister(Reg))
869       return true;
870     lex();
871     LaneBitmask Mask = LaneBitmask::getAll();
872     if (consumeIfPresent(MIToken::colon)) {
873       // Parse lane mask.
874       if (Token.isNot(MIToken::IntegerLiteral) &&
875           Token.isNot(MIToken::HexLiteral))
876         return error("expected a lane mask");
877       static_assert(sizeof(LaneBitmask::Type) == sizeof(uint64_t),
878                     "Use correct get-function for lane mask");
879       LaneBitmask::Type V;
880       if (getUint64(V))
881         return error("invalid lane mask value");
882       Mask = LaneBitmask(V);
883       lex();
884     }
885     MBB.addLiveIn(Reg, Mask);
886   } while (consumeIfPresent(MIToken::comma));
887   return false;
888 }
889 
890 bool MIParser::parseBasicBlockSuccessors(MachineBasicBlock &MBB) {
891   assert(Token.is(MIToken::kw_successors));
892   lex();
893   if (expectAndConsume(MIToken::colon))
894     return true;
895   if (Token.isNewlineOrEOF()) // Allow an empty list of successors.
896     return false;
897   do {
898     if (Token.isNot(MIToken::MachineBasicBlock))
899       return error("expected a machine basic block reference");
900     MachineBasicBlock *SuccMBB = nullptr;
901     if (parseMBBReference(SuccMBB))
902       return true;
903     lex();
904     unsigned Weight = 0;
905     if (consumeIfPresent(MIToken::lparen)) {
906       if (Token.isNot(MIToken::IntegerLiteral) &&
907           Token.isNot(MIToken::HexLiteral))
908         return error("expected an integer literal after '('");
909       if (getUnsigned(Weight))
910         return true;
911       lex();
912       if (expectAndConsume(MIToken::rparen))
913         return true;
914     }
915     MBB.addSuccessor(SuccMBB, BranchProbability::getRaw(Weight));
916   } while (consumeIfPresent(MIToken::comma));
917   MBB.normalizeSuccProbs();
918   return false;
919 }
920 
921 bool MIParser::parseBasicBlock(MachineBasicBlock &MBB,
922                                MachineBasicBlock *&AddFalthroughFrom) {
923   // Skip the definition.
924   assert(Token.is(MIToken::MachineBasicBlockLabel));
925   lex();
926   if (consumeIfPresent(MIToken::lparen)) {
927     while (Token.isNot(MIToken::rparen) && !Token.isErrorOrEOF())
928       lex();
929     consumeIfPresent(MIToken::rparen);
930   }
931   consumeIfPresent(MIToken::colon);
932 
933   // Parse the liveins and successors.
934   // N.B: Multiple lists of successors and liveins are allowed and they're
935   // merged into one.
936   // Example:
937   //   liveins: $edi
938   //   liveins: $esi
939   //
940   // is equivalent to
941   //   liveins: $edi, $esi
942   bool ExplicitSuccessors = false;
943   while (true) {
944     if (Token.is(MIToken::kw_successors)) {
945       if (parseBasicBlockSuccessors(MBB))
946         return true;
947       ExplicitSuccessors = true;
948     } else if (Token.is(MIToken::kw_liveins)) {
949       if (parseBasicBlockLiveins(MBB))
950         return true;
951     } else if (consumeIfPresent(MIToken::Newline)) {
952       continue;
953     } else
954       break;
955     if (!Token.isNewlineOrEOF())
956       return error("expected line break at the end of a list");
957     lex();
958   }
959 
960   // Parse the instructions.
961   bool IsInBundle = false;
962   MachineInstr *PrevMI = nullptr;
963   while (!Token.is(MIToken::MachineBasicBlockLabel) &&
964          !Token.is(MIToken::Eof)) {
965     if (consumeIfPresent(MIToken::Newline))
966       continue;
967     if (consumeIfPresent(MIToken::rbrace)) {
968       // The first parsing pass should verify that all closing '}' have an
969       // opening '{'.
970       assert(IsInBundle);
971       IsInBundle = false;
972       continue;
973     }
974     MachineInstr *MI = nullptr;
975     if (parse(MI))
976       return true;
977     MBB.insert(MBB.end(), MI);
978     if (IsInBundle) {
979       PrevMI->setFlag(MachineInstr::BundledSucc);
980       MI->setFlag(MachineInstr::BundledPred);
981     }
982     PrevMI = MI;
983     if (Token.is(MIToken::lbrace)) {
984       if (IsInBundle)
985         return error("nested instruction bundles are not allowed");
986       lex();
987       // This instruction is the start of the bundle.
988       MI->setFlag(MachineInstr::BundledSucc);
989       IsInBundle = true;
990       if (!Token.is(MIToken::Newline))
991         // The next instruction can be on the same line.
992         continue;
993     }
994     assert(Token.isNewlineOrEOF() && "MI is not fully parsed");
995     lex();
996   }
997 
998   // Construct successor list by searching for basic block machine operands.
999   if (!ExplicitSuccessors) {
1000     SmallVector<MachineBasicBlock*,4> Successors;
1001     bool IsFallthrough;
1002     guessSuccessors(MBB, Successors, IsFallthrough);
1003     for (MachineBasicBlock *Succ : Successors)
1004       MBB.addSuccessor(Succ);
1005 
1006     if (IsFallthrough) {
1007       AddFalthroughFrom = &MBB;
1008     } else {
1009       MBB.normalizeSuccProbs();
1010     }
1011   }
1012 
1013   return false;
1014 }
1015 
1016 bool MIParser::parseBasicBlocks() {
1017   lex();
1018   // Skip until the first machine basic block.
1019   while (Token.is(MIToken::Newline))
1020     lex();
1021   if (Token.isErrorOrEOF())
1022     return Token.isError();
1023   // The first parsing pass should have verified that this token is a MBB label
1024   // in the 'parseBasicBlockDefinitions' method.
1025   assert(Token.is(MIToken::MachineBasicBlockLabel));
1026   MachineBasicBlock *AddFalthroughFrom = nullptr;
1027   do {
1028     MachineBasicBlock *MBB = nullptr;
1029     if (parseMBBReference(MBB))
1030       return true;
1031     if (AddFalthroughFrom) {
1032       if (!AddFalthroughFrom->isSuccessor(MBB))
1033         AddFalthroughFrom->addSuccessor(MBB);
1034       AddFalthroughFrom->normalizeSuccProbs();
1035       AddFalthroughFrom = nullptr;
1036     }
1037     if (parseBasicBlock(*MBB, AddFalthroughFrom))
1038       return true;
1039     // The method 'parseBasicBlock' should parse the whole block until the next
1040     // block or the end of file.
1041     assert(Token.is(MIToken::MachineBasicBlockLabel) || Token.is(MIToken::Eof));
1042   } while (Token.isNot(MIToken::Eof));
1043   return false;
1044 }
1045 
1046 bool MIParser::parse(MachineInstr *&MI) {
1047   // Parse any register operands before '='
1048   MachineOperand MO = MachineOperand::CreateImm(0);
1049   SmallVector<ParsedMachineOperand, 8> Operands;
1050   while (Token.isRegister() || Token.isRegisterFlag()) {
1051     auto Loc = Token.location();
1052     std::optional<unsigned> TiedDefIdx;
1053     if (parseRegisterOperand(MO, TiedDefIdx, /*IsDef=*/true))
1054       return true;
1055     Operands.push_back(
1056         ParsedMachineOperand(MO, Loc, Token.location(), TiedDefIdx));
1057     if (Token.isNot(MIToken::comma))
1058       break;
1059     lex();
1060   }
1061   if (!Operands.empty() && expectAndConsume(MIToken::equal))
1062     return true;
1063 
1064   unsigned OpCode, Flags = 0;
1065   if (Token.isError() || parseInstruction(OpCode, Flags))
1066     return true;
1067 
1068   // Parse the remaining machine operands.
1069   while (!Token.isNewlineOrEOF() && Token.isNot(MIToken::kw_pre_instr_symbol) &&
1070          Token.isNot(MIToken::kw_post_instr_symbol) &&
1071          Token.isNot(MIToken::kw_heap_alloc_marker) &&
1072          Token.isNot(MIToken::kw_pcsections) &&
1073          Token.isNot(MIToken::kw_cfi_type) &&
1074          Token.isNot(MIToken::kw_debug_location) &&
1075          Token.isNot(MIToken::kw_debug_instr_number) &&
1076          Token.isNot(MIToken::coloncolon) && Token.isNot(MIToken::lbrace)) {
1077     auto Loc = Token.location();
1078     std::optional<unsigned> TiedDefIdx;
1079     if (parseMachineOperandAndTargetFlags(OpCode, Operands.size(), MO, TiedDefIdx))
1080       return true;
1081     Operands.push_back(
1082         ParsedMachineOperand(MO, Loc, Token.location(), TiedDefIdx));
1083     if (Token.isNewlineOrEOF() || Token.is(MIToken::coloncolon) ||
1084         Token.is(MIToken::lbrace))
1085       break;
1086     if (Token.isNot(MIToken::comma))
1087       return error("expected ',' before the next machine operand");
1088     lex();
1089   }
1090 
1091   MCSymbol *PreInstrSymbol = nullptr;
1092   if (Token.is(MIToken::kw_pre_instr_symbol))
1093     if (parsePreOrPostInstrSymbol(PreInstrSymbol))
1094       return true;
1095   MCSymbol *PostInstrSymbol = nullptr;
1096   if (Token.is(MIToken::kw_post_instr_symbol))
1097     if (parsePreOrPostInstrSymbol(PostInstrSymbol))
1098       return true;
1099   MDNode *HeapAllocMarker = nullptr;
1100   if (Token.is(MIToken::kw_heap_alloc_marker))
1101     if (parseHeapAllocMarker(HeapAllocMarker))
1102       return true;
1103   MDNode *PCSections = nullptr;
1104   if (Token.is(MIToken::kw_pcsections))
1105     if (parsePCSections(PCSections))
1106       return true;
1107 
1108   unsigned CFIType = 0;
1109   if (Token.is(MIToken::kw_cfi_type)) {
1110     lex();
1111     if (Token.isNot(MIToken::IntegerLiteral))
1112       return error("expected an integer literal after 'cfi-type'");
1113     // getUnsigned is sufficient for 32-bit integers.
1114     if (getUnsigned(CFIType))
1115       return true;
1116     lex();
1117     // Lex past trailing comma if present.
1118     if (Token.is(MIToken::comma))
1119       lex();
1120   }
1121 
1122   unsigned InstrNum = 0;
1123   if (Token.is(MIToken::kw_debug_instr_number)) {
1124     lex();
1125     if (Token.isNot(MIToken::IntegerLiteral))
1126       return error("expected an integer literal after 'debug-instr-number'");
1127     if (getUnsigned(InstrNum))
1128       return true;
1129     lex();
1130     // Lex past trailing comma if present.
1131     if (Token.is(MIToken::comma))
1132       lex();
1133   }
1134 
1135   DebugLoc DebugLocation;
1136   if (Token.is(MIToken::kw_debug_location)) {
1137     lex();
1138     MDNode *Node = nullptr;
1139     if (Token.is(MIToken::exclaim)) {
1140       if (parseMDNode(Node))
1141         return true;
1142     } else if (Token.is(MIToken::md_dilocation)) {
1143       if (parseDILocation(Node))
1144         return true;
1145     } else
1146       return error("expected a metadata node after 'debug-location'");
1147     if (!isa<DILocation>(Node))
1148       return error("referenced metadata is not a DILocation");
1149     DebugLocation = DebugLoc(Node);
1150   }
1151 
1152   // Parse the machine memory operands.
1153   SmallVector<MachineMemOperand *, 2> MemOperands;
1154   if (Token.is(MIToken::coloncolon)) {
1155     lex();
1156     while (!Token.isNewlineOrEOF()) {
1157       MachineMemOperand *MemOp = nullptr;
1158       if (parseMachineMemoryOperand(MemOp))
1159         return true;
1160       MemOperands.push_back(MemOp);
1161       if (Token.isNewlineOrEOF())
1162         break;
1163       if (Token.isNot(MIToken::comma))
1164         return error("expected ',' before the next machine memory operand");
1165       lex();
1166     }
1167   }
1168 
1169   const auto &MCID = MF.getSubtarget().getInstrInfo()->get(OpCode);
1170   if (!MCID.isVariadic()) {
1171     // FIXME: Move the implicit operand verification to the machine verifier.
1172     if (verifyImplicitOperands(Operands, MCID))
1173       return true;
1174   }
1175 
1176   MI = MF.CreateMachineInstr(MCID, DebugLocation, /*NoImplicit=*/true);
1177   MI->setFlags(Flags);
1178 
1179   // Don't check the operands make sense, let the verifier catch any
1180   // improprieties.
1181   for (const auto &Operand : Operands)
1182     MI->addOperand(MF, Operand.Operand);
1183 
1184   if (assignRegisterTies(*MI, Operands))
1185     return true;
1186   if (PreInstrSymbol)
1187     MI->setPreInstrSymbol(MF, PreInstrSymbol);
1188   if (PostInstrSymbol)
1189     MI->setPostInstrSymbol(MF, PostInstrSymbol);
1190   if (HeapAllocMarker)
1191     MI->setHeapAllocMarker(MF, HeapAllocMarker);
1192   if (PCSections)
1193     MI->setPCSections(MF, PCSections);
1194   if (CFIType)
1195     MI->setCFIType(MF, CFIType);
1196   if (!MemOperands.empty())
1197     MI->setMemRefs(MF, MemOperands);
1198   if (InstrNum)
1199     MI->setDebugInstrNum(InstrNum);
1200   return false;
1201 }
1202 
1203 bool MIParser::parseStandaloneMBB(MachineBasicBlock *&MBB) {
1204   lex();
1205   if (Token.isNot(MIToken::MachineBasicBlock))
1206     return error("expected a machine basic block reference");
1207   if (parseMBBReference(MBB))
1208     return true;
1209   lex();
1210   if (Token.isNot(MIToken::Eof))
1211     return error(
1212         "expected end of string after the machine basic block reference");
1213   return false;
1214 }
1215 
1216 bool MIParser::parseStandaloneNamedRegister(Register &Reg) {
1217   lex();
1218   if (Token.isNot(MIToken::NamedRegister))
1219     return error("expected a named register");
1220   if (parseNamedRegister(Reg))
1221     return true;
1222   lex();
1223   if (Token.isNot(MIToken::Eof))
1224     return error("expected end of string after the register reference");
1225   return false;
1226 }
1227 
1228 bool MIParser::parseStandaloneVirtualRegister(VRegInfo *&Info) {
1229   lex();
1230   if (Token.isNot(MIToken::VirtualRegister))
1231     return error("expected a virtual register");
1232   if (parseVirtualRegister(Info))
1233     return true;
1234   lex();
1235   if (Token.isNot(MIToken::Eof))
1236     return error("expected end of string after the register reference");
1237   return false;
1238 }
1239 
1240 bool MIParser::parseStandaloneRegister(Register &Reg) {
1241   lex();
1242   if (Token.isNot(MIToken::NamedRegister) &&
1243       Token.isNot(MIToken::VirtualRegister))
1244     return error("expected either a named or virtual register");
1245 
1246   VRegInfo *Info;
1247   if (parseRegister(Reg, Info))
1248     return true;
1249 
1250   lex();
1251   if (Token.isNot(MIToken::Eof))
1252     return error("expected end of string after the register reference");
1253   return false;
1254 }
1255 
1256 bool MIParser::parseStandaloneStackObject(int &FI) {
1257   lex();
1258   if (Token.isNot(MIToken::StackObject))
1259     return error("expected a stack object");
1260   if (parseStackFrameIndex(FI))
1261     return true;
1262   if (Token.isNot(MIToken::Eof))
1263     return error("expected end of string after the stack object reference");
1264   return false;
1265 }
1266 
1267 bool MIParser::parseStandaloneMDNode(MDNode *&Node) {
1268   lex();
1269   if (Token.is(MIToken::exclaim)) {
1270     if (parseMDNode(Node))
1271       return true;
1272   } else if (Token.is(MIToken::md_diexpr)) {
1273     if (parseDIExpression(Node))
1274       return true;
1275   } else if (Token.is(MIToken::md_dilocation)) {
1276     if (parseDILocation(Node))
1277       return true;
1278   } else
1279     return error("expected a metadata node");
1280   if (Token.isNot(MIToken::Eof))
1281     return error("expected end of string after the metadata node");
1282   return false;
1283 }
1284 
1285 bool MIParser::parseMachineMetadata() {
1286   lex();
1287   if (Token.isNot(MIToken::exclaim))
1288     return error("expected a metadata node");
1289 
1290   lex();
1291   if (Token.isNot(MIToken::IntegerLiteral) || Token.integerValue().isSigned())
1292     return error("expected metadata id after '!'");
1293   unsigned ID = 0;
1294   if (getUnsigned(ID))
1295     return true;
1296   lex();
1297   if (expectAndConsume(MIToken::equal))
1298     return true;
1299   bool IsDistinct = Token.is(MIToken::kw_distinct);
1300   if (IsDistinct)
1301     lex();
1302   if (Token.isNot(MIToken::exclaim))
1303     return error("expected a metadata node");
1304   lex();
1305 
1306   MDNode *MD;
1307   if (parseMDTuple(MD, IsDistinct))
1308     return true;
1309 
1310   auto FI = PFS.MachineForwardRefMDNodes.find(ID);
1311   if (FI != PFS.MachineForwardRefMDNodes.end()) {
1312     FI->second.first->replaceAllUsesWith(MD);
1313     PFS.MachineForwardRefMDNodes.erase(FI);
1314 
1315     assert(PFS.MachineMetadataNodes[ID] == MD && "Tracking VH didn't work");
1316   } else {
1317     if (PFS.MachineMetadataNodes.count(ID))
1318       return error("Metadata id is already used");
1319     PFS.MachineMetadataNodes[ID].reset(MD);
1320   }
1321 
1322   return false;
1323 }
1324 
1325 bool MIParser::parseMDTuple(MDNode *&MD, bool IsDistinct) {
1326   SmallVector<Metadata *, 16> Elts;
1327   if (parseMDNodeVector(Elts))
1328     return true;
1329   MD = (IsDistinct ? MDTuple::getDistinct
1330                    : MDTuple::get)(MF.getFunction().getContext(), Elts);
1331   return false;
1332 }
1333 
1334 bool MIParser::parseMDNodeVector(SmallVectorImpl<Metadata *> &Elts) {
1335   if (Token.isNot(MIToken::lbrace))
1336     return error("expected '{' here");
1337   lex();
1338 
1339   if (Token.is(MIToken::rbrace)) {
1340     lex();
1341     return false;
1342   }
1343 
1344   do {
1345     Metadata *MD;
1346     if (parseMetadata(MD))
1347       return true;
1348 
1349     Elts.push_back(MD);
1350 
1351     if (Token.isNot(MIToken::comma))
1352       break;
1353     lex();
1354   } while (true);
1355 
1356   if (Token.isNot(MIToken::rbrace))
1357     return error("expected end of metadata node");
1358   lex();
1359 
1360   return false;
1361 }
1362 
1363 // ::= !42
1364 // ::= !"string"
1365 bool MIParser::parseMetadata(Metadata *&MD) {
1366   if (Token.isNot(MIToken::exclaim))
1367     return error("expected '!' here");
1368   lex();
1369 
1370   if (Token.is(MIToken::StringConstant)) {
1371     std::string Str;
1372     if (parseStringConstant(Str))
1373       return true;
1374     MD = MDString::get(MF.getFunction().getContext(), Str);
1375     return false;
1376   }
1377 
1378   if (Token.isNot(MIToken::IntegerLiteral) || Token.integerValue().isSigned())
1379     return error("expected metadata id after '!'");
1380 
1381   SMLoc Loc = mapSMLoc(Token.location());
1382 
1383   unsigned ID = 0;
1384   if (getUnsigned(ID))
1385     return true;
1386   lex();
1387 
1388   auto NodeInfo = PFS.IRSlots.MetadataNodes.find(ID);
1389   if (NodeInfo != PFS.IRSlots.MetadataNodes.end()) {
1390     MD = NodeInfo->second.get();
1391     return false;
1392   }
1393   // Check machine metadata.
1394   NodeInfo = PFS.MachineMetadataNodes.find(ID);
1395   if (NodeInfo != PFS.MachineMetadataNodes.end()) {
1396     MD = NodeInfo->second.get();
1397     return false;
1398   }
1399   // Forward reference.
1400   auto &FwdRef = PFS.MachineForwardRefMDNodes[ID];
1401   FwdRef = std::make_pair(
1402       MDTuple::getTemporary(MF.getFunction().getContext(), std::nullopt), Loc);
1403   PFS.MachineMetadataNodes[ID].reset(FwdRef.first.get());
1404   MD = FwdRef.first.get();
1405 
1406   return false;
1407 }
1408 
1409 static const char *printImplicitRegisterFlag(const MachineOperand &MO) {
1410   assert(MO.isImplicit());
1411   return MO.isDef() ? "implicit-def" : "implicit";
1412 }
1413 
1414 static std::string getRegisterName(const TargetRegisterInfo *TRI,
1415                                    Register Reg) {
1416   assert(Reg.isPhysical() && "expected phys reg");
1417   return StringRef(TRI->getName(Reg)).lower();
1418 }
1419 
1420 /// Return true if the parsed machine operands contain a given machine operand.
1421 static bool isImplicitOperandIn(const MachineOperand &ImplicitOperand,
1422                                 ArrayRef<ParsedMachineOperand> Operands) {
1423   for (const auto &I : Operands) {
1424     if (ImplicitOperand.isIdenticalTo(I.Operand))
1425       return true;
1426   }
1427   return false;
1428 }
1429 
1430 bool MIParser::verifyImplicitOperands(ArrayRef<ParsedMachineOperand> Operands,
1431                                       const MCInstrDesc &MCID) {
1432   if (MCID.isCall())
1433     // We can't verify call instructions as they can contain arbitrary implicit
1434     // register and register mask operands.
1435     return false;
1436 
1437   // Gather all the expected implicit operands.
1438   SmallVector<MachineOperand, 4> ImplicitOperands;
1439   for (MCPhysReg ImpDef : MCID.implicit_defs())
1440     ImplicitOperands.push_back(MachineOperand::CreateReg(ImpDef, true, true));
1441   for (MCPhysReg ImpUse : MCID.implicit_uses())
1442     ImplicitOperands.push_back(MachineOperand::CreateReg(ImpUse, false, true));
1443 
1444   const auto *TRI = MF.getSubtarget().getRegisterInfo();
1445   assert(TRI && "Expected target register info");
1446   for (const auto &I : ImplicitOperands) {
1447     if (isImplicitOperandIn(I, Operands))
1448       continue;
1449     return error(Operands.empty() ? Token.location() : Operands.back().End,
1450                  Twine("missing implicit register operand '") +
1451                      printImplicitRegisterFlag(I) + " $" +
1452                      getRegisterName(TRI, I.getReg()) + "'");
1453   }
1454   return false;
1455 }
1456 
1457 bool MIParser::parseInstruction(unsigned &OpCode, unsigned &Flags) {
1458   // Allow frame and fast math flags for OPCODE
1459   // clang-format off
1460   while (Token.is(MIToken::kw_frame_setup) ||
1461          Token.is(MIToken::kw_frame_destroy) ||
1462          Token.is(MIToken::kw_nnan) ||
1463          Token.is(MIToken::kw_ninf) ||
1464          Token.is(MIToken::kw_nsz) ||
1465          Token.is(MIToken::kw_arcp) ||
1466          Token.is(MIToken::kw_contract) ||
1467          Token.is(MIToken::kw_afn) ||
1468          Token.is(MIToken::kw_reassoc) ||
1469          Token.is(MIToken::kw_nuw) ||
1470          Token.is(MIToken::kw_nsw) ||
1471          Token.is(MIToken::kw_exact) ||
1472          Token.is(MIToken::kw_nofpexcept) ||
1473          Token.is(MIToken::kw_noconvergent) ||
1474          Token.is(MIToken::kw_unpredictable)) {
1475     // clang-format on
1476     // Mine frame and fast math flags
1477     if (Token.is(MIToken::kw_frame_setup))
1478       Flags |= MachineInstr::FrameSetup;
1479     if (Token.is(MIToken::kw_frame_destroy))
1480       Flags |= MachineInstr::FrameDestroy;
1481     if (Token.is(MIToken::kw_nnan))
1482       Flags |= MachineInstr::FmNoNans;
1483     if (Token.is(MIToken::kw_ninf))
1484       Flags |= MachineInstr::FmNoInfs;
1485     if (Token.is(MIToken::kw_nsz))
1486       Flags |= MachineInstr::FmNsz;
1487     if (Token.is(MIToken::kw_arcp))
1488       Flags |= MachineInstr::FmArcp;
1489     if (Token.is(MIToken::kw_contract))
1490       Flags |= MachineInstr::FmContract;
1491     if (Token.is(MIToken::kw_afn))
1492       Flags |= MachineInstr::FmAfn;
1493     if (Token.is(MIToken::kw_reassoc))
1494       Flags |= MachineInstr::FmReassoc;
1495     if (Token.is(MIToken::kw_nuw))
1496       Flags |= MachineInstr::NoUWrap;
1497     if (Token.is(MIToken::kw_nsw))
1498       Flags |= MachineInstr::NoSWrap;
1499     if (Token.is(MIToken::kw_exact))
1500       Flags |= MachineInstr::IsExact;
1501     if (Token.is(MIToken::kw_nofpexcept))
1502       Flags |= MachineInstr::NoFPExcept;
1503     if (Token.is(MIToken::kw_unpredictable))
1504       Flags |= MachineInstr::Unpredictable;
1505     if (Token.is(MIToken::kw_noconvergent))
1506       Flags |= MachineInstr::NoConvergent;
1507 
1508     lex();
1509   }
1510   if (Token.isNot(MIToken::Identifier))
1511     return error("expected a machine instruction");
1512   StringRef InstrName = Token.stringValue();
1513   if (PFS.Target.parseInstrName(InstrName, OpCode))
1514     return error(Twine("unknown machine instruction name '") + InstrName + "'");
1515   lex();
1516   return false;
1517 }
1518 
1519 bool MIParser::parseNamedRegister(Register &Reg) {
1520   assert(Token.is(MIToken::NamedRegister) && "Needs NamedRegister token");
1521   StringRef Name = Token.stringValue();
1522   if (PFS.Target.getRegisterByName(Name, Reg))
1523     return error(Twine("unknown register name '") + Name + "'");
1524   return false;
1525 }
1526 
1527 bool MIParser::parseNamedVirtualRegister(VRegInfo *&Info) {
1528   assert(Token.is(MIToken::NamedVirtualRegister) && "Expected NamedVReg token");
1529   StringRef Name = Token.stringValue();
1530   // TODO: Check that the VReg name is not the same as a physical register name.
1531   //       If it is, then print a warning (when warnings are implemented).
1532   Info = &PFS.getVRegInfoNamed(Name);
1533   return false;
1534 }
1535 
1536 bool MIParser::parseVirtualRegister(VRegInfo *&Info) {
1537   if (Token.is(MIToken::NamedVirtualRegister))
1538     return parseNamedVirtualRegister(Info);
1539   assert(Token.is(MIToken::VirtualRegister) && "Needs VirtualRegister token");
1540   unsigned ID;
1541   if (getUnsigned(ID))
1542     return true;
1543   Info = &PFS.getVRegInfo(ID);
1544   return false;
1545 }
1546 
1547 bool MIParser::parseRegister(Register &Reg, VRegInfo *&Info) {
1548   switch (Token.kind()) {
1549   case MIToken::underscore:
1550     Reg = 0;
1551     return false;
1552   case MIToken::NamedRegister:
1553     return parseNamedRegister(Reg);
1554   case MIToken::NamedVirtualRegister:
1555   case MIToken::VirtualRegister:
1556     if (parseVirtualRegister(Info))
1557       return true;
1558     Reg = Info->VReg;
1559     return false;
1560   // TODO: Parse other register kinds.
1561   default:
1562     llvm_unreachable("The current token should be a register");
1563   }
1564 }
1565 
1566 bool MIParser::parseRegisterClassOrBank(VRegInfo &RegInfo) {
1567   if (Token.isNot(MIToken::Identifier) && Token.isNot(MIToken::underscore))
1568     return error("expected '_', register class, or register bank name");
1569   StringRef::iterator Loc = Token.location();
1570   StringRef Name = Token.stringValue();
1571 
1572   // Was it a register class?
1573   const TargetRegisterClass *RC = PFS.Target.getRegClass(Name);
1574   if (RC) {
1575     lex();
1576 
1577     switch (RegInfo.Kind) {
1578     case VRegInfo::UNKNOWN:
1579     case VRegInfo::NORMAL:
1580       RegInfo.Kind = VRegInfo::NORMAL;
1581       if (RegInfo.Explicit && RegInfo.D.RC != RC) {
1582         const TargetRegisterInfo &TRI = *MF.getSubtarget().getRegisterInfo();
1583         return error(Loc, Twine("conflicting register classes, previously: ") +
1584                      Twine(TRI.getRegClassName(RegInfo.D.RC)));
1585       }
1586       RegInfo.D.RC = RC;
1587       RegInfo.Explicit = true;
1588       return false;
1589 
1590     case VRegInfo::GENERIC:
1591     case VRegInfo::REGBANK:
1592       return error(Loc, "register class specification on generic register");
1593     }
1594     llvm_unreachable("Unexpected register kind");
1595   }
1596 
1597   // Should be a register bank or a generic register.
1598   const RegisterBank *RegBank = nullptr;
1599   if (Name != "_") {
1600     RegBank = PFS.Target.getRegBank(Name);
1601     if (!RegBank)
1602       return error(Loc, "expected '_', register class, or register bank name");
1603   }
1604 
1605   lex();
1606 
1607   switch (RegInfo.Kind) {
1608   case VRegInfo::UNKNOWN:
1609   case VRegInfo::GENERIC:
1610   case VRegInfo::REGBANK:
1611     RegInfo.Kind = RegBank ? VRegInfo::REGBANK : VRegInfo::GENERIC;
1612     if (RegInfo.Explicit && RegInfo.D.RegBank != RegBank)
1613       return error(Loc, "conflicting generic register banks");
1614     RegInfo.D.RegBank = RegBank;
1615     RegInfo.Explicit = true;
1616     return false;
1617 
1618   case VRegInfo::NORMAL:
1619     return error(Loc, "register bank specification on normal register");
1620   }
1621   llvm_unreachable("Unexpected register kind");
1622 }
1623 
1624 bool MIParser::parseRegisterFlag(unsigned &Flags) {
1625   const unsigned OldFlags = Flags;
1626   switch (Token.kind()) {
1627   case MIToken::kw_implicit:
1628     Flags |= RegState::Implicit;
1629     break;
1630   case MIToken::kw_implicit_define:
1631     Flags |= RegState::ImplicitDefine;
1632     break;
1633   case MIToken::kw_def:
1634     Flags |= RegState::Define;
1635     break;
1636   case MIToken::kw_dead:
1637     Flags |= RegState::Dead;
1638     break;
1639   case MIToken::kw_killed:
1640     Flags |= RegState::Kill;
1641     break;
1642   case MIToken::kw_undef:
1643     Flags |= RegState::Undef;
1644     break;
1645   case MIToken::kw_internal:
1646     Flags |= RegState::InternalRead;
1647     break;
1648   case MIToken::kw_early_clobber:
1649     Flags |= RegState::EarlyClobber;
1650     break;
1651   case MIToken::kw_debug_use:
1652     Flags |= RegState::Debug;
1653     break;
1654   case MIToken::kw_renamable:
1655     Flags |= RegState::Renamable;
1656     break;
1657   default:
1658     llvm_unreachable("The current token should be a register flag");
1659   }
1660   if (OldFlags == Flags)
1661     // We know that the same flag is specified more than once when the flags
1662     // weren't modified.
1663     return error("duplicate '" + Token.stringValue() + "' register flag");
1664   lex();
1665   return false;
1666 }
1667 
1668 bool MIParser::parseSubRegisterIndex(unsigned &SubReg) {
1669   assert(Token.is(MIToken::dot));
1670   lex();
1671   if (Token.isNot(MIToken::Identifier))
1672     return error("expected a subregister index after '.'");
1673   auto Name = Token.stringValue();
1674   SubReg = PFS.Target.getSubRegIndex(Name);
1675   if (!SubReg)
1676     return error(Twine("use of unknown subregister index '") + Name + "'");
1677   lex();
1678   return false;
1679 }
1680 
1681 bool MIParser::parseRegisterTiedDefIndex(unsigned &TiedDefIdx) {
1682   if (!consumeIfPresent(MIToken::kw_tied_def))
1683     return true;
1684   if (Token.isNot(MIToken::IntegerLiteral))
1685     return error("expected an integer literal after 'tied-def'");
1686   if (getUnsigned(TiedDefIdx))
1687     return true;
1688   lex();
1689   if (expectAndConsume(MIToken::rparen))
1690     return true;
1691   return false;
1692 }
1693 
1694 bool MIParser::assignRegisterTies(MachineInstr &MI,
1695                                   ArrayRef<ParsedMachineOperand> Operands) {
1696   SmallVector<std::pair<unsigned, unsigned>, 4> TiedRegisterPairs;
1697   for (unsigned I = 0, E = Operands.size(); I != E; ++I) {
1698     if (!Operands[I].TiedDefIdx)
1699       continue;
1700     // The parser ensures that this operand is a register use, so we just have
1701     // to check the tied-def operand.
1702     unsigned DefIdx = *Operands[I].TiedDefIdx;
1703     if (DefIdx >= E)
1704       return error(Operands[I].Begin,
1705                    Twine("use of invalid tied-def operand index '" +
1706                          Twine(DefIdx) + "'; instruction has only ") +
1707                        Twine(E) + " operands");
1708     const auto &DefOperand = Operands[DefIdx].Operand;
1709     if (!DefOperand.isReg() || !DefOperand.isDef())
1710       // FIXME: add note with the def operand.
1711       return error(Operands[I].Begin,
1712                    Twine("use of invalid tied-def operand index '") +
1713                        Twine(DefIdx) + "'; the operand #" + Twine(DefIdx) +
1714                        " isn't a defined register");
1715     // Check that the tied-def operand wasn't tied elsewhere.
1716     for (const auto &TiedPair : TiedRegisterPairs) {
1717       if (TiedPair.first == DefIdx)
1718         return error(Operands[I].Begin,
1719                      Twine("the tied-def operand #") + Twine(DefIdx) +
1720                          " is already tied with another register operand");
1721     }
1722     TiedRegisterPairs.push_back(std::make_pair(DefIdx, I));
1723   }
1724   // FIXME: Verify that for non INLINEASM instructions, the def and use tied
1725   // indices must be less than tied max.
1726   for (const auto &TiedPair : TiedRegisterPairs)
1727     MI.tieOperands(TiedPair.first, TiedPair.second);
1728   return false;
1729 }
1730 
1731 bool MIParser::parseRegisterOperand(MachineOperand &Dest,
1732                                     std::optional<unsigned> &TiedDefIdx,
1733                                     bool IsDef) {
1734   unsigned Flags = IsDef ? RegState::Define : 0;
1735   while (Token.isRegisterFlag()) {
1736     if (parseRegisterFlag(Flags))
1737       return true;
1738   }
1739   if (!Token.isRegister())
1740     return error("expected a register after register flags");
1741   Register Reg;
1742   VRegInfo *RegInfo;
1743   if (parseRegister(Reg, RegInfo))
1744     return true;
1745   lex();
1746   unsigned SubReg = 0;
1747   if (Token.is(MIToken::dot)) {
1748     if (parseSubRegisterIndex(SubReg))
1749       return true;
1750     if (!Reg.isVirtual())
1751       return error("subregister index expects a virtual register");
1752   }
1753   if (Token.is(MIToken::colon)) {
1754     if (!Reg.isVirtual())
1755       return error("register class specification expects a virtual register");
1756     lex();
1757     if (parseRegisterClassOrBank(*RegInfo))
1758         return true;
1759   }
1760   MachineRegisterInfo &MRI = MF.getRegInfo();
1761   if ((Flags & RegState::Define) == 0) {
1762     if (consumeIfPresent(MIToken::lparen)) {
1763       unsigned Idx;
1764       if (!parseRegisterTiedDefIndex(Idx))
1765         TiedDefIdx = Idx;
1766       else {
1767         // Try a redundant low-level type.
1768         LLT Ty;
1769         if (parseLowLevelType(Token.location(), Ty))
1770           return error("expected tied-def or low-level type after '('");
1771 
1772         if (expectAndConsume(MIToken::rparen))
1773           return true;
1774 
1775         if (MRI.getType(Reg).isValid() && MRI.getType(Reg) != Ty)
1776           return error("inconsistent type for generic virtual register");
1777 
1778         MRI.setRegClassOrRegBank(Reg, static_cast<RegisterBank *>(nullptr));
1779         MRI.setType(Reg, Ty);
1780       }
1781     }
1782   } else if (consumeIfPresent(MIToken::lparen)) {
1783     // Virtual registers may have a tpe with GlobalISel.
1784     if (!Reg.isVirtual())
1785       return error("unexpected type on physical register");
1786 
1787     LLT Ty;
1788     if (parseLowLevelType(Token.location(), Ty))
1789       return true;
1790 
1791     if (expectAndConsume(MIToken::rparen))
1792       return true;
1793 
1794     if (MRI.getType(Reg).isValid() && MRI.getType(Reg) != Ty)
1795       return error("inconsistent type for generic virtual register");
1796 
1797     MRI.setRegClassOrRegBank(Reg, static_cast<RegisterBank *>(nullptr));
1798     MRI.setType(Reg, Ty);
1799   } else if (Reg.isVirtual()) {
1800     // Generic virtual registers must have a type.
1801     // If we end up here this means the type hasn't been specified and
1802     // this is bad!
1803     if (RegInfo->Kind == VRegInfo::GENERIC ||
1804         RegInfo->Kind == VRegInfo::REGBANK)
1805       return error("generic virtual registers must have a type");
1806   }
1807 
1808   if (Flags & RegState::Define) {
1809     if (Flags & RegState::Kill)
1810       return error("cannot have a killed def operand");
1811   } else {
1812     if (Flags & RegState::Dead)
1813       return error("cannot have a dead use operand");
1814   }
1815 
1816   Dest = MachineOperand::CreateReg(
1817       Reg, Flags & RegState::Define, Flags & RegState::Implicit,
1818       Flags & RegState::Kill, Flags & RegState::Dead, Flags & RegState::Undef,
1819       Flags & RegState::EarlyClobber, SubReg, Flags & RegState::Debug,
1820       Flags & RegState::InternalRead, Flags & RegState::Renamable);
1821 
1822   return false;
1823 }
1824 
1825 bool MIParser::parseImmediateOperand(MachineOperand &Dest) {
1826   assert(Token.is(MIToken::IntegerLiteral));
1827   const APSInt &Int = Token.integerValue();
1828   if (auto SImm = Int.trySExtValue(); Int.isSigned() && SImm.has_value())
1829     Dest = MachineOperand::CreateImm(*SImm);
1830   else if (auto UImm = Int.tryZExtValue(); !Int.isSigned() && UImm.has_value())
1831     Dest = MachineOperand::CreateImm(*UImm);
1832   else
1833     return error("integer literal is too large to be an immediate operand");
1834   lex();
1835   return false;
1836 }
1837 
1838 bool MIParser::parseTargetImmMnemonic(const unsigned OpCode,
1839                                       const unsigned OpIdx,
1840                                       MachineOperand &Dest,
1841                                       const MIRFormatter &MF) {
1842   assert(Token.is(MIToken::dot));
1843   auto Loc = Token.location(); // record start position
1844   size_t Len = 1;              // for "."
1845   lex();
1846 
1847   // Handle the case that mnemonic starts with number.
1848   if (Token.is(MIToken::IntegerLiteral)) {
1849     Len += Token.range().size();
1850     lex();
1851   }
1852 
1853   StringRef Src;
1854   if (Token.is(MIToken::comma))
1855     Src = StringRef(Loc, Len);
1856   else {
1857     assert(Token.is(MIToken::Identifier));
1858     Src = StringRef(Loc, Len + Token.stringValue().size());
1859   }
1860   int64_t Val;
1861   if (MF.parseImmMnemonic(OpCode, OpIdx, Src, Val,
1862                           [this](StringRef::iterator Loc, const Twine &Msg)
1863                               -> bool { return error(Loc, Msg); }))
1864     return true;
1865 
1866   Dest = MachineOperand::CreateImm(Val);
1867   if (!Token.is(MIToken::comma))
1868     lex();
1869   return false;
1870 }
1871 
1872 static bool parseIRConstant(StringRef::iterator Loc, StringRef StringValue,
1873                             PerFunctionMIParsingState &PFS, const Constant *&C,
1874                             ErrorCallbackType ErrCB) {
1875   auto Source = StringValue.str(); // The source has to be null terminated.
1876   SMDiagnostic Err;
1877   C = parseConstantValue(Source, Err, *PFS.MF.getFunction().getParent(),
1878                          &PFS.IRSlots);
1879   if (!C)
1880     return ErrCB(Loc + Err.getColumnNo(), Err.getMessage());
1881   return false;
1882 }
1883 
1884 bool MIParser::parseIRConstant(StringRef::iterator Loc, StringRef StringValue,
1885                                const Constant *&C) {
1886   return ::parseIRConstant(
1887       Loc, StringValue, PFS, C,
1888       [this](StringRef::iterator Loc, const Twine &Msg) -> bool {
1889         return error(Loc, Msg);
1890       });
1891 }
1892 
1893 bool MIParser::parseIRConstant(StringRef::iterator Loc, const Constant *&C) {
1894   if (parseIRConstant(Loc, StringRef(Loc, Token.range().end() - Loc), C))
1895     return true;
1896   lex();
1897   return false;
1898 }
1899 
1900 // See LLT implementation for bit size limits.
1901 static bool verifyScalarSize(uint64_t Size) {
1902   return Size != 0 && isUInt<16>(Size);
1903 }
1904 
1905 static bool verifyVectorElementCount(uint64_t NumElts) {
1906   return NumElts != 0 && isUInt<16>(NumElts);
1907 }
1908 
1909 static bool verifyAddrSpace(uint64_t AddrSpace) {
1910   return isUInt<24>(AddrSpace);
1911 }
1912 
1913 bool MIParser::parseLowLevelType(StringRef::iterator Loc, LLT &Ty) {
1914   if (Token.range().front() == 's' || Token.range().front() == 'p') {
1915     StringRef SizeStr = Token.range().drop_front();
1916     if (SizeStr.size() == 0 || !llvm::all_of(SizeStr, isdigit))
1917       return error("expected integers after 's'/'p' type character");
1918   }
1919 
1920   if (Token.range().front() == 's') {
1921     auto ScalarSize = APSInt(Token.range().drop_front()).getZExtValue();
1922     if (!verifyScalarSize(ScalarSize))
1923       return error("invalid size for scalar type");
1924 
1925     Ty = LLT::scalar(ScalarSize);
1926     lex();
1927     return false;
1928   } else if (Token.range().front() == 'p') {
1929     const DataLayout &DL = MF.getDataLayout();
1930     uint64_t AS = APSInt(Token.range().drop_front()).getZExtValue();
1931     if (!verifyAddrSpace(AS))
1932       return error("invalid address space number");
1933 
1934     Ty = LLT::pointer(AS, DL.getPointerSizeInBits(AS));
1935     lex();
1936     return false;
1937   }
1938 
1939   // Now we're looking for a vector.
1940   if (Token.isNot(MIToken::less))
1941     return error(Loc, "expected sN, pA, <M x sN>, <M x pA>, <vscale x M x sN>, "
1942                       "or <vscale x M x pA> for GlobalISel type");
1943   lex();
1944 
1945   bool HasVScale =
1946       Token.is(MIToken::Identifier) && Token.stringValue() == "vscale";
1947   if (HasVScale) {
1948     lex();
1949     if (Token.isNot(MIToken::Identifier) || Token.stringValue() != "x")
1950       return error("expected <vscale x M x sN> or <vscale x M x pA>");
1951     lex();
1952   }
1953 
1954   auto GetError = [this, &HasVScale, Loc]() {
1955     if (HasVScale)
1956       return error(
1957           Loc, "expected <vscale x M x sN> or <vscale M x pA> for vector type");
1958     return error(Loc, "expected <M x sN> or <M x pA> for vector type");
1959   };
1960 
1961   if (Token.isNot(MIToken::IntegerLiteral))
1962     return GetError();
1963   uint64_t NumElements = Token.integerValue().getZExtValue();
1964   if (!verifyVectorElementCount(NumElements))
1965     return error("invalid number of vector elements");
1966 
1967   lex();
1968 
1969   if (Token.isNot(MIToken::Identifier) || Token.stringValue() != "x")
1970     return GetError();
1971   lex();
1972 
1973   if (Token.range().front() != 's' && Token.range().front() != 'p')
1974     return GetError();
1975 
1976   StringRef SizeStr = Token.range().drop_front();
1977   if (SizeStr.size() == 0 || !llvm::all_of(SizeStr, isdigit))
1978     return error("expected integers after 's'/'p' type character");
1979 
1980   if (Token.range().front() == 's') {
1981     auto ScalarSize = APSInt(Token.range().drop_front()).getZExtValue();
1982     if (!verifyScalarSize(ScalarSize))
1983       return error("invalid size for scalar type");
1984     Ty = LLT::scalar(ScalarSize);
1985   } else if (Token.range().front() == 'p') {
1986     const DataLayout &DL = MF.getDataLayout();
1987     uint64_t AS = APSInt(Token.range().drop_front()).getZExtValue();
1988     if (!verifyAddrSpace(AS))
1989       return error("invalid address space number");
1990 
1991     Ty = LLT::pointer(AS, DL.getPointerSizeInBits(AS));
1992   } else
1993     return GetError();
1994   lex();
1995 
1996   if (Token.isNot(MIToken::greater))
1997     return GetError();
1998 
1999   lex();
2000 
2001   Ty = LLT::vector(ElementCount::get(NumElements, HasVScale), Ty);
2002   return false;
2003 }
2004 
2005 bool MIParser::parseTypedImmediateOperand(MachineOperand &Dest) {
2006   assert(Token.is(MIToken::Identifier));
2007   StringRef TypeStr = Token.range();
2008   if (TypeStr.front() != 'i' && TypeStr.front() != 's' &&
2009       TypeStr.front() != 'p')
2010     return error(
2011         "a typed immediate operand should start with one of 'i', 's', or 'p'");
2012   StringRef SizeStr = Token.range().drop_front();
2013   if (SizeStr.size() == 0 || !llvm::all_of(SizeStr, isdigit))
2014     return error("expected integers after 'i'/'s'/'p' type character");
2015 
2016   auto Loc = Token.location();
2017   lex();
2018   if (Token.isNot(MIToken::IntegerLiteral)) {
2019     if (Token.isNot(MIToken::Identifier) ||
2020         !(Token.range() == "true" || Token.range() == "false"))
2021       return error("expected an integer literal");
2022   }
2023   const Constant *C = nullptr;
2024   if (parseIRConstant(Loc, C))
2025     return true;
2026   Dest = MachineOperand::CreateCImm(cast<ConstantInt>(C));
2027   return false;
2028 }
2029 
2030 bool MIParser::parseFPImmediateOperand(MachineOperand &Dest) {
2031   auto Loc = Token.location();
2032   lex();
2033   if (Token.isNot(MIToken::FloatingPointLiteral) &&
2034       Token.isNot(MIToken::HexLiteral))
2035     return error("expected a floating point literal");
2036   const Constant *C = nullptr;
2037   if (parseIRConstant(Loc, C))
2038     return true;
2039   Dest = MachineOperand::CreateFPImm(cast<ConstantFP>(C));
2040   return false;
2041 }
2042 
2043 static bool getHexUint(const MIToken &Token, APInt &Result) {
2044   assert(Token.is(MIToken::HexLiteral));
2045   StringRef S = Token.range();
2046   assert(S[0] == '0' && tolower(S[1]) == 'x');
2047   // This could be a floating point literal with a special prefix.
2048   if (!isxdigit(S[2]))
2049     return true;
2050   StringRef V = S.substr(2);
2051   APInt A(V.size()*4, V, 16);
2052 
2053   // If A is 0, then A.getActiveBits() is 0. This isn't a valid bitwidth. Make
2054   // sure it isn't the case before constructing result.
2055   unsigned NumBits = (A == 0) ? 32 : A.getActiveBits();
2056   Result = APInt(NumBits, ArrayRef<uint64_t>(A.getRawData(), A.getNumWords()));
2057   return false;
2058 }
2059 
2060 static bool getUnsigned(const MIToken &Token, unsigned &Result,
2061                         ErrorCallbackType ErrCB) {
2062   if (Token.hasIntegerValue()) {
2063     const uint64_t Limit = uint64_t(std::numeric_limits<unsigned>::max()) + 1;
2064     uint64_t Val64 = Token.integerValue().getLimitedValue(Limit);
2065     if (Val64 == Limit)
2066       return ErrCB(Token.location(), "expected 32-bit integer (too large)");
2067     Result = Val64;
2068     return false;
2069   }
2070   if (Token.is(MIToken::HexLiteral)) {
2071     APInt A;
2072     if (getHexUint(Token, A))
2073       return true;
2074     if (A.getBitWidth() > 32)
2075       return ErrCB(Token.location(), "expected 32-bit integer (too large)");
2076     Result = A.getZExtValue();
2077     return false;
2078   }
2079   return true;
2080 }
2081 
2082 bool MIParser::getUnsigned(unsigned &Result) {
2083   return ::getUnsigned(
2084       Token, Result, [this](StringRef::iterator Loc, const Twine &Msg) -> bool {
2085         return error(Loc, Msg);
2086       });
2087 }
2088 
2089 bool MIParser::parseMBBReference(MachineBasicBlock *&MBB) {
2090   assert(Token.is(MIToken::MachineBasicBlock) ||
2091          Token.is(MIToken::MachineBasicBlockLabel));
2092   unsigned Number;
2093   if (getUnsigned(Number))
2094     return true;
2095   auto MBBInfo = PFS.MBBSlots.find(Number);
2096   if (MBBInfo == PFS.MBBSlots.end())
2097     return error(Twine("use of undefined machine basic block #") +
2098                  Twine(Number));
2099   MBB = MBBInfo->second;
2100   // TODO: Only parse the name if it's a MachineBasicBlockLabel. Deprecate once
2101   // we drop the <irname> from the bb.<id>.<irname> format.
2102   if (!Token.stringValue().empty() && Token.stringValue() != MBB->getName())
2103     return error(Twine("the name of machine basic block #") + Twine(Number) +
2104                  " isn't '" + Token.stringValue() + "'");
2105   return false;
2106 }
2107 
2108 bool MIParser::parseMBBOperand(MachineOperand &Dest) {
2109   MachineBasicBlock *MBB;
2110   if (parseMBBReference(MBB))
2111     return true;
2112   Dest = MachineOperand::CreateMBB(MBB);
2113   lex();
2114   return false;
2115 }
2116 
2117 bool MIParser::parseStackFrameIndex(int &FI) {
2118   assert(Token.is(MIToken::StackObject));
2119   unsigned ID;
2120   if (getUnsigned(ID))
2121     return true;
2122   auto ObjectInfo = PFS.StackObjectSlots.find(ID);
2123   if (ObjectInfo == PFS.StackObjectSlots.end())
2124     return error(Twine("use of undefined stack object '%stack.") + Twine(ID) +
2125                  "'");
2126   StringRef Name;
2127   if (const auto *Alloca =
2128           MF.getFrameInfo().getObjectAllocation(ObjectInfo->second))
2129     Name = Alloca->getName();
2130   if (!Token.stringValue().empty() && Token.stringValue() != Name)
2131     return error(Twine("the name of the stack object '%stack.") + Twine(ID) +
2132                  "' isn't '" + Token.stringValue() + "'");
2133   lex();
2134   FI = ObjectInfo->second;
2135   return false;
2136 }
2137 
2138 bool MIParser::parseStackObjectOperand(MachineOperand &Dest) {
2139   int FI;
2140   if (parseStackFrameIndex(FI))
2141     return true;
2142   Dest = MachineOperand::CreateFI(FI);
2143   return false;
2144 }
2145 
2146 bool MIParser::parseFixedStackFrameIndex(int &FI) {
2147   assert(Token.is(MIToken::FixedStackObject));
2148   unsigned ID;
2149   if (getUnsigned(ID))
2150     return true;
2151   auto ObjectInfo = PFS.FixedStackObjectSlots.find(ID);
2152   if (ObjectInfo == PFS.FixedStackObjectSlots.end())
2153     return error(Twine("use of undefined fixed stack object '%fixed-stack.") +
2154                  Twine(ID) + "'");
2155   lex();
2156   FI = ObjectInfo->second;
2157   return false;
2158 }
2159 
2160 bool MIParser::parseFixedStackObjectOperand(MachineOperand &Dest) {
2161   int FI;
2162   if (parseFixedStackFrameIndex(FI))
2163     return true;
2164   Dest = MachineOperand::CreateFI(FI);
2165   return false;
2166 }
2167 
2168 static bool parseGlobalValue(const MIToken &Token,
2169                              PerFunctionMIParsingState &PFS, GlobalValue *&GV,
2170                              ErrorCallbackType ErrCB) {
2171   switch (Token.kind()) {
2172   case MIToken::NamedGlobalValue: {
2173     const Module *M = PFS.MF.getFunction().getParent();
2174     GV = M->getNamedValue(Token.stringValue());
2175     if (!GV)
2176       return ErrCB(Token.location(), Twine("use of undefined global value '") +
2177                                          Token.range() + "'");
2178     break;
2179   }
2180   case MIToken::GlobalValue: {
2181     unsigned GVIdx;
2182     if (getUnsigned(Token, GVIdx, ErrCB))
2183       return true;
2184     if (GVIdx >= PFS.IRSlots.GlobalValues.size())
2185       return ErrCB(Token.location(), Twine("use of undefined global value '@") +
2186                                          Twine(GVIdx) + "'");
2187     GV = PFS.IRSlots.GlobalValues[GVIdx];
2188     break;
2189   }
2190   default:
2191     llvm_unreachable("The current token should be a global value");
2192   }
2193   return false;
2194 }
2195 
2196 bool MIParser::parseGlobalValue(GlobalValue *&GV) {
2197   return ::parseGlobalValue(
2198       Token, PFS, GV,
2199       [this](StringRef::iterator Loc, const Twine &Msg) -> bool {
2200         return error(Loc, Msg);
2201       });
2202 }
2203 
2204 bool MIParser::parseGlobalAddressOperand(MachineOperand &Dest) {
2205   GlobalValue *GV = nullptr;
2206   if (parseGlobalValue(GV))
2207     return true;
2208   lex();
2209   Dest = MachineOperand::CreateGA(GV, /*Offset=*/0);
2210   if (parseOperandsOffset(Dest))
2211     return true;
2212   return false;
2213 }
2214 
2215 bool MIParser::parseConstantPoolIndexOperand(MachineOperand &Dest) {
2216   assert(Token.is(MIToken::ConstantPoolItem));
2217   unsigned ID;
2218   if (getUnsigned(ID))
2219     return true;
2220   auto ConstantInfo = PFS.ConstantPoolSlots.find(ID);
2221   if (ConstantInfo == PFS.ConstantPoolSlots.end())
2222     return error("use of undefined constant '%const." + Twine(ID) + "'");
2223   lex();
2224   Dest = MachineOperand::CreateCPI(ID, /*Offset=*/0);
2225   if (parseOperandsOffset(Dest))
2226     return true;
2227   return false;
2228 }
2229 
2230 bool MIParser::parseJumpTableIndexOperand(MachineOperand &Dest) {
2231   assert(Token.is(MIToken::JumpTableIndex));
2232   unsigned ID;
2233   if (getUnsigned(ID))
2234     return true;
2235   auto JumpTableEntryInfo = PFS.JumpTableSlots.find(ID);
2236   if (JumpTableEntryInfo == PFS.JumpTableSlots.end())
2237     return error("use of undefined jump table '%jump-table." + Twine(ID) + "'");
2238   lex();
2239   Dest = MachineOperand::CreateJTI(JumpTableEntryInfo->second);
2240   return false;
2241 }
2242 
2243 bool MIParser::parseExternalSymbolOperand(MachineOperand &Dest) {
2244   assert(Token.is(MIToken::ExternalSymbol));
2245   const char *Symbol = MF.createExternalSymbolName(Token.stringValue());
2246   lex();
2247   Dest = MachineOperand::CreateES(Symbol);
2248   if (parseOperandsOffset(Dest))
2249     return true;
2250   return false;
2251 }
2252 
2253 bool MIParser::parseMCSymbolOperand(MachineOperand &Dest) {
2254   assert(Token.is(MIToken::MCSymbol));
2255   MCSymbol *Symbol = getOrCreateMCSymbol(Token.stringValue());
2256   lex();
2257   Dest = MachineOperand::CreateMCSymbol(Symbol);
2258   if (parseOperandsOffset(Dest))
2259     return true;
2260   return false;
2261 }
2262 
2263 bool MIParser::parseSubRegisterIndexOperand(MachineOperand &Dest) {
2264   assert(Token.is(MIToken::SubRegisterIndex));
2265   StringRef Name = Token.stringValue();
2266   unsigned SubRegIndex = PFS.Target.getSubRegIndex(Token.stringValue());
2267   if (SubRegIndex == 0)
2268     return error(Twine("unknown subregister index '") + Name + "'");
2269   lex();
2270   Dest = MachineOperand::CreateImm(SubRegIndex);
2271   return false;
2272 }
2273 
2274 bool MIParser::parseMDNode(MDNode *&Node) {
2275   assert(Token.is(MIToken::exclaim));
2276 
2277   auto Loc = Token.location();
2278   lex();
2279   if (Token.isNot(MIToken::IntegerLiteral) || Token.integerValue().isSigned())
2280     return error("expected metadata id after '!'");
2281   unsigned ID;
2282   if (getUnsigned(ID))
2283     return true;
2284   auto NodeInfo = PFS.IRSlots.MetadataNodes.find(ID);
2285   if (NodeInfo == PFS.IRSlots.MetadataNodes.end()) {
2286     NodeInfo = PFS.MachineMetadataNodes.find(ID);
2287     if (NodeInfo == PFS.MachineMetadataNodes.end())
2288       return error(Loc, "use of undefined metadata '!" + Twine(ID) + "'");
2289   }
2290   lex();
2291   Node = NodeInfo->second.get();
2292   return false;
2293 }
2294 
2295 bool MIParser::parseDIExpression(MDNode *&Expr) {
2296   assert(Token.is(MIToken::md_diexpr));
2297   lex();
2298 
2299   // FIXME: Share this parsing with the IL parser.
2300   SmallVector<uint64_t, 8> Elements;
2301 
2302   if (expectAndConsume(MIToken::lparen))
2303     return true;
2304 
2305   if (Token.isNot(MIToken::rparen)) {
2306     do {
2307       if (Token.is(MIToken::Identifier)) {
2308         if (unsigned Op = dwarf::getOperationEncoding(Token.stringValue())) {
2309           lex();
2310           Elements.push_back(Op);
2311           continue;
2312         }
2313         if (unsigned Enc = dwarf::getAttributeEncoding(Token.stringValue())) {
2314           lex();
2315           Elements.push_back(Enc);
2316           continue;
2317         }
2318         return error(Twine("invalid DWARF op '") + Token.stringValue() + "'");
2319       }
2320 
2321       if (Token.isNot(MIToken::IntegerLiteral) ||
2322           Token.integerValue().isSigned())
2323         return error("expected unsigned integer");
2324 
2325       auto &U = Token.integerValue();
2326       if (U.ugt(UINT64_MAX))
2327         return error("element too large, limit is " + Twine(UINT64_MAX));
2328       Elements.push_back(U.getZExtValue());
2329       lex();
2330 
2331     } while (consumeIfPresent(MIToken::comma));
2332   }
2333 
2334   if (expectAndConsume(MIToken::rparen))
2335     return true;
2336 
2337   Expr = DIExpression::get(MF.getFunction().getContext(), Elements);
2338   return false;
2339 }
2340 
2341 bool MIParser::parseDILocation(MDNode *&Loc) {
2342   assert(Token.is(MIToken::md_dilocation));
2343   lex();
2344 
2345   bool HaveLine = false;
2346   unsigned Line = 0;
2347   unsigned Column = 0;
2348   MDNode *Scope = nullptr;
2349   MDNode *InlinedAt = nullptr;
2350   bool ImplicitCode = false;
2351 
2352   if (expectAndConsume(MIToken::lparen))
2353     return true;
2354 
2355   if (Token.isNot(MIToken::rparen)) {
2356     do {
2357       if (Token.is(MIToken::Identifier)) {
2358         if (Token.stringValue() == "line") {
2359           lex();
2360           if (expectAndConsume(MIToken::colon))
2361             return true;
2362           if (Token.isNot(MIToken::IntegerLiteral) ||
2363               Token.integerValue().isSigned())
2364             return error("expected unsigned integer");
2365           Line = Token.integerValue().getZExtValue();
2366           HaveLine = true;
2367           lex();
2368           continue;
2369         }
2370         if (Token.stringValue() == "column") {
2371           lex();
2372           if (expectAndConsume(MIToken::colon))
2373             return true;
2374           if (Token.isNot(MIToken::IntegerLiteral) ||
2375               Token.integerValue().isSigned())
2376             return error("expected unsigned integer");
2377           Column = Token.integerValue().getZExtValue();
2378           lex();
2379           continue;
2380         }
2381         if (Token.stringValue() == "scope") {
2382           lex();
2383           if (expectAndConsume(MIToken::colon))
2384             return true;
2385           if (parseMDNode(Scope))
2386             return error("expected metadata node");
2387           if (!isa<DIScope>(Scope))
2388             return error("expected DIScope node");
2389           continue;
2390         }
2391         if (Token.stringValue() == "inlinedAt") {
2392           lex();
2393           if (expectAndConsume(MIToken::colon))
2394             return true;
2395           if (Token.is(MIToken::exclaim)) {
2396             if (parseMDNode(InlinedAt))
2397               return true;
2398           } else if (Token.is(MIToken::md_dilocation)) {
2399             if (parseDILocation(InlinedAt))
2400               return true;
2401           } else
2402             return error("expected metadata node");
2403           if (!isa<DILocation>(InlinedAt))
2404             return error("expected DILocation node");
2405           continue;
2406         }
2407         if (Token.stringValue() == "isImplicitCode") {
2408           lex();
2409           if (expectAndConsume(MIToken::colon))
2410             return true;
2411           if (!Token.is(MIToken::Identifier))
2412             return error("expected true/false");
2413           // As far as I can see, we don't have any existing need for parsing
2414           // true/false in MIR yet. Do it ad-hoc until there's something else
2415           // that needs it.
2416           if (Token.stringValue() == "true")
2417             ImplicitCode = true;
2418           else if (Token.stringValue() == "false")
2419             ImplicitCode = false;
2420           else
2421             return error("expected true/false");
2422           lex();
2423           continue;
2424         }
2425       }
2426       return error(Twine("invalid DILocation argument '") +
2427                    Token.stringValue() + "'");
2428     } while (consumeIfPresent(MIToken::comma));
2429   }
2430 
2431   if (expectAndConsume(MIToken::rparen))
2432     return true;
2433 
2434   if (!HaveLine)
2435     return error("DILocation requires line number");
2436   if (!Scope)
2437     return error("DILocation requires a scope");
2438 
2439   Loc = DILocation::get(MF.getFunction().getContext(), Line, Column, Scope,
2440                         InlinedAt, ImplicitCode);
2441   return false;
2442 }
2443 
2444 bool MIParser::parseMetadataOperand(MachineOperand &Dest) {
2445   MDNode *Node = nullptr;
2446   if (Token.is(MIToken::exclaim)) {
2447     if (parseMDNode(Node))
2448       return true;
2449   } else if (Token.is(MIToken::md_diexpr)) {
2450     if (parseDIExpression(Node))
2451       return true;
2452   }
2453   Dest = MachineOperand::CreateMetadata(Node);
2454   return false;
2455 }
2456 
2457 bool MIParser::parseCFIOffset(int &Offset) {
2458   if (Token.isNot(MIToken::IntegerLiteral))
2459     return error("expected a cfi offset");
2460   if (Token.integerValue().getSignificantBits() > 32)
2461     return error("expected a 32 bit integer (the cfi offset is too large)");
2462   Offset = (int)Token.integerValue().getExtValue();
2463   lex();
2464   return false;
2465 }
2466 
2467 bool MIParser::parseCFIRegister(Register &Reg) {
2468   if (Token.isNot(MIToken::NamedRegister))
2469     return error("expected a cfi register");
2470   Register LLVMReg;
2471   if (parseNamedRegister(LLVMReg))
2472     return true;
2473   const auto *TRI = MF.getSubtarget().getRegisterInfo();
2474   assert(TRI && "Expected target register info");
2475   int DwarfReg = TRI->getDwarfRegNum(LLVMReg, true);
2476   if (DwarfReg < 0)
2477     return error("invalid DWARF register");
2478   Reg = (unsigned)DwarfReg;
2479   lex();
2480   return false;
2481 }
2482 
2483 bool MIParser::parseCFIAddressSpace(unsigned &AddressSpace) {
2484   if (Token.isNot(MIToken::IntegerLiteral))
2485     return error("expected a cfi address space literal");
2486   if (Token.integerValue().isSigned())
2487     return error("expected an unsigned integer (cfi address space)");
2488   AddressSpace = Token.integerValue().getZExtValue();
2489   lex();
2490   return false;
2491 }
2492 
2493 bool MIParser::parseCFIEscapeValues(std::string &Values) {
2494   do {
2495     if (Token.isNot(MIToken::HexLiteral))
2496       return error("expected a hexadecimal literal");
2497     unsigned Value;
2498     if (getUnsigned(Value))
2499       return true;
2500     if (Value > UINT8_MAX)
2501       return error("expected a 8-bit integer (too large)");
2502     Values.push_back(static_cast<uint8_t>(Value));
2503     lex();
2504   } while (consumeIfPresent(MIToken::comma));
2505   return false;
2506 }
2507 
2508 bool MIParser::parseCFIOperand(MachineOperand &Dest) {
2509   auto Kind = Token.kind();
2510   lex();
2511   int Offset;
2512   Register Reg;
2513   unsigned AddressSpace;
2514   unsigned CFIIndex;
2515   switch (Kind) {
2516   case MIToken::kw_cfi_same_value:
2517     if (parseCFIRegister(Reg))
2518       return true;
2519     CFIIndex = MF.addFrameInst(MCCFIInstruction::createSameValue(nullptr, Reg));
2520     break;
2521   case MIToken::kw_cfi_offset:
2522     if (parseCFIRegister(Reg) || expectAndConsume(MIToken::comma) ||
2523         parseCFIOffset(Offset))
2524       return true;
2525     CFIIndex =
2526         MF.addFrameInst(MCCFIInstruction::createOffset(nullptr, Reg, Offset));
2527     break;
2528   case MIToken::kw_cfi_rel_offset:
2529     if (parseCFIRegister(Reg) || expectAndConsume(MIToken::comma) ||
2530         parseCFIOffset(Offset))
2531       return true;
2532     CFIIndex = MF.addFrameInst(
2533         MCCFIInstruction::createRelOffset(nullptr, Reg, Offset));
2534     break;
2535   case MIToken::kw_cfi_def_cfa_register:
2536     if (parseCFIRegister(Reg))
2537       return true;
2538     CFIIndex =
2539         MF.addFrameInst(MCCFIInstruction::createDefCfaRegister(nullptr, Reg));
2540     break;
2541   case MIToken::kw_cfi_def_cfa_offset:
2542     if (parseCFIOffset(Offset))
2543       return true;
2544     CFIIndex =
2545         MF.addFrameInst(MCCFIInstruction::cfiDefCfaOffset(nullptr, Offset));
2546     break;
2547   case MIToken::kw_cfi_adjust_cfa_offset:
2548     if (parseCFIOffset(Offset))
2549       return true;
2550     CFIIndex = MF.addFrameInst(
2551         MCCFIInstruction::createAdjustCfaOffset(nullptr, Offset));
2552     break;
2553   case MIToken::kw_cfi_def_cfa:
2554     if (parseCFIRegister(Reg) || expectAndConsume(MIToken::comma) ||
2555         parseCFIOffset(Offset))
2556       return true;
2557     CFIIndex =
2558         MF.addFrameInst(MCCFIInstruction::cfiDefCfa(nullptr, Reg, Offset));
2559     break;
2560   case MIToken::kw_cfi_llvm_def_aspace_cfa:
2561     if (parseCFIRegister(Reg) || expectAndConsume(MIToken::comma) ||
2562         parseCFIOffset(Offset) || expectAndConsume(MIToken::comma) ||
2563         parseCFIAddressSpace(AddressSpace))
2564       return true;
2565     CFIIndex = MF.addFrameInst(MCCFIInstruction::createLLVMDefAspaceCfa(
2566         nullptr, Reg, Offset, AddressSpace, SMLoc()));
2567     break;
2568   case MIToken::kw_cfi_remember_state:
2569     CFIIndex = MF.addFrameInst(MCCFIInstruction::createRememberState(nullptr));
2570     break;
2571   case MIToken::kw_cfi_restore:
2572     if (parseCFIRegister(Reg))
2573       return true;
2574     CFIIndex = MF.addFrameInst(MCCFIInstruction::createRestore(nullptr, Reg));
2575     break;
2576   case MIToken::kw_cfi_restore_state:
2577     CFIIndex = MF.addFrameInst(MCCFIInstruction::createRestoreState(nullptr));
2578     break;
2579   case MIToken::kw_cfi_undefined:
2580     if (parseCFIRegister(Reg))
2581       return true;
2582     CFIIndex = MF.addFrameInst(MCCFIInstruction::createUndefined(nullptr, Reg));
2583     break;
2584   case MIToken::kw_cfi_register: {
2585     Register Reg2;
2586     if (parseCFIRegister(Reg) || expectAndConsume(MIToken::comma) ||
2587         parseCFIRegister(Reg2))
2588       return true;
2589 
2590     CFIIndex =
2591         MF.addFrameInst(MCCFIInstruction::createRegister(nullptr, Reg, Reg2));
2592     break;
2593   }
2594   case MIToken::kw_cfi_window_save:
2595     CFIIndex = MF.addFrameInst(MCCFIInstruction::createWindowSave(nullptr));
2596     break;
2597   case MIToken::kw_cfi_aarch64_negate_ra_sign_state:
2598     CFIIndex = MF.addFrameInst(MCCFIInstruction::createNegateRAState(nullptr));
2599     break;
2600   case MIToken::kw_cfi_escape: {
2601     std::string Values;
2602     if (parseCFIEscapeValues(Values))
2603       return true;
2604     CFIIndex = MF.addFrameInst(MCCFIInstruction::createEscape(nullptr, Values));
2605     break;
2606   }
2607   default:
2608     // TODO: Parse the other CFI operands.
2609     llvm_unreachable("The current token should be a cfi operand");
2610   }
2611   Dest = MachineOperand::CreateCFIIndex(CFIIndex);
2612   return false;
2613 }
2614 
2615 bool MIParser::parseIRBlock(BasicBlock *&BB, const Function &F) {
2616   switch (Token.kind()) {
2617   case MIToken::NamedIRBlock: {
2618     BB = dyn_cast_or_null<BasicBlock>(
2619         F.getValueSymbolTable()->lookup(Token.stringValue()));
2620     if (!BB)
2621       return error(Twine("use of undefined IR block '") + Token.range() + "'");
2622     break;
2623   }
2624   case MIToken::IRBlock: {
2625     unsigned SlotNumber = 0;
2626     if (getUnsigned(SlotNumber))
2627       return true;
2628     BB = const_cast<BasicBlock *>(getIRBlock(SlotNumber, F));
2629     if (!BB)
2630       return error(Twine("use of undefined IR block '%ir-block.") +
2631                    Twine(SlotNumber) + "'");
2632     break;
2633   }
2634   default:
2635     llvm_unreachable("The current token should be an IR block reference");
2636   }
2637   return false;
2638 }
2639 
2640 bool MIParser::parseBlockAddressOperand(MachineOperand &Dest) {
2641   assert(Token.is(MIToken::kw_blockaddress));
2642   lex();
2643   if (expectAndConsume(MIToken::lparen))
2644     return true;
2645   if (Token.isNot(MIToken::GlobalValue) &&
2646       Token.isNot(MIToken::NamedGlobalValue))
2647     return error("expected a global value");
2648   GlobalValue *GV = nullptr;
2649   if (parseGlobalValue(GV))
2650     return true;
2651   auto *F = dyn_cast<Function>(GV);
2652   if (!F)
2653     return error("expected an IR function reference");
2654   lex();
2655   if (expectAndConsume(MIToken::comma))
2656     return true;
2657   BasicBlock *BB = nullptr;
2658   if (Token.isNot(MIToken::IRBlock) && Token.isNot(MIToken::NamedIRBlock))
2659     return error("expected an IR block reference");
2660   if (parseIRBlock(BB, *F))
2661     return true;
2662   lex();
2663   if (expectAndConsume(MIToken::rparen))
2664     return true;
2665   Dest = MachineOperand::CreateBA(BlockAddress::get(F, BB), /*Offset=*/0);
2666   if (parseOperandsOffset(Dest))
2667     return true;
2668   return false;
2669 }
2670 
2671 bool MIParser::parseIntrinsicOperand(MachineOperand &Dest) {
2672   assert(Token.is(MIToken::kw_intrinsic));
2673   lex();
2674   if (expectAndConsume(MIToken::lparen))
2675     return error("expected syntax intrinsic(@llvm.whatever)");
2676 
2677   if (Token.isNot(MIToken::NamedGlobalValue))
2678     return error("expected syntax intrinsic(@llvm.whatever)");
2679 
2680   std::string Name = std::string(Token.stringValue());
2681   lex();
2682 
2683   if (expectAndConsume(MIToken::rparen))
2684     return error("expected ')' to terminate intrinsic name");
2685 
2686   // Find out what intrinsic we're dealing with, first try the global namespace
2687   // and then the target's private intrinsics if that fails.
2688   const TargetIntrinsicInfo *TII = MF.getTarget().getIntrinsicInfo();
2689   Intrinsic::ID ID = Function::lookupIntrinsicID(Name);
2690   if (ID == Intrinsic::not_intrinsic && TII)
2691     ID = static_cast<Intrinsic::ID>(TII->lookupName(Name));
2692 
2693   if (ID == Intrinsic::not_intrinsic)
2694     return error("unknown intrinsic name");
2695   Dest = MachineOperand::CreateIntrinsicID(ID);
2696 
2697   return false;
2698 }
2699 
2700 bool MIParser::parsePredicateOperand(MachineOperand &Dest) {
2701   assert(Token.is(MIToken::kw_intpred) || Token.is(MIToken::kw_floatpred));
2702   bool IsFloat = Token.is(MIToken::kw_floatpred);
2703   lex();
2704 
2705   if (expectAndConsume(MIToken::lparen))
2706     return error("expected syntax intpred(whatever) or floatpred(whatever");
2707 
2708   if (Token.isNot(MIToken::Identifier))
2709     return error("whatever");
2710 
2711   CmpInst::Predicate Pred;
2712   if (IsFloat) {
2713     Pred = StringSwitch<CmpInst::Predicate>(Token.stringValue())
2714                .Case("false", CmpInst::FCMP_FALSE)
2715                .Case("oeq", CmpInst::FCMP_OEQ)
2716                .Case("ogt", CmpInst::FCMP_OGT)
2717                .Case("oge", CmpInst::FCMP_OGE)
2718                .Case("olt", CmpInst::FCMP_OLT)
2719                .Case("ole", CmpInst::FCMP_OLE)
2720                .Case("one", CmpInst::FCMP_ONE)
2721                .Case("ord", CmpInst::FCMP_ORD)
2722                .Case("uno", CmpInst::FCMP_UNO)
2723                .Case("ueq", CmpInst::FCMP_UEQ)
2724                .Case("ugt", CmpInst::FCMP_UGT)
2725                .Case("uge", CmpInst::FCMP_UGE)
2726                .Case("ult", CmpInst::FCMP_ULT)
2727                .Case("ule", CmpInst::FCMP_ULE)
2728                .Case("une", CmpInst::FCMP_UNE)
2729                .Case("true", CmpInst::FCMP_TRUE)
2730                .Default(CmpInst::BAD_FCMP_PREDICATE);
2731     if (!CmpInst::isFPPredicate(Pred))
2732       return error("invalid floating-point predicate");
2733   } else {
2734     Pred = StringSwitch<CmpInst::Predicate>(Token.stringValue())
2735                .Case("eq", CmpInst::ICMP_EQ)
2736                .Case("ne", CmpInst::ICMP_NE)
2737                .Case("sgt", CmpInst::ICMP_SGT)
2738                .Case("sge", CmpInst::ICMP_SGE)
2739                .Case("slt", CmpInst::ICMP_SLT)
2740                .Case("sle", CmpInst::ICMP_SLE)
2741                .Case("ugt", CmpInst::ICMP_UGT)
2742                .Case("uge", CmpInst::ICMP_UGE)
2743                .Case("ult", CmpInst::ICMP_ULT)
2744                .Case("ule", CmpInst::ICMP_ULE)
2745                .Default(CmpInst::BAD_ICMP_PREDICATE);
2746     if (!CmpInst::isIntPredicate(Pred))
2747       return error("invalid integer predicate");
2748   }
2749 
2750   lex();
2751   Dest = MachineOperand::CreatePredicate(Pred);
2752   if (expectAndConsume(MIToken::rparen))
2753     return error("predicate should be terminated by ')'.");
2754 
2755   return false;
2756 }
2757 
2758 bool MIParser::parseShuffleMaskOperand(MachineOperand &Dest) {
2759   assert(Token.is(MIToken::kw_shufflemask));
2760 
2761   lex();
2762   if (expectAndConsume(MIToken::lparen))
2763     return error("expected syntax shufflemask(<integer or undef>, ...)");
2764 
2765   SmallVector<int, 32> ShufMask;
2766   do {
2767     if (Token.is(MIToken::kw_undef)) {
2768       ShufMask.push_back(-1);
2769     } else if (Token.is(MIToken::IntegerLiteral)) {
2770       const APSInt &Int = Token.integerValue();
2771       ShufMask.push_back(Int.getExtValue());
2772     } else
2773       return error("expected integer constant");
2774 
2775     lex();
2776   } while (consumeIfPresent(MIToken::comma));
2777 
2778   if (expectAndConsume(MIToken::rparen))
2779     return error("shufflemask should be terminated by ')'.");
2780 
2781   ArrayRef<int> MaskAlloc = MF.allocateShuffleMask(ShufMask);
2782   Dest = MachineOperand::CreateShuffleMask(MaskAlloc);
2783   return false;
2784 }
2785 
2786 bool MIParser::parseDbgInstrRefOperand(MachineOperand &Dest) {
2787   assert(Token.is(MIToken::kw_dbg_instr_ref));
2788 
2789   lex();
2790   if (expectAndConsume(MIToken::lparen))
2791     return error("expected syntax dbg-instr-ref(<unsigned>, <unsigned>)");
2792 
2793   if (Token.isNot(MIToken::IntegerLiteral) || Token.integerValue().isNegative())
2794     return error("expected unsigned integer for instruction index");
2795   uint64_t InstrIdx = Token.integerValue().getZExtValue();
2796   assert(InstrIdx <= std::numeric_limits<unsigned>::max() &&
2797          "Instruction reference's instruction index is too large");
2798   lex();
2799 
2800   if (expectAndConsume(MIToken::comma))
2801     return error("expected syntax dbg-instr-ref(<unsigned>, <unsigned>)");
2802 
2803   if (Token.isNot(MIToken::IntegerLiteral) || Token.integerValue().isNegative())
2804     return error("expected unsigned integer for operand index");
2805   uint64_t OpIdx = Token.integerValue().getZExtValue();
2806   assert(OpIdx <= std::numeric_limits<unsigned>::max() &&
2807          "Instruction reference's operand index is too large");
2808   lex();
2809 
2810   if (expectAndConsume(MIToken::rparen))
2811     return error("expected syntax dbg-instr-ref(<unsigned>, <unsigned>)");
2812 
2813   Dest = MachineOperand::CreateDbgInstrRef(InstrIdx, OpIdx);
2814   return false;
2815 }
2816 
2817 bool MIParser::parseTargetIndexOperand(MachineOperand &Dest) {
2818   assert(Token.is(MIToken::kw_target_index));
2819   lex();
2820   if (expectAndConsume(MIToken::lparen))
2821     return true;
2822   if (Token.isNot(MIToken::Identifier))
2823     return error("expected the name of the target index");
2824   int Index = 0;
2825   if (PFS.Target.getTargetIndex(Token.stringValue(), Index))
2826     return error("use of undefined target index '" + Token.stringValue() + "'");
2827   lex();
2828   if (expectAndConsume(MIToken::rparen))
2829     return true;
2830   Dest = MachineOperand::CreateTargetIndex(unsigned(Index), /*Offset=*/0);
2831   if (parseOperandsOffset(Dest))
2832     return true;
2833   return false;
2834 }
2835 
2836 bool MIParser::parseCustomRegisterMaskOperand(MachineOperand &Dest) {
2837   assert(Token.stringValue() == "CustomRegMask" && "Expected a custom RegMask");
2838   lex();
2839   if (expectAndConsume(MIToken::lparen))
2840     return true;
2841 
2842   uint32_t *Mask = MF.allocateRegMask();
2843   do {
2844     if (Token.isNot(MIToken::rparen)) {
2845       if (Token.isNot(MIToken::NamedRegister))
2846         return error("expected a named register");
2847       Register Reg;
2848       if (parseNamedRegister(Reg))
2849         return true;
2850       lex();
2851       Mask[Reg / 32] |= 1U << (Reg % 32);
2852     }
2853 
2854     // TODO: Report an error if the same register is used more than once.
2855   } while (consumeIfPresent(MIToken::comma));
2856 
2857   if (expectAndConsume(MIToken::rparen))
2858     return true;
2859   Dest = MachineOperand::CreateRegMask(Mask);
2860   return false;
2861 }
2862 
2863 bool MIParser::parseLiveoutRegisterMaskOperand(MachineOperand &Dest) {
2864   assert(Token.is(MIToken::kw_liveout));
2865   uint32_t *Mask = MF.allocateRegMask();
2866   lex();
2867   if (expectAndConsume(MIToken::lparen))
2868     return true;
2869   while (true) {
2870     if (Token.isNot(MIToken::NamedRegister))
2871       return error("expected a named register");
2872     Register Reg;
2873     if (parseNamedRegister(Reg))
2874       return true;
2875     lex();
2876     Mask[Reg / 32] |= 1U << (Reg % 32);
2877     // TODO: Report an error if the same register is used more than once.
2878     if (Token.isNot(MIToken::comma))
2879       break;
2880     lex();
2881   }
2882   if (expectAndConsume(MIToken::rparen))
2883     return true;
2884   Dest = MachineOperand::CreateRegLiveOut(Mask);
2885   return false;
2886 }
2887 
2888 bool MIParser::parseMachineOperand(const unsigned OpCode, const unsigned OpIdx,
2889                                    MachineOperand &Dest,
2890                                    std::optional<unsigned> &TiedDefIdx) {
2891   switch (Token.kind()) {
2892   case MIToken::kw_implicit:
2893   case MIToken::kw_implicit_define:
2894   case MIToken::kw_def:
2895   case MIToken::kw_dead:
2896   case MIToken::kw_killed:
2897   case MIToken::kw_undef:
2898   case MIToken::kw_internal:
2899   case MIToken::kw_early_clobber:
2900   case MIToken::kw_debug_use:
2901   case MIToken::kw_renamable:
2902   case MIToken::underscore:
2903   case MIToken::NamedRegister:
2904   case MIToken::VirtualRegister:
2905   case MIToken::NamedVirtualRegister:
2906     return parseRegisterOperand(Dest, TiedDefIdx);
2907   case MIToken::IntegerLiteral:
2908     return parseImmediateOperand(Dest);
2909   case MIToken::kw_half:
2910   case MIToken::kw_float:
2911   case MIToken::kw_double:
2912   case MIToken::kw_x86_fp80:
2913   case MIToken::kw_fp128:
2914   case MIToken::kw_ppc_fp128:
2915     return parseFPImmediateOperand(Dest);
2916   case MIToken::MachineBasicBlock:
2917     return parseMBBOperand(Dest);
2918   case MIToken::StackObject:
2919     return parseStackObjectOperand(Dest);
2920   case MIToken::FixedStackObject:
2921     return parseFixedStackObjectOperand(Dest);
2922   case MIToken::GlobalValue:
2923   case MIToken::NamedGlobalValue:
2924     return parseGlobalAddressOperand(Dest);
2925   case MIToken::ConstantPoolItem:
2926     return parseConstantPoolIndexOperand(Dest);
2927   case MIToken::JumpTableIndex:
2928     return parseJumpTableIndexOperand(Dest);
2929   case MIToken::ExternalSymbol:
2930     return parseExternalSymbolOperand(Dest);
2931   case MIToken::MCSymbol:
2932     return parseMCSymbolOperand(Dest);
2933   case MIToken::SubRegisterIndex:
2934     return parseSubRegisterIndexOperand(Dest);
2935   case MIToken::md_diexpr:
2936   case MIToken::exclaim:
2937     return parseMetadataOperand(Dest);
2938   case MIToken::kw_cfi_same_value:
2939   case MIToken::kw_cfi_offset:
2940   case MIToken::kw_cfi_rel_offset:
2941   case MIToken::kw_cfi_def_cfa_register:
2942   case MIToken::kw_cfi_def_cfa_offset:
2943   case MIToken::kw_cfi_adjust_cfa_offset:
2944   case MIToken::kw_cfi_escape:
2945   case MIToken::kw_cfi_def_cfa:
2946   case MIToken::kw_cfi_llvm_def_aspace_cfa:
2947   case MIToken::kw_cfi_register:
2948   case MIToken::kw_cfi_remember_state:
2949   case MIToken::kw_cfi_restore:
2950   case MIToken::kw_cfi_restore_state:
2951   case MIToken::kw_cfi_undefined:
2952   case MIToken::kw_cfi_window_save:
2953   case MIToken::kw_cfi_aarch64_negate_ra_sign_state:
2954     return parseCFIOperand(Dest);
2955   case MIToken::kw_blockaddress:
2956     return parseBlockAddressOperand(Dest);
2957   case MIToken::kw_intrinsic:
2958     return parseIntrinsicOperand(Dest);
2959   case MIToken::kw_target_index:
2960     return parseTargetIndexOperand(Dest);
2961   case MIToken::kw_liveout:
2962     return parseLiveoutRegisterMaskOperand(Dest);
2963   case MIToken::kw_floatpred:
2964   case MIToken::kw_intpred:
2965     return parsePredicateOperand(Dest);
2966   case MIToken::kw_shufflemask:
2967     return parseShuffleMaskOperand(Dest);
2968   case MIToken::kw_dbg_instr_ref:
2969     return parseDbgInstrRefOperand(Dest);
2970   case MIToken::Error:
2971     return true;
2972   case MIToken::Identifier:
2973     if (const auto *RegMask = PFS.Target.getRegMask(Token.stringValue())) {
2974       Dest = MachineOperand::CreateRegMask(RegMask);
2975       lex();
2976       break;
2977     } else if (Token.stringValue() == "CustomRegMask") {
2978       return parseCustomRegisterMaskOperand(Dest);
2979     } else
2980       return parseTypedImmediateOperand(Dest);
2981   case MIToken::dot: {
2982     const auto *TII = MF.getSubtarget().getInstrInfo();
2983     if (const auto *Formatter = TII->getMIRFormatter()) {
2984       return parseTargetImmMnemonic(OpCode, OpIdx, Dest, *Formatter);
2985     }
2986     [[fallthrough]];
2987   }
2988   default:
2989     // FIXME: Parse the MCSymbol machine operand.
2990     return error("expected a machine operand");
2991   }
2992   return false;
2993 }
2994 
2995 bool MIParser::parseMachineOperandAndTargetFlags(
2996     const unsigned OpCode, const unsigned OpIdx, MachineOperand &Dest,
2997     std::optional<unsigned> &TiedDefIdx) {
2998   unsigned TF = 0;
2999   bool HasTargetFlags = false;
3000   if (Token.is(MIToken::kw_target_flags)) {
3001     HasTargetFlags = true;
3002     lex();
3003     if (expectAndConsume(MIToken::lparen))
3004       return true;
3005     if (Token.isNot(MIToken::Identifier))
3006       return error("expected the name of the target flag");
3007     if (PFS.Target.getDirectTargetFlag(Token.stringValue(), TF)) {
3008       if (PFS.Target.getBitmaskTargetFlag(Token.stringValue(), TF))
3009         return error("use of undefined target flag '" + Token.stringValue() +
3010                      "'");
3011     }
3012     lex();
3013     while (Token.is(MIToken::comma)) {
3014       lex();
3015       if (Token.isNot(MIToken::Identifier))
3016         return error("expected the name of the target flag");
3017       unsigned BitFlag = 0;
3018       if (PFS.Target.getBitmaskTargetFlag(Token.stringValue(), BitFlag))
3019         return error("use of undefined target flag '" + Token.stringValue() +
3020                      "'");
3021       // TODO: Report an error when using a duplicate bit target flag.
3022       TF |= BitFlag;
3023       lex();
3024     }
3025     if (expectAndConsume(MIToken::rparen))
3026       return true;
3027   }
3028   auto Loc = Token.location();
3029   if (parseMachineOperand(OpCode, OpIdx, Dest, TiedDefIdx))
3030     return true;
3031   if (!HasTargetFlags)
3032     return false;
3033   if (Dest.isReg())
3034     return error(Loc, "register operands can't have target flags");
3035   Dest.setTargetFlags(TF);
3036   return false;
3037 }
3038 
3039 bool MIParser::parseOffset(int64_t &Offset) {
3040   if (Token.isNot(MIToken::plus) && Token.isNot(MIToken::minus))
3041     return false;
3042   StringRef Sign = Token.range();
3043   bool IsNegative = Token.is(MIToken::minus);
3044   lex();
3045   if (Token.isNot(MIToken::IntegerLiteral))
3046     return error("expected an integer literal after '" + Sign + "'");
3047   if (Token.integerValue().getSignificantBits() > 64)
3048     return error("expected 64-bit integer (too large)");
3049   Offset = Token.integerValue().getExtValue();
3050   if (IsNegative)
3051     Offset = -Offset;
3052   lex();
3053   return false;
3054 }
3055 
3056 bool MIParser::parseIRBlockAddressTaken(BasicBlock *&BB) {
3057   assert(Token.is(MIToken::kw_ir_block_address_taken));
3058   lex();
3059   if (Token.isNot(MIToken::IRBlock) && Token.isNot(MIToken::NamedIRBlock))
3060     return error("expected basic block after 'ir_block_address_taken'");
3061 
3062   if (parseIRBlock(BB, MF.getFunction()))
3063     return true;
3064 
3065   lex();
3066   return false;
3067 }
3068 
3069 bool MIParser::parseAlignment(uint64_t &Alignment) {
3070   assert(Token.is(MIToken::kw_align) || Token.is(MIToken::kw_basealign));
3071   lex();
3072   if (Token.isNot(MIToken::IntegerLiteral) || Token.integerValue().isSigned())
3073     return error("expected an integer literal after 'align'");
3074   if (getUint64(Alignment))
3075     return true;
3076   lex();
3077 
3078   if (!isPowerOf2_64(Alignment))
3079     return error("expected a power-of-2 literal after 'align'");
3080 
3081   return false;
3082 }
3083 
3084 bool MIParser::parseAddrspace(unsigned &Addrspace) {
3085   assert(Token.is(MIToken::kw_addrspace));
3086   lex();
3087   if (Token.isNot(MIToken::IntegerLiteral) || Token.integerValue().isSigned())
3088     return error("expected an integer literal after 'addrspace'");
3089   if (getUnsigned(Addrspace))
3090     return true;
3091   lex();
3092   return false;
3093 }
3094 
3095 bool MIParser::parseOperandsOffset(MachineOperand &Op) {
3096   int64_t Offset = 0;
3097   if (parseOffset(Offset))
3098     return true;
3099   Op.setOffset(Offset);
3100   return false;
3101 }
3102 
3103 static bool parseIRValue(const MIToken &Token, PerFunctionMIParsingState &PFS,
3104                          const Value *&V, ErrorCallbackType ErrCB) {
3105   switch (Token.kind()) {
3106   case MIToken::NamedIRValue: {
3107     V = PFS.MF.getFunction().getValueSymbolTable()->lookup(Token.stringValue());
3108     break;
3109   }
3110   case MIToken::IRValue: {
3111     unsigned SlotNumber = 0;
3112     if (getUnsigned(Token, SlotNumber, ErrCB))
3113       return true;
3114     V = PFS.getIRValue(SlotNumber);
3115     break;
3116   }
3117   case MIToken::NamedGlobalValue:
3118   case MIToken::GlobalValue: {
3119     GlobalValue *GV = nullptr;
3120     if (parseGlobalValue(Token, PFS, GV, ErrCB))
3121       return true;
3122     V = GV;
3123     break;
3124   }
3125   case MIToken::QuotedIRValue: {
3126     const Constant *C = nullptr;
3127     if (parseIRConstant(Token.location(), Token.stringValue(), PFS, C, ErrCB))
3128       return true;
3129     V = C;
3130     break;
3131   }
3132   case MIToken::kw_unknown_address:
3133     V = nullptr;
3134     return false;
3135   default:
3136     llvm_unreachable("The current token should be an IR block reference");
3137   }
3138   if (!V)
3139     return ErrCB(Token.location(), Twine("use of undefined IR value '") + Token.range() + "'");
3140   return false;
3141 }
3142 
3143 bool MIParser::parseIRValue(const Value *&V) {
3144   return ::parseIRValue(
3145       Token, PFS, V, [this](StringRef::iterator Loc, const Twine &Msg) -> bool {
3146         return error(Loc, Msg);
3147       });
3148 }
3149 
3150 bool MIParser::getUint64(uint64_t &Result) {
3151   if (Token.hasIntegerValue()) {
3152     if (Token.integerValue().getActiveBits() > 64)
3153       return error("expected 64-bit integer (too large)");
3154     Result = Token.integerValue().getZExtValue();
3155     return false;
3156   }
3157   if (Token.is(MIToken::HexLiteral)) {
3158     APInt A;
3159     if (getHexUint(A))
3160       return true;
3161     if (A.getBitWidth() > 64)
3162       return error("expected 64-bit integer (too large)");
3163     Result = A.getZExtValue();
3164     return false;
3165   }
3166   return true;
3167 }
3168 
3169 bool MIParser::getHexUint(APInt &Result) {
3170   return ::getHexUint(Token, Result);
3171 }
3172 
3173 bool MIParser::parseMemoryOperandFlag(MachineMemOperand::Flags &Flags) {
3174   const auto OldFlags = Flags;
3175   switch (Token.kind()) {
3176   case MIToken::kw_volatile:
3177     Flags |= MachineMemOperand::MOVolatile;
3178     break;
3179   case MIToken::kw_non_temporal:
3180     Flags |= MachineMemOperand::MONonTemporal;
3181     break;
3182   case MIToken::kw_dereferenceable:
3183     Flags |= MachineMemOperand::MODereferenceable;
3184     break;
3185   case MIToken::kw_invariant:
3186     Flags |= MachineMemOperand::MOInvariant;
3187     break;
3188   case MIToken::StringConstant: {
3189     MachineMemOperand::Flags TF;
3190     if (PFS.Target.getMMOTargetFlag(Token.stringValue(), TF))
3191       return error("use of undefined target MMO flag '" + Token.stringValue() +
3192                    "'");
3193     Flags |= TF;
3194     break;
3195   }
3196   default:
3197     llvm_unreachable("The current token should be a memory operand flag");
3198   }
3199   if (OldFlags == Flags)
3200     // We know that the same flag is specified more than once when the flags
3201     // weren't modified.
3202     return error("duplicate '" + Token.stringValue() + "' memory operand flag");
3203   lex();
3204   return false;
3205 }
3206 
3207 bool MIParser::parseMemoryPseudoSourceValue(const PseudoSourceValue *&PSV) {
3208   switch (Token.kind()) {
3209   case MIToken::kw_stack:
3210     PSV = MF.getPSVManager().getStack();
3211     break;
3212   case MIToken::kw_got:
3213     PSV = MF.getPSVManager().getGOT();
3214     break;
3215   case MIToken::kw_jump_table:
3216     PSV = MF.getPSVManager().getJumpTable();
3217     break;
3218   case MIToken::kw_constant_pool:
3219     PSV = MF.getPSVManager().getConstantPool();
3220     break;
3221   case MIToken::FixedStackObject: {
3222     int FI;
3223     if (parseFixedStackFrameIndex(FI))
3224       return true;
3225     PSV = MF.getPSVManager().getFixedStack(FI);
3226     // The token was already consumed, so use return here instead of break.
3227     return false;
3228   }
3229   case MIToken::StackObject: {
3230     int FI;
3231     if (parseStackFrameIndex(FI))
3232       return true;
3233     PSV = MF.getPSVManager().getFixedStack(FI);
3234     // The token was already consumed, so use return here instead of break.
3235     return false;
3236   }
3237   case MIToken::kw_call_entry:
3238     lex();
3239     switch (Token.kind()) {
3240     case MIToken::GlobalValue:
3241     case MIToken::NamedGlobalValue: {
3242       GlobalValue *GV = nullptr;
3243       if (parseGlobalValue(GV))
3244         return true;
3245       PSV = MF.getPSVManager().getGlobalValueCallEntry(GV);
3246       break;
3247     }
3248     case MIToken::ExternalSymbol:
3249       PSV = MF.getPSVManager().getExternalSymbolCallEntry(
3250           MF.createExternalSymbolName(Token.stringValue()));
3251       break;
3252     default:
3253       return error(
3254           "expected a global value or an external symbol after 'call-entry'");
3255     }
3256     break;
3257   case MIToken::kw_custom: {
3258     lex();
3259     const auto *TII = MF.getSubtarget().getInstrInfo();
3260     if (const auto *Formatter = TII->getMIRFormatter()) {
3261       if (Formatter->parseCustomPseudoSourceValue(
3262               Token.stringValue(), MF, PFS, PSV,
3263               [this](StringRef::iterator Loc, const Twine &Msg) -> bool {
3264                 return error(Loc, Msg);
3265               }))
3266         return true;
3267     } else
3268       return error("unable to parse target custom pseudo source value");
3269     break;
3270   }
3271   default:
3272     llvm_unreachable("The current token should be pseudo source value");
3273   }
3274   lex();
3275   return false;
3276 }
3277 
3278 bool MIParser::parseMachinePointerInfo(MachinePointerInfo &Dest) {
3279   if (Token.is(MIToken::kw_constant_pool) || Token.is(MIToken::kw_stack) ||
3280       Token.is(MIToken::kw_got) || Token.is(MIToken::kw_jump_table) ||
3281       Token.is(MIToken::FixedStackObject) || Token.is(MIToken::StackObject) ||
3282       Token.is(MIToken::kw_call_entry) || Token.is(MIToken::kw_custom)) {
3283     const PseudoSourceValue *PSV = nullptr;
3284     if (parseMemoryPseudoSourceValue(PSV))
3285       return true;
3286     int64_t Offset = 0;
3287     if (parseOffset(Offset))
3288       return true;
3289     Dest = MachinePointerInfo(PSV, Offset);
3290     return false;
3291   }
3292   if (Token.isNot(MIToken::NamedIRValue) && Token.isNot(MIToken::IRValue) &&
3293       Token.isNot(MIToken::GlobalValue) &&
3294       Token.isNot(MIToken::NamedGlobalValue) &&
3295       Token.isNot(MIToken::QuotedIRValue) &&
3296       Token.isNot(MIToken::kw_unknown_address))
3297     return error("expected an IR value reference");
3298   const Value *V = nullptr;
3299   if (parseIRValue(V))
3300     return true;
3301   if (V && !V->getType()->isPointerTy())
3302     return error("expected a pointer IR value");
3303   lex();
3304   int64_t Offset = 0;
3305   if (parseOffset(Offset))
3306     return true;
3307   Dest = MachinePointerInfo(V, Offset);
3308   return false;
3309 }
3310 
3311 bool MIParser::parseOptionalScope(LLVMContext &Context,
3312                                   SyncScope::ID &SSID) {
3313   SSID = SyncScope::System;
3314   if (Token.is(MIToken::Identifier) && Token.stringValue() == "syncscope") {
3315     lex();
3316     if (expectAndConsume(MIToken::lparen))
3317       return error("expected '(' in syncscope");
3318 
3319     std::string SSN;
3320     if (parseStringConstant(SSN))
3321       return true;
3322 
3323     SSID = Context.getOrInsertSyncScopeID(SSN);
3324     if (expectAndConsume(MIToken::rparen))
3325       return error("expected ')' in syncscope");
3326   }
3327 
3328   return false;
3329 }
3330 
3331 bool MIParser::parseOptionalAtomicOrdering(AtomicOrdering &Order) {
3332   Order = AtomicOrdering::NotAtomic;
3333   if (Token.isNot(MIToken::Identifier))
3334     return false;
3335 
3336   Order = StringSwitch<AtomicOrdering>(Token.stringValue())
3337               .Case("unordered", AtomicOrdering::Unordered)
3338               .Case("monotonic", AtomicOrdering::Monotonic)
3339               .Case("acquire", AtomicOrdering::Acquire)
3340               .Case("release", AtomicOrdering::Release)
3341               .Case("acq_rel", AtomicOrdering::AcquireRelease)
3342               .Case("seq_cst", AtomicOrdering::SequentiallyConsistent)
3343               .Default(AtomicOrdering::NotAtomic);
3344 
3345   if (Order != AtomicOrdering::NotAtomic) {
3346     lex();
3347     return false;
3348   }
3349 
3350   return error("expected an atomic scope, ordering or a size specification");
3351 }
3352 
3353 bool MIParser::parseMachineMemoryOperand(MachineMemOperand *&Dest) {
3354   if (expectAndConsume(MIToken::lparen))
3355     return true;
3356   MachineMemOperand::Flags Flags = MachineMemOperand::MONone;
3357   while (Token.isMemoryOperandFlag()) {
3358     if (parseMemoryOperandFlag(Flags))
3359       return true;
3360   }
3361   if (Token.isNot(MIToken::Identifier) ||
3362       (Token.stringValue() != "load" && Token.stringValue() != "store"))
3363     return error("expected 'load' or 'store' memory operation");
3364   if (Token.stringValue() == "load")
3365     Flags |= MachineMemOperand::MOLoad;
3366   else
3367     Flags |= MachineMemOperand::MOStore;
3368   lex();
3369 
3370   // Optional 'store' for operands that both load and store.
3371   if (Token.is(MIToken::Identifier) && Token.stringValue() == "store") {
3372     Flags |= MachineMemOperand::MOStore;
3373     lex();
3374   }
3375 
3376   // Optional synchronization scope.
3377   SyncScope::ID SSID;
3378   if (parseOptionalScope(MF.getFunction().getContext(), SSID))
3379     return true;
3380 
3381   // Up to two atomic orderings (cmpxchg provides guarantees on failure).
3382   AtomicOrdering Order, FailureOrder;
3383   if (parseOptionalAtomicOrdering(Order))
3384     return true;
3385 
3386   if (parseOptionalAtomicOrdering(FailureOrder))
3387     return true;
3388 
3389   LLT MemoryType;
3390   if (Token.isNot(MIToken::IntegerLiteral) &&
3391       Token.isNot(MIToken::kw_unknown_size) &&
3392       Token.isNot(MIToken::lparen))
3393     return error("expected memory LLT, the size integer literal or 'unknown-size' after "
3394                  "memory operation");
3395 
3396   uint64_t Size = MemoryLocation::UnknownSize;
3397   if (Token.is(MIToken::IntegerLiteral)) {
3398     if (getUint64(Size))
3399       return true;
3400 
3401     // Convert from bytes to bits for storage.
3402     MemoryType = LLT::scalar(8 * Size);
3403     lex();
3404   } else if (Token.is(MIToken::kw_unknown_size)) {
3405     Size = MemoryLocation::UnknownSize;
3406     lex();
3407   } else {
3408     if (expectAndConsume(MIToken::lparen))
3409       return true;
3410     if (parseLowLevelType(Token.location(), MemoryType))
3411       return true;
3412     if (expectAndConsume(MIToken::rparen))
3413       return true;
3414 
3415     Size = MemoryType.getSizeInBytes();
3416   }
3417 
3418   MachinePointerInfo Ptr = MachinePointerInfo();
3419   if (Token.is(MIToken::Identifier)) {
3420     const char *Word =
3421         ((Flags & MachineMemOperand::MOLoad) &&
3422          (Flags & MachineMemOperand::MOStore))
3423             ? "on"
3424             : Flags & MachineMemOperand::MOLoad ? "from" : "into";
3425     if (Token.stringValue() != Word)
3426       return error(Twine("expected '") + Word + "'");
3427     lex();
3428 
3429     if (parseMachinePointerInfo(Ptr))
3430       return true;
3431   }
3432   uint64_t BaseAlignment =
3433       (Size != MemoryLocation::UnknownSize ? PowerOf2Ceil(Size) : 1);
3434   AAMDNodes AAInfo;
3435   MDNode *Range = nullptr;
3436   while (consumeIfPresent(MIToken::comma)) {
3437     switch (Token.kind()) {
3438     case MIToken::kw_align: {
3439       // align is printed if it is different than size.
3440       uint64_t Alignment;
3441       if (parseAlignment(Alignment))
3442         return true;
3443       if (Ptr.Offset & (Alignment - 1)) {
3444         // MachineMemOperand::getAlign never returns a value greater than the
3445         // alignment of offset, so this just guards against hand-written MIR
3446         // that specifies a large "align" value when it should probably use
3447         // "basealign" instead.
3448         return error("specified alignment is more aligned than offset");
3449       }
3450       BaseAlignment = Alignment;
3451       break;
3452     }
3453     case MIToken::kw_basealign:
3454       // basealign is printed if it is different than align.
3455       if (parseAlignment(BaseAlignment))
3456         return true;
3457       break;
3458     case MIToken::kw_addrspace:
3459       if (parseAddrspace(Ptr.AddrSpace))
3460         return true;
3461       break;
3462     case MIToken::md_tbaa:
3463       lex();
3464       if (parseMDNode(AAInfo.TBAA))
3465         return true;
3466       break;
3467     case MIToken::md_alias_scope:
3468       lex();
3469       if (parseMDNode(AAInfo.Scope))
3470         return true;
3471       break;
3472     case MIToken::md_noalias:
3473       lex();
3474       if (parseMDNode(AAInfo.NoAlias))
3475         return true;
3476       break;
3477     case MIToken::md_range:
3478       lex();
3479       if (parseMDNode(Range))
3480         return true;
3481       break;
3482     // TODO: Report an error on duplicate metadata nodes.
3483     default:
3484       return error("expected 'align' or '!tbaa' or '!alias.scope' or "
3485                    "'!noalias' or '!range'");
3486     }
3487   }
3488   if (expectAndConsume(MIToken::rparen))
3489     return true;
3490   Dest = MF.getMachineMemOperand(Ptr, Flags, MemoryType, Align(BaseAlignment),
3491                                  AAInfo, Range, SSID, Order, FailureOrder);
3492   return false;
3493 }
3494 
3495 bool MIParser::parsePreOrPostInstrSymbol(MCSymbol *&Symbol) {
3496   assert((Token.is(MIToken::kw_pre_instr_symbol) ||
3497           Token.is(MIToken::kw_post_instr_symbol)) &&
3498          "Invalid token for a pre- post-instruction symbol!");
3499   lex();
3500   if (Token.isNot(MIToken::MCSymbol))
3501     return error("expected a symbol after 'pre-instr-symbol'");
3502   Symbol = getOrCreateMCSymbol(Token.stringValue());
3503   lex();
3504   if (Token.isNewlineOrEOF() || Token.is(MIToken::coloncolon) ||
3505       Token.is(MIToken::lbrace))
3506     return false;
3507   if (Token.isNot(MIToken::comma))
3508     return error("expected ',' before the next machine operand");
3509   lex();
3510   return false;
3511 }
3512 
3513 bool MIParser::parseHeapAllocMarker(MDNode *&Node) {
3514   assert(Token.is(MIToken::kw_heap_alloc_marker) &&
3515          "Invalid token for a heap alloc marker!");
3516   lex();
3517   if (parseMDNode(Node))
3518     return true;
3519   if (!Node)
3520     return error("expected a MDNode after 'heap-alloc-marker'");
3521   if (Token.isNewlineOrEOF() || Token.is(MIToken::coloncolon) ||
3522       Token.is(MIToken::lbrace))
3523     return false;
3524   if (Token.isNot(MIToken::comma))
3525     return error("expected ',' before the next machine operand");
3526   lex();
3527   return false;
3528 }
3529 
3530 bool MIParser::parsePCSections(MDNode *&Node) {
3531   assert(Token.is(MIToken::kw_pcsections) &&
3532          "Invalid token for a PC sections!");
3533   lex();
3534   if (parseMDNode(Node))
3535     return true;
3536   if (!Node)
3537     return error("expected a MDNode after 'pcsections'");
3538   if (Token.isNewlineOrEOF() || Token.is(MIToken::coloncolon) ||
3539       Token.is(MIToken::lbrace))
3540     return false;
3541   if (Token.isNot(MIToken::comma))
3542     return error("expected ',' before the next machine operand");
3543   lex();
3544   return false;
3545 }
3546 
3547 static void initSlots2BasicBlocks(
3548     const Function &F,
3549     DenseMap<unsigned, const BasicBlock *> &Slots2BasicBlocks) {
3550   ModuleSlotTracker MST(F.getParent(), /*ShouldInitializeAllMetadata=*/false);
3551   MST.incorporateFunction(F);
3552   for (const auto &BB : F) {
3553     if (BB.hasName())
3554       continue;
3555     int Slot = MST.getLocalSlot(&BB);
3556     if (Slot == -1)
3557       continue;
3558     Slots2BasicBlocks.insert(std::make_pair(unsigned(Slot), &BB));
3559   }
3560 }
3561 
3562 static const BasicBlock *getIRBlockFromSlot(
3563     unsigned Slot,
3564     const DenseMap<unsigned, const BasicBlock *> &Slots2BasicBlocks) {
3565   return Slots2BasicBlocks.lookup(Slot);
3566 }
3567 
3568 const BasicBlock *MIParser::getIRBlock(unsigned Slot) {
3569   if (Slots2BasicBlocks.empty())
3570     initSlots2BasicBlocks(MF.getFunction(), Slots2BasicBlocks);
3571   return getIRBlockFromSlot(Slot, Slots2BasicBlocks);
3572 }
3573 
3574 const BasicBlock *MIParser::getIRBlock(unsigned Slot, const Function &F) {
3575   if (&F == &MF.getFunction())
3576     return getIRBlock(Slot);
3577   DenseMap<unsigned, const BasicBlock *> CustomSlots2BasicBlocks;
3578   initSlots2BasicBlocks(F, CustomSlots2BasicBlocks);
3579   return getIRBlockFromSlot(Slot, CustomSlots2BasicBlocks);
3580 }
3581 
3582 MCSymbol *MIParser::getOrCreateMCSymbol(StringRef Name) {
3583   // FIXME: Currently we can't recognize temporary or local symbols and call all
3584   // of the appropriate forms to create them. However, this handles basic cases
3585   // well as most of the special aspects are recognized by a prefix on their
3586   // name, and the input names should already be unique. For test cases, keeping
3587   // the symbol name out of the symbol table isn't terribly important.
3588   return MF.getContext().getOrCreateSymbol(Name);
3589 }
3590 
3591 bool MIParser::parseStringConstant(std::string &Result) {
3592   if (Token.isNot(MIToken::StringConstant))
3593     return error("expected string constant");
3594   Result = std::string(Token.stringValue());
3595   lex();
3596   return false;
3597 }
3598 
3599 bool llvm::parseMachineBasicBlockDefinitions(PerFunctionMIParsingState &PFS,
3600                                              StringRef Src,
3601                                              SMDiagnostic &Error) {
3602   return MIParser(PFS, Error, Src).parseBasicBlockDefinitions(PFS.MBBSlots);
3603 }
3604 
3605 bool llvm::parseMachineInstructions(PerFunctionMIParsingState &PFS,
3606                                     StringRef Src, SMDiagnostic &Error) {
3607   return MIParser(PFS, Error, Src).parseBasicBlocks();
3608 }
3609 
3610 bool llvm::parseMBBReference(PerFunctionMIParsingState &PFS,
3611                              MachineBasicBlock *&MBB, StringRef Src,
3612                              SMDiagnostic &Error) {
3613   return MIParser(PFS, Error, Src).parseStandaloneMBB(MBB);
3614 }
3615 
3616 bool llvm::parseRegisterReference(PerFunctionMIParsingState &PFS,
3617                                   Register &Reg, StringRef Src,
3618                                   SMDiagnostic &Error) {
3619   return MIParser(PFS, Error, Src).parseStandaloneRegister(Reg);
3620 }
3621 
3622 bool llvm::parseNamedRegisterReference(PerFunctionMIParsingState &PFS,
3623                                        Register &Reg, StringRef Src,
3624                                        SMDiagnostic &Error) {
3625   return MIParser(PFS, Error, Src).parseStandaloneNamedRegister(Reg);
3626 }
3627 
3628 bool llvm::parseVirtualRegisterReference(PerFunctionMIParsingState &PFS,
3629                                          VRegInfo *&Info, StringRef Src,
3630                                          SMDiagnostic &Error) {
3631   return MIParser(PFS, Error, Src).parseStandaloneVirtualRegister(Info);
3632 }
3633 
3634 bool llvm::parseStackObjectReference(PerFunctionMIParsingState &PFS,
3635                                      int &FI, StringRef Src,
3636                                      SMDiagnostic &Error) {
3637   return MIParser(PFS, Error, Src).parseStandaloneStackObject(FI);
3638 }
3639 
3640 bool llvm::parseMDNode(PerFunctionMIParsingState &PFS,
3641                        MDNode *&Node, StringRef Src, SMDiagnostic &Error) {
3642   return MIParser(PFS, Error, Src).parseStandaloneMDNode(Node);
3643 }
3644 
3645 bool llvm::parseMachineMetadata(PerFunctionMIParsingState &PFS, StringRef Src,
3646                                 SMRange SrcRange, SMDiagnostic &Error) {
3647   return MIParser(PFS, Error, Src, SrcRange).parseMachineMetadata();
3648 }
3649 
3650 bool MIRFormatter::parseIRValue(StringRef Src, MachineFunction &MF,
3651                                 PerFunctionMIParsingState &PFS, const Value *&V,
3652                                 ErrorCallbackType ErrorCallback) {
3653   MIToken Token;
3654   Src = lexMIToken(Src, Token, [&](StringRef::iterator Loc, const Twine &Msg) {
3655     ErrorCallback(Loc, Msg);
3656   });
3657   V = nullptr;
3658 
3659   return ::parseIRValue(Token, PFS, V, ErrorCallback);
3660 }
3661