xref: /freebsd/contrib/llvm-project/llvm/lib/CodeGen/MIRParser/MIParser.cpp (revision d5b0e70f7e04d971691517ce1304d86a1e367e2e)
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/None.h"
20 #include "llvm/ADT/Optional.h"
21 #include "llvm/ADT/SmallVector.h"
22 #include "llvm/ADT/StringMap.h"
23 #include "llvm/ADT/StringRef.h"
24 #include "llvm/ADT/StringSwitch.h"
25 #include "llvm/ADT/Twine.h"
26 #include "llvm/Analysis/MemoryLocation.h"
27 #include "llvm/AsmParser/Parser.h"
28 #include "llvm/AsmParser/SlotMapping.h"
29 #include "llvm/CodeGen/GlobalISel/RegisterBank.h"
30 #include "llvm/CodeGen/GlobalISel/RegisterBankInfo.h"
31 #include "llvm/CodeGen/MIRFormatter.h"
32 #include "llvm/CodeGen/MIRPrinter.h"
33 #include "llvm/CodeGen/MachineBasicBlock.h"
34 #include "llvm/CodeGen/MachineFrameInfo.h"
35 #include "llvm/CodeGen/MachineFunction.h"
36 #include "llvm/CodeGen/MachineInstr.h"
37 #include "llvm/CodeGen/MachineInstrBuilder.h"
38 #include "llvm/CodeGen/MachineMemOperand.h"
39 #include "llvm/CodeGen/MachineOperand.h"
40 #include "llvm/CodeGen/MachineRegisterInfo.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/MC/MCRegisterInfo.h"
64 #include "llvm/Support/AtomicOrdering.h"
65 #include "llvm/Support/BranchProbability.h"
66 #include "llvm/Support/Casting.h"
67 #include "llvm/Support/ErrorHandling.h"
68 #include "llvm/Support/LowLevelTypeImpl.h"
69 #include "llvm/Support/MemoryBuffer.h"
70 #include "llvm/Support/SMLoc.h"
71 #include "llvm/Support/SourceMgr.h"
72 #include "llvm/Support/raw_ostream.h"
73 #include "llvm/Target/TargetIntrinsicInfo.h"
74 #include "llvm/Target/TargetMachine.h"
75 #include <algorithm>
76 #include <cassert>
77 #include <cctype>
78 #include <cstddef>
79 #include <cstdint>
80 #include <limits>
81 #include <string>
82 #include <utility>
83 
84 using namespace llvm;
85 
86 void PerTargetMIParsingState::setTarget(
87   const TargetSubtargetInfo &NewSubtarget) {
88 
89   // If the subtarget changed, over conservatively assume everything is invalid.
90   if (&Subtarget == &NewSubtarget)
91     return;
92 
93   Names2InstrOpCodes.clear();
94   Names2Regs.clear();
95   Names2RegMasks.clear();
96   Names2SubRegIndices.clear();
97   Names2TargetIndices.clear();
98   Names2DirectTargetFlags.clear();
99   Names2BitmaskTargetFlags.clear();
100   Names2MMOTargetFlags.clear();
101 
102   initNames2RegClasses();
103   initNames2RegBanks();
104 }
105 
106 void PerTargetMIParsingState::initNames2Regs() {
107   if (!Names2Regs.empty())
108     return;
109 
110   // The '%noreg' register is the register 0.
111   Names2Regs.insert(std::make_pair("noreg", 0));
112   const auto *TRI = Subtarget.getRegisterInfo();
113   assert(TRI && "Expected target register info");
114 
115   for (unsigned I = 0, E = TRI->getNumRegs(); I < E; ++I) {
116     bool WasInserted =
117         Names2Regs.insert(std::make_pair(StringRef(TRI->getName(I)).lower(), I))
118             .second;
119     (void)WasInserted;
120     assert(WasInserted && "Expected registers to be unique case-insensitively");
121   }
122 }
123 
124 bool PerTargetMIParsingState::getRegisterByName(StringRef RegName,
125                                                 Register &Reg) {
126   initNames2Regs();
127   auto RegInfo = Names2Regs.find(RegName);
128   if (RegInfo == Names2Regs.end())
129     return true;
130   Reg = RegInfo->getValue();
131   return false;
132 }
133 
134 void PerTargetMIParsingState::initNames2InstrOpCodes() {
135   if (!Names2InstrOpCodes.empty())
136     return;
137   const auto *TII = Subtarget.getInstrInfo();
138   assert(TII && "Expected target instruction info");
139   for (unsigned I = 0, E = TII->getNumOpcodes(); I < E; ++I)
140     Names2InstrOpCodes.insert(std::make_pair(StringRef(TII->getName(I)), I));
141 }
142 
143 bool PerTargetMIParsingState::parseInstrName(StringRef InstrName,
144                                              unsigned &OpCode) {
145   initNames2InstrOpCodes();
146   auto InstrInfo = Names2InstrOpCodes.find(InstrName);
147   if (InstrInfo == Names2InstrOpCodes.end())
148     return true;
149   OpCode = InstrInfo->getValue();
150   return false;
151 }
152 
153 void PerTargetMIParsingState::initNames2RegMasks() {
154   if (!Names2RegMasks.empty())
155     return;
156   const auto *TRI = Subtarget.getRegisterInfo();
157   assert(TRI && "Expected target register info");
158   ArrayRef<const uint32_t *> RegMasks = TRI->getRegMasks();
159   ArrayRef<const char *> RegMaskNames = TRI->getRegMaskNames();
160   assert(RegMasks.size() == RegMaskNames.size());
161   for (size_t I = 0, E = RegMasks.size(); I < E; ++I)
162     Names2RegMasks.insert(
163         std::make_pair(StringRef(RegMaskNames[I]).lower(), RegMasks[I]));
164 }
165 
166 const uint32_t *PerTargetMIParsingState::getRegMask(StringRef Identifier) {
167   initNames2RegMasks();
168   auto RegMaskInfo = Names2RegMasks.find(Identifier);
169   if (RegMaskInfo == Names2RegMasks.end())
170     return nullptr;
171   return RegMaskInfo->getValue();
172 }
173 
174 void PerTargetMIParsingState::initNames2SubRegIndices() {
175   if (!Names2SubRegIndices.empty())
176     return;
177   const TargetRegisterInfo *TRI = Subtarget.getRegisterInfo();
178   for (unsigned I = 1, E = TRI->getNumSubRegIndices(); I < E; ++I)
179     Names2SubRegIndices.insert(
180         std::make_pair(TRI->getSubRegIndexName(I), I));
181 }
182 
183 unsigned PerTargetMIParsingState::getSubRegIndex(StringRef Name) {
184   initNames2SubRegIndices();
185   auto SubRegInfo = Names2SubRegIndices.find(Name);
186   if (SubRegInfo == Names2SubRegIndices.end())
187     return 0;
188   return SubRegInfo->getValue();
189 }
190 
191 void PerTargetMIParsingState::initNames2TargetIndices() {
192   if (!Names2TargetIndices.empty())
193     return;
194   const auto *TII = Subtarget.getInstrInfo();
195   assert(TII && "Expected target instruction info");
196   auto Indices = TII->getSerializableTargetIndices();
197   for (const auto &I : Indices)
198     Names2TargetIndices.insert(std::make_pair(StringRef(I.second), I.first));
199 }
200 
201 bool PerTargetMIParsingState::getTargetIndex(StringRef Name, int &Index) {
202   initNames2TargetIndices();
203   auto IndexInfo = Names2TargetIndices.find(Name);
204   if (IndexInfo == Names2TargetIndices.end())
205     return true;
206   Index = IndexInfo->second;
207   return false;
208 }
209 
210 void PerTargetMIParsingState::initNames2DirectTargetFlags() {
211   if (!Names2DirectTargetFlags.empty())
212     return;
213 
214   const auto *TII = Subtarget.getInstrInfo();
215   assert(TII && "Expected target instruction info");
216   auto Flags = TII->getSerializableDirectMachineOperandTargetFlags();
217   for (const auto &I : Flags)
218     Names2DirectTargetFlags.insert(
219         std::make_pair(StringRef(I.second), I.first));
220 }
221 
222 bool PerTargetMIParsingState::getDirectTargetFlag(StringRef Name,
223                                                   unsigned &Flag) {
224   initNames2DirectTargetFlags();
225   auto FlagInfo = Names2DirectTargetFlags.find(Name);
226   if (FlagInfo == Names2DirectTargetFlags.end())
227     return true;
228   Flag = FlagInfo->second;
229   return false;
230 }
231 
232 void PerTargetMIParsingState::initNames2BitmaskTargetFlags() {
233   if (!Names2BitmaskTargetFlags.empty())
234     return;
235 
236   const auto *TII = Subtarget.getInstrInfo();
237   assert(TII && "Expected target instruction info");
238   auto Flags = TII->getSerializableBitmaskMachineOperandTargetFlags();
239   for (const auto &I : Flags)
240     Names2BitmaskTargetFlags.insert(
241         std::make_pair(StringRef(I.second), I.first));
242 }
243 
244 bool PerTargetMIParsingState::getBitmaskTargetFlag(StringRef Name,
245                                                    unsigned &Flag) {
246   initNames2BitmaskTargetFlags();
247   auto FlagInfo = Names2BitmaskTargetFlags.find(Name);
248   if (FlagInfo == Names2BitmaskTargetFlags.end())
249     return true;
250   Flag = FlagInfo->second;
251   return false;
252 }
253 
254 void PerTargetMIParsingState::initNames2MMOTargetFlags() {
255   if (!Names2MMOTargetFlags.empty())
256     return;
257 
258   const auto *TII = Subtarget.getInstrInfo();
259   assert(TII && "Expected target instruction info");
260   auto Flags = TII->getSerializableMachineMemOperandTargetFlags();
261   for (const auto &I : Flags)
262     Names2MMOTargetFlags.insert(std::make_pair(StringRef(I.second), I.first));
263 }
264 
265 bool PerTargetMIParsingState::getMMOTargetFlag(StringRef Name,
266                                                MachineMemOperand::Flags &Flag) {
267   initNames2MMOTargetFlags();
268   auto FlagInfo = Names2MMOTargetFlags.find(Name);
269   if (FlagInfo == Names2MMOTargetFlags.end())
270     return true;
271   Flag = FlagInfo->second;
272   return false;
273 }
274 
275 void PerTargetMIParsingState::initNames2RegClasses() {
276   if (!Names2RegClasses.empty())
277     return;
278 
279   const TargetRegisterInfo *TRI = Subtarget.getRegisterInfo();
280   for (unsigned I = 0, E = TRI->getNumRegClasses(); I < E; ++I) {
281     const auto *RC = TRI->getRegClass(I);
282     Names2RegClasses.insert(
283         std::make_pair(StringRef(TRI->getRegClassName(RC)).lower(), RC));
284   }
285 }
286 
287 void PerTargetMIParsingState::initNames2RegBanks() {
288   if (!Names2RegBanks.empty())
289     return;
290 
291   const RegisterBankInfo *RBI = Subtarget.getRegBankInfo();
292   // If the target does not support GlobalISel, we may not have a
293   // register bank info.
294   if (!RBI)
295     return;
296 
297   for (unsigned I = 0, E = RBI->getNumRegBanks(); I < E; ++I) {
298     const auto &RegBank = RBI->getRegBank(I);
299     Names2RegBanks.insert(
300         std::make_pair(StringRef(RegBank.getName()).lower(), &RegBank));
301   }
302 }
303 
304 const TargetRegisterClass *
305 PerTargetMIParsingState::getRegClass(StringRef Name) {
306   auto RegClassInfo = Names2RegClasses.find(Name);
307   if (RegClassInfo == Names2RegClasses.end())
308     return nullptr;
309   return RegClassInfo->getValue();
310 }
311 
312 const RegisterBank *PerTargetMIParsingState::getRegBank(StringRef Name) {
313   auto RegBankInfo = Names2RegBanks.find(Name);
314   if (RegBankInfo == Names2RegBanks.end())
315     return nullptr;
316   return RegBankInfo->getValue();
317 }
318 
319 PerFunctionMIParsingState::PerFunctionMIParsingState(MachineFunction &MF,
320     SourceMgr &SM, const SlotMapping &IRSlots, PerTargetMIParsingState &T)
321   : MF(MF), SM(&SM), IRSlots(IRSlots), Target(T) {
322 }
323 
324 VRegInfo &PerFunctionMIParsingState::getVRegInfo(Register Num) {
325   auto I = VRegInfos.insert(std::make_pair(Num, nullptr));
326   if (I.second) {
327     MachineRegisterInfo &MRI = MF.getRegInfo();
328     VRegInfo *Info = new (Allocator) VRegInfo;
329     Info->VReg = MRI.createIncompleteVirtualRegister();
330     I.first->second = Info;
331   }
332   return *I.first->second;
333 }
334 
335 VRegInfo &PerFunctionMIParsingState::getVRegInfoNamed(StringRef RegName) {
336   assert(RegName != "" && "Expected named reg.");
337 
338   auto I = VRegInfosNamed.insert(std::make_pair(RegName.str(), nullptr));
339   if (I.second) {
340     VRegInfo *Info = new (Allocator) VRegInfo;
341     Info->VReg = MF.getRegInfo().createIncompleteVirtualRegister(RegName);
342     I.first->second = Info;
343   }
344   return *I.first->second;
345 }
346 
347 static void mapValueToSlot(const Value *V, ModuleSlotTracker &MST,
348                            DenseMap<unsigned, const Value *> &Slots2Values) {
349   int Slot = MST.getLocalSlot(V);
350   if (Slot == -1)
351     return;
352   Slots2Values.insert(std::make_pair(unsigned(Slot), V));
353 }
354 
355 /// Creates the mapping from slot numbers to function's unnamed IR values.
356 static void initSlots2Values(const Function &F,
357                              DenseMap<unsigned, const Value *> &Slots2Values) {
358   ModuleSlotTracker MST(F.getParent(), /*ShouldInitializeAllMetadata=*/false);
359   MST.incorporateFunction(F);
360   for (const auto &Arg : F.args())
361     mapValueToSlot(&Arg, MST, Slots2Values);
362   for (const auto &BB : F) {
363     mapValueToSlot(&BB, MST, Slots2Values);
364     for (const auto &I : BB)
365       mapValueToSlot(&I, MST, Slots2Values);
366   }
367 }
368 
369 const Value* PerFunctionMIParsingState::getIRValue(unsigned Slot) {
370   if (Slots2Values.empty())
371     initSlots2Values(MF.getFunction(), Slots2Values);
372   return Slots2Values.lookup(Slot);
373 }
374 
375 namespace {
376 
377 /// A wrapper struct around the 'MachineOperand' struct that includes a source
378 /// range and other attributes.
379 struct ParsedMachineOperand {
380   MachineOperand Operand;
381   StringRef::iterator Begin;
382   StringRef::iterator End;
383   Optional<unsigned> TiedDefIdx;
384 
385   ParsedMachineOperand(const MachineOperand &Operand, StringRef::iterator Begin,
386                        StringRef::iterator End, Optional<unsigned> &TiedDefIdx)
387       : Operand(Operand), Begin(Begin), End(End), TiedDefIdx(TiedDefIdx) {
388     if (TiedDefIdx)
389       assert(Operand.isReg() && Operand.isUse() &&
390              "Only used register operands can be tied");
391   }
392 };
393 
394 class MIParser {
395   MachineFunction &MF;
396   SMDiagnostic &Error;
397   StringRef Source, CurrentSource;
398   SMRange SourceRange;
399   MIToken Token;
400   PerFunctionMIParsingState &PFS;
401   /// Maps from slot numbers to function's unnamed basic blocks.
402   DenseMap<unsigned, const BasicBlock *> Slots2BasicBlocks;
403 
404 public:
405   MIParser(PerFunctionMIParsingState &PFS, SMDiagnostic &Error,
406            StringRef Source);
407   MIParser(PerFunctionMIParsingState &PFS, SMDiagnostic &Error,
408            StringRef Source, SMRange SourceRange);
409 
410   /// \p SkipChar gives the number of characters to skip before looking
411   /// for the next token.
412   void lex(unsigned SkipChar = 0);
413 
414   /// Report an error at the current location with the given message.
415   ///
416   /// This function always return true.
417   bool error(const Twine &Msg);
418 
419   /// Report an error at the given location with the given message.
420   ///
421   /// This function always return true.
422   bool error(StringRef::iterator Loc, const Twine &Msg);
423 
424   bool
425   parseBasicBlockDefinitions(DenseMap<unsigned, MachineBasicBlock *> &MBBSlots);
426   bool parseBasicBlocks();
427   bool parse(MachineInstr *&MI);
428   bool parseStandaloneMBB(MachineBasicBlock *&MBB);
429   bool parseStandaloneNamedRegister(Register &Reg);
430   bool parseStandaloneVirtualRegister(VRegInfo *&Info);
431   bool parseStandaloneRegister(Register &Reg);
432   bool parseStandaloneStackObject(int &FI);
433   bool parseStandaloneMDNode(MDNode *&Node);
434   bool parseMachineMetadata();
435   bool parseMDTuple(MDNode *&MD, bool IsDistinct);
436   bool parseMDNodeVector(SmallVectorImpl<Metadata *> &Elts);
437   bool parseMetadata(Metadata *&MD);
438 
439   bool
440   parseBasicBlockDefinition(DenseMap<unsigned, MachineBasicBlock *> &MBBSlots);
441   bool parseBasicBlock(MachineBasicBlock &MBB,
442                        MachineBasicBlock *&AddFalthroughFrom);
443   bool parseBasicBlockLiveins(MachineBasicBlock &MBB);
444   bool parseBasicBlockSuccessors(MachineBasicBlock &MBB);
445 
446   bool parseNamedRegister(Register &Reg);
447   bool parseVirtualRegister(VRegInfo *&Info);
448   bool parseNamedVirtualRegister(VRegInfo *&Info);
449   bool parseRegister(Register &Reg, VRegInfo *&VRegInfo);
450   bool parseRegisterFlag(unsigned &Flags);
451   bool parseRegisterClassOrBank(VRegInfo &RegInfo);
452   bool parseSubRegisterIndex(unsigned &SubReg);
453   bool parseRegisterTiedDefIndex(unsigned &TiedDefIdx);
454   bool parseRegisterOperand(MachineOperand &Dest,
455                             Optional<unsigned> &TiedDefIdx, bool IsDef = false);
456   bool parseImmediateOperand(MachineOperand &Dest);
457   bool parseIRConstant(StringRef::iterator Loc, StringRef StringValue,
458                        const Constant *&C);
459   bool parseIRConstant(StringRef::iterator Loc, const Constant *&C);
460   bool parseLowLevelType(StringRef::iterator Loc, LLT &Ty);
461   bool parseTypedImmediateOperand(MachineOperand &Dest);
462   bool parseFPImmediateOperand(MachineOperand &Dest);
463   bool parseMBBReference(MachineBasicBlock *&MBB);
464   bool parseMBBOperand(MachineOperand &Dest);
465   bool parseStackFrameIndex(int &FI);
466   bool parseStackObjectOperand(MachineOperand &Dest);
467   bool parseFixedStackFrameIndex(int &FI);
468   bool parseFixedStackObjectOperand(MachineOperand &Dest);
469   bool parseGlobalValue(GlobalValue *&GV);
470   bool parseGlobalAddressOperand(MachineOperand &Dest);
471   bool parseConstantPoolIndexOperand(MachineOperand &Dest);
472   bool parseSubRegisterIndexOperand(MachineOperand &Dest);
473   bool parseJumpTableIndexOperand(MachineOperand &Dest);
474   bool parseExternalSymbolOperand(MachineOperand &Dest);
475   bool parseMCSymbolOperand(MachineOperand &Dest);
476   bool parseMDNode(MDNode *&Node);
477   bool parseDIExpression(MDNode *&Expr);
478   bool parseDILocation(MDNode *&Expr);
479   bool parseMetadataOperand(MachineOperand &Dest);
480   bool parseCFIOffset(int &Offset);
481   bool parseCFIRegister(Register &Reg);
482   bool parseCFIAddressSpace(unsigned &AddressSpace);
483   bool parseCFIEscapeValues(std::string& Values);
484   bool parseCFIOperand(MachineOperand &Dest);
485   bool parseIRBlock(BasicBlock *&BB, const Function &F);
486   bool parseBlockAddressOperand(MachineOperand &Dest);
487   bool parseIntrinsicOperand(MachineOperand &Dest);
488   bool parsePredicateOperand(MachineOperand &Dest);
489   bool parseShuffleMaskOperand(MachineOperand &Dest);
490   bool parseTargetIndexOperand(MachineOperand &Dest);
491   bool parseCustomRegisterMaskOperand(MachineOperand &Dest);
492   bool parseLiveoutRegisterMaskOperand(MachineOperand &Dest);
493   bool parseMachineOperand(const unsigned OpCode, const unsigned OpIdx,
494                            MachineOperand &Dest,
495                            Optional<unsigned> &TiedDefIdx);
496   bool parseMachineOperandAndTargetFlags(const unsigned OpCode,
497                                          const unsigned OpIdx,
498                                          MachineOperand &Dest,
499                                          Optional<unsigned> &TiedDefIdx);
500   bool parseOffset(int64_t &Offset);
501   bool parseAlignment(uint64_t &Alignment);
502   bool parseAddrspace(unsigned &Addrspace);
503   bool parseSectionID(Optional<MBBSectionID> &SID);
504   bool parseOperandsOffset(MachineOperand &Op);
505   bool parseIRValue(const Value *&V);
506   bool parseMemoryOperandFlag(MachineMemOperand::Flags &Flags);
507   bool parseMemoryPseudoSourceValue(const PseudoSourceValue *&PSV);
508   bool parseMachinePointerInfo(MachinePointerInfo &Dest);
509   bool parseOptionalScope(LLVMContext &Context, SyncScope::ID &SSID);
510   bool parseOptionalAtomicOrdering(AtomicOrdering &Order);
511   bool parseMachineMemoryOperand(MachineMemOperand *&Dest);
512   bool parsePreOrPostInstrSymbol(MCSymbol *&Symbol);
513   bool parseHeapAllocMarker(MDNode *&Node);
514 
515   bool parseTargetImmMnemonic(const unsigned OpCode, const unsigned OpIdx,
516                               MachineOperand &Dest, const MIRFormatter &MF);
517 
518 private:
519   /// Convert the integer literal in the current token into an unsigned integer.
520   ///
521   /// Return true if an error occurred.
522   bool getUnsigned(unsigned &Result);
523 
524   /// Convert the integer literal in the current token into an uint64.
525   ///
526   /// Return true if an error occurred.
527   bool getUint64(uint64_t &Result);
528 
529   /// Convert the hexadecimal literal in the current token into an unsigned
530   ///  APInt with a minimum bitwidth required to represent the value.
531   ///
532   /// Return true if the literal does not represent an integer value.
533   bool getHexUint(APInt &Result);
534 
535   /// If the current token is of the given kind, consume it and return false.
536   /// Otherwise report an error and return true.
537   bool expectAndConsume(MIToken::TokenKind TokenKind);
538 
539   /// If the current token is of the given kind, consume it and return true.
540   /// Otherwise return false.
541   bool consumeIfPresent(MIToken::TokenKind TokenKind);
542 
543   bool parseInstruction(unsigned &OpCode, unsigned &Flags);
544 
545   bool assignRegisterTies(MachineInstr &MI,
546                           ArrayRef<ParsedMachineOperand> Operands);
547 
548   bool verifyImplicitOperands(ArrayRef<ParsedMachineOperand> Operands,
549                               const MCInstrDesc &MCID);
550 
551   const BasicBlock *getIRBlock(unsigned Slot);
552   const BasicBlock *getIRBlock(unsigned Slot, const Function &F);
553 
554   /// Get or create an MCSymbol for a given name.
555   MCSymbol *getOrCreateMCSymbol(StringRef Name);
556 
557   /// parseStringConstant
558   ///   ::= StringConstant
559   bool parseStringConstant(std::string &Result);
560 
561   /// Map the location in the MI string to the corresponding location specified
562   /// in `SourceRange`.
563   SMLoc mapSMLoc(StringRef::iterator Loc);
564 };
565 
566 } // end anonymous namespace
567 
568 MIParser::MIParser(PerFunctionMIParsingState &PFS, SMDiagnostic &Error,
569                    StringRef Source)
570     : MF(PFS.MF), Error(Error), Source(Source), CurrentSource(Source), PFS(PFS)
571 {}
572 
573 MIParser::MIParser(PerFunctionMIParsingState &PFS, SMDiagnostic &Error,
574                    StringRef Source, SMRange SourceRange)
575     : MF(PFS.MF), Error(Error), Source(Source), CurrentSource(Source),
576       SourceRange(SourceRange), PFS(PFS) {}
577 
578 void MIParser::lex(unsigned SkipChar) {
579   CurrentSource = lexMIToken(
580       CurrentSource.slice(SkipChar, StringRef::npos), Token,
581       [this](StringRef::iterator Loc, const Twine &Msg) { error(Loc, Msg); });
582 }
583 
584 bool MIParser::error(const Twine &Msg) { return error(Token.location(), Msg); }
585 
586 bool MIParser::error(StringRef::iterator Loc, const Twine &Msg) {
587   const SourceMgr &SM = *PFS.SM;
588   assert(Loc >= Source.data() && Loc <= (Source.data() + Source.size()));
589   const MemoryBuffer &Buffer = *SM.getMemoryBuffer(SM.getMainFileID());
590   if (Loc >= Buffer.getBufferStart() && Loc <= Buffer.getBufferEnd()) {
591     // Create an ordinary diagnostic when the source manager's buffer is the
592     // source string.
593     Error = SM.GetMessage(SMLoc::getFromPointer(Loc), SourceMgr::DK_Error, Msg);
594     return true;
595   }
596   // Create a diagnostic for a YAML string literal.
597   Error = SMDiagnostic(SM, SMLoc(), Buffer.getBufferIdentifier(), 1,
598                        Loc - Source.data(), SourceMgr::DK_Error, Msg.str(),
599                        Source, None, None);
600   return true;
601 }
602 
603 SMLoc MIParser::mapSMLoc(StringRef::iterator Loc) {
604   assert(SourceRange.isValid() && "Invalid source range");
605   assert(Loc >= Source.data() && Loc <= (Source.data() + Source.size()));
606   return SMLoc::getFromPointer(SourceRange.Start.getPointer() +
607                                (Loc - Source.data()));
608 }
609 
610 typedef function_ref<bool(StringRef::iterator Loc, const Twine &)>
611     ErrorCallbackType;
612 
613 static const char *toString(MIToken::TokenKind TokenKind) {
614   switch (TokenKind) {
615   case MIToken::comma:
616     return "','";
617   case MIToken::equal:
618     return "'='";
619   case MIToken::colon:
620     return "':'";
621   case MIToken::lparen:
622     return "'('";
623   case MIToken::rparen:
624     return "')'";
625   default:
626     return "<unknown token>";
627   }
628 }
629 
630 bool MIParser::expectAndConsume(MIToken::TokenKind TokenKind) {
631   if (Token.isNot(TokenKind))
632     return error(Twine("expected ") + toString(TokenKind));
633   lex();
634   return false;
635 }
636 
637 bool MIParser::consumeIfPresent(MIToken::TokenKind TokenKind) {
638   if (Token.isNot(TokenKind))
639     return false;
640   lex();
641   return true;
642 }
643 
644 // Parse Machine Basic Block Section ID.
645 bool MIParser::parseSectionID(Optional<MBBSectionID> &SID) {
646   assert(Token.is(MIToken::kw_bbsections));
647   lex();
648   if (Token.is(MIToken::IntegerLiteral)) {
649     unsigned Value = 0;
650     if (getUnsigned(Value))
651       return error("Unknown Section ID");
652     SID = MBBSectionID{Value};
653   } else {
654     const StringRef &S = Token.stringValue();
655     if (S == "Exception")
656       SID = MBBSectionID::ExceptionSectionID;
657     else if (S == "Cold")
658       SID = MBBSectionID::ColdSectionID;
659     else
660       return error("Unknown Section ID");
661   }
662   lex();
663   return false;
664 }
665 
666 bool MIParser::parseBasicBlockDefinition(
667     DenseMap<unsigned, MachineBasicBlock *> &MBBSlots) {
668   assert(Token.is(MIToken::MachineBasicBlockLabel));
669   unsigned ID = 0;
670   if (getUnsigned(ID))
671     return true;
672   auto Loc = Token.location();
673   auto Name = Token.stringValue();
674   lex();
675   bool HasAddressTaken = false;
676   bool IsLandingPad = false;
677   bool IsInlineAsmBrIndirectTarget = false;
678   bool IsEHFuncletEntry = false;
679   Optional<MBBSectionID> SectionID;
680   uint64_t Alignment = 0;
681   BasicBlock *BB = nullptr;
682   if (consumeIfPresent(MIToken::lparen)) {
683     do {
684       // TODO: Report an error when multiple same attributes are specified.
685       switch (Token.kind()) {
686       case MIToken::kw_address_taken:
687         HasAddressTaken = true;
688         lex();
689         break;
690       case MIToken::kw_landing_pad:
691         IsLandingPad = true;
692         lex();
693         break;
694       case MIToken::kw_inlineasm_br_indirect_target:
695         IsInlineAsmBrIndirectTarget = true;
696         lex();
697         break;
698       case MIToken::kw_ehfunclet_entry:
699         IsEHFuncletEntry = true;
700         lex();
701         break;
702       case MIToken::kw_align:
703         if (parseAlignment(Alignment))
704           return true;
705         break;
706       case MIToken::IRBlock:
707         // TODO: Report an error when both name and ir block are specified.
708         if (parseIRBlock(BB, MF.getFunction()))
709           return true;
710         lex();
711         break;
712       case MIToken::kw_bbsections:
713         if (parseSectionID(SectionID))
714           return true;
715         break;
716       default:
717         break;
718       }
719     } while (consumeIfPresent(MIToken::comma));
720     if (expectAndConsume(MIToken::rparen))
721       return true;
722   }
723   if (expectAndConsume(MIToken::colon))
724     return true;
725 
726   if (!Name.empty()) {
727     BB = dyn_cast_or_null<BasicBlock>(
728         MF.getFunction().getValueSymbolTable()->lookup(Name));
729     if (!BB)
730       return error(Loc, Twine("basic block '") + Name +
731                             "' is not defined in the function '" +
732                             MF.getName() + "'");
733   }
734   auto *MBB = MF.CreateMachineBasicBlock(BB);
735   MF.insert(MF.end(), MBB);
736   bool WasInserted = MBBSlots.insert(std::make_pair(ID, MBB)).second;
737   if (!WasInserted)
738     return error(Loc, Twine("redefinition of machine basic block with id #") +
739                           Twine(ID));
740   if (Alignment)
741     MBB->setAlignment(Align(Alignment));
742   if (HasAddressTaken)
743     MBB->setHasAddressTaken();
744   MBB->setIsEHPad(IsLandingPad);
745   MBB->setIsInlineAsmBrIndirectTarget(IsInlineAsmBrIndirectTarget);
746   MBB->setIsEHFuncletEntry(IsEHFuncletEntry);
747   if (SectionID.hasValue()) {
748     MBB->setSectionID(SectionID.getValue());
749     MF.setBBSectionsType(BasicBlockSection::List);
750   }
751   return false;
752 }
753 
754 bool MIParser::parseBasicBlockDefinitions(
755     DenseMap<unsigned, MachineBasicBlock *> &MBBSlots) {
756   lex();
757   // Skip until the first machine basic block.
758   while (Token.is(MIToken::Newline))
759     lex();
760   if (Token.isErrorOrEOF())
761     return Token.isError();
762   if (Token.isNot(MIToken::MachineBasicBlockLabel))
763     return error("expected a basic block definition before instructions");
764   unsigned BraceDepth = 0;
765   do {
766     if (parseBasicBlockDefinition(MBBSlots))
767       return true;
768     bool IsAfterNewline = false;
769     // Skip until the next machine basic block.
770     while (true) {
771       if ((Token.is(MIToken::MachineBasicBlockLabel) && IsAfterNewline) ||
772           Token.isErrorOrEOF())
773         break;
774       else if (Token.is(MIToken::MachineBasicBlockLabel))
775         return error("basic block definition should be located at the start of "
776                      "the line");
777       else if (consumeIfPresent(MIToken::Newline)) {
778         IsAfterNewline = true;
779         continue;
780       }
781       IsAfterNewline = false;
782       if (Token.is(MIToken::lbrace))
783         ++BraceDepth;
784       if (Token.is(MIToken::rbrace)) {
785         if (!BraceDepth)
786           return error("extraneous closing brace ('}')");
787         --BraceDepth;
788       }
789       lex();
790     }
791     // Verify that we closed all of the '{' at the end of a file or a block.
792     if (!Token.isError() && BraceDepth)
793       return error("expected '}'"); // FIXME: Report a note that shows '{'.
794   } while (!Token.isErrorOrEOF());
795   return Token.isError();
796 }
797 
798 bool MIParser::parseBasicBlockLiveins(MachineBasicBlock &MBB) {
799   assert(Token.is(MIToken::kw_liveins));
800   lex();
801   if (expectAndConsume(MIToken::colon))
802     return true;
803   if (Token.isNewlineOrEOF()) // Allow an empty list of liveins.
804     return false;
805   do {
806     if (Token.isNot(MIToken::NamedRegister))
807       return error("expected a named register");
808     Register Reg;
809     if (parseNamedRegister(Reg))
810       return true;
811     lex();
812     LaneBitmask Mask = LaneBitmask::getAll();
813     if (consumeIfPresent(MIToken::colon)) {
814       // Parse lane mask.
815       if (Token.isNot(MIToken::IntegerLiteral) &&
816           Token.isNot(MIToken::HexLiteral))
817         return error("expected a lane mask");
818       static_assert(sizeof(LaneBitmask::Type) == sizeof(uint64_t),
819                     "Use correct get-function for lane mask");
820       LaneBitmask::Type V;
821       if (getUint64(V))
822         return error("invalid lane mask value");
823       Mask = LaneBitmask(V);
824       lex();
825     }
826     MBB.addLiveIn(Reg, Mask);
827   } while (consumeIfPresent(MIToken::comma));
828   return false;
829 }
830 
831 bool MIParser::parseBasicBlockSuccessors(MachineBasicBlock &MBB) {
832   assert(Token.is(MIToken::kw_successors));
833   lex();
834   if (expectAndConsume(MIToken::colon))
835     return true;
836   if (Token.isNewlineOrEOF()) // Allow an empty list of successors.
837     return false;
838   do {
839     if (Token.isNot(MIToken::MachineBasicBlock))
840       return error("expected a machine basic block reference");
841     MachineBasicBlock *SuccMBB = nullptr;
842     if (parseMBBReference(SuccMBB))
843       return true;
844     lex();
845     unsigned Weight = 0;
846     if (consumeIfPresent(MIToken::lparen)) {
847       if (Token.isNot(MIToken::IntegerLiteral) &&
848           Token.isNot(MIToken::HexLiteral))
849         return error("expected an integer literal after '('");
850       if (getUnsigned(Weight))
851         return true;
852       lex();
853       if (expectAndConsume(MIToken::rparen))
854         return true;
855     }
856     MBB.addSuccessor(SuccMBB, BranchProbability::getRaw(Weight));
857   } while (consumeIfPresent(MIToken::comma));
858   MBB.normalizeSuccProbs();
859   return false;
860 }
861 
862 bool MIParser::parseBasicBlock(MachineBasicBlock &MBB,
863                                MachineBasicBlock *&AddFalthroughFrom) {
864   // Skip the definition.
865   assert(Token.is(MIToken::MachineBasicBlockLabel));
866   lex();
867   if (consumeIfPresent(MIToken::lparen)) {
868     while (Token.isNot(MIToken::rparen) && !Token.isErrorOrEOF())
869       lex();
870     consumeIfPresent(MIToken::rparen);
871   }
872   consumeIfPresent(MIToken::colon);
873 
874   // Parse the liveins and successors.
875   // N.B: Multiple lists of successors and liveins are allowed and they're
876   // merged into one.
877   // Example:
878   //   liveins: $edi
879   //   liveins: $esi
880   //
881   // is equivalent to
882   //   liveins: $edi, $esi
883   bool ExplicitSuccessors = false;
884   while (true) {
885     if (Token.is(MIToken::kw_successors)) {
886       if (parseBasicBlockSuccessors(MBB))
887         return true;
888       ExplicitSuccessors = true;
889     } else if (Token.is(MIToken::kw_liveins)) {
890       if (parseBasicBlockLiveins(MBB))
891         return true;
892     } else if (consumeIfPresent(MIToken::Newline)) {
893       continue;
894     } else
895       break;
896     if (!Token.isNewlineOrEOF())
897       return error("expected line break at the end of a list");
898     lex();
899   }
900 
901   // Parse the instructions.
902   bool IsInBundle = false;
903   MachineInstr *PrevMI = nullptr;
904   while (!Token.is(MIToken::MachineBasicBlockLabel) &&
905          !Token.is(MIToken::Eof)) {
906     if (consumeIfPresent(MIToken::Newline))
907       continue;
908     if (consumeIfPresent(MIToken::rbrace)) {
909       // The first parsing pass should verify that all closing '}' have an
910       // opening '{'.
911       assert(IsInBundle);
912       IsInBundle = false;
913       continue;
914     }
915     MachineInstr *MI = nullptr;
916     if (parse(MI))
917       return true;
918     MBB.insert(MBB.end(), MI);
919     if (IsInBundle) {
920       PrevMI->setFlag(MachineInstr::BundledSucc);
921       MI->setFlag(MachineInstr::BundledPred);
922     }
923     PrevMI = MI;
924     if (Token.is(MIToken::lbrace)) {
925       if (IsInBundle)
926         return error("nested instruction bundles are not allowed");
927       lex();
928       // This instruction is the start of the bundle.
929       MI->setFlag(MachineInstr::BundledSucc);
930       IsInBundle = true;
931       if (!Token.is(MIToken::Newline))
932         // The next instruction can be on the same line.
933         continue;
934     }
935     assert(Token.isNewlineOrEOF() && "MI is not fully parsed");
936     lex();
937   }
938 
939   // Construct successor list by searching for basic block machine operands.
940   if (!ExplicitSuccessors) {
941     SmallVector<MachineBasicBlock*,4> Successors;
942     bool IsFallthrough;
943     guessSuccessors(MBB, Successors, IsFallthrough);
944     for (MachineBasicBlock *Succ : Successors)
945       MBB.addSuccessor(Succ);
946 
947     if (IsFallthrough) {
948       AddFalthroughFrom = &MBB;
949     } else {
950       MBB.normalizeSuccProbs();
951     }
952   }
953 
954   return false;
955 }
956 
957 bool MIParser::parseBasicBlocks() {
958   lex();
959   // Skip until the first machine basic block.
960   while (Token.is(MIToken::Newline))
961     lex();
962   if (Token.isErrorOrEOF())
963     return Token.isError();
964   // The first parsing pass should have verified that this token is a MBB label
965   // in the 'parseBasicBlockDefinitions' method.
966   assert(Token.is(MIToken::MachineBasicBlockLabel));
967   MachineBasicBlock *AddFalthroughFrom = nullptr;
968   do {
969     MachineBasicBlock *MBB = nullptr;
970     if (parseMBBReference(MBB))
971       return true;
972     if (AddFalthroughFrom) {
973       if (!AddFalthroughFrom->isSuccessor(MBB))
974         AddFalthroughFrom->addSuccessor(MBB);
975       AddFalthroughFrom->normalizeSuccProbs();
976       AddFalthroughFrom = nullptr;
977     }
978     if (parseBasicBlock(*MBB, AddFalthroughFrom))
979       return true;
980     // The method 'parseBasicBlock' should parse the whole block until the next
981     // block or the end of file.
982     assert(Token.is(MIToken::MachineBasicBlockLabel) || Token.is(MIToken::Eof));
983   } while (Token.isNot(MIToken::Eof));
984   return false;
985 }
986 
987 bool MIParser::parse(MachineInstr *&MI) {
988   // Parse any register operands before '='
989   MachineOperand MO = MachineOperand::CreateImm(0);
990   SmallVector<ParsedMachineOperand, 8> Operands;
991   while (Token.isRegister() || Token.isRegisterFlag()) {
992     auto Loc = Token.location();
993     Optional<unsigned> TiedDefIdx;
994     if (parseRegisterOperand(MO, TiedDefIdx, /*IsDef=*/true))
995       return true;
996     Operands.push_back(
997         ParsedMachineOperand(MO, Loc, Token.location(), TiedDefIdx));
998     if (Token.isNot(MIToken::comma))
999       break;
1000     lex();
1001   }
1002   if (!Operands.empty() && expectAndConsume(MIToken::equal))
1003     return true;
1004 
1005   unsigned OpCode, Flags = 0;
1006   if (Token.isError() || parseInstruction(OpCode, Flags))
1007     return true;
1008 
1009   // Parse the remaining machine operands.
1010   while (!Token.isNewlineOrEOF() && Token.isNot(MIToken::kw_pre_instr_symbol) &&
1011          Token.isNot(MIToken::kw_post_instr_symbol) &&
1012          Token.isNot(MIToken::kw_heap_alloc_marker) &&
1013          Token.isNot(MIToken::kw_debug_location) &&
1014          Token.isNot(MIToken::kw_debug_instr_number) &&
1015          Token.isNot(MIToken::coloncolon) && Token.isNot(MIToken::lbrace)) {
1016     auto Loc = Token.location();
1017     Optional<unsigned> TiedDefIdx;
1018     if (parseMachineOperandAndTargetFlags(OpCode, Operands.size(), MO, TiedDefIdx))
1019       return true;
1020     Operands.push_back(
1021         ParsedMachineOperand(MO, Loc, Token.location(), TiedDefIdx));
1022     if (Token.isNewlineOrEOF() || Token.is(MIToken::coloncolon) ||
1023         Token.is(MIToken::lbrace))
1024       break;
1025     if (Token.isNot(MIToken::comma))
1026       return error("expected ',' before the next machine operand");
1027     lex();
1028   }
1029 
1030   MCSymbol *PreInstrSymbol = nullptr;
1031   if (Token.is(MIToken::kw_pre_instr_symbol))
1032     if (parsePreOrPostInstrSymbol(PreInstrSymbol))
1033       return true;
1034   MCSymbol *PostInstrSymbol = nullptr;
1035   if (Token.is(MIToken::kw_post_instr_symbol))
1036     if (parsePreOrPostInstrSymbol(PostInstrSymbol))
1037       return true;
1038   MDNode *HeapAllocMarker = nullptr;
1039   if (Token.is(MIToken::kw_heap_alloc_marker))
1040     if (parseHeapAllocMarker(HeapAllocMarker))
1041       return true;
1042 
1043   unsigned InstrNum = 0;
1044   if (Token.is(MIToken::kw_debug_instr_number)) {
1045     lex();
1046     if (Token.isNot(MIToken::IntegerLiteral))
1047       return error("expected an integer literal after 'debug-instr-number'");
1048     if (getUnsigned(InstrNum))
1049       return true;
1050     lex();
1051     // Lex past trailing comma if present.
1052     if (Token.is(MIToken::comma))
1053       lex();
1054   }
1055 
1056   DebugLoc DebugLocation;
1057   if (Token.is(MIToken::kw_debug_location)) {
1058     lex();
1059     MDNode *Node = nullptr;
1060     if (Token.is(MIToken::exclaim)) {
1061       if (parseMDNode(Node))
1062         return true;
1063     } else if (Token.is(MIToken::md_dilocation)) {
1064       if (parseDILocation(Node))
1065         return true;
1066     } else
1067       return error("expected a metadata node after 'debug-location'");
1068     if (!isa<DILocation>(Node))
1069       return error("referenced metadata is not a DILocation");
1070     DebugLocation = DebugLoc(Node);
1071   }
1072 
1073   // Parse the machine memory operands.
1074   SmallVector<MachineMemOperand *, 2> MemOperands;
1075   if (Token.is(MIToken::coloncolon)) {
1076     lex();
1077     while (!Token.isNewlineOrEOF()) {
1078       MachineMemOperand *MemOp = nullptr;
1079       if (parseMachineMemoryOperand(MemOp))
1080         return true;
1081       MemOperands.push_back(MemOp);
1082       if (Token.isNewlineOrEOF())
1083         break;
1084       if (Token.isNot(MIToken::comma))
1085         return error("expected ',' before the next machine memory operand");
1086       lex();
1087     }
1088   }
1089 
1090   const auto &MCID = MF.getSubtarget().getInstrInfo()->get(OpCode);
1091   if (!MCID.isVariadic()) {
1092     // FIXME: Move the implicit operand verification to the machine verifier.
1093     if (verifyImplicitOperands(Operands, MCID))
1094       return true;
1095   }
1096 
1097   // TODO: Check for extraneous machine operands.
1098   MI = MF.CreateMachineInstr(MCID, DebugLocation, /*NoImplicit=*/true);
1099   MI->setFlags(Flags);
1100   for (const auto &Operand : Operands)
1101     MI->addOperand(MF, Operand.Operand);
1102   if (assignRegisterTies(*MI, Operands))
1103     return true;
1104   if (PreInstrSymbol)
1105     MI->setPreInstrSymbol(MF, PreInstrSymbol);
1106   if (PostInstrSymbol)
1107     MI->setPostInstrSymbol(MF, PostInstrSymbol);
1108   if (HeapAllocMarker)
1109     MI->setHeapAllocMarker(MF, HeapAllocMarker);
1110   if (!MemOperands.empty())
1111     MI->setMemRefs(MF, MemOperands);
1112   if (InstrNum)
1113     MI->setDebugInstrNum(InstrNum);
1114   return false;
1115 }
1116 
1117 bool MIParser::parseStandaloneMBB(MachineBasicBlock *&MBB) {
1118   lex();
1119   if (Token.isNot(MIToken::MachineBasicBlock))
1120     return error("expected a machine basic block reference");
1121   if (parseMBBReference(MBB))
1122     return true;
1123   lex();
1124   if (Token.isNot(MIToken::Eof))
1125     return error(
1126         "expected end of string after the machine basic block reference");
1127   return false;
1128 }
1129 
1130 bool MIParser::parseStandaloneNamedRegister(Register &Reg) {
1131   lex();
1132   if (Token.isNot(MIToken::NamedRegister))
1133     return error("expected a named register");
1134   if (parseNamedRegister(Reg))
1135     return true;
1136   lex();
1137   if (Token.isNot(MIToken::Eof))
1138     return error("expected end of string after the register reference");
1139   return false;
1140 }
1141 
1142 bool MIParser::parseStandaloneVirtualRegister(VRegInfo *&Info) {
1143   lex();
1144   if (Token.isNot(MIToken::VirtualRegister))
1145     return error("expected a virtual register");
1146   if (parseVirtualRegister(Info))
1147     return true;
1148   lex();
1149   if (Token.isNot(MIToken::Eof))
1150     return error("expected end of string after the register reference");
1151   return false;
1152 }
1153 
1154 bool MIParser::parseStandaloneRegister(Register &Reg) {
1155   lex();
1156   if (Token.isNot(MIToken::NamedRegister) &&
1157       Token.isNot(MIToken::VirtualRegister))
1158     return error("expected either a named or virtual register");
1159 
1160   VRegInfo *Info;
1161   if (parseRegister(Reg, Info))
1162     return true;
1163 
1164   lex();
1165   if (Token.isNot(MIToken::Eof))
1166     return error("expected end of string after the register reference");
1167   return false;
1168 }
1169 
1170 bool MIParser::parseStandaloneStackObject(int &FI) {
1171   lex();
1172   if (Token.isNot(MIToken::StackObject))
1173     return error("expected a stack object");
1174   if (parseStackFrameIndex(FI))
1175     return true;
1176   if (Token.isNot(MIToken::Eof))
1177     return error("expected end of string after the stack object reference");
1178   return false;
1179 }
1180 
1181 bool MIParser::parseStandaloneMDNode(MDNode *&Node) {
1182   lex();
1183   if (Token.is(MIToken::exclaim)) {
1184     if (parseMDNode(Node))
1185       return true;
1186   } else if (Token.is(MIToken::md_diexpr)) {
1187     if (parseDIExpression(Node))
1188       return true;
1189   } else if (Token.is(MIToken::md_dilocation)) {
1190     if (parseDILocation(Node))
1191       return true;
1192   } else
1193     return error("expected a metadata node");
1194   if (Token.isNot(MIToken::Eof))
1195     return error("expected end of string after the metadata node");
1196   return false;
1197 }
1198 
1199 bool MIParser::parseMachineMetadata() {
1200   lex();
1201   if (Token.isNot(MIToken::exclaim))
1202     return error("expected a metadata node");
1203 
1204   lex();
1205   if (Token.isNot(MIToken::IntegerLiteral) || Token.integerValue().isSigned())
1206     return error("expected metadata id after '!'");
1207   unsigned ID = 0;
1208   if (getUnsigned(ID))
1209     return true;
1210   lex();
1211   if (expectAndConsume(MIToken::equal))
1212     return true;
1213   bool IsDistinct = Token.is(MIToken::kw_distinct);
1214   if (IsDistinct)
1215     lex();
1216   if (Token.isNot(MIToken::exclaim))
1217     return error("expected a metadata node");
1218   lex();
1219 
1220   MDNode *MD;
1221   if (parseMDTuple(MD, IsDistinct))
1222     return true;
1223 
1224   auto FI = PFS.MachineForwardRefMDNodes.find(ID);
1225   if (FI != PFS.MachineForwardRefMDNodes.end()) {
1226     FI->second.first->replaceAllUsesWith(MD);
1227     PFS.MachineForwardRefMDNodes.erase(FI);
1228 
1229     assert(PFS.MachineMetadataNodes[ID] == MD && "Tracking VH didn't work");
1230   } else {
1231     if (PFS.MachineMetadataNodes.count(ID))
1232       return error("Metadata id is already used");
1233     PFS.MachineMetadataNodes[ID].reset(MD);
1234   }
1235 
1236   return false;
1237 }
1238 
1239 bool MIParser::parseMDTuple(MDNode *&MD, bool IsDistinct) {
1240   SmallVector<Metadata *, 16> Elts;
1241   if (parseMDNodeVector(Elts))
1242     return true;
1243   MD = (IsDistinct ? MDTuple::getDistinct
1244                    : MDTuple::get)(MF.getFunction().getContext(), Elts);
1245   return false;
1246 }
1247 
1248 bool MIParser::parseMDNodeVector(SmallVectorImpl<Metadata *> &Elts) {
1249   if (Token.isNot(MIToken::lbrace))
1250     return error("expected '{' here");
1251   lex();
1252 
1253   if (Token.is(MIToken::rbrace)) {
1254     lex();
1255     return false;
1256   }
1257 
1258   do {
1259     Metadata *MD;
1260     if (parseMetadata(MD))
1261       return true;
1262 
1263     Elts.push_back(MD);
1264 
1265     if (Token.isNot(MIToken::comma))
1266       break;
1267     lex();
1268   } while (true);
1269 
1270   if (Token.isNot(MIToken::rbrace))
1271     return error("expected end of metadata node");
1272   lex();
1273 
1274   return false;
1275 }
1276 
1277 // ::= !42
1278 // ::= !"string"
1279 bool MIParser::parseMetadata(Metadata *&MD) {
1280   if (Token.isNot(MIToken::exclaim))
1281     return error("expected '!' here");
1282   lex();
1283 
1284   if (Token.is(MIToken::StringConstant)) {
1285     std::string Str;
1286     if (parseStringConstant(Str))
1287       return true;
1288     MD = MDString::get(MF.getFunction().getContext(), Str);
1289     return false;
1290   }
1291 
1292   if (Token.isNot(MIToken::IntegerLiteral) || Token.integerValue().isSigned())
1293     return error("expected metadata id after '!'");
1294 
1295   SMLoc Loc = mapSMLoc(Token.location());
1296 
1297   unsigned ID = 0;
1298   if (getUnsigned(ID))
1299     return true;
1300   lex();
1301 
1302   auto NodeInfo = PFS.IRSlots.MetadataNodes.find(ID);
1303   if (NodeInfo != PFS.IRSlots.MetadataNodes.end()) {
1304     MD = NodeInfo->second.get();
1305     return false;
1306   }
1307   // Check machine metadata.
1308   NodeInfo = PFS.MachineMetadataNodes.find(ID);
1309   if (NodeInfo != PFS.MachineMetadataNodes.end()) {
1310     MD = NodeInfo->second.get();
1311     return false;
1312   }
1313   // Forward reference.
1314   auto &FwdRef = PFS.MachineForwardRefMDNodes[ID];
1315   FwdRef = std::make_pair(
1316       MDTuple::getTemporary(MF.getFunction().getContext(), None), Loc);
1317   PFS.MachineMetadataNodes[ID].reset(FwdRef.first.get());
1318   MD = FwdRef.first.get();
1319 
1320   return false;
1321 }
1322 
1323 static const char *printImplicitRegisterFlag(const MachineOperand &MO) {
1324   assert(MO.isImplicit());
1325   return MO.isDef() ? "implicit-def" : "implicit";
1326 }
1327 
1328 static std::string getRegisterName(const TargetRegisterInfo *TRI,
1329                                    Register Reg) {
1330   assert(Register::isPhysicalRegister(Reg) && "expected phys reg");
1331   return StringRef(TRI->getName(Reg)).lower();
1332 }
1333 
1334 /// Return true if the parsed machine operands contain a given machine operand.
1335 static bool isImplicitOperandIn(const MachineOperand &ImplicitOperand,
1336                                 ArrayRef<ParsedMachineOperand> Operands) {
1337   for (const auto &I : Operands) {
1338     if (ImplicitOperand.isIdenticalTo(I.Operand))
1339       return true;
1340   }
1341   return false;
1342 }
1343 
1344 bool MIParser::verifyImplicitOperands(ArrayRef<ParsedMachineOperand> Operands,
1345                                       const MCInstrDesc &MCID) {
1346   if (MCID.isCall())
1347     // We can't verify call instructions as they can contain arbitrary implicit
1348     // register and register mask operands.
1349     return false;
1350 
1351   // Gather all the expected implicit operands.
1352   SmallVector<MachineOperand, 4> ImplicitOperands;
1353   if (MCID.ImplicitDefs)
1354     for (const MCPhysReg *ImpDefs = MCID.getImplicitDefs(); *ImpDefs; ++ImpDefs)
1355       ImplicitOperands.push_back(
1356           MachineOperand::CreateReg(*ImpDefs, true, true));
1357   if (MCID.ImplicitUses)
1358     for (const MCPhysReg *ImpUses = MCID.getImplicitUses(); *ImpUses; ++ImpUses)
1359       ImplicitOperands.push_back(
1360           MachineOperand::CreateReg(*ImpUses, false, true));
1361 
1362   const auto *TRI = MF.getSubtarget().getRegisterInfo();
1363   assert(TRI && "Expected target register info");
1364   for (const auto &I : ImplicitOperands) {
1365     if (isImplicitOperandIn(I, Operands))
1366       continue;
1367     return error(Operands.empty() ? Token.location() : Operands.back().End,
1368                  Twine("missing implicit register operand '") +
1369                      printImplicitRegisterFlag(I) + " $" +
1370                      getRegisterName(TRI, I.getReg()) + "'");
1371   }
1372   return false;
1373 }
1374 
1375 bool MIParser::parseInstruction(unsigned &OpCode, unsigned &Flags) {
1376   // Allow frame and fast math flags for OPCODE
1377   while (Token.is(MIToken::kw_frame_setup) ||
1378          Token.is(MIToken::kw_frame_destroy) ||
1379          Token.is(MIToken::kw_nnan) ||
1380          Token.is(MIToken::kw_ninf) ||
1381          Token.is(MIToken::kw_nsz) ||
1382          Token.is(MIToken::kw_arcp) ||
1383          Token.is(MIToken::kw_contract) ||
1384          Token.is(MIToken::kw_afn) ||
1385          Token.is(MIToken::kw_reassoc) ||
1386          Token.is(MIToken::kw_nuw) ||
1387          Token.is(MIToken::kw_nsw) ||
1388          Token.is(MIToken::kw_exact) ||
1389          Token.is(MIToken::kw_nofpexcept)) {
1390     // Mine frame and fast math flags
1391     if (Token.is(MIToken::kw_frame_setup))
1392       Flags |= MachineInstr::FrameSetup;
1393     if (Token.is(MIToken::kw_frame_destroy))
1394       Flags |= MachineInstr::FrameDestroy;
1395     if (Token.is(MIToken::kw_nnan))
1396       Flags |= MachineInstr::FmNoNans;
1397     if (Token.is(MIToken::kw_ninf))
1398       Flags |= MachineInstr::FmNoInfs;
1399     if (Token.is(MIToken::kw_nsz))
1400       Flags |= MachineInstr::FmNsz;
1401     if (Token.is(MIToken::kw_arcp))
1402       Flags |= MachineInstr::FmArcp;
1403     if (Token.is(MIToken::kw_contract))
1404       Flags |= MachineInstr::FmContract;
1405     if (Token.is(MIToken::kw_afn))
1406       Flags |= MachineInstr::FmAfn;
1407     if (Token.is(MIToken::kw_reassoc))
1408       Flags |= MachineInstr::FmReassoc;
1409     if (Token.is(MIToken::kw_nuw))
1410       Flags |= MachineInstr::NoUWrap;
1411     if (Token.is(MIToken::kw_nsw))
1412       Flags |= MachineInstr::NoSWrap;
1413     if (Token.is(MIToken::kw_exact))
1414       Flags |= MachineInstr::IsExact;
1415     if (Token.is(MIToken::kw_nofpexcept))
1416       Flags |= MachineInstr::NoFPExcept;
1417 
1418     lex();
1419   }
1420   if (Token.isNot(MIToken::Identifier))
1421     return error("expected a machine instruction");
1422   StringRef InstrName = Token.stringValue();
1423   if (PFS.Target.parseInstrName(InstrName, OpCode))
1424     return error(Twine("unknown machine instruction name '") + InstrName + "'");
1425   lex();
1426   return false;
1427 }
1428 
1429 bool MIParser::parseNamedRegister(Register &Reg) {
1430   assert(Token.is(MIToken::NamedRegister) && "Needs NamedRegister token");
1431   StringRef Name = Token.stringValue();
1432   if (PFS.Target.getRegisterByName(Name, Reg))
1433     return error(Twine("unknown register name '") + Name + "'");
1434   return false;
1435 }
1436 
1437 bool MIParser::parseNamedVirtualRegister(VRegInfo *&Info) {
1438   assert(Token.is(MIToken::NamedVirtualRegister) && "Expected NamedVReg token");
1439   StringRef Name = Token.stringValue();
1440   // TODO: Check that the VReg name is not the same as a physical register name.
1441   //       If it is, then print a warning (when warnings are implemented).
1442   Info = &PFS.getVRegInfoNamed(Name);
1443   return false;
1444 }
1445 
1446 bool MIParser::parseVirtualRegister(VRegInfo *&Info) {
1447   if (Token.is(MIToken::NamedVirtualRegister))
1448     return parseNamedVirtualRegister(Info);
1449   assert(Token.is(MIToken::VirtualRegister) && "Needs VirtualRegister token");
1450   unsigned ID;
1451   if (getUnsigned(ID))
1452     return true;
1453   Info = &PFS.getVRegInfo(ID);
1454   return false;
1455 }
1456 
1457 bool MIParser::parseRegister(Register &Reg, VRegInfo *&Info) {
1458   switch (Token.kind()) {
1459   case MIToken::underscore:
1460     Reg = 0;
1461     return false;
1462   case MIToken::NamedRegister:
1463     return parseNamedRegister(Reg);
1464   case MIToken::NamedVirtualRegister:
1465   case MIToken::VirtualRegister:
1466     if (parseVirtualRegister(Info))
1467       return true;
1468     Reg = Info->VReg;
1469     return false;
1470   // TODO: Parse other register kinds.
1471   default:
1472     llvm_unreachable("The current token should be a register");
1473   }
1474 }
1475 
1476 bool MIParser::parseRegisterClassOrBank(VRegInfo &RegInfo) {
1477   if (Token.isNot(MIToken::Identifier) && Token.isNot(MIToken::underscore))
1478     return error("expected '_', register class, or register bank name");
1479   StringRef::iterator Loc = Token.location();
1480   StringRef Name = Token.stringValue();
1481 
1482   // Was it a register class?
1483   const TargetRegisterClass *RC = PFS.Target.getRegClass(Name);
1484   if (RC) {
1485     lex();
1486 
1487     switch (RegInfo.Kind) {
1488     case VRegInfo::UNKNOWN:
1489     case VRegInfo::NORMAL:
1490       RegInfo.Kind = VRegInfo::NORMAL;
1491       if (RegInfo.Explicit && RegInfo.D.RC != RC) {
1492         const TargetRegisterInfo &TRI = *MF.getSubtarget().getRegisterInfo();
1493         return error(Loc, Twine("conflicting register classes, previously: ") +
1494                      Twine(TRI.getRegClassName(RegInfo.D.RC)));
1495       }
1496       RegInfo.D.RC = RC;
1497       RegInfo.Explicit = true;
1498       return false;
1499 
1500     case VRegInfo::GENERIC:
1501     case VRegInfo::REGBANK:
1502       return error(Loc, "register class specification on generic register");
1503     }
1504     llvm_unreachable("Unexpected register kind");
1505   }
1506 
1507   // Should be a register bank or a generic register.
1508   const RegisterBank *RegBank = nullptr;
1509   if (Name != "_") {
1510     RegBank = PFS.Target.getRegBank(Name);
1511     if (!RegBank)
1512       return error(Loc, "expected '_', register class, or register bank name");
1513   }
1514 
1515   lex();
1516 
1517   switch (RegInfo.Kind) {
1518   case VRegInfo::UNKNOWN:
1519   case VRegInfo::GENERIC:
1520   case VRegInfo::REGBANK:
1521     RegInfo.Kind = RegBank ? VRegInfo::REGBANK : VRegInfo::GENERIC;
1522     if (RegInfo.Explicit && RegInfo.D.RegBank != RegBank)
1523       return error(Loc, "conflicting generic register banks");
1524     RegInfo.D.RegBank = RegBank;
1525     RegInfo.Explicit = true;
1526     return false;
1527 
1528   case VRegInfo::NORMAL:
1529     return error(Loc, "register bank specification on normal register");
1530   }
1531   llvm_unreachable("Unexpected register kind");
1532 }
1533 
1534 bool MIParser::parseRegisterFlag(unsigned &Flags) {
1535   const unsigned OldFlags = Flags;
1536   switch (Token.kind()) {
1537   case MIToken::kw_implicit:
1538     Flags |= RegState::Implicit;
1539     break;
1540   case MIToken::kw_implicit_define:
1541     Flags |= RegState::ImplicitDefine;
1542     break;
1543   case MIToken::kw_def:
1544     Flags |= RegState::Define;
1545     break;
1546   case MIToken::kw_dead:
1547     Flags |= RegState::Dead;
1548     break;
1549   case MIToken::kw_killed:
1550     Flags |= RegState::Kill;
1551     break;
1552   case MIToken::kw_undef:
1553     Flags |= RegState::Undef;
1554     break;
1555   case MIToken::kw_internal:
1556     Flags |= RegState::InternalRead;
1557     break;
1558   case MIToken::kw_early_clobber:
1559     Flags |= RegState::EarlyClobber;
1560     break;
1561   case MIToken::kw_debug_use:
1562     Flags |= RegState::Debug;
1563     break;
1564   case MIToken::kw_renamable:
1565     Flags |= RegState::Renamable;
1566     break;
1567   default:
1568     llvm_unreachable("The current token should be a register flag");
1569   }
1570   if (OldFlags == Flags)
1571     // We know that the same flag is specified more than once when the flags
1572     // weren't modified.
1573     return error("duplicate '" + Token.stringValue() + "' register flag");
1574   lex();
1575   return false;
1576 }
1577 
1578 bool MIParser::parseSubRegisterIndex(unsigned &SubReg) {
1579   assert(Token.is(MIToken::dot));
1580   lex();
1581   if (Token.isNot(MIToken::Identifier))
1582     return error("expected a subregister index after '.'");
1583   auto Name = Token.stringValue();
1584   SubReg = PFS.Target.getSubRegIndex(Name);
1585   if (!SubReg)
1586     return error(Twine("use of unknown subregister index '") + Name + "'");
1587   lex();
1588   return false;
1589 }
1590 
1591 bool MIParser::parseRegisterTiedDefIndex(unsigned &TiedDefIdx) {
1592   if (!consumeIfPresent(MIToken::kw_tied_def))
1593     return true;
1594   if (Token.isNot(MIToken::IntegerLiteral))
1595     return error("expected an integer literal after 'tied-def'");
1596   if (getUnsigned(TiedDefIdx))
1597     return true;
1598   lex();
1599   if (expectAndConsume(MIToken::rparen))
1600     return true;
1601   return false;
1602 }
1603 
1604 bool MIParser::assignRegisterTies(MachineInstr &MI,
1605                                   ArrayRef<ParsedMachineOperand> Operands) {
1606   SmallVector<std::pair<unsigned, unsigned>, 4> TiedRegisterPairs;
1607   for (unsigned I = 0, E = Operands.size(); I != E; ++I) {
1608     if (!Operands[I].TiedDefIdx)
1609       continue;
1610     // The parser ensures that this operand is a register use, so we just have
1611     // to check the tied-def operand.
1612     unsigned DefIdx = Operands[I].TiedDefIdx.getValue();
1613     if (DefIdx >= E)
1614       return error(Operands[I].Begin,
1615                    Twine("use of invalid tied-def operand index '" +
1616                          Twine(DefIdx) + "'; instruction has only ") +
1617                        Twine(E) + " operands");
1618     const auto &DefOperand = Operands[DefIdx].Operand;
1619     if (!DefOperand.isReg() || !DefOperand.isDef())
1620       // FIXME: add note with the def operand.
1621       return error(Operands[I].Begin,
1622                    Twine("use of invalid tied-def operand index '") +
1623                        Twine(DefIdx) + "'; the operand #" + Twine(DefIdx) +
1624                        " isn't a defined register");
1625     // Check that the tied-def operand wasn't tied elsewhere.
1626     for (const auto &TiedPair : TiedRegisterPairs) {
1627       if (TiedPair.first == DefIdx)
1628         return error(Operands[I].Begin,
1629                      Twine("the tied-def operand #") + Twine(DefIdx) +
1630                          " is already tied with another register operand");
1631     }
1632     TiedRegisterPairs.push_back(std::make_pair(DefIdx, I));
1633   }
1634   // FIXME: Verify that for non INLINEASM instructions, the def and use tied
1635   // indices must be less than tied max.
1636   for (const auto &TiedPair : TiedRegisterPairs)
1637     MI.tieOperands(TiedPair.first, TiedPair.second);
1638   return false;
1639 }
1640 
1641 bool MIParser::parseRegisterOperand(MachineOperand &Dest,
1642                                     Optional<unsigned> &TiedDefIdx,
1643                                     bool IsDef) {
1644   unsigned Flags = IsDef ? RegState::Define : 0;
1645   while (Token.isRegisterFlag()) {
1646     if (parseRegisterFlag(Flags))
1647       return true;
1648   }
1649   if (!Token.isRegister())
1650     return error("expected a register after register flags");
1651   Register Reg;
1652   VRegInfo *RegInfo;
1653   if (parseRegister(Reg, RegInfo))
1654     return true;
1655   lex();
1656   unsigned SubReg = 0;
1657   if (Token.is(MIToken::dot)) {
1658     if (parseSubRegisterIndex(SubReg))
1659       return true;
1660     if (!Register::isVirtualRegister(Reg))
1661       return error("subregister index expects a virtual register");
1662   }
1663   if (Token.is(MIToken::colon)) {
1664     if (!Register::isVirtualRegister(Reg))
1665       return error("register class specification expects a virtual register");
1666     lex();
1667     if (parseRegisterClassOrBank(*RegInfo))
1668         return true;
1669   }
1670   MachineRegisterInfo &MRI = MF.getRegInfo();
1671   if ((Flags & RegState::Define) == 0) {
1672     if (consumeIfPresent(MIToken::lparen)) {
1673       unsigned Idx;
1674       if (!parseRegisterTiedDefIndex(Idx))
1675         TiedDefIdx = Idx;
1676       else {
1677         // Try a redundant low-level type.
1678         LLT Ty;
1679         if (parseLowLevelType(Token.location(), Ty))
1680           return error("expected tied-def or low-level type after '('");
1681 
1682         if (expectAndConsume(MIToken::rparen))
1683           return true;
1684 
1685         if (MRI.getType(Reg).isValid() && MRI.getType(Reg) != Ty)
1686           return error("inconsistent type for generic virtual register");
1687 
1688         MRI.setRegClassOrRegBank(Reg, static_cast<RegisterBank *>(nullptr));
1689         MRI.setType(Reg, Ty);
1690       }
1691     }
1692   } else if (consumeIfPresent(MIToken::lparen)) {
1693     // Virtual registers may have a tpe with GlobalISel.
1694     if (!Register::isVirtualRegister(Reg))
1695       return error("unexpected type on physical register");
1696 
1697     LLT Ty;
1698     if (parseLowLevelType(Token.location(), Ty))
1699       return true;
1700 
1701     if (expectAndConsume(MIToken::rparen))
1702       return true;
1703 
1704     if (MRI.getType(Reg).isValid() && MRI.getType(Reg) != Ty)
1705       return error("inconsistent type for generic virtual register");
1706 
1707     MRI.setRegClassOrRegBank(Reg, static_cast<RegisterBank *>(nullptr));
1708     MRI.setType(Reg, Ty);
1709   } else if (Register::isVirtualRegister(Reg)) {
1710     // Generic virtual registers must have a type.
1711     // If we end up here this means the type hasn't been specified and
1712     // this is bad!
1713     if (RegInfo->Kind == VRegInfo::GENERIC ||
1714         RegInfo->Kind == VRegInfo::REGBANK)
1715       return error("generic virtual registers must have a type");
1716   }
1717   Dest = MachineOperand::CreateReg(
1718       Reg, Flags & RegState::Define, Flags & RegState::Implicit,
1719       Flags & RegState::Kill, Flags & RegState::Dead, Flags & RegState::Undef,
1720       Flags & RegState::EarlyClobber, SubReg, Flags & RegState::Debug,
1721       Flags & RegState::InternalRead, Flags & RegState::Renamable);
1722 
1723   return false;
1724 }
1725 
1726 bool MIParser::parseImmediateOperand(MachineOperand &Dest) {
1727   assert(Token.is(MIToken::IntegerLiteral));
1728   const APSInt &Int = Token.integerValue();
1729   if (Int.getMinSignedBits() > 64)
1730     return error("integer literal is too large to be an immediate operand");
1731   Dest = MachineOperand::CreateImm(Int.getExtValue());
1732   lex();
1733   return false;
1734 }
1735 
1736 bool MIParser::parseTargetImmMnemonic(const unsigned OpCode,
1737                                       const unsigned OpIdx,
1738                                       MachineOperand &Dest,
1739                                       const MIRFormatter &MF) {
1740   assert(Token.is(MIToken::dot));
1741   auto Loc = Token.location(); // record start position
1742   size_t Len = 1;              // for "."
1743   lex();
1744 
1745   // Handle the case that mnemonic starts with number.
1746   if (Token.is(MIToken::IntegerLiteral)) {
1747     Len += Token.range().size();
1748     lex();
1749   }
1750 
1751   StringRef Src;
1752   if (Token.is(MIToken::comma))
1753     Src = StringRef(Loc, Len);
1754   else {
1755     assert(Token.is(MIToken::Identifier));
1756     Src = StringRef(Loc, Len + Token.stringValue().size());
1757   }
1758   int64_t Val;
1759   if (MF.parseImmMnemonic(OpCode, OpIdx, Src, Val,
1760                           [this](StringRef::iterator Loc, const Twine &Msg)
1761                               -> bool { return error(Loc, Msg); }))
1762     return true;
1763 
1764   Dest = MachineOperand::CreateImm(Val);
1765   if (!Token.is(MIToken::comma))
1766     lex();
1767   return false;
1768 }
1769 
1770 static bool parseIRConstant(StringRef::iterator Loc, StringRef StringValue,
1771                             PerFunctionMIParsingState &PFS, const Constant *&C,
1772                             ErrorCallbackType ErrCB) {
1773   auto Source = StringValue.str(); // The source has to be null terminated.
1774   SMDiagnostic Err;
1775   C = parseConstantValue(Source, Err, *PFS.MF.getFunction().getParent(),
1776                          &PFS.IRSlots);
1777   if (!C)
1778     return ErrCB(Loc + Err.getColumnNo(), Err.getMessage());
1779   return false;
1780 }
1781 
1782 bool MIParser::parseIRConstant(StringRef::iterator Loc, StringRef StringValue,
1783                                const Constant *&C) {
1784   return ::parseIRConstant(
1785       Loc, StringValue, PFS, C,
1786       [this](StringRef::iterator Loc, const Twine &Msg) -> bool {
1787         return error(Loc, Msg);
1788       });
1789 }
1790 
1791 bool MIParser::parseIRConstant(StringRef::iterator Loc, const Constant *&C) {
1792   if (parseIRConstant(Loc, StringRef(Loc, Token.range().end() - Loc), C))
1793     return true;
1794   lex();
1795   return false;
1796 }
1797 
1798 // See LLT implemntation for bit size limits.
1799 static bool verifyScalarSize(uint64_t Size) {
1800   return Size != 0 && isUInt<16>(Size);
1801 }
1802 
1803 static bool verifyVectorElementCount(uint64_t NumElts) {
1804   return NumElts != 0 && isUInt<16>(NumElts);
1805 }
1806 
1807 static bool verifyAddrSpace(uint64_t AddrSpace) {
1808   return isUInt<24>(AddrSpace);
1809 }
1810 
1811 bool MIParser::parseLowLevelType(StringRef::iterator Loc, LLT &Ty) {
1812   if (Token.range().front() == 's' || Token.range().front() == 'p') {
1813     StringRef SizeStr = Token.range().drop_front();
1814     if (SizeStr.size() == 0 || !llvm::all_of(SizeStr, isdigit))
1815       return error("expected integers after 's'/'p' type character");
1816   }
1817 
1818   if (Token.range().front() == 's') {
1819     auto ScalarSize = APSInt(Token.range().drop_front()).getZExtValue();
1820     if (!verifyScalarSize(ScalarSize))
1821       return error("invalid size for scalar type");
1822 
1823     Ty = LLT::scalar(ScalarSize);
1824     lex();
1825     return false;
1826   } else if (Token.range().front() == 'p') {
1827     const DataLayout &DL = MF.getDataLayout();
1828     uint64_t AS = APSInt(Token.range().drop_front()).getZExtValue();
1829     if (!verifyAddrSpace(AS))
1830       return error("invalid address space number");
1831 
1832     Ty = LLT::pointer(AS, DL.getPointerSizeInBits(AS));
1833     lex();
1834     return false;
1835   }
1836 
1837   // Now we're looking for a vector.
1838   if (Token.isNot(MIToken::less))
1839     return error(Loc,
1840                  "expected sN, pA, <M x sN>, or <M x pA> for GlobalISel type");
1841   lex();
1842 
1843   if (Token.isNot(MIToken::IntegerLiteral))
1844     return error(Loc, "expected <M x sN> or <M x pA> for vector type");
1845   uint64_t NumElements = Token.integerValue().getZExtValue();
1846   if (!verifyVectorElementCount(NumElements))
1847     return error("invalid number of vector elements");
1848 
1849   lex();
1850 
1851   if (Token.isNot(MIToken::Identifier) || Token.stringValue() != "x")
1852     return error(Loc, "expected <M x sN> or <M x pA> for vector type");
1853   lex();
1854 
1855   if (Token.range().front() != 's' && Token.range().front() != 'p')
1856     return error(Loc, "expected <M x sN> or <M x pA> for vector type");
1857   StringRef SizeStr = Token.range().drop_front();
1858   if (SizeStr.size() == 0 || !llvm::all_of(SizeStr, isdigit))
1859     return error("expected integers after 's'/'p' type character");
1860 
1861   if (Token.range().front() == 's') {
1862     auto ScalarSize = APSInt(Token.range().drop_front()).getZExtValue();
1863     if (!verifyScalarSize(ScalarSize))
1864       return error("invalid size for scalar type");
1865     Ty = LLT::scalar(ScalarSize);
1866   } else if (Token.range().front() == 'p') {
1867     const DataLayout &DL = MF.getDataLayout();
1868     uint64_t AS = APSInt(Token.range().drop_front()).getZExtValue();
1869     if (!verifyAddrSpace(AS))
1870       return error("invalid address space number");
1871 
1872     Ty = LLT::pointer(AS, DL.getPointerSizeInBits(AS));
1873   } else
1874     return error(Loc, "expected <M x sN> or <M x pA> for vector type");
1875   lex();
1876 
1877   if (Token.isNot(MIToken::greater))
1878     return error(Loc, "expected <M x sN> or <M x pA> for vector type");
1879   lex();
1880 
1881   Ty = LLT::fixed_vector(NumElements, Ty);
1882   return false;
1883 }
1884 
1885 bool MIParser::parseTypedImmediateOperand(MachineOperand &Dest) {
1886   assert(Token.is(MIToken::Identifier));
1887   StringRef TypeStr = Token.range();
1888   if (TypeStr.front() != 'i' && TypeStr.front() != 's' &&
1889       TypeStr.front() != 'p')
1890     return error(
1891         "a typed immediate operand should start with one of 'i', 's', or 'p'");
1892   StringRef SizeStr = Token.range().drop_front();
1893   if (SizeStr.size() == 0 || !llvm::all_of(SizeStr, isdigit))
1894     return error("expected integers after 'i'/'s'/'p' type character");
1895 
1896   auto Loc = Token.location();
1897   lex();
1898   if (Token.isNot(MIToken::IntegerLiteral)) {
1899     if (Token.isNot(MIToken::Identifier) ||
1900         !(Token.range() == "true" || Token.range() == "false"))
1901       return error("expected an integer literal");
1902   }
1903   const Constant *C = nullptr;
1904   if (parseIRConstant(Loc, C))
1905     return true;
1906   Dest = MachineOperand::CreateCImm(cast<ConstantInt>(C));
1907   return false;
1908 }
1909 
1910 bool MIParser::parseFPImmediateOperand(MachineOperand &Dest) {
1911   auto Loc = Token.location();
1912   lex();
1913   if (Token.isNot(MIToken::FloatingPointLiteral) &&
1914       Token.isNot(MIToken::HexLiteral))
1915     return error("expected a floating point literal");
1916   const Constant *C = nullptr;
1917   if (parseIRConstant(Loc, C))
1918     return true;
1919   Dest = MachineOperand::CreateFPImm(cast<ConstantFP>(C));
1920   return false;
1921 }
1922 
1923 static bool getHexUint(const MIToken &Token, APInt &Result) {
1924   assert(Token.is(MIToken::HexLiteral));
1925   StringRef S = Token.range();
1926   assert(S[0] == '0' && tolower(S[1]) == 'x');
1927   // This could be a floating point literal with a special prefix.
1928   if (!isxdigit(S[2]))
1929     return true;
1930   StringRef V = S.substr(2);
1931   APInt A(V.size()*4, V, 16);
1932 
1933   // If A is 0, then A.getActiveBits() is 0. This isn't a valid bitwidth. Make
1934   // sure it isn't the case before constructing result.
1935   unsigned NumBits = (A == 0) ? 32 : A.getActiveBits();
1936   Result = APInt(NumBits, ArrayRef<uint64_t>(A.getRawData(), A.getNumWords()));
1937   return false;
1938 }
1939 
1940 static bool getUnsigned(const MIToken &Token, unsigned &Result,
1941                         ErrorCallbackType ErrCB) {
1942   if (Token.hasIntegerValue()) {
1943     const uint64_t Limit = uint64_t(std::numeric_limits<unsigned>::max()) + 1;
1944     uint64_t Val64 = Token.integerValue().getLimitedValue(Limit);
1945     if (Val64 == Limit)
1946       return ErrCB(Token.location(), "expected 32-bit integer (too large)");
1947     Result = Val64;
1948     return false;
1949   }
1950   if (Token.is(MIToken::HexLiteral)) {
1951     APInt A;
1952     if (getHexUint(Token, A))
1953       return true;
1954     if (A.getBitWidth() > 32)
1955       return ErrCB(Token.location(), "expected 32-bit integer (too large)");
1956     Result = A.getZExtValue();
1957     return false;
1958   }
1959   return true;
1960 }
1961 
1962 bool MIParser::getUnsigned(unsigned &Result) {
1963   return ::getUnsigned(
1964       Token, Result, [this](StringRef::iterator Loc, const Twine &Msg) -> bool {
1965         return error(Loc, Msg);
1966       });
1967 }
1968 
1969 bool MIParser::parseMBBReference(MachineBasicBlock *&MBB) {
1970   assert(Token.is(MIToken::MachineBasicBlock) ||
1971          Token.is(MIToken::MachineBasicBlockLabel));
1972   unsigned Number;
1973   if (getUnsigned(Number))
1974     return true;
1975   auto MBBInfo = PFS.MBBSlots.find(Number);
1976   if (MBBInfo == PFS.MBBSlots.end())
1977     return error(Twine("use of undefined machine basic block #") +
1978                  Twine(Number));
1979   MBB = MBBInfo->second;
1980   // TODO: Only parse the name if it's a MachineBasicBlockLabel. Deprecate once
1981   // we drop the <irname> from the bb.<id>.<irname> format.
1982   if (!Token.stringValue().empty() && Token.stringValue() != MBB->getName())
1983     return error(Twine("the name of machine basic block #") + Twine(Number) +
1984                  " isn't '" + Token.stringValue() + "'");
1985   return false;
1986 }
1987 
1988 bool MIParser::parseMBBOperand(MachineOperand &Dest) {
1989   MachineBasicBlock *MBB;
1990   if (parseMBBReference(MBB))
1991     return true;
1992   Dest = MachineOperand::CreateMBB(MBB);
1993   lex();
1994   return false;
1995 }
1996 
1997 bool MIParser::parseStackFrameIndex(int &FI) {
1998   assert(Token.is(MIToken::StackObject));
1999   unsigned ID;
2000   if (getUnsigned(ID))
2001     return true;
2002   auto ObjectInfo = PFS.StackObjectSlots.find(ID);
2003   if (ObjectInfo == PFS.StackObjectSlots.end())
2004     return error(Twine("use of undefined stack object '%stack.") + Twine(ID) +
2005                  "'");
2006   StringRef Name;
2007   if (const auto *Alloca =
2008           MF.getFrameInfo().getObjectAllocation(ObjectInfo->second))
2009     Name = Alloca->getName();
2010   if (!Token.stringValue().empty() && Token.stringValue() != Name)
2011     return error(Twine("the name of the stack object '%stack.") + Twine(ID) +
2012                  "' isn't '" + Token.stringValue() + "'");
2013   lex();
2014   FI = ObjectInfo->second;
2015   return false;
2016 }
2017 
2018 bool MIParser::parseStackObjectOperand(MachineOperand &Dest) {
2019   int FI;
2020   if (parseStackFrameIndex(FI))
2021     return true;
2022   Dest = MachineOperand::CreateFI(FI);
2023   return false;
2024 }
2025 
2026 bool MIParser::parseFixedStackFrameIndex(int &FI) {
2027   assert(Token.is(MIToken::FixedStackObject));
2028   unsigned ID;
2029   if (getUnsigned(ID))
2030     return true;
2031   auto ObjectInfo = PFS.FixedStackObjectSlots.find(ID);
2032   if (ObjectInfo == PFS.FixedStackObjectSlots.end())
2033     return error(Twine("use of undefined fixed stack object '%fixed-stack.") +
2034                  Twine(ID) + "'");
2035   lex();
2036   FI = ObjectInfo->second;
2037   return false;
2038 }
2039 
2040 bool MIParser::parseFixedStackObjectOperand(MachineOperand &Dest) {
2041   int FI;
2042   if (parseFixedStackFrameIndex(FI))
2043     return true;
2044   Dest = MachineOperand::CreateFI(FI);
2045   return false;
2046 }
2047 
2048 static bool parseGlobalValue(const MIToken &Token,
2049                              PerFunctionMIParsingState &PFS, GlobalValue *&GV,
2050                              ErrorCallbackType ErrCB) {
2051   switch (Token.kind()) {
2052   case MIToken::NamedGlobalValue: {
2053     const Module *M = PFS.MF.getFunction().getParent();
2054     GV = M->getNamedValue(Token.stringValue());
2055     if (!GV)
2056       return ErrCB(Token.location(), Twine("use of undefined global value '") +
2057                                          Token.range() + "'");
2058     break;
2059   }
2060   case MIToken::GlobalValue: {
2061     unsigned GVIdx;
2062     if (getUnsigned(Token, GVIdx, ErrCB))
2063       return true;
2064     if (GVIdx >= PFS.IRSlots.GlobalValues.size())
2065       return ErrCB(Token.location(), Twine("use of undefined global value '@") +
2066                                          Twine(GVIdx) + "'");
2067     GV = PFS.IRSlots.GlobalValues[GVIdx];
2068     break;
2069   }
2070   default:
2071     llvm_unreachable("The current token should be a global value");
2072   }
2073   return false;
2074 }
2075 
2076 bool MIParser::parseGlobalValue(GlobalValue *&GV) {
2077   return ::parseGlobalValue(
2078       Token, PFS, GV,
2079       [this](StringRef::iterator Loc, const Twine &Msg) -> bool {
2080         return error(Loc, Msg);
2081       });
2082 }
2083 
2084 bool MIParser::parseGlobalAddressOperand(MachineOperand &Dest) {
2085   GlobalValue *GV = nullptr;
2086   if (parseGlobalValue(GV))
2087     return true;
2088   lex();
2089   Dest = MachineOperand::CreateGA(GV, /*Offset=*/0);
2090   if (parseOperandsOffset(Dest))
2091     return true;
2092   return false;
2093 }
2094 
2095 bool MIParser::parseConstantPoolIndexOperand(MachineOperand &Dest) {
2096   assert(Token.is(MIToken::ConstantPoolItem));
2097   unsigned ID;
2098   if (getUnsigned(ID))
2099     return true;
2100   auto ConstantInfo = PFS.ConstantPoolSlots.find(ID);
2101   if (ConstantInfo == PFS.ConstantPoolSlots.end())
2102     return error("use of undefined constant '%const." + Twine(ID) + "'");
2103   lex();
2104   Dest = MachineOperand::CreateCPI(ID, /*Offset=*/0);
2105   if (parseOperandsOffset(Dest))
2106     return true;
2107   return false;
2108 }
2109 
2110 bool MIParser::parseJumpTableIndexOperand(MachineOperand &Dest) {
2111   assert(Token.is(MIToken::JumpTableIndex));
2112   unsigned ID;
2113   if (getUnsigned(ID))
2114     return true;
2115   auto JumpTableEntryInfo = PFS.JumpTableSlots.find(ID);
2116   if (JumpTableEntryInfo == PFS.JumpTableSlots.end())
2117     return error("use of undefined jump table '%jump-table." + Twine(ID) + "'");
2118   lex();
2119   Dest = MachineOperand::CreateJTI(JumpTableEntryInfo->second);
2120   return false;
2121 }
2122 
2123 bool MIParser::parseExternalSymbolOperand(MachineOperand &Dest) {
2124   assert(Token.is(MIToken::ExternalSymbol));
2125   const char *Symbol = MF.createExternalSymbolName(Token.stringValue());
2126   lex();
2127   Dest = MachineOperand::CreateES(Symbol);
2128   if (parseOperandsOffset(Dest))
2129     return true;
2130   return false;
2131 }
2132 
2133 bool MIParser::parseMCSymbolOperand(MachineOperand &Dest) {
2134   assert(Token.is(MIToken::MCSymbol));
2135   MCSymbol *Symbol = getOrCreateMCSymbol(Token.stringValue());
2136   lex();
2137   Dest = MachineOperand::CreateMCSymbol(Symbol);
2138   if (parseOperandsOffset(Dest))
2139     return true;
2140   return false;
2141 }
2142 
2143 bool MIParser::parseSubRegisterIndexOperand(MachineOperand &Dest) {
2144   assert(Token.is(MIToken::SubRegisterIndex));
2145   StringRef Name = Token.stringValue();
2146   unsigned SubRegIndex = PFS.Target.getSubRegIndex(Token.stringValue());
2147   if (SubRegIndex == 0)
2148     return error(Twine("unknown subregister index '") + Name + "'");
2149   lex();
2150   Dest = MachineOperand::CreateImm(SubRegIndex);
2151   return false;
2152 }
2153 
2154 bool MIParser::parseMDNode(MDNode *&Node) {
2155   assert(Token.is(MIToken::exclaim));
2156 
2157   auto Loc = Token.location();
2158   lex();
2159   if (Token.isNot(MIToken::IntegerLiteral) || Token.integerValue().isSigned())
2160     return error("expected metadata id after '!'");
2161   unsigned ID;
2162   if (getUnsigned(ID))
2163     return true;
2164   auto NodeInfo = PFS.IRSlots.MetadataNodes.find(ID);
2165   if (NodeInfo == PFS.IRSlots.MetadataNodes.end()) {
2166     NodeInfo = PFS.MachineMetadataNodes.find(ID);
2167     if (NodeInfo == PFS.MachineMetadataNodes.end())
2168       return error(Loc, "use of undefined metadata '!" + Twine(ID) + "'");
2169   }
2170   lex();
2171   Node = NodeInfo->second.get();
2172   return false;
2173 }
2174 
2175 bool MIParser::parseDIExpression(MDNode *&Expr) {
2176   assert(Token.is(MIToken::md_diexpr));
2177   lex();
2178 
2179   // FIXME: Share this parsing with the IL parser.
2180   SmallVector<uint64_t, 8> Elements;
2181 
2182   if (expectAndConsume(MIToken::lparen))
2183     return true;
2184 
2185   if (Token.isNot(MIToken::rparen)) {
2186     do {
2187       if (Token.is(MIToken::Identifier)) {
2188         if (unsigned Op = dwarf::getOperationEncoding(Token.stringValue())) {
2189           lex();
2190           Elements.push_back(Op);
2191           continue;
2192         }
2193         if (unsigned Enc = dwarf::getAttributeEncoding(Token.stringValue())) {
2194           lex();
2195           Elements.push_back(Enc);
2196           continue;
2197         }
2198         return error(Twine("invalid DWARF op '") + Token.stringValue() + "'");
2199       }
2200 
2201       if (Token.isNot(MIToken::IntegerLiteral) ||
2202           Token.integerValue().isSigned())
2203         return error("expected unsigned integer");
2204 
2205       auto &U = Token.integerValue();
2206       if (U.ugt(UINT64_MAX))
2207         return error("element too large, limit is " + Twine(UINT64_MAX));
2208       Elements.push_back(U.getZExtValue());
2209       lex();
2210 
2211     } while (consumeIfPresent(MIToken::comma));
2212   }
2213 
2214   if (expectAndConsume(MIToken::rparen))
2215     return true;
2216 
2217   Expr = DIExpression::get(MF.getFunction().getContext(), Elements);
2218   return false;
2219 }
2220 
2221 bool MIParser::parseDILocation(MDNode *&Loc) {
2222   assert(Token.is(MIToken::md_dilocation));
2223   lex();
2224 
2225   bool HaveLine = false;
2226   unsigned Line = 0;
2227   unsigned Column = 0;
2228   MDNode *Scope = nullptr;
2229   MDNode *InlinedAt = nullptr;
2230   bool ImplicitCode = false;
2231 
2232   if (expectAndConsume(MIToken::lparen))
2233     return true;
2234 
2235   if (Token.isNot(MIToken::rparen)) {
2236     do {
2237       if (Token.is(MIToken::Identifier)) {
2238         if (Token.stringValue() == "line") {
2239           lex();
2240           if (expectAndConsume(MIToken::colon))
2241             return true;
2242           if (Token.isNot(MIToken::IntegerLiteral) ||
2243               Token.integerValue().isSigned())
2244             return error("expected unsigned integer");
2245           Line = Token.integerValue().getZExtValue();
2246           HaveLine = true;
2247           lex();
2248           continue;
2249         }
2250         if (Token.stringValue() == "column") {
2251           lex();
2252           if (expectAndConsume(MIToken::colon))
2253             return true;
2254           if (Token.isNot(MIToken::IntegerLiteral) ||
2255               Token.integerValue().isSigned())
2256             return error("expected unsigned integer");
2257           Column = Token.integerValue().getZExtValue();
2258           lex();
2259           continue;
2260         }
2261         if (Token.stringValue() == "scope") {
2262           lex();
2263           if (expectAndConsume(MIToken::colon))
2264             return true;
2265           if (parseMDNode(Scope))
2266             return error("expected metadata node");
2267           if (!isa<DIScope>(Scope))
2268             return error("expected DIScope node");
2269           continue;
2270         }
2271         if (Token.stringValue() == "inlinedAt") {
2272           lex();
2273           if (expectAndConsume(MIToken::colon))
2274             return true;
2275           if (Token.is(MIToken::exclaim)) {
2276             if (parseMDNode(InlinedAt))
2277               return true;
2278           } else if (Token.is(MIToken::md_dilocation)) {
2279             if (parseDILocation(InlinedAt))
2280               return true;
2281           } else
2282             return error("expected metadata node");
2283           if (!isa<DILocation>(InlinedAt))
2284             return error("expected DILocation node");
2285           continue;
2286         }
2287         if (Token.stringValue() == "isImplicitCode") {
2288           lex();
2289           if (expectAndConsume(MIToken::colon))
2290             return true;
2291           if (!Token.is(MIToken::Identifier))
2292             return error("expected true/false");
2293           // As far as I can see, we don't have any existing need for parsing
2294           // true/false in MIR yet. Do it ad-hoc until there's something else
2295           // that needs it.
2296           if (Token.stringValue() == "true")
2297             ImplicitCode = true;
2298           else if (Token.stringValue() == "false")
2299             ImplicitCode = false;
2300           else
2301             return error("expected true/false");
2302           lex();
2303           continue;
2304         }
2305       }
2306       return error(Twine("invalid DILocation argument '") +
2307                    Token.stringValue() + "'");
2308     } while (consumeIfPresent(MIToken::comma));
2309   }
2310 
2311   if (expectAndConsume(MIToken::rparen))
2312     return true;
2313 
2314   if (!HaveLine)
2315     return error("DILocation requires line number");
2316   if (!Scope)
2317     return error("DILocation requires a scope");
2318 
2319   Loc = DILocation::get(MF.getFunction().getContext(), Line, Column, Scope,
2320                         InlinedAt, ImplicitCode);
2321   return false;
2322 }
2323 
2324 bool MIParser::parseMetadataOperand(MachineOperand &Dest) {
2325   MDNode *Node = nullptr;
2326   if (Token.is(MIToken::exclaim)) {
2327     if (parseMDNode(Node))
2328       return true;
2329   } else if (Token.is(MIToken::md_diexpr)) {
2330     if (parseDIExpression(Node))
2331       return true;
2332   }
2333   Dest = MachineOperand::CreateMetadata(Node);
2334   return false;
2335 }
2336 
2337 bool MIParser::parseCFIOffset(int &Offset) {
2338   if (Token.isNot(MIToken::IntegerLiteral))
2339     return error("expected a cfi offset");
2340   if (Token.integerValue().getMinSignedBits() > 32)
2341     return error("expected a 32 bit integer (the cfi offset is too large)");
2342   Offset = (int)Token.integerValue().getExtValue();
2343   lex();
2344   return false;
2345 }
2346 
2347 bool MIParser::parseCFIRegister(Register &Reg) {
2348   if (Token.isNot(MIToken::NamedRegister))
2349     return error("expected a cfi register");
2350   Register LLVMReg;
2351   if (parseNamedRegister(LLVMReg))
2352     return true;
2353   const auto *TRI = MF.getSubtarget().getRegisterInfo();
2354   assert(TRI && "Expected target register info");
2355   int DwarfReg = TRI->getDwarfRegNum(LLVMReg, true);
2356   if (DwarfReg < 0)
2357     return error("invalid DWARF register");
2358   Reg = (unsigned)DwarfReg;
2359   lex();
2360   return false;
2361 }
2362 
2363 bool MIParser::parseCFIAddressSpace(unsigned &AddressSpace) {
2364   if (Token.isNot(MIToken::IntegerLiteral))
2365     return error("expected a cfi address space literal");
2366   if (Token.integerValue().isSigned())
2367     return error("expected an unsigned integer (cfi address space)");
2368   AddressSpace = Token.integerValue().getZExtValue();
2369   lex();
2370   return false;
2371 }
2372 
2373 bool MIParser::parseCFIEscapeValues(std::string &Values) {
2374   do {
2375     if (Token.isNot(MIToken::HexLiteral))
2376       return error("expected a hexadecimal literal");
2377     unsigned Value;
2378     if (getUnsigned(Value))
2379       return true;
2380     if (Value > UINT8_MAX)
2381       return error("expected a 8-bit integer (too large)");
2382     Values.push_back(static_cast<uint8_t>(Value));
2383     lex();
2384   } while (consumeIfPresent(MIToken::comma));
2385   return false;
2386 }
2387 
2388 bool MIParser::parseCFIOperand(MachineOperand &Dest) {
2389   auto Kind = Token.kind();
2390   lex();
2391   int Offset;
2392   Register Reg;
2393   unsigned AddressSpace;
2394   unsigned CFIIndex;
2395   switch (Kind) {
2396   case MIToken::kw_cfi_same_value:
2397     if (parseCFIRegister(Reg))
2398       return true;
2399     CFIIndex = MF.addFrameInst(MCCFIInstruction::createSameValue(nullptr, Reg));
2400     break;
2401   case MIToken::kw_cfi_offset:
2402     if (parseCFIRegister(Reg) || expectAndConsume(MIToken::comma) ||
2403         parseCFIOffset(Offset))
2404       return true;
2405     CFIIndex =
2406         MF.addFrameInst(MCCFIInstruction::createOffset(nullptr, Reg, Offset));
2407     break;
2408   case MIToken::kw_cfi_rel_offset:
2409     if (parseCFIRegister(Reg) || expectAndConsume(MIToken::comma) ||
2410         parseCFIOffset(Offset))
2411       return true;
2412     CFIIndex = MF.addFrameInst(
2413         MCCFIInstruction::createRelOffset(nullptr, Reg, Offset));
2414     break;
2415   case MIToken::kw_cfi_def_cfa_register:
2416     if (parseCFIRegister(Reg))
2417       return true;
2418     CFIIndex =
2419         MF.addFrameInst(MCCFIInstruction::createDefCfaRegister(nullptr, Reg));
2420     break;
2421   case MIToken::kw_cfi_def_cfa_offset:
2422     if (parseCFIOffset(Offset))
2423       return true;
2424     CFIIndex =
2425         MF.addFrameInst(MCCFIInstruction::cfiDefCfaOffset(nullptr, Offset));
2426     break;
2427   case MIToken::kw_cfi_adjust_cfa_offset:
2428     if (parseCFIOffset(Offset))
2429       return true;
2430     CFIIndex = MF.addFrameInst(
2431         MCCFIInstruction::createAdjustCfaOffset(nullptr, Offset));
2432     break;
2433   case MIToken::kw_cfi_def_cfa:
2434     if (parseCFIRegister(Reg) || expectAndConsume(MIToken::comma) ||
2435         parseCFIOffset(Offset))
2436       return true;
2437     CFIIndex =
2438         MF.addFrameInst(MCCFIInstruction::cfiDefCfa(nullptr, Reg, Offset));
2439     break;
2440   case MIToken::kw_cfi_llvm_def_aspace_cfa:
2441     if (parseCFIRegister(Reg) || expectAndConsume(MIToken::comma) ||
2442         parseCFIOffset(Offset) || expectAndConsume(MIToken::comma) ||
2443         parseCFIAddressSpace(AddressSpace))
2444       return true;
2445     CFIIndex = MF.addFrameInst(MCCFIInstruction::createLLVMDefAspaceCfa(
2446         nullptr, Reg, Offset, AddressSpace));
2447     break;
2448   case MIToken::kw_cfi_remember_state:
2449     CFIIndex = MF.addFrameInst(MCCFIInstruction::createRememberState(nullptr));
2450     break;
2451   case MIToken::kw_cfi_restore:
2452     if (parseCFIRegister(Reg))
2453       return true;
2454     CFIIndex = MF.addFrameInst(MCCFIInstruction::createRestore(nullptr, Reg));
2455     break;
2456   case MIToken::kw_cfi_restore_state:
2457     CFIIndex = MF.addFrameInst(MCCFIInstruction::createRestoreState(nullptr));
2458     break;
2459   case MIToken::kw_cfi_undefined:
2460     if (parseCFIRegister(Reg))
2461       return true;
2462     CFIIndex = MF.addFrameInst(MCCFIInstruction::createUndefined(nullptr, Reg));
2463     break;
2464   case MIToken::kw_cfi_register: {
2465     Register Reg2;
2466     if (parseCFIRegister(Reg) || expectAndConsume(MIToken::comma) ||
2467         parseCFIRegister(Reg2))
2468       return true;
2469 
2470     CFIIndex =
2471         MF.addFrameInst(MCCFIInstruction::createRegister(nullptr, Reg, Reg2));
2472     break;
2473   }
2474   case MIToken::kw_cfi_window_save:
2475     CFIIndex = MF.addFrameInst(MCCFIInstruction::createWindowSave(nullptr));
2476     break;
2477   case MIToken::kw_cfi_aarch64_negate_ra_sign_state:
2478     CFIIndex = MF.addFrameInst(MCCFIInstruction::createNegateRAState(nullptr));
2479     break;
2480   case MIToken::kw_cfi_escape: {
2481     std::string Values;
2482     if (parseCFIEscapeValues(Values))
2483       return true;
2484     CFIIndex = MF.addFrameInst(MCCFIInstruction::createEscape(nullptr, Values));
2485     break;
2486   }
2487   default:
2488     // TODO: Parse the other CFI operands.
2489     llvm_unreachable("The current token should be a cfi operand");
2490   }
2491   Dest = MachineOperand::CreateCFIIndex(CFIIndex);
2492   return false;
2493 }
2494 
2495 bool MIParser::parseIRBlock(BasicBlock *&BB, const Function &F) {
2496   switch (Token.kind()) {
2497   case MIToken::NamedIRBlock: {
2498     BB = dyn_cast_or_null<BasicBlock>(
2499         F.getValueSymbolTable()->lookup(Token.stringValue()));
2500     if (!BB)
2501       return error(Twine("use of undefined IR block '") + Token.range() + "'");
2502     break;
2503   }
2504   case MIToken::IRBlock: {
2505     unsigned SlotNumber = 0;
2506     if (getUnsigned(SlotNumber))
2507       return true;
2508     BB = const_cast<BasicBlock *>(getIRBlock(SlotNumber, F));
2509     if (!BB)
2510       return error(Twine("use of undefined IR block '%ir-block.") +
2511                    Twine(SlotNumber) + "'");
2512     break;
2513   }
2514   default:
2515     llvm_unreachable("The current token should be an IR block reference");
2516   }
2517   return false;
2518 }
2519 
2520 bool MIParser::parseBlockAddressOperand(MachineOperand &Dest) {
2521   assert(Token.is(MIToken::kw_blockaddress));
2522   lex();
2523   if (expectAndConsume(MIToken::lparen))
2524     return true;
2525   if (Token.isNot(MIToken::GlobalValue) &&
2526       Token.isNot(MIToken::NamedGlobalValue))
2527     return error("expected a global value");
2528   GlobalValue *GV = nullptr;
2529   if (parseGlobalValue(GV))
2530     return true;
2531   auto *F = dyn_cast<Function>(GV);
2532   if (!F)
2533     return error("expected an IR function reference");
2534   lex();
2535   if (expectAndConsume(MIToken::comma))
2536     return true;
2537   BasicBlock *BB = nullptr;
2538   if (Token.isNot(MIToken::IRBlock) && Token.isNot(MIToken::NamedIRBlock))
2539     return error("expected an IR block reference");
2540   if (parseIRBlock(BB, *F))
2541     return true;
2542   lex();
2543   if (expectAndConsume(MIToken::rparen))
2544     return true;
2545   Dest = MachineOperand::CreateBA(BlockAddress::get(F, BB), /*Offset=*/0);
2546   if (parseOperandsOffset(Dest))
2547     return true;
2548   return false;
2549 }
2550 
2551 bool MIParser::parseIntrinsicOperand(MachineOperand &Dest) {
2552   assert(Token.is(MIToken::kw_intrinsic));
2553   lex();
2554   if (expectAndConsume(MIToken::lparen))
2555     return error("expected syntax intrinsic(@llvm.whatever)");
2556 
2557   if (Token.isNot(MIToken::NamedGlobalValue))
2558     return error("expected syntax intrinsic(@llvm.whatever)");
2559 
2560   std::string Name = std::string(Token.stringValue());
2561   lex();
2562 
2563   if (expectAndConsume(MIToken::rparen))
2564     return error("expected ')' to terminate intrinsic name");
2565 
2566   // Find out what intrinsic we're dealing with, first try the global namespace
2567   // and then the target's private intrinsics if that fails.
2568   const TargetIntrinsicInfo *TII = MF.getTarget().getIntrinsicInfo();
2569   Intrinsic::ID ID = Function::lookupIntrinsicID(Name);
2570   if (ID == Intrinsic::not_intrinsic && TII)
2571     ID = static_cast<Intrinsic::ID>(TII->lookupName(Name));
2572 
2573   if (ID == Intrinsic::not_intrinsic)
2574     return error("unknown intrinsic name");
2575   Dest = MachineOperand::CreateIntrinsicID(ID);
2576 
2577   return false;
2578 }
2579 
2580 bool MIParser::parsePredicateOperand(MachineOperand &Dest) {
2581   assert(Token.is(MIToken::kw_intpred) || Token.is(MIToken::kw_floatpred));
2582   bool IsFloat = Token.is(MIToken::kw_floatpred);
2583   lex();
2584 
2585   if (expectAndConsume(MIToken::lparen))
2586     return error("expected syntax intpred(whatever) or floatpred(whatever");
2587 
2588   if (Token.isNot(MIToken::Identifier))
2589     return error("whatever");
2590 
2591   CmpInst::Predicate Pred;
2592   if (IsFloat) {
2593     Pred = StringSwitch<CmpInst::Predicate>(Token.stringValue())
2594                .Case("false", CmpInst::FCMP_FALSE)
2595                .Case("oeq", CmpInst::FCMP_OEQ)
2596                .Case("ogt", CmpInst::FCMP_OGT)
2597                .Case("oge", CmpInst::FCMP_OGE)
2598                .Case("olt", CmpInst::FCMP_OLT)
2599                .Case("ole", CmpInst::FCMP_OLE)
2600                .Case("one", CmpInst::FCMP_ONE)
2601                .Case("ord", CmpInst::FCMP_ORD)
2602                .Case("uno", CmpInst::FCMP_UNO)
2603                .Case("ueq", CmpInst::FCMP_UEQ)
2604                .Case("ugt", CmpInst::FCMP_UGT)
2605                .Case("uge", CmpInst::FCMP_UGE)
2606                .Case("ult", CmpInst::FCMP_ULT)
2607                .Case("ule", CmpInst::FCMP_ULE)
2608                .Case("une", CmpInst::FCMP_UNE)
2609                .Case("true", CmpInst::FCMP_TRUE)
2610                .Default(CmpInst::BAD_FCMP_PREDICATE);
2611     if (!CmpInst::isFPPredicate(Pred))
2612       return error("invalid floating-point predicate");
2613   } else {
2614     Pred = StringSwitch<CmpInst::Predicate>(Token.stringValue())
2615                .Case("eq", CmpInst::ICMP_EQ)
2616                .Case("ne", CmpInst::ICMP_NE)
2617                .Case("sgt", CmpInst::ICMP_SGT)
2618                .Case("sge", CmpInst::ICMP_SGE)
2619                .Case("slt", CmpInst::ICMP_SLT)
2620                .Case("sle", CmpInst::ICMP_SLE)
2621                .Case("ugt", CmpInst::ICMP_UGT)
2622                .Case("uge", CmpInst::ICMP_UGE)
2623                .Case("ult", CmpInst::ICMP_ULT)
2624                .Case("ule", CmpInst::ICMP_ULE)
2625                .Default(CmpInst::BAD_ICMP_PREDICATE);
2626     if (!CmpInst::isIntPredicate(Pred))
2627       return error("invalid integer predicate");
2628   }
2629 
2630   lex();
2631   Dest = MachineOperand::CreatePredicate(Pred);
2632   if (expectAndConsume(MIToken::rparen))
2633     return error("predicate should be terminated by ')'.");
2634 
2635   return false;
2636 }
2637 
2638 bool MIParser::parseShuffleMaskOperand(MachineOperand &Dest) {
2639   assert(Token.is(MIToken::kw_shufflemask));
2640 
2641   lex();
2642   if (expectAndConsume(MIToken::lparen))
2643     return error("expected syntax shufflemask(<integer or undef>, ...)");
2644 
2645   SmallVector<int, 32> ShufMask;
2646   do {
2647     if (Token.is(MIToken::kw_undef)) {
2648       ShufMask.push_back(-1);
2649     } else if (Token.is(MIToken::IntegerLiteral)) {
2650       const APSInt &Int = Token.integerValue();
2651       ShufMask.push_back(Int.getExtValue());
2652     } else
2653       return error("expected integer constant");
2654 
2655     lex();
2656   } while (consumeIfPresent(MIToken::comma));
2657 
2658   if (expectAndConsume(MIToken::rparen))
2659     return error("shufflemask should be terminated by ')'.");
2660 
2661   ArrayRef<int> MaskAlloc = MF.allocateShuffleMask(ShufMask);
2662   Dest = MachineOperand::CreateShuffleMask(MaskAlloc);
2663   return false;
2664 }
2665 
2666 bool MIParser::parseTargetIndexOperand(MachineOperand &Dest) {
2667   assert(Token.is(MIToken::kw_target_index));
2668   lex();
2669   if (expectAndConsume(MIToken::lparen))
2670     return true;
2671   if (Token.isNot(MIToken::Identifier))
2672     return error("expected the name of the target index");
2673   int Index = 0;
2674   if (PFS.Target.getTargetIndex(Token.stringValue(), Index))
2675     return error("use of undefined target index '" + Token.stringValue() + "'");
2676   lex();
2677   if (expectAndConsume(MIToken::rparen))
2678     return true;
2679   Dest = MachineOperand::CreateTargetIndex(unsigned(Index), /*Offset=*/0);
2680   if (parseOperandsOffset(Dest))
2681     return true;
2682   return false;
2683 }
2684 
2685 bool MIParser::parseCustomRegisterMaskOperand(MachineOperand &Dest) {
2686   assert(Token.stringValue() == "CustomRegMask" && "Expected a custom RegMask");
2687   lex();
2688   if (expectAndConsume(MIToken::lparen))
2689     return true;
2690 
2691   uint32_t *Mask = MF.allocateRegMask();
2692   while (true) {
2693     if (Token.isNot(MIToken::NamedRegister))
2694       return error("expected a named register");
2695     Register Reg;
2696     if (parseNamedRegister(Reg))
2697       return true;
2698     lex();
2699     Mask[Reg / 32] |= 1U << (Reg % 32);
2700     // TODO: Report an error if the same register is used more than once.
2701     if (Token.isNot(MIToken::comma))
2702       break;
2703     lex();
2704   }
2705 
2706   if (expectAndConsume(MIToken::rparen))
2707     return true;
2708   Dest = MachineOperand::CreateRegMask(Mask);
2709   return false;
2710 }
2711 
2712 bool MIParser::parseLiveoutRegisterMaskOperand(MachineOperand &Dest) {
2713   assert(Token.is(MIToken::kw_liveout));
2714   uint32_t *Mask = MF.allocateRegMask();
2715   lex();
2716   if (expectAndConsume(MIToken::lparen))
2717     return true;
2718   while (true) {
2719     if (Token.isNot(MIToken::NamedRegister))
2720       return error("expected a named register");
2721     Register Reg;
2722     if (parseNamedRegister(Reg))
2723       return true;
2724     lex();
2725     Mask[Reg / 32] |= 1U << (Reg % 32);
2726     // TODO: Report an error if the same register is used more than once.
2727     if (Token.isNot(MIToken::comma))
2728       break;
2729     lex();
2730   }
2731   if (expectAndConsume(MIToken::rparen))
2732     return true;
2733   Dest = MachineOperand::CreateRegLiveOut(Mask);
2734   return false;
2735 }
2736 
2737 bool MIParser::parseMachineOperand(const unsigned OpCode, const unsigned OpIdx,
2738                                    MachineOperand &Dest,
2739                                    Optional<unsigned> &TiedDefIdx) {
2740   switch (Token.kind()) {
2741   case MIToken::kw_implicit:
2742   case MIToken::kw_implicit_define:
2743   case MIToken::kw_def:
2744   case MIToken::kw_dead:
2745   case MIToken::kw_killed:
2746   case MIToken::kw_undef:
2747   case MIToken::kw_internal:
2748   case MIToken::kw_early_clobber:
2749   case MIToken::kw_debug_use:
2750   case MIToken::kw_renamable:
2751   case MIToken::underscore:
2752   case MIToken::NamedRegister:
2753   case MIToken::VirtualRegister:
2754   case MIToken::NamedVirtualRegister:
2755     return parseRegisterOperand(Dest, TiedDefIdx);
2756   case MIToken::IntegerLiteral:
2757     return parseImmediateOperand(Dest);
2758   case MIToken::kw_half:
2759   case MIToken::kw_float:
2760   case MIToken::kw_double:
2761   case MIToken::kw_x86_fp80:
2762   case MIToken::kw_fp128:
2763   case MIToken::kw_ppc_fp128:
2764     return parseFPImmediateOperand(Dest);
2765   case MIToken::MachineBasicBlock:
2766     return parseMBBOperand(Dest);
2767   case MIToken::StackObject:
2768     return parseStackObjectOperand(Dest);
2769   case MIToken::FixedStackObject:
2770     return parseFixedStackObjectOperand(Dest);
2771   case MIToken::GlobalValue:
2772   case MIToken::NamedGlobalValue:
2773     return parseGlobalAddressOperand(Dest);
2774   case MIToken::ConstantPoolItem:
2775     return parseConstantPoolIndexOperand(Dest);
2776   case MIToken::JumpTableIndex:
2777     return parseJumpTableIndexOperand(Dest);
2778   case MIToken::ExternalSymbol:
2779     return parseExternalSymbolOperand(Dest);
2780   case MIToken::MCSymbol:
2781     return parseMCSymbolOperand(Dest);
2782   case MIToken::SubRegisterIndex:
2783     return parseSubRegisterIndexOperand(Dest);
2784   case MIToken::md_diexpr:
2785   case MIToken::exclaim:
2786     return parseMetadataOperand(Dest);
2787   case MIToken::kw_cfi_same_value:
2788   case MIToken::kw_cfi_offset:
2789   case MIToken::kw_cfi_rel_offset:
2790   case MIToken::kw_cfi_def_cfa_register:
2791   case MIToken::kw_cfi_def_cfa_offset:
2792   case MIToken::kw_cfi_adjust_cfa_offset:
2793   case MIToken::kw_cfi_escape:
2794   case MIToken::kw_cfi_def_cfa:
2795   case MIToken::kw_cfi_llvm_def_aspace_cfa:
2796   case MIToken::kw_cfi_register:
2797   case MIToken::kw_cfi_remember_state:
2798   case MIToken::kw_cfi_restore:
2799   case MIToken::kw_cfi_restore_state:
2800   case MIToken::kw_cfi_undefined:
2801   case MIToken::kw_cfi_window_save:
2802   case MIToken::kw_cfi_aarch64_negate_ra_sign_state:
2803     return parseCFIOperand(Dest);
2804   case MIToken::kw_blockaddress:
2805     return parseBlockAddressOperand(Dest);
2806   case MIToken::kw_intrinsic:
2807     return parseIntrinsicOperand(Dest);
2808   case MIToken::kw_target_index:
2809     return parseTargetIndexOperand(Dest);
2810   case MIToken::kw_liveout:
2811     return parseLiveoutRegisterMaskOperand(Dest);
2812   case MIToken::kw_floatpred:
2813   case MIToken::kw_intpred:
2814     return parsePredicateOperand(Dest);
2815   case MIToken::kw_shufflemask:
2816     return parseShuffleMaskOperand(Dest);
2817   case MIToken::Error:
2818     return true;
2819   case MIToken::Identifier:
2820     if (const auto *RegMask = PFS.Target.getRegMask(Token.stringValue())) {
2821       Dest = MachineOperand::CreateRegMask(RegMask);
2822       lex();
2823       break;
2824     } else if (Token.stringValue() == "CustomRegMask") {
2825       return parseCustomRegisterMaskOperand(Dest);
2826     } else
2827       return parseTypedImmediateOperand(Dest);
2828   case MIToken::dot: {
2829     const auto *TII = MF.getSubtarget().getInstrInfo();
2830     if (const auto *Formatter = TII->getMIRFormatter()) {
2831       return parseTargetImmMnemonic(OpCode, OpIdx, Dest, *Formatter);
2832     }
2833     LLVM_FALLTHROUGH;
2834   }
2835   default:
2836     // FIXME: Parse the MCSymbol machine operand.
2837     return error("expected a machine operand");
2838   }
2839   return false;
2840 }
2841 
2842 bool MIParser::parseMachineOperandAndTargetFlags(
2843     const unsigned OpCode, const unsigned OpIdx, MachineOperand &Dest,
2844     Optional<unsigned> &TiedDefIdx) {
2845   unsigned TF = 0;
2846   bool HasTargetFlags = false;
2847   if (Token.is(MIToken::kw_target_flags)) {
2848     HasTargetFlags = true;
2849     lex();
2850     if (expectAndConsume(MIToken::lparen))
2851       return true;
2852     if (Token.isNot(MIToken::Identifier))
2853       return error("expected the name of the target flag");
2854     if (PFS.Target.getDirectTargetFlag(Token.stringValue(), TF)) {
2855       if (PFS.Target.getBitmaskTargetFlag(Token.stringValue(), TF))
2856         return error("use of undefined target flag '" + Token.stringValue() +
2857                      "'");
2858     }
2859     lex();
2860     while (Token.is(MIToken::comma)) {
2861       lex();
2862       if (Token.isNot(MIToken::Identifier))
2863         return error("expected the name of the target flag");
2864       unsigned BitFlag = 0;
2865       if (PFS.Target.getBitmaskTargetFlag(Token.stringValue(), BitFlag))
2866         return error("use of undefined target flag '" + Token.stringValue() +
2867                      "'");
2868       // TODO: Report an error when using a duplicate bit target flag.
2869       TF |= BitFlag;
2870       lex();
2871     }
2872     if (expectAndConsume(MIToken::rparen))
2873       return true;
2874   }
2875   auto Loc = Token.location();
2876   if (parseMachineOperand(OpCode, OpIdx, Dest, TiedDefIdx))
2877     return true;
2878   if (!HasTargetFlags)
2879     return false;
2880   if (Dest.isReg())
2881     return error(Loc, "register operands can't have target flags");
2882   Dest.setTargetFlags(TF);
2883   return false;
2884 }
2885 
2886 bool MIParser::parseOffset(int64_t &Offset) {
2887   if (Token.isNot(MIToken::plus) && Token.isNot(MIToken::minus))
2888     return false;
2889   StringRef Sign = Token.range();
2890   bool IsNegative = Token.is(MIToken::minus);
2891   lex();
2892   if (Token.isNot(MIToken::IntegerLiteral))
2893     return error("expected an integer literal after '" + Sign + "'");
2894   if (Token.integerValue().getMinSignedBits() > 64)
2895     return error("expected 64-bit integer (too large)");
2896   Offset = Token.integerValue().getExtValue();
2897   if (IsNegative)
2898     Offset = -Offset;
2899   lex();
2900   return false;
2901 }
2902 
2903 bool MIParser::parseAlignment(uint64_t &Alignment) {
2904   assert(Token.is(MIToken::kw_align) || Token.is(MIToken::kw_basealign));
2905   lex();
2906   if (Token.isNot(MIToken::IntegerLiteral) || Token.integerValue().isSigned())
2907     return error("expected an integer literal after 'align'");
2908   if (getUint64(Alignment))
2909     return true;
2910   lex();
2911 
2912   if (!isPowerOf2_64(Alignment))
2913     return error("expected a power-of-2 literal after 'align'");
2914 
2915   return false;
2916 }
2917 
2918 bool MIParser::parseAddrspace(unsigned &Addrspace) {
2919   assert(Token.is(MIToken::kw_addrspace));
2920   lex();
2921   if (Token.isNot(MIToken::IntegerLiteral) || Token.integerValue().isSigned())
2922     return error("expected an integer literal after 'addrspace'");
2923   if (getUnsigned(Addrspace))
2924     return true;
2925   lex();
2926   return false;
2927 }
2928 
2929 bool MIParser::parseOperandsOffset(MachineOperand &Op) {
2930   int64_t Offset = 0;
2931   if (parseOffset(Offset))
2932     return true;
2933   Op.setOffset(Offset);
2934   return false;
2935 }
2936 
2937 static bool parseIRValue(const MIToken &Token, PerFunctionMIParsingState &PFS,
2938                          const Value *&V, ErrorCallbackType ErrCB) {
2939   switch (Token.kind()) {
2940   case MIToken::NamedIRValue: {
2941     V = PFS.MF.getFunction().getValueSymbolTable()->lookup(Token.stringValue());
2942     break;
2943   }
2944   case MIToken::IRValue: {
2945     unsigned SlotNumber = 0;
2946     if (getUnsigned(Token, SlotNumber, ErrCB))
2947       return true;
2948     V = PFS.getIRValue(SlotNumber);
2949     break;
2950   }
2951   case MIToken::NamedGlobalValue:
2952   case MIToken::GlobalValue: {
2953     GlobalValue *GV = nullptr;
2954     if (parseGlobalValue(Token, PFS, GV, ErrCB))
2955       return true;
2956     V = GV;
2957     break;
2958   }
2959   case MIToken::QuotedIRValue: {
2960     const Constant *C = nullptr;
2961     if (parseIRConstant(Token.location(), Token.stringValue(), PFS, C, ErrCB))
2962       return true;
2963     V = C;
2964     break;
2965   }
2966   case MIToken::kw_unknown_address:
2967     V = nullptr;
2968     return false;
2969   default:
2970     llvm_unreachable("The current token should be an IR block reference");
2971   }
2972   if (!V)
2973     return ErrCB(Token.location(), Twine("use of undefined IR value '") + Token.range() + "'");
2974   return false;
2975 }
2976 
2977 bool MIParser::parseIRValue(const Value *&V) {
2978   return ::parseIRValue(
2979       Token, PFS, V, [this](StringRef::iterator Loc, const Twine &Msg) -> bool {
2980         return error(Loc, Msg);
2981       });
2982 }
2983 
2984 bool MIParser::getUint64(uint64_t &Result) {
2985   if (Token.hasIntegerValue()) {
2986     if (Token.integerValue().getActiveBits() > 64)
2987       return error("expected 64-bit integer (too large)");
2988     Result = Token.integerValue().getZExtValue();
2989     return false;
2990   }
2991   if (Token.is(MIToken::HexLiteral)) {
2992     APInt A;
2993     if (getHexUint(A))
2994       return true;
2995     if (A.getBitWidth() > 64)
2996       return error("expected 64-bit integer (too large)");
2997     Result = A.getZExtValue();
2998     return false;
2999   }
3000   return true;
3001 }
3002 
3003 bool MIParser::getHexUint(APInt &Result) {
3004   return ::getHexUint(Token, Result);
3005 }
3006 
3007 bool MIParser::parseMemoryOperandFlag(MachineMemOperand::Flags &Flags) {
3008   const auto OldFlags = Flags;
3009   switch (Token.kind()) {
3010   case MIToken::kw_volatile:
3011     Flags |= MachineMemOperand::MOVolatile;
3012     break;
3013   case MIToken::kw_non_temporal:
3014     Flags |= MachineMemOperand::MONonTemporal;
3015     break;
3016   case MIToken::kw_dereferenceable:
3017     Flags |= MachineMemOperand::MODereferenceable;
3018     break;
3019   case MIToken::kw_invariant:
3020     Flags |= MachineMemOperand::MOInvariant;
3021     break;
3022   case MIToken::StringConstant: {
3023     MachineMemOperand::Flags TF;
3024     if (PFS.Target.getMMOTargetFlag(Token.stringValue(), TF))
3025       return error("use of undefined target MMO flag '" + Token.stringValue() +
3026                    "'");
3027     Flags |= TF;
3028     break;
3029   }
3030   default:
3031     llvm_unreachable("The current token should be a memory operand flag");
3032   }
3033   if (OldFlags == Flags)
3034     // We know that the same flag is specified more than once when the flags
3035     // weren't modified.
3036     return error("duplicate '" + Token.stringValue() + "' memory operand flag");
3037   lex();
3038   return false;
3039 }
3040 
3041 bool MIParser::parseMemoryPseudoSourceValue(const PseudoSourceValue *&PSV) {
3042   switch (Token.kind()) {
3043   case MIToken::kw_stack:
3044     PSV = MF.getPSVManager().getStack();
3045     break;
3046   case MIToken::kw_got:
3047     PSV = MF.getPSVManager().getGOT();
3048     break;
3049   case MIToken::kw_jump_table:
3050     PSV = MF.getPSVManager().getJumpTable();
3051     break;
3052   case MIToken::kw_constant_pool:
3053     PSV = MF.getPSVManager().getConstantPool();
3054     break;
3055   case MIToken::FixedStackObject: {
3056     int FI;
3057     if (parseFixedStackFrameIndex(FI))
3058       return true;
3059     PSV = MF.getPSVManager().getFixedStack(FI);
3060     // The token was already consumed, so use return here instead of break.
3061     return false;
3062   }
3063   case MIToken::StackObject: {
3064     int FI;
3065     if (parseStackFrameIndex(FI))
3066       return true;
3067     PSV = MF.getPSVManager().getFixedStack(FI);
3068     // The token was already consumed, so use return here instead of break.
3069     return false;
3070   }
3071   case MIToken::kw_call_entry:
3072     lex();
3073     switch (Token.kind()) {
3074     case MIToken::GlobalValue:
3075     case MIToken::NamedGlobalValue: {
3076       GlobalValue *GV = nullptr;
3077       if (parseGlobalValue(GV))
3078         return true;
3079       PSV = MF.getPSVManager().getGlobalValueCallEntry(GV);
3080       break;
3081     }
3082     case MIToken::ExternalSymbol:
3083       PSV = MF.getPSVManager().getExternalSymbolCallEntry(
3084           MF.createExternalSymbolName(Token.stringValue()));
3085       break;
3086     default:
3087       return error(
3088           "expected a global value or an external symbol after 'call-entry'");
3089     }
3090     break;
3091   case MIToken::kw_custom: {
3092     lex();
3093     const auto *TII = MF.getSubtarget().getInstrInfo();
3094     if (const auto *Formatter = TII->getMIRFormatter()) {
3095       if (Formatter->parseCustomPseudoSourceValue(
3096               Token.stringValue(), MF, PFS, PSV,
3097               [this](StringRef::iterator Loc, const Twine &Msg) -> bool {
3098                 return error(Loc, Msg);
3099               }))
3100         return true;
3101     } else
3102       return error("unable to parse target custom pseudo source value");
3103     break;
3104   }
3105   default:
3106     llvm_unreachable("The current token should be pseudo source value");
3107   }
3108   lex();
3109   return false;
3110 }
3111 
3112 bool MIParser::parseMachinePointerInfo(MachinePointerInfo &Dest) {
3113   if (Token.is(MIToken::kw_constant_pool) || Token.is(MIToken::kw_stack) ||
3114       Token.is(MIToken::kw_got) || Token.is(MIToken::kw_jump_table) ||
3115       Token.is(MIToken::FixedStackObject) || Token.is(MIToken::StackObject) ||
3116       Token.is(MIToken::kw_call_entry) || Token.is(MIToken::kw_custom)) {
3117     const PseudoSourceValue *PSV = nullptr;
3118     if (parseMemoryPseudoSourceValue(PSV))
3119       return true;
3120     int64_t Offset = 0;
3121     if (parseOffset(Offset))
3122       return true;
3123     Dest = MachinePointerInfo(PSV, Offset);
3124     return false;
3125   }
3126   if (Token.isNot(MIToken::NamedIRValue) && Token.isNot(MIToken::IRValue) &&
3127       Token.isNot(MIToken::GlobalValue) &&
3128       Token.isNot(MIToken::NamedGlobalValue) &&
3129       Token.isNot(MIToken::QuotedIRValue) &&
3130       Token.isNot(MIToken::kw_unknown_address))
3131     return error("expected an IR value reference");
3132   const Value *V = nullptr;
3133   if (parseIRValue(V))
3134     return true;
3135   if (V && !V->getType()->isPointerTy())
3136     return error("expected a pointer IR value");
3137   lex();
3138   int64_t Offset = 0;
3139   if (parseOffset(Offset))
3140     return true;
3141   Dest = MachinePointerInfo(V, Offset);
3142   return false;
3143 }
3144 
3145 bool MIParser::parseOptionalScope(LLVMContext &Context,
3146                                   SyncScope::ID &SSID) {
3147   SSID = SyncScope::System;
3148   if (Token.is(MIToken::Identifier) && Token.stringValue() == "syncscope") {
3149     lex();
3150     if (expectAndConsume(MIToken::lparen))
3151       return error("expected '(' in syncscope");
3152 
3153     std::string SSN;
3154     if (parseStringConstant(SSN))
3155       return true;
3156 
3157     SSID = Context.getOrInsertSyncScopeID(SSN);
3158     if (expectAndConsume(MIToken::rparen))
3159       return error("expected ')' in syncscope");
3160   }
3161 
3162   return false;
3163 }
3164 
3165 bool MIParser::parseOptionalAtomicOrdering(AtomicOrdering &Order) {
3166   Order = AtomicOrdering::NotAtomic;
3167   if (Token.isNot(MIToken::Identifier))
3168     return false;
3169 
3170   Order = StringSwitch<AtomicOrdering>(Token.stringValue())
3171               .Case("unordered", AtomicOrdering::Unordered)
3172               .Case("monotonic", AtomicOrdering::Monotonic)
3173               .Case("acquire", AtomicOrdering::Acquire)
3174               .Case("release", AtomicOrdering::Release)
3175               .Case("acq_rel", AtomicOrdering::AcquireRelease)
3176               .Case("seq_cst", AtomicOrdering::SequentiallyConsistent)
3177               .Default(AtomicOrdering::NotAtomic);
3178 
3179   if (Order != AtomicOrdering::NotAtomic) {
3180     lex();
3181     return false;
3182   }
3183 
3184   return error("expected an atomic scope, ordering or a size specification");
3185 }
3186 
3187 bool MIParser::parseMachineMemoryOperand(MachineMemOperand *&Dest) {
3188   if (expectAndConsume(MIToken::lparen))
3189     return true;
3190   MachineMemOperand::Flags Flags = MachineMemOperand::MONone;
3191   while (Token.isMemoryOperandFlag()) {
3192     if (parseMemoryOperandFlag(Flags))
3193       return true;
3194   }
3195   if (Token.isNot(MIToken::Identifier) ||
3196       (Token.stringValue() != "load" && Token.stringValue() != "store"))
3197     return error("expected 'load' or 'store' memory operation");
3198   if (Token.stringValue() == "load")
3199     Flags |= MachineMemOperand::MOLoad;
3200   else
3201     Flags |= MachineMemOperand::MOStore;
3202   lex();
3203 
3204   // Optional 'store' for operands that both load and store.
3205   if (Token.is(MIToken::Identifier) && Token.stringValue() == "store") {
3206     Flags |= MachineMemOperand::MOStore;
3207     lex();
3208   }
3209 
3210   // Optional synchronization scope.
3211   SyncScope::ID SSID;
3212   if (parseOptionalScope(MF.getFunction().getContext(), SSID))
3213     return true;
3214 
3215   // Up to two atomic orderings (cmpxchg provides guarantees on failure).
3216   AtomicOrdering Order, FailureOrder;
3217   if (parseOptionalAtomicOrdering(Order))
3218     return true;
3219 
3220   if (parseOptionalAtomicOrdering(FailureOrder))
3221     return true;
3222 
3223   LLT MemoryType;
3224   if (Token.isNot(MIToken::IntegerLiteral) &&
3225       Token.isNot(MIToken::kw_unknown_size) &&
3226       Token.isNot(MIToken::lparen))
3227     return error("expected memory LLT, the size integer literal or 'unknown-size' after "
3228                  "memory operation");
3229 
3230   uint64_t Size = MemoryLocation::UnknownSize;
3231   if (Token.is(MIToken::IntegerLiteral)) {
3232     if (getUint64(Size))
3233       return true;
3234 
3235     // Convert from bytes to bits for storage.
3236     MemoryType = LLT::scalar(8 * Size);
3237     lex();
3238   } else if (Token.is(MIToken::kw_unknown_size)) {
3239     Size = MemoryLocation::UnknownSize;
3240     lex();
3241   } else {
3242     if (expectAndConsume(MIToken::lparen))
3243       return true;
3244     if (parseLowLevelType(Token.location(), MemoryType))
3245       return true;
3246     if (expectAndConsume(MIToken::rparen))
3247       return true;
3248 
3249     Size = MemoryType.getSizeInBytes();
3250   }
3251 
3252   MachinePointerInfo Ptr = MachinePointerInfo();
3253   if (Token.is(MIToken::Identifier)) {
3254     const char *Word =
3255         ((Flags & MachineMemOperand::MOLoad) &&
3256          (Flags & MachineMemOperand::MOStore))
3257             ? "on"
3258             : Flags & MachineMemOperand::MOLoad ? "from" : "into";
3259     if (Token.stringValue() != Word)
3260       return error(Twine("expected '") + Word + "'");
3261     lex();
3262 
3263     if (parseMachinePointerInfo(Ptr))
3264       return true;
3265   }
3266   uint64_t BaseAlignment =
3267       (Size != MemoryLocation::UnknownSize ? PowerOf2Ceil(Size) : 1);
3268   AAMDNodes AAInfo;
3269   MDNode *Range = nullptr;
3270   while (consumeIfPresent(MIToken::comma)) {
3271     switch (Token.kind()) {
3272     case MIToken::kw_align:
3273       // align is printed if it is different than size.
3274       if (parseAlignment(BaseAlignment))
3275         return true;
3276       break;
3277     case MIToken::kw_basealign:
3278       // basealign is printed if it is different than align.
3279       if (parseAlignment(BaseAlignment))
3280         return true;
3281       break;
3282     case MIToken::kw_addrspace:
3283       if (parseAddrspace(Ptr.AddrSpace))
3284         return true;
3285       break;
3286     case MIToken::md_tbaa:
3287       lex();
3288       if (parseMDNode(AAInfo.TBAA))
3289         return true;
3290       break;
3291     case MIToken::md_alias_scope:
3292       lex();
3293       if (parseMDNode(AAInfo.Scope))
3294         return true;
3295       break;
3296     case MIToken::md_noalias:
3297       lex();
3298       if (parseMDNode(AAInfo.NoAlias))
3299         return true;
3300       break;
3301     case MIToken::md_range:
3302       lex();
3303       if (parseMDNode(Range))
3304         return true;
3305       break;
3306     // TODO: Report an error on duplicate metadata nodes.
3307     default:
3308       return error("expected 'align' or '!tbaa' or '!alias.scope' or "
3309                    "'!noalias' or '!range'");
3310     }
3311   }
3312   if (expectAndConsume(MIToken::rparen))
3313     return true;
3314   Dest = MF.getMachineMemOperand(Ptr, Flags, MemoryType, Align(BaseAlignment),
3315                                  AAInfo, Range, SSID, Order, FailureOrder);
3316   return false;
3317 }
3318 
3319 bool MIParser::parsePreOrPostInstrSymbol(MCSymbol *&Symbol) {
3320   assert((Token.is(MIToken::kw_pre_instr_symbol) ||
3321           Token.is(MIToken::kw_post_instr_symbol)) &&
3322          "Invalid token for a pre- post-instruction symbol!");
3323   lex();
3324   if (Token.isNot(MIToken::MCSymbol))
3325     return error("expected a symbol after 'pre-instr-symbol'");
3326   Symbol = getOrCreateMCSymbol(Token.stringValue());
3327   lex();
3328   if (Token.isNewlineOrEOF() || Token.is(MIToken::coloncolon) ||
3329       Token.is(MIToken::lbrace))
3330     return false;
3331   if (Token.isNot(MIToken::comma))
3332     return error("expected ',' before the next machine operand");
3333   lex();
3334   return false;
3335 }
3336 
3337 bool MIParser::parseHeapAllocMarker(MDNode *&Node) {
3338   assert(Token.is(MIToken::kw_heap_alloc_marker) &&
3339          "Invalid token for a heap alloc marker!");
3340   lex();
3341   parseMDNode(Node);
3342   if (!Node)
3343     return error("expected a MDNode after 'heap-alloc-marker'");
3344   if (Token.isNewlineOrEOF() || Token.is(MIToken::coloncolon) ||
3345       Token.is(MIToken::lbrace))
3346     return false;
3347   if (Token.isNot(MIToken::comma))
3348     return error("expected ',' before the next machine operand");
3349   lex();
3350   return false;
3351 }
3352 
3353 static void initSlots2BasicBlocks(
3354     const Function &F,
3355     DenseMap<unsigned, const BasicBlock *> &Slots2BasicBlocks) {
3356   ModuleSlotTracker MST(F.getParent(), /*ShouldInitializeAllMetadata=*/false);
3357   MST.incorporateFunction(F);
3358   for (auto &BB : F) {
3359     if (BB.hasName())
3360       continue;
3361     int Slot = MST.getLocalSlot(&BB);
3362     if (Slot == -1)
3363       continue;
3364     Slots2BasicBlocks.insert(std::make_pair(unsigned(Slot), &BB));
3365   }
3366 }
3367 
3368 static const BasicBlock *getIRBlockFromSlot(
3369     unsigned Slot,
3370     const DenseMap<unsigned, const BasicBlock *> &Slots2BasicBlocks) {
3371   return Slots2BasicBlocks.lookup(Slot);
3372 }
3373 
3374 const BasicBlock *MIParser::getIRBlock(unsigned Slot) {
3375   if (Slots2BasicBlocks.empty())
3376     initSlots2BasicBlocks(MF.getFunction(), Slots2BasicBlocks);
3377   return getIRBlockFromSlot(Slot, Slots2BasicBlocks);
3378 }
3379 
3380 const BasicBlock *MIParser::getIRBlock(unsigned Slot, const Function &F) {
3381   if (&F == &MF.getFunction())
3382     return getIRBlock(Slot);
3383   DenseMap<unsigned, const BasicBlock *> CustomSlots2BasicBlocks;
3384   initSlots2BasicBlocks(F, CustomSlots2BasicBlocks);
3385   return getIRBlockFromSlot(Slot, CustomSlots2BasicBlocks);
3386 }
3387 
3388 MCSymbol *MIParser::getOrCreateMCSymbol(StringRef Name) {
3389   // FIXME: Currently we can't recognize temporary or local symbols and call all
3390   // of the appropriate forms to create them. However, this handles basic cases
3391   // well as most of the special aspects are recognized by a prefix on their
3392   // name, and the input names should already be unique. For test cases, keeping
3393   // the symbol name out of the symbol table isn't terribly important.
3394   return MF.getContext().getOrCreateSymbol(Name);
3395 }
3396 
3397 bool MIParser::parseStringConstant(std::string &Result) {
3398   if (Token.isNot(MIToken::StringConstant))
3399     return error("expected string constant");
3400   Result = std::string(Token.stringValue());
3401   lex();
3402   return false;
3403 }
3404 
3405 bool llvm::parseMachineBasicBlockDefinitions(PerFunctionMIParsingState &PFS,
3406                                              StringRef Src,
3407                                              SMDiagnostic &Error) {
3408   return MIParser(PFS, Error, Src).parseBasicBlockDefinitions(PFS.MBBSlots);
3409 }
3410 
3411 bool llvm::parseMachineInstructions(PerFunctionMIParsingState &PFS,
3412                                     StringRef Src, SMDiagnostic &Error) {
3413   return MIParser(PFS, Error, Src).parseBasicBlocks();
3414 }
3415 
3416 bool llvm::parseMBBReference(PerFunctionMIParsingState &PFS,
3417                              MachineBasicBlock *&MBB, StringRef Src,
3418                              SMDiagnostic &Error) {
3419   return MIParser(PFS, Error, Src).parseStandaloneMBB(MBB);
3420 }
3421 
3422 bool llvm::parseRegisterReference(PerFunctionMIParsingState &PFS,
3423                                   Register &Reg, StringRef Src,
3424                                   SMDiagnostic &Error) {
3425   return MIParser(PFS, Error, Src).parseStandaloneRegister(Reg);
3426 }
3427 
3428 bool llvm::parseNamedRegisterReference(PerFunctionMIParsingState &PFS,
3429                                        Register &Reg, StringRef Src,
3430                                        SMDiagnostic &Error) {
3431   return MIParser(PFS, Error, Src).parseStandaloneNamedRegister(Reg);
3432 }
3433 
3434 bool llvm::parseVirtualRegisterReference(PerFunctionMIParsingState &PFS,
3435                                          VRegInfo *&Info, StringRef Src,
3436                                          SMDiagnostic &Error) {
3437   return MIParser(PFS, Error, Src).parseStandaloneVirtualRegister(Info);
3438 }
3439 
3440 bool llvm::parseStackObjectReference(PerFunctionMIParsingState &PFS,
3441                                      int &FI, StringRef Src,
3442                                      SMDiagnostic &Error) {
3443   return MIParser(PFS, Error, Src).parseStandaloneStackObject(FI);
3444 }
3445 
3446 bool llvm::parseMDNode(PerFunctionMIParsingState &PFS,
3447                        MDNode *&Node, StringRef Src, SMDiagnostic &Error) {
3448   return MIParser(PFS, Error, Src).parseStandaloneMDNode(Node);
3449 }
3450 
3451 bool llvm::parseMachineMetadata(PerFunctionMIParsingState &PFS, StringRef Src,
3452                                 SMRange SrcRange, SMDiagnostic &Error) {
3453   return MIParser(PFS, Error, Src, SrcRange).parseMachineMetadata();
3454 }
3455 
3456 bool MIRFormatter::parseIRValue(StringRef Src, MachineFunction &MF,
3457                                 PerFunctionMIParsingState &PFS, const Value *&V,
3458                                 ErrorCallbackType ErrorCallback) {
3459   MIToken Token;
3460   Src = lexMIToken(Src, Token, [&](StringRef::iterator Loc, const Twine &Msg) {
3461     ErrorCallback(Loc, Msg);
3462   });
3463   V = nullptr;
3464 
3465   return ::parseIRValue(Token, PFS, V, ErrorCallback);
3466 }
3467