1 //===- BitcodeReader.cpp - Internal BitcodeReader implementation ----------===//
2 //
3 //                     The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 
10 #include "llvm/Bitcode/ReaderWriter.h"
11 #include "llvm/ADT/STLExtras.h"
12 #include "llvm/ADT/SmallString.h"
13 #include "llvm/ADT/SmallVector.h"
14 #include "llvm/ADT/Triple.h"
15 #include "llvm/Bitcode/BitstreamReader.h"
16 #include "llvm/Bitcode/LLVMBitCodes.h"
17 #include "llvm/IR/AutoUpgrade.h"
18 #include "llvm/IR/Constants.h"
19 #include "llvm/IR/DebugInfo.h"
20 #include "llvm/IR/DebugInfoMetadata.h"
21 #include "llvm/IR/DerivedTypes.h"
22 #include "llvm/IR/DiagnosticPrinter.h"
23 #include "llvm/IR/GVMaterializer.h"
24 #include "llvm/IR/InlineAsm.h"
25 #include "llvm/IR/IntrinsicInst.h"
26 #include "llvm/IR/LLVMContext.h"
27 #include "llvm/IR/Module.h"
28 #include "llvm/IR/OperandTraits.h"
29 #include "llvm/IR/Operator.h"
30 #include "llvm/IR/FunctionInfo.h"
31 #include "llvm/IR/ValueHandle.h"
32 #include "llvm/Support/DataStream.h"
33 #include "llvm/Support/ManagedStatic.h"
34 #include "llvm/Support/MathExtras.h"
35 #include "llvm/Support/MemoryBuffer.h"
36 #include "llvm/Support/raw_ostream.h"
37 #include <deque>
38 using namespace llvm;
39 
40 namespace {
41 enum {
42   SWITCH_INST_MAGIC = 0x4B5 // May 2012 => 1205 => Hex
43 };
44 
45 class BitcodeReaderValueList {
46   std::vector<WeakVH> ValuePtrs;
47 
48   /// As we resolve forward-referenced constants, we add information about them
49   /// to this vector.  This allows us to resolve them in bulk instead of
50   /// resolving each reference at a time.  See the code in
51   /// ResolveConstantForwardRefs for more information about this.
52   ///
53   /// The key of this vector is the placeholder constant, the value is the slot
54   /// number that holds the resolved value.
55   typedef std::vector<std::pair<Constant*, unsigned> > ResolveConstantsTy;
56   ResolveConstantsTy ResolveConstants;
57   LLVMContext &Context;
58 public:
BitcodeReaderValueList(LLVMContext & C)59   BitcodeReaderValueList(LLVMContext &C) : Context(C) {}
~BitcodeReaderValueList()60   ~BitcodeReaderValueList() {
61     assert(ResolveConstants.empty() && "Constants not resolved?");
62   }
63 
64   // vector compatibility methods
size() const65   unsigned size() const { return ValuePtrs.size(); }
resize(unsigned N)66   void resize(unsigned N) { ValuePtrs.resize(N); }
push_back(Value * V)67   void push_back(Value *V) { ValuePtrs.emplace_back(V); }
68 
clear()69   void clear() {
70     assert(ResolveConstants.empty() && "Constants not resolved?");
71     ValuePtrs.clear();
72   }
73 
operator [](unsigned i) const74   Value *operator[](unsigned i) const {
75     assert(i < ValuePtrs.size());
76     return ValuePtrs[i];
77   }
78 
back() const79   Value *back() const { return ValuePtrs.back(); }
pop_back()80     void pop_back() { ValuePtrs.pop_back(); }
empty() const81   bool empty() const { return ValuePtrs.empty(); }
shrinkTo(unsigned N)82   void shrinkTo(unsigned N) {
83     assert(N <= size() && "Invalid shrinkTo request!");
84     ValuePtrs.resize(N);
85   }
86 
87   Constant *getConstantFwdRef(unsigned Idx, Type *Ty);
88   Value *getValueFwdRef(unsigned Idx, Type *Ty);
89 
90   void assignValue(Value *V, unsigned Idx);
91 
92   /// Once all constants are read, this method bulk resolves any forward
93   /// references.
94   void resolveConstantForwardRefs();
95 };
96 
97 class BitcodeReaderMDValueList {
98   unsigned NumFwdRefs;
99   bool AnyFwdRefs;
100   unsigned MinFwdRef;
101   unsigned MaxFwdRef;
102   std::vector<TrackingMDRef> MDValuePtrs;
103 
104   LLVMContext &Context;
105 public:
BitcodeReaderMDValueList(LLVMContext & C)106   BitcodeReaderMDValueList(LLVMContext &C)
107       : NumFwdRefs(0), AnyFwdRefs(false), Context(C) {}
108 
109   // vector compatibility methods
size() const110   unsigned size() const       { return MDValuePtrs.size(); }
resize(unsigned N)111   void resize(unsigned N)     { MDValuePtrs.resize(N); }
push_back(Metadata * MD)112   void push_back(Metadata *MD) { MDValuePtrs.emplace_back(MD); }
clear()113   void clear()                { MDValuePtrs.clear();  }
back() const114   Metadata *back() const      { return MDValuePtrs.back(); }
pop_back()115   void pop_back()             { MDValuePtrs.pop_back(); }
empty() const116   bool empty() const          { return MDValuePtrs.empty(); }
117 
operator [](unsigned i) const118   Metadata *operator[](unsigned i) const {
119     assert(i < MDValuePtrs.size());
120     return MDValuePtrs[i];
121   }
122 
shrinkTo(unsigned N)123   void shrinkTo(unsigned N) {
124     assert(N <= size() && "Invalid shrinkTo request!");
125     MDValuePtrs.resize(N);
126   }
127 
128   Metadata *getValueFwdRef(unsigned Idx);
129   void assignValue(Metadata *MD, unsigned Idx);
130   void tryToResolveCycles();
131 };
132 
133 class BitcodeReader : public GVMaterializer {
134   LLVMContext &Context;
135   Module *TheModule = nullptr;
136   std::unique_ptr<MemoryBuffer> Buffer;
137   std::unique_ptr<BitstreamReader> StreamFile;
138   BitstreamCursor Stream;
139   // Next offset to start scanning for lazy parsing of function bodies.
140   uint64_t NextUnreadBit = 0;
141   // Last function offset found in the VST.
142   uint64_t LastFunctionBlockBit = 0;
143   bool SeenValueSymbolTable = false;
144   uint64_t VSTOffset = 0;
145   // Contains an arbitrary and optional string identifying the bitcode producer
146   std::string ProducerIdentification;
147   // Number of module level metadata records specified by the
148   // MODULE_CODE_METADATA_VALUES record.
149   unsigned NumModuleMDs = 0;
150   // Support older bitcode without the MODULE_CODE_METADATA_VALUES record.
151   bool SeenModuleValuesRecord = false;
152 
153   std::vector<Type*> TypeList;
154   BitcodeReaderValueList ValueList;
155   BitcodeReaderMDValueList MDValueList;
156   std::vector<Comdat *> ComdatList;
157   SmallVector<Instruction *, 64> InstructionList;
158 
159   std::vector<std::pair<GlobalVariable*, unsigned> > GlobalInits;
160   std::vector<std::pair<GlobalAlias*, unsigned> > AliasInits;
161   std::vector<std::pair<Function*, unsigned> > FunctionPrefixes;
162   std::vector<std::pair<Function*, unsigned> > FunctionPrologues;
163   std::vector<std::pair<Function*, unsigned> > FunctionPersonalityFns;
164 
165   SmallVector<Instruction*, 64> InstsWithTBAATag;
166 
167   /// The set of attributes by index.  Index zero in the file is for null, and
168   /// is thus not represented here.  As such all indices are off by one.
169   std::vector<AttributeSet> MAttributes;
170 
171   /// The set of attribute groups.
172   std::map<unsigned, AttributeSet> MAttributeGroups;
173 
174   /// While parsing a function body, this is a list of the basic blocks for the
175   /// function.
176   std::vector<BasicBlock*> FunctionBBs;
177 
178   // When reading the module header, this list is populated with functions that
179   // have bodies later in the file.
180   std::vector<Function*> FunctionsWithBodies;
181 
182   // When intrinsic functions are encountered which require upgrading they are
183   // stored here with their replacement function.
184   typedef DenseMap<Function*, Function*> UpgradedIntrinsicMap;
185   UpgradedIntrinsicMap UpgradedIntrinsics;
186 
187   // Map the bitcode's custom MDKind ID to the Module's MDKind ID.
188   DenseMap<unsigned, unsigned> MDKindMap;
189 
190   // Several operations happen after the module header has been read, but
191   // before function bodies are processed. This keeps track of whether
192   // we've done this yet.
193   bool SeenFirstFunctionBody = false;
194 
195   /// When function bodies are initially scanned, this map contains info about
196   /// where to find deferred function body in the stream.
197   DenseMap<Function*, uint64_t> DeferredFunctionInfo;
198 
199   /// When Metadata block is initially scanned when parsing the module, we may
200   /// choose to defer parsing of the metadata. This vector contains info about
201   /// which Metadata blocks are deferred.
202   std::vector<uint64_t> DeferredMetadataInfo;
203 
204   /// These are basic blocks forward-referenced by block addresses.  They are
205   /// inserted lazily into functions when they're loaded.  The basic block ID is
206   /// its index into the vector.
207   DenseMap<Function *, std::vector<BasicBlock *>> BasicBlockFwdRefs;
208   std::deque<Function *> BasicBlockFwdRefQueue;
209 
210   /// Indicates that we are using a new encoding for instruction operands where
211   /// most operands in the current FUNCTION_BLOCK are encoded relative to the
212   /// instruction number, for a more compact encoding.  Some instruction
213   /// operands are not relative to the instruction ID: basic block numbers, and
214   /// types. Once the old style function blocks have been phased out, we would
215   /// not need this flag.
216   bool UseRelativeIDs = false;
217 
218   /// True if all functions will be materialized, negating the need to process
219   /// (e.g.) blockaddress forward references.
220   bool WillMaterializeAllForwardRefs = false;
221 
222   /// True if any Metadata block has been materialized.
223   bool IsMetadataMaterialized = false;
224 
225   bool StripDebugInfo = false;
226 
227   /// Functions that need to be matched with subprograms when upgrading old
228   /// metadata.
229   SmallDenseMap<Function *, DISubprogram *, 16> FunctionsWithSPs;
230 
231   std::vector<std::string> BundleTags;
232 
233 public:
234   std::error_code error(BitcodeError E, const Twine &Message);
235   std::error_code error(BitcodeError E);
236   std::error_code error(const Twine &Message);
237 
238   BitcodeReader(MemoryBuffer *Buffer, LLVMContext &Context);
239   BitcodeReader(LLVMContext &Context);
~BitcodeReader()240   ~BitcodeReader() override { freeState(); }
241 
242   std::error_code materializeForwardReferencedFunctions();
243 
244   void freeState();
245 
246   void releaseBuffer();
247 
248   std::error_code materialize(GlobalValue *GV) override;
249   std::error_code materializeModule() override;
250   std::vector<StructType *> getIdentifiedStructTypes() const override;
251 
252   /// \brief Main interface to parsing a bitcode buffer.
253   /// \returns true if an error occurred.
254   std::error_code parseBitcodeInto(std::unique_ptr<DataStreamer> Streamer,
255                                    Module *M,
256                                    bool ShouldLazyLoadMetadata = false);
257 
258   /// \brief Cheap mechanism to just extract module triple
259   /// \returns true if an error occurred.
260   ErrorOr<std::string> parseTriple();
261 
262   /// Cheap mechanism to just extract the identification block out of bitcode.
263   ErrorOr<std::string> parseIdentificationBlock();
264 
265   static uint64_t decodeSignRotatedValue(uint64_t V);
266 
267   /// Materialize any deferred Metadata block.
268   std::error_code materializeMetadata() override;
269 
270   void setStripDebugInfo() override;
271 
272   /// Save the mapping between the metadata values and the corresponding
273   /// value id that were recorded in the MDValueList during parsing. If
274   /// OnlyTempMD is true, then only record those entries that are still
275   /// temporary metadata. This interface is used when metadata linking is
276   /// performed as a postpass, such as during function importing.
277   void saveMDValueList(DenseMap<const Metadata *, unsigned> &MDValueToValIDMap,
278                        bool OnlyTempMD) override;
279 
280 private:
281   /// Parse the "IDENTIFICATION_BLOCK_ID" block, populate the
282   // ProducerIdentification data member, and do some basic enforcement on the
283   // "epoch" encoded in the bitcode.
284   std::error_code parseBitcodeVersion();
285 
286   std::vector<StructType *> IdentifiedStructTypes;
287   StructType *createIdentifiedStructType(LLVMContext &Context, StringRef Name);
288   StructType *createIdentifiedStructType(LLVMContext &Context);
289 
290   Type *getTypeByID(unsigned ID);
getFnValueByID(unsigned ID,Type * Ty)291   Value *getFnValueByID(unsigned ID, Type *Ty) {
292     if (Ty && Ty->isMetadataTy())
293       return MetadataAsValue::get(Ty->getContext(), getFnMetadataByID(ID));
294     return ValueList.getValueFwdRef(ID, Ty);
295   }
getFnMetadataByID(unsigned ID)296   Metadata *getFnMetadataByID(unsigned ID) {
297     return MDValueList.getValueFwdRef(ID);
298   }
getBasicBlock(unsigned ID) const299   BasicBlock *getBasicBlock(unsigned ID) const {
300     if (ID >= FunctionBBs.size()) return nullptr; // Invalid ID
301     return FunctionBBs[ID];
302   }
getAttributes(unsigned i) const303   AttributeSet getAttributes(unsigned i) const {
304     if (i-1 < MAttributes.size())
305       return MAttributes[i-1];
306     return AttributeSet();
307   }
308 
309   /// Read a value/type pair out of the specified record from slot 'Slot'.
310   /// Increment Slot past the number of slots used in the record. Return true on
311   /// failure.
getValueTypePair(SmallVectorImpl<uint64_t> & Record,unsigned & Slot,unsigned InstNum,Value * & ResVal)312   bool getValueTypePair(SmallVectorImpl<uint64_t> &Record, unsigned &Slot,
313                         unsigned InstNum, Value *&ResVal) {
314     if (Slot == Record.size()) return true;
315     unsigned ValNo = (unsigned)Record[Slot++];
316     // Adjust the ValNo, if it was encoded relative to the InstNum.
317     if (UseRelativeIDs)
318       ValNo = InstNum - ValNo;
319     if (ValNo < InstNum) {
320       // If this is not a forward reference, just return the value we already
321       // have.
322       ResVal = getFnValueByID(ValNo, nullptr);
323       return ResVal == nullptr;
324     }
325     if (Slot == Record.size())
326       return true;
327 
328     unsigned TypeNo = (unsigned)Record[Slot++];
329     ResVal = getFnValueByID(ValNo, getTypeByID(TypeNo));
330     return ResVal == nullptr;
331   }
332 
333   /// Read a value out of the specified record from slot 'Slot'. Increment Slot
334   /// past the number of slots used by the value in the record. Return true if
335   /// there is an error.
popValue(SmallVectorImpl<uint64_t> & Record,unsigned & Slot,unsigned InstNum,Type * Ty,Value * & ResVal)336   bool popValue(SmallVectorImpl<uint64_t> &Record, unsigned &Slot,
337                 unsigned InstNum, Type *Ty, Value *&ResVal) {
338     if (getValue(Record, Slot, InstNum, Ty, ResVal))
339       return true;
340     // All values currently take a single record slot.
341     ++Slot;
342     return false;
343   }
344 
345   /// Like popValue, but does not increment the Slot number.
getValue(SmallVectorImpl<uint64_t> & Record,unsigned Slot,unsigned InstNum,Type * Ty,Value * & ResVal)346   bool getValue(SmallVectorImpl<uint64_t> &Record, unsigned Slot,
347                 unsigned InstNum, Type *Ty, Value *&ResVal) {
348     ResVal = getValue(Record, Slot, InstNum, Ty);
349     return ResVal == nullptr;
350   }
351 
352   /// Version of getValue that returns ResVal directly, or 0 if there is an
353   /// error.
getValue(SmallVectorImpl<uint64_t> & Record,unsigned Slot,unsigned InstNum,Type * Ty)354   Value *getValue(SmallVectorImpl<uint64_t> &Record, unsigned Slot,
355                   unsigned InstNum, Type *Ty) {
356     if (Slot == Record.size()) return nullptr;
357     unsigned ValNo = (unsigned)Record[Slot];
358     // Adjust the ValNo, if it was encoded relative to the InstNum.
359     if (UseRelativeIDs)
360       ValNo = InstNum - ValNo;
361     return getFnValueByID(ValNo, Ty);
362   }
363 
364   /// Like getValue, but decodes signed VBRs.
getValueSigned(SmallVectorImpl<uint64_t> & Record,unsigned Slot,unsigned InstNum,Type * Ty)365   Value *getValueSigned(SmallVectorImpl<uint64_t> &Record, unsigned Slot,
366                         unsigned InstNum, Type *Ty) {
367     if (Slot == Record.size()) return nullptr;
368     unsigned ValNo = (unsigned)decodeSignRotatedValue(Record[Slot]);
369     // Adjust the ValNo, if it was encoded relative to the InstNum.
370     if (UseRelativeIDs)
371       ValNo = InstNum - ValNo;
372     return getFnValueByID(ValNo, Ty);
373   }
374 
375   /// Converts alignment exponent (i.e. power of two (or zero)) to the
376   /// corresponding alignment to use. If alignment is too large, returns
377   /// a corresponding error code.
378   std::error_code parseAlignmentValue(uint64_t Exponent, unsigned &Alignment);
379   std::error_code parseAttrKind(uint64_t Code, Attribute::AttrKind *Kind);
380   std::error_code parseModule(uint64_t ResumeBit,
381                               bool ShouldLazyLoadMetadata = false);
382   std::error_code parseAttributeBlock();
383   std::error_code parseAttributeGroupBlock();
384   std::error_code parseTypeTable();
385   std::error_code parseTypeTableBody();
386   std::error_code parseOperandBundleTags();
387 
388   ErrorOr<Value *> recordValue(SmallVectorImpl<uint64_t> &Record,
389                                unsigned NameIndex, Triple &TT);
390   std::error_code parseValueSymbolTable(uint64_t Offset = 0);
391   std::error_code parseConstants();
392   std::error_code rememberAndSkipFunctionBodies();
393   std::error_code rememberAndSkipFunctionBody();
394   /// Save the positions of the Metadata blocks and skip parsing the blocks.
395   std::error_code rememberAndSkipMetadata();
396   std::error_code parseFunctionBody(Function *F);
397   std::error_code globalCleanup();
398   std::error_code resolveGlobalAndAliasInits();
399   std::error_code parseMetadata(bool ModuleLevel = false);
400   std::error_code parseMetadataKinds();
401   std::error_code parseMetadataKindRecord(SmallVectorImpl<uint64_t> &Record);
402   std::error_code parseMetadataAttachment(Function &F);
403   ErrorOr<std::string> parseModuleTriple();
404   std::error_code parseUseLists();
405   std::error_code initStream(std::unique_ptr<DataStreamer> Streamer);
406   std::error_code initStreamFromBuffer();
407   std::error_code initLazyStream(std::unique_ptr<DataStreamer> Streamer);
408   std::error_code findFunctionInStream(
409       Function *F,
410       DenseMap<Function *, uint64_t>::iterator DeferredFunctionInfoIterator);
411 };
412 
413 /// Class to manage reading and parsing function summary index bitcode
414 /// files/sections.
415 class FunctionIndexBitcodeReader {
416   DiagnosticHandlerFunction DiagnosticHandler;
417 
418   /// Eventually points to the function index built during parsing.
419   FunctionInfoIndex *TheIndex = nullptr;
420 
421   std::unique_ptr<MemoryBuffer> Buffer;
422   std::unique_ptr<BitstreamReader> StreamFile;
423   BitstreamCursor Stream;
424 
425   /// \brief Used to indicate whether we are doing lazy parsing of summary data.
426   ///
427   /// If false, the summary section is fully parsed into the index during
428   /// the initial parse. Otherwise, if true, the caller is expected to
429   /// invoke \a readFunctionSummary for each summary needed, and the summary
430   /// section is thus parsed lazily.
431   bool IsLazy = false;
432 
433   /// Used to indicate whether caller only wants to check for the presence
434   /// of the function summary bitcode section. All blocks are skipped,
435   /// but the SeenFuncSummary boolean is set.
436   bool CheckFuncSummaryPresenceOnly = false;
437 
438   /// Indicates whether we have encountered a function summary section
439   /// yet during parsing, used when checking if file contains function
440   /// summary section.
441   bool SeenFuncSummary = false;
442 
443   /// \brief Map populated during function summary section parsing, and
444   /// consumed during ValueSymbolTable parsing.
445   ///
446   /// Used to correlate summary records with VST entries. For the per-module
447   /// index this maps the ValueID to the parsed function summary, and
448   /// for the combined index this maps the summary record's bitcode
449   /// offset to the function summary (since in the combined index the
450   /// VST records do not hold value IDs but rather hold the function
451   /// summary record offset).
452   DenseMap<uint64_t, std::unique_ptr<FunctionSummary>> SummaryMap;
453 
454   /// Map populated during module path string table parsing, from the
455   /// module ID to a string reference owned by the index's module
456   /// path string table, used to correlate with combined index function
457   /// summary records.
458   DenseMap<uint64_t, StringRef> ModuleIdMap;
459 
460 public:
461   std::error_code error(BitcodeError E, const Twine &Message);
462   std::error_code error(BitcodeError E);
463   std::error_code error(const Twine &Message);
464 
465   FunctionIndexBitcodeReader(MemoryBuffer *Buffer,
466                              DiagnosticHandlerFunction DiagnosticHandler,
467                              bool IsLazy = false,
468                              bool CheckFuncSummaryPresenceOnly = false);
469   FunctionIndexBitcodeReader(DiagnosticHandlerFunction DiagnosticHandler,
470                              bool IsLazy = false,
471                              bool CheckFuncSummaryPresenceOnly = false);
~FunctionIndexBitcodeReader()472   ~FunctionIndexBitcodeReader() { freeState(); }
473 
474   void freeState();
475 
476   void releaseBuffer();
477 
478   /// Check if the parser has encountered a function summary section.
foundFuncSummary()479   bool foundFuncSummary() { return SeenFuncSummary; }
480 
481   /// \brief Main interface to parsing a bitcode buffer.
482   /// \returns true if an error occurred.
483   std::error_code parseSummaryIndexInto(std::unique_ptr<DataStreamer> Streamer,
484                                         FunctionInfoIndex *I);
485 
486   /// \brief Interface for parsing a function summary lazily.
487   std::error_code parseFunctionSummary(std::unique_ptr<DataStreamer> Streamer,
488                                        FunctionInfoIndex *I,
489                                        size_t FunctionSummaryOffset);
490 
491 private:
492   std::error_code parseModule();
493   std::error_code parseValueSymbolTable();
494   std::error_code parseEntireSummary();
495   std::error_code parseModuleStringTable();
496   std::error_code initStream(std::unique_ptr<DataStreamer> Streamer);
497   std::error_code initStreamFromBuffer();
498   std::error_code initLazyStream(std::unique_ptr<DataStreamer> Streamer);
499 };
500 } // namespace
501 
BitcodeDiagnosticInfo(std::error_code EC,DiagnosticSeverity Severity,const Twine & Msg)502 BitcodeDiagnosticInfo::BitcodeDiagnosticInfo(std::error_code EC,
503                                              DiagnosticSeverity Severity,
504                                              const Twine &Msg)
505     : DiagnosticInfo(DK_Bitcode, Severity), Msg(Msg), EC(EC) {}
506 
print(DiagnosticPrinter & DP) const507 void BitcodeDiagnosticInfo::print(DiagnosticPrinter &DP) const { DP << Msg; }
508 
error(DiagnosticHandlerFunction DiagnosticHandler,std::error_code EC,const Twine & Message)509 static std::error_code error(DiagnosticHandlerFunction DiagnosticHandler,
510                              std::error_code EC, const Twine &Message) {
511   BitcodeDiagnosticInfo DI(EC, DS_Error, Message);
512   DiagnosticHandler(DI);
513   return EC;
514 }
515 
error(DiagnosticHandlerFunction DiagnosticHandler,std::error_code EC)516 static std::error_code error(DiagnosticHandlerFunction DiagnosticHandler,
517                              std::error_code EC) {
518   return error(DiagnosticHandler, EC, EC.message());
519 }
520 
error(LLVMContext & Context,std::error_code EC,const Twine & Message)521 static std::error_code error(LLVMContext &Context, std::error_code EC,
522                              const Twine &Message) {
523   return error([&](const DiagnosticInfo &DI) { Context.diagnose(DI); }, EC,
524                Message);
525 }
526 
error(LLVMContext & Context,std::error_code EC)527 static std::error_code error(LLVMContext &Context, std::error_code EC) {
528   return error(Context, EC, EC.message());
529 }
530 
error(LLVMContext & Context,const Twine & Message)531 static std::error_code error(LLVMContext &Context, const Twine &Message) {
532   return error(Context, make_error_code(BitcodeError::CorruptedBitcode),
533                Message);
534 }
535 
error(BitcodeError E,const Twine & Message)536 std::error_code BitcodeReader::error(BitcodeError E, const Twine &Message) {
537   if (!ProducerIdentification.empty()) {
538     return ::error(Context, make_error_code(E),
539                    Message + " (Producer: '" + ProducerIdentification +
540                        "' Reader: 'LLVM " + LLVM_VERSION_STRING "')");
541   }
542   return ::error(Context, make_error_code(E), Message);
543 }
544 
error(const Twine & Message)545 std::error_code BitcodeReader::error(const Twine &Message) {
546   if (!ProducerIdentification.empty()) {
547     return ::error(Context, make_error_code(BitcodeError::CorruptedBitcode),
548                    Message + " (Producer: '" + ProducerIdentification +
549                        "' Reader: 'LLVM " + LLVM_VERSION_STRING "')");
550   }
551   return ::error(Context, make_error_code(BitcodeError::CorruptedBitcode),
552                  Message);
553 }
554 
error(BitcodeError E)555 std::error_code BitcodeReader::error(BitcodeError E) {
556   return ::error(Context, make_error_code(E));
557 }
558 
BitcodeReader(MemoryBuffer * Buffer,LLVMContext & Context)559 BitcodeReader::BitcodeReader(MemoryBuffer *Buffer, LLVMContext &Context)
560     : Context(Context), Buffer(Buffer), ValueList(Context),
561       MDValueList(Context) {}
562 
BitcodeReader(LLVMContext & Context)563 BitcodeReader::BitcodeReader(LLVMContext &Context)
564     : Context(Context), Buffer(nullptr), ValueList(Context),
565       MDValueList(Context) {}
566 
materializeForwardReferencedFunctions()567 std::error_code BitcodeReader::materializeForwardReferencedFunctions() {
568   if (WillMaterializeAllForwardRefs)
569     return std::error_code();
570 
571   // Prevent recursion.
572   WillMaterializeAllForwardRefs = true;
573 
574   while (!BasicBlockFwdRefQueue.empty()) {
575     Function *F = BasicBlockFwdRefQueue.front();
576     BasicBlockFwdRefQueue.pop_front();
577     assert(F && "Expected valid function");
578     if (!BasicBlockFwdRefs.count(F))
579       // Already materialized.
580       continue;
581 
582     // Check for a function that isn't materializable to prevent an infinite
583     // loop.  When parsing a blockaddress stored in a global variable, there
584     // isn't a trivial way to check if a function will have a body without a
585     // linear search through FunctionsWithBodies, so just check it here.
586     if (!F->isMaterializable())
587       return error("Never resolved function from blockaddress");
588 
589     // Try to materialize F.
590     if (std::error_code EC = materialize(F))
591       return EC;
592   }
593   assert(BasicBlockFwdRefs.empty() && "Function missing from queue");
594 
595   // Reset state.
596   WillMaterializeAllForwardRefs = false;
597   return std::error_code();
598 }
599 
freeState()600 void BitcodeReader::freeState() {
601   Buffer = nullptr;
602   std::vector<Type*>().swap(TypeList);
603   ValueList.clear();
604   MDValueList.clear();
605   std::vector<Comdat *>().swap(ComdatList);
606 
607   std::vector<AttributeSet>().swap(MAttributes);
608   std::vector<BasicBlock*>().swap(FunctionBBs);
609   std::vector<Function*>().swap(FunctionsWithBodies);
610   DeferredFunctionInfo.clear();
611   DeferredMetadataInfo.clear();
612   MDKindMap.clear();
613 
614   assert(BasicBlockFwdRefs.empty() && "Unresolved blockaddress fwd references");
615   BasicBlockFwdRefQueue.clear();
616 }
617 
618 //===----------------------------------------------------------------------===//
619 //  Helper functions to implement forward reference resolution, etc.
620 //===----------------------------------------------------------------------===//
621 
622 /// Convert a string from a record into an std::string, return true on failure.
623 template <typename StrTy>
convertToString(ArrayRef<uint64_t> Record,unsigned Idx,StrTy & Result)624 static bool convertToString(ArrayRef<uint64_t> Record, unsigned Idx,
625                             StrTy &Result) {
626   if (Idx > Record.size())
627     return true;
628 
629   for (unsigned i = Idx, e = Record.size(); i != e; ++i)
630     Result += (char)Record[i];
631   return false;
632 }
633 
hasImplicitComdat(size_t Val)634 static bool hasImplicitComdat(size_t Val) {
635   switch (Val) {
636   default:
637     return false;
638   case 1:  // Old WeakAnyLinkage
639   case 4:  // Old LinkOnceAnyLinkage
640   case 10: // Old WeakODRLinkage
641   case 11: // Old LinkOnceODRLinkage
642     return true;
643   }
644 }
645 
getDecodedLinkage(unsigned Val)646 static GlobalValue::LinkageTypes getDecodedLinkage(unsigned Val) {
647   switch (Val) {
648   default: // Map unknown/new linkages to external
649   case 0:
650     return GlobalValue::ExternalLinkage;
651   case 2:
652     return GlobalValue::AppendingLinkage;
653   case 3:
654     return GlobalValue::InternalLinkage;
655   case 5:
656     return GlobalValue::ExternalLinkage; // Obsolete DLLImportLinkage
657   case 6:
658     return GlobalValue::ExternalLinkage; // Obsolete DLLExportLinkage
659   case 7:
660     return GlobalValue::ExternalWeakLinkage;
661   case 8:
662     return GlobalValue::CommonLinkage;
663   case 9:
664     return GlobalValue::PrivateLinkage;
665   case 12:
666     return GlobalValue::AvailableExternallyLinkage;
667   case 13:
668     return GlobalValue::PrivateLinkage; // Obsolete LinkerPrivateLinkage
669   case 14:
670     return GlobalValue::PrivateLinkage; // Obsolete LinkerPrivateWeakLinkage
671   case 15:
672     return GlobalValue::ExternalLinkage; // Obsolete LinkOnceODRAutoHideLinkage
673   case 1: // Old value with implicit comdat.
674   case 16:
675     return GlobalValue::WeakAnyLinkage;
676   case 10: // Old value with implicit comdat.
677   case 17:
678     return GlobalValue::WeakODRLinkage;
679   case 4: // Old value with implicit comdat.
680   case 18:
681     return GlobalValue::LinkOnceAnyLinkage;
682   case 11: // Old value with implicit comdat.
683   case 19:
684     return GlobalValue::LinkOnceODRLinkage;
685   }
686 }
687 
getDecodedVisibility(unsigned Val)688 static GlobalValue::VisibilityTypes getDecodedVisibility(unsigned Val) {
689   switch (Val) {
690   default: // Map unknown visibilities to default.
691   case 0: return GlobalValue::DefaultVisibility;
692   case 1: return GlobalValue::HiddenVisibility;
693   case 2: return GlobalValue::ProtectedVisibility;
694   }
695 }
696 
697 static GlobalValue::DLLStorageClassTypes
getDecodedDLLStorageClass(unsigned Val)698 getDecodedDLLStorageClass(unsigned Val) {
699   switch (Val) {
700   default: // Map unknown values to default.
701   case 0: return GlobalValue::DefaultStorageClass;
702   case 1: return GlobalValue::DLLImportStorageClass;
703   case 2: return GlobalValue::DLLExportStorageClass;
704   }
705 }
706 
getDecodedThreadLocalMode(unsigned Val)707 static GlobalVariable::ThreadLocalMode getDecodedThreadLocalMode(unsigned Val) {
708   switch (Val) {
709     case 0: return GlobalVariable::NotThreadLocal;
710     default: // Map unknown non-zero value to general dynamic.
711     case 1: return GlobalVariable::GeneralDynamicTLSModel;
712     case 2: return GlobalVariable::LocalDynamicTLSModel;
713     case 3: return GlobalVariable::InitialExecTLSModel;
714     case 4: return GlobalVariable::LocalExecTLSModel;
715   }
716 }
717 
getDecodedCastOpcode(unsigned Val)718 static int getDecodedCastOpcode(unsigned Val) {
719   switch (Val) {
720   default: return -1;
721   case bitc::CAST_TRUNC   : return Instruction::Trunc;
722   case bitc::CAST_ZEXT    : return Instruction::ZExt;
723   case bitc::CAST_SEXT    : return Instruction::SExt;
724   case bitc::CAST_FPTOUI  : return Instruction::FPToUI;
725   case bitc::CAST_FPTOSI  : return Instruction::FPToSI;
726   case bitc::CAST_UITOFP  : return Instruction::UIToFP;
727   case bitc::CAST_SITOFP  : return Instruction::SIToFP;
728   case bitc::CAST_FPTRUNC : return Instruction::FPTrunc;
729   case bitc::CAST_FPEXT   : return Instruction::FPExt;
730   case bitc::CAST_PTRTOINT: return Instruction::PtrToInt;
731   case bitc::CAST_INTTOPTR: return Instruction::IntToPtr;
732   case bitc::CAST_BITCAST : return Instruction::BitCast;
733   case bitc::CAST_ADDRSPACECAST: return Instruction::AddrSpaceCast;
734   }
735 }
736 
getDecodedBinaryOpcode(unsigned Val,Type * Ty)737 static int getDecodedBinaryOpcode(unsigned Val, Type *Ty) {
738   bool IsFP = Ty->isFPOrFPVectorTy();
739   // BinOps are only valid for int/fp or vector of int/fp types
740   if (!IsFP && !Ty->isIntOrIntVectorTy())
741     return -1;
742 
743   switch (Val) {
744   default:
745     return -1;
746   case bitc::BINOP_ADD:
747     return IsFP ? Instruction::FAdd : Instruction::Add;
748   case bitc::BINOP_SUB:
749     return IsFP ? Instruction::FSub : Instruction::Sub;
750   case bitc::BINOP_MUL:
751     return IsFP ? Instruction::FMul : Instruction::Mul;
752   case bitc::BINOP_UDIV:
753     return IsFP ? -1 : Instruction::UDiv;
754   case bitc::BINOP_SDIV:
755     return IsFP ? Instruction::FDiv : Instruction::SDiv;
756   case bitc::BINOP_UREM:
757     return IsFP ? -1 : Instruction::URem;
758   case bitc::BINOP_SREM:
759     return IsFP ? Instruction::FRem : Instruction::SRem;
760   case bitc::BINOP_SHL:
761     return IsFP ? -1 : Instruction::Shl;
762   case bitc::BINOP_LSHR:
763     return IsFP ? -1 : Instruction::LShr;
764   case bitc::BINOP_ASHR:
765     return IsFP ? -1 : Instruction::AShr;
766   case bitc::BINOP_AND:
767     return IsFP ? -1 : Instruction::And;
768   case bitc::BINOP_OR:
769     return IsFP ? -1 : Instruction::Or;
770   case bitc::BINOP_XOR:
771     return IsFP ? -1 : Instruction::Xor;
772   }
773 }
774 
getDecodedRMWOperation(unsigned Val)775 static AtomicRMWInst::BinOp getDecodedRMWOperation(unsigned Val) {
776   switch (Val) {
777   default: return AtomicRMWInst::BAD_BINOP;
778   case bitc::RMW_XCHG: return AtomicRMWInst::Xchg;
779   case bitc::RMW_ADD: return AtomicRMWInst::Add;
780   case bitc::RMW_SUB: return AtomicRMWInst::Sub;
781   case bitc::RMW_AND: return AtomicRMWInst::And;
782   case bitc::RMW_NAND: return AtomicRMWInst::Nand;
783   case bitc::RMW_OR: return AtomicRMWInst::Or;
784   case bitc::RMW_XOR: return AtomicRMWInst::Xor;
785   case bitc::RMW_MAX: return AtomicRMWInst::Max;
786   case bitc::RMW_MIN: return AtomicRMWInst::Min;
787   case bitc::RMW_UMAX: return AtomicRMWInst::UMax;
788   case bitc::RMW_UMIN: return AtomicRMWInst::UMin;
789   }
790 }
791 
getDecodedOrdering(unsigned Val)792 static AtomicOrdering getDecodedOrdering(unsigned Val) {
793   switch (Val) {
794   case bitc::ORDERING_NOTATOMIC: return NotAtomic;
795   case bitc::ORDERING_UNORDERED: return Unordered;
796   case bitc::ORDERING_MONOTONIC: return Monotonic;
797   case bitc::ORDERING_ACQUIRE: return Acquire;
798   case bitc::ORDERING_RELEASE: return Release;
799   case bitc::ORDERING_ACQREL: return AcquireRelease;
800   default: // Map unknown orderings to sequentially-consistent.
801   case bitc::ORDERING_SEQCST: return SequentiallyConsistent;
802   }
803 }
804 
getDecodedSynchScope(unsigned Val)805 static SynchronizationScope getDecodedSynchScope(unsigned Val) {
806   switch (Val) {
807   case bitc::SYNCHSCOPE_SINGLETHREAD: return SingleThread;
808   default: // Map unknown scopes to cross-thread.
809   case bitc::SYNCHSCOPE_CROSSTHREAD: return CrossThread;
810   }
811 }
812 
getDecodedComdatSelectionKind(unsigned Val)813 static Comdat::SelectionKind getDecodedComdatSelectionKind(unsigned Val) {
814   switch (Val) {
815   default: // Map unknown selection kinds to any.
816   case bitc::COMDAT_SELECTION_KIND_ANY:
817     return Comdat::Any;
818   case bitc::COMDAT_SELECTION_KIND_EXACT_MATCH:
819     return Comdat::ExactMatch;
820   case bitc::COMDAT_SELECTION_KIND_LARGEST:
821     return Comdat::Largest;
822   case bitc::COMDAT_SELECTION_KIND_NO_DUPLICATES:
823     return Comdat::NoDuplicates;
824   case bitc::COMDAT_SELECTION_KIND_SAME_SIZE:
825     return Comdat::SameSize;
826   }
827 }
828 
getDecodedFastMathFlags(unsigned Val)829 static FastMathFlags getDecodedFastMathFlags(unsigned Val) {
830   FastMathFlags FMF;
831   if (0 != (Val & FastMathFlags::UnsafeAlgebra))
832     FMF.setUnsafeAlgebra();
833   if (0 != (Val & FastMathFlags::NoNaNs))
834     FMF.setNoNaNs();
835   if (0 != (Val & FastMathFlags::NoInfs))
836     FMF.setNoInfs();
837   if (0 != (Val & FastMathFlags::NoSignedZeros))
838     FMF.setNoSignedZeros();
839   if (0 != (Val & FastMathFlags::AllowReciprocal))
840     FMF.setAllowReciprocal();
841   return FMF;
842 }
843 
upgradeDLLImportExportLinkage(llvm::GlobalValue * GV,unsigned Val)844 static void upgradeDLLImportExportLinkage(llvm::GlobalValue *GV, unsigned Val) {
845   switch (Val) {
846   case 5: GV->setDLLStorageClass(GlobalValue::DLLImportStorageClass); break;
847   case 6: GV->setDLLStorageClass(GlobalValue::DLLExportStorageClass); break;
848   }
849 }
850 
851 namespace llvm {
852 namespace {
853 /// \brief A class for maintaining the slot number definition
854 /// as a placeholder for the actual definition for forward constants defs.
855 class ConstantPlaceHolder : public ConstantExpr {
856   void operator=(const ConstantPlaceHolder &) = delete;
857 
858 public:
859   // allocate space for exactly one operand
operator new(size_t s)860   void *operator new(size_t s) { return User::operator new(s, 1); }
ConstantPlaceHolder(Type * Ty,LLVMContext & Context)861   explicit ConstantPlaceHolder(Type *Ty, LLVMContext &Context)
862       : ConstantExpr(Ty, Instruction::UserOp1, &Op<0>(), 1) {
863     Op<0>() = UndefValue::get(Type::getInt32Ty(Context));
864   }
865 
866   /// \brief Methods to support type inquiry through isa, cast, and dyn_cast.
classof(const Value * V)867   static bool classof(const Value *V) {
868     return isa<ConstantExpr>(V) &&
869            cast<ConstantExpr>(V)->getOpcode() == Instruction::UserOp1;
870   }
871 
872   /// Provide fast operand accessors
873   DECLARE_TRANSPARENT_OPERAND_ACCESSORS(Value);
874 };
875 }
876 
877 // FIXME: can we inherit this from ConstantExpr?
878 template <>
879 struct OperandTraits<ConstantPlaceHolder> :
880   public FixedNumOperandTraits<ConstantPlaceHolder, 1> {
881 };
882 DEFINE_TRANSPARENT_OPERAND_ACCESSORS(ConstantPlaceHolder, Value)
883 }
884 
assignValue(Value * V,unsigned Idx)885 void BitcodeReaderValueList::assignValue(Value *V, unsigned Idx) {
886   if (Idx == size()) {
887     push_back(V);
888     return;
889   }
890 
891   if (Idx >= size())
892     resize(Idx+1);
893 
894   WeakVH &OldV = ValuePtrs[Idx];
895   if (!OldV) {
896     OldV = V;
897     return;
898   }
899 
900   // Handle constants and non-constants (e.g. instrs) differently for
901   // efficiency.
902   if (Constant *PHC = dyn_cast<Constant>(&*OldV)) {
903     ResolveConstants.push_back(std::make_pair(PHC, Idx));
904     OldV = V;
905   } else {
906     // If there was a forward reference to this value, replace it.
907     Value *PrevVal = OldV;
908     OldV->replaceAllUsesWith(V);
909     delete PrevVal;
910   }
911 
912   return;
913 }
914 
915 
getConstantFwdRef(unsigned Idx,Type * Ty)916 Constant *BitcodeReaderValueList::getConstantFwdRef(unsigned Idx,
917                                                     Type *Ty) {
918   if (Idx >= size())
919     resize(Idx + 1);
920 
921   if (Value *V = ValuePtrs[Idx]) {
922     if (Ty != V->getType())
923       report_fatal_error("Type mismatch in constant table!");
924     return cast<Constant>(V);
925   }
926 
927   // Create and return a placeholder, which will later be RAUW'd.
928   Constant *C = new ConstantPlaceHolder(Ty, Context);
929   ValuePtrs[Idx] = C;
930   return C;
931 }
932 
getValueFwdRef(unsigned Idx,Type * Ty)933 Value *BitcodeReaderValueList::getValueFwdRef(unsigned Idx, Type *Ty) {
934   // Bail out for a clearly invalid value. This would make us call resize(0)
935   if (Idx == UINT_MAX)
936     return nullptr;
937 
938   if (Idx >= size())
939     resize(Idx + 1);
940 
941   if (Value *V = ValuePtrs[Idx]) {
942     // If the types don't match, it's invalid.
943     if (Ty && Ty != V->getType())
944       return nullptr;
945     return V;
946   }
947 
948   // No type specified, must be invalid reference.
949   if (!Ty) return nullptr;
950 
951   // Create and return a placeholder, which will later be RAUW'd.
952   Value *V = new Argument(Ty);
953   ValuePtrs[Idx] = V;
954   return V;
955 }
956 
957 /// Once all constants are read, this method bulk resolves any forward
958 /// references.  The idea behind this is that we sometimes get constants (such
959 /// as large arrays) which reference *many* forward ref constants.  Replacing
960 /// each of these causes a lot of thrashing when building/reuniquing the
961 /// constant.  Instead of doing this, we look at all the uses and rewrite all
962 /// the place holders at once for any constant that uses a placeholder.
resolveConstantForwardRefs()963 void BitcodeReaderValueList::resolveConstantForwardRefs() {
964   // Sort the values by-pointer so that they are efficient to look up with a
965   // binary search.
966   std::sort(ResolveConstants.begin(), ResolveConstants.end());
967 
968   SmallVector<Constant*, 64> NewOps;
969 
970   while (!ResolveConstants.empty()) {
971     Value *RealVal = operator[](ResolveConstants.back().second);
972     Constant *Placeholder = ResolveConstants.back().first;
973     ResolveConstants.pop_back();
974 
975     // Loop over all users of the placeholder, updating them to reference the
976     // new value.  If they reference more than one placeholder, update them all
977     // at once.
978     while (!Placeholder->use_empty()) {
979       auto UI = Placeholder->user_begin();
980       User *U = *UI;
981 
982       // If the using object isn't uniqued, just update the operands.  This
983       // handles instructions and initializers for global variables.
984       if (!isa<Constant>(U) || isa<GlobalValue>(U)) {
985         UI.getUse().set(RealVal);
986         continue;
987       }
988 
989       // Otherwise, we have a constant that uses the placeholder.  Replace that
990       // constant with a new constant that has *all* placeholder uses updated.
991       Constant *UserC = cast<Constant>(U);
992       for (User::op_iterator I = UserC->op_begin(), E = UserC->op_end();
993            I != E; ++I) {
994         Value *NewOp;
995         if (!isa<ConstantPlaceHolder>(*I)) {
996           // Not a placeholder reference.
997           NewOp = *I;
998         } else if (*I == Placeholder) {
999           // Common case is that it just references this one placeholder.
1000           NewOp = RealVal;
1001         } else {
1002           // Otherwise, look up the placeholder in ResolveConstants.
1003           ResolveConstantsTy::iterator It =
1004             std::lower_bound(ResolveConstants.begin(), ResolveConstants.end(),
1005                              std::pair<Constant*, unsigned>(cast<Constant>(*I),
1006                                                             0));
1007           assert(It != ResolveConstants.end() && It->first == *I);
1008           NewOp = operator[](It->second);
1009         }
1010 
1011         NewOps.push_back(cast<Constant>(NewOp));
1012       }
1013 
1014       // Make the new constant.
1015       Constant *NewC;
1016       if (ConstantArray *UserCA = dyn_cast<ConstantArray>(UserC)) {
1017         NewC = ConstantArray::get(UserCA->getType(), NewOps);
1018       } else if (ConstantStruct *UserCS = dyn_cast<ConstantStruct>(UserC)) {
1019         NewC = ConstantStruct::get(UserCS->getType(), NewOps);
1020       } else if (isa<ConstantVector>(UserC)) {
1021         NewC = ConstantVector::get(NewOps);
1022       } else {
1023         assert(isa<ConstantExpr>(UserC) && "Must be a ConstantExpr.");
1024         NewC = cast<ConstantExpr>(UserC)->getWithOperands(NewOps);
1025       }
1026 
1027       UserC->replaceAllUsesWith(NewC);
1028       UserC->destroyConstant();
1029       NewOps.clear();
1030     }
1031 
1032     // Update all ValueHandles, they should be the only users at this point.
1033     Placeholder->replaceAllUsesWith(RealVal);
1034     delete Placeholder;
1035   }
1036 }
1037 
assignValue(Metadata * MD,unsigned Idx)1038 void BitcodeReaderMDValueList::assignValue(Metadata *MD, unsigned Idx) {
1039   if (Idx == size()) {
1040     push_back(MD);
1041     return;
1042   }
1043 
1044   if (Idx >= size())
1045     resize(Idx+1);
1046 
1047   TrackingMDRef &OldMD = MDValuePtrs[Idx];
1048   if (!OldMD) {
1049     OldMD.reset(MD);
1050     return;
1051   }
1052 
1053   // If there was a forward reference to this value, replace it.
1054   TempMDTuple PrevMD(cast<MDTuple>(OldMD.get()));
1055   PrevMD->replaceAllUsesWith(MD);
1056   --NumFwdRefs;
1057 }
1058 
getValueFwdRef(unsigned Idx)1059 Metadata *BitcodeReaderMDValueList::getValueFwdRef(unsigned Idx) {
1060   if (Idx >= size())
1061     resize(Idx + 1);
1062 
1063   if (Metadata *MD = MDValuePtrs[Idx])
1064     return MD;
1065 
1066   // Track forward refs to be resolved later.
1067   if (AnyFwdRefs) {
1068     MinFwdRef = std::min(MinFwdRef, Idx);
1069     MaxFwdRef = std::max(MaxFwdRef, Idx);
1070   } else {
1071     AnyFwdRefs = true;
1072     MinFwdRef = MaxFwdRef = Idx;
1073   }
1074   ++NumFwdRefs;
1075 
1076   // Create and return a placeholder, which will later be RAUW'd.
1077   Metadata *MD = MDNode::getTemporary(Context, None).release();
1078   MDValuePtrs[Idx].reset(MD);
1079   return MD;
1080 }
1081 
tryToResolveCycles()1082 void BitcodeReaderMDValueList::tryToResolveCycles() {
1083   if (!AnyFwdRefs)
1084     // Nothing to do.
1085     return;
1086 
1087   if (NumFwdRefs)
1088     // Still forward references... can't resolve cycles.
1089     return;
1090 
1091   // Resolve any cycles.
1092   for (unsigned I = MinFwdRef, E = MaxFwdRef + 1; I != E; ++I) {
1093     auto &MD = MDValuePtrs[I];
1094     auto *N = dyn_cast_or_null<MDNode>(MD);
1095     if (!N)
1096       continue;
1097 
1098     assert(!N->isTemporary() && "Unexpected forward reference");
1099     N->resolveCycles();
1100   }
1101 
1102   // Make sure we return early again until there's another forward ref.
1103   AnyFwdRefs = false;
1104 }
1105 
getTypeByID(unsigned ID)1106 Type *BitcodeReader::getTypeByID(unsigned ID) {
1107   // The type table size is always specified correctly.
1108   if (ID >= TypeList.size())
1109     return nullptr;
1110 
1111   if (Type *Ty = TypeList[ID])
1112     return Ty;
1113 
1114   // If we have a forward reference, the only possible case is when it is to a
1115   // named struct.  Just create a placeholder for now.
1116   return TypeList[ID] = createIdentifiedStructType(Context);
1117 }
1118 
createIdentifiedStructType(LLVMContext & Context,StringRef Name)1119 StructType *BitcodeReader::createIdentifiedStructType(LLVMContext &Context,
1120                                                       StringRef Name) {
1121   auto *Ret = StructType::create(Context, Name);
1122   IdentifiedStructTypes.push_back(Ret);
1123   return Ret;
1124 }
1125 
createIdentifiedStructType(LLVMContext & Context)1126 StructType *BitcodeReader::createIdentifiedStructType(LLVMContext &Context) {
1127   auto *Ret = StructType::create(Context);
1128   IdentifiedStructTypes.push_back(Ret);
1129   return Ret;
1130 }
1131 
1132 
1133 //===----------------------------------------------------------------------===//
1134 //  Functions for parsing blocks from the bitcode file
1135 //===----------------------------------------------------------------------===//
1136 
1137 
1138 /// \brief This fills an AttrBuilder object with the LLVM attributes that have
1139 /// been decoded from the given integer. This function must stay in sync with
1140 /// 'encodeLLVMAttributesForBitcode'.
decodeLLVMAttributesForBitcode(AttrBuilder & B,uint64_t EncodedAttrs)1141 static void decodeLLVMAttributesForBitcode(AttrBuilder &B,
1142                                            uint64_t EncodedAttrs) {
1143   // FIXME: Remove in 4.0.
1144 
1145   // The alignment is stored as a 16-bit raw value from bits 31--16.  We shift
1146   // the bits above 31 down by 11 bits.
1147   unsigned Alignment = (EncodedAttrs & (0xffffULL << 16)) >> 16;
1148   assert((!Alignment || isPowerOf2_32(Alignment)) &&
1149          "Alignment must be a power of two.");
1150 
1151   if (Alignment)
1152     B.addAlignmentAttr(Alignment);
1153   B.addRawValue(((EncodedAttrs & (0xfffffULL << 32)) >> 11) |
1154                 (EncodedAttrs & 0xffff));
1155 }
1156 
parseAttributeBlock()1157 std::error_code BitcodeReader::parseAttributeBlock() {
1158   if (Stream.EnterSubBlock(bitc::PARAMATTR_BLOCK_ID))
1159     return error("Invalid record");
1160 
1161   if (!MAttributes.empty())
1162     return error("Invalid multiple blocks");
1163 
1164   SmallVector<uint64_t, 64> Record;
1165 
1166   SmallVector<AttributeSet, 8> Attrs;
1167 
1168   // Read all the records.
1169   while (1) {
1170     BitstreamEntry Entry = Stream.advanceSkippingSubblocks();
1171 
1172     switch (Entry.Kind) {
1173     case BitstreamEntry::SubBlock: // Handled for us already.
1174     case BitstreamEntry::Error:
1175       return error("Malformed block");
1176     case BitstreamEntry::EndBlock:
1177       return std::error_code();
1178     case BitstreamEntry::Record:
1179       // The interesting case.
1180       break;
1181     }
1182 
1183     // Read a record.
1184     Record.clear();
1185     switch (Stream.readRecord(Entry.ID, Record)) {
1186     default:  // Default behavior: ignore.
1187       break;
1188     case bitc::PARAMATTR_CODE_ENTRY_OLD: { // ENTRY: [paramidx0, attr0, ...]
1189       // FIXME: Remove in 4.0.
1190       if (Record.size() & 1)
1191         return error("Invalid record");
1192 
1193       for (unsigned i = 0, e = Record.size(); i != e; i += 2) {
1194         AttrBuilder B;
1195         decodeLLVMAttributesForBitcode(B, Record[i+1]);
1196         Attrs.push_back(AttributeSet::get(Context, Record[i], B));
1197       }
1198 
1199       MAttributes.push_back(AttributeSet::get(Context, Attrs));
1200       Attrs.clear();
1201       break;
1202     }
1203     case bitc::PARAMATTR_CODE_ENTRY: { // ENTRY: [attrgrp0, attrgrp1, ...]
1204       for (unsigned i = 0, e = Record.size(); i != e; ++i)
1205         Attrs.push_back(MAttributeGroups[Record[i]]);
1206 
1207       MAttributes.push_back(AttributeSet::get(Context, Attrs));
1208       Attrs.clear();
1209       break;
1210     }
1211     }
1212   }
1213 }
1214 
1215 // Returns Attribute::None on unrecognized codes.
getAttrFromCode(uint64_t Code)1216 static Attribute::AttrKind getAttrFromCode(uint64_t Code) {
1217   switch (Code) {
1218   default:
1219     return Attribute::None;
1220   case bitc::ATTR_KIND_ALIGNMENT:
1221     return Attribute::Alignment;
1222   case bitc::ATTR_KIND_ALWAYS_INLINE:
1223     return Attribute::AlwaysInline;
1224   case bitc::ATTR_KIND_ARGMEMONLY:
1225     return Attribute::ArgMemOnly;
1226   case bitc::ATTR_KIND_BUILTIN:
1227     return Attribute::Builtin;
1228   case bitc::ATTR_KIND_BY_VAL:
1229     return Attribute::ByVal;
1230   case bitc::ATTR_KIND_IN_ALLOCA:
1231     return Attribute::InAlloca;
1232   case bitc::ATTR_KIND_COLD:
1233     return Attribute::Cold;
1234   case bitc::ATTR_KIND_CONVERGENT:
1235     return Attribute::Convergent;
1236   case bitc::ATTR_KIND_INACCESSIBLEMEM_ONLY:
1237     return Attribute::InaccessibleMemOnly;
1238   case bitc::ATTR_KIND_INACCESSIBLEMEM_OR_ARGMEMONLY:
1239     return Attribute::InaccessibleMemOrArgMemOnly;
1240   case bitc::ATTR_KIND_INLINE_HINT:
1241     return Attribute::InlineHint;
1242   case bitc::ATTR_KIND_IN_REG:
1243     return Attribute::InReg;
1244   case bitc::ATTR_KIND_JUMP_TABLE:
1245     return Attribute::JumpTable;
1246   case bitc::ATTR_KIND_MIN_SIZE:
1247     return Attribute::MinSize;
1248   case bitc::ATTR_KIND_NAKED:
1249     return Attribute::Naked;
1250   case bitc::ATTR_KIND_NEST:
1251     return Attribute::Nest;
1252   case bitc::ATTR_KIND_NO_ALIAS:
1253     return Attribute::NoAlias;
1254   case bitc::ATTR_KIND_NO_BUILTIN:
1255     return Attribute::NoBuiltin;
1256   case bitc::ATTR_KIND_NO_CAPTURE:
1257     return Attribute::NoCapture;
1258   case bitc::ATTR_KIND_NO_DUPLICATE:
1259     return Attribute::NoDuplicate;
1260   case bitc::ATTR_KIND_NO_IMPLICIT_FLOAT:
1261     return Attribute::NoImplicitFloat;
1262   case bitc::ATTR_KIND_NO_INLINE:
1263     return Attribute::NoInline;
1264   case bitc::ATTR_KIND_NO_RECURSE:
1265     return Attribute::NoRecurse;
1266   case bitc::ATTR_KIND_NON_LAZY_BIND:
1267     return Attribute::NonLazyBind;
1268   case bitc::ATTR_KIND_NON_NULL:
1269     return Attribute::NonNull;
1270   case bitc::ATTR_KIND_DEREFERENCEABLE:
1271     return Attribute::Dereferenceable;
1272   case bitc::ATTR_KIND_DEREFERENCEABLE_OR_NULL:
1273     return Attribute::DereferenceableOrNull;
1274   case bitc::ATTR_KIND_NO_RED_ZONE:
1275     return Attribute::NoRedZone;
1276   case bitc::ATTR_KIND_NO_RETURN:
1277     return Attribute::NoReturn;
1278   case bitc::ATTR_KIND_NO_UNWIND:
1279     return Attribute::NoUnwind;
1280   case bitc::ATTR_KIND_OPTIMIZE_FOR_SIZE:
1281     return Attribute::OptimizeForSize;
1282   case bitc::ATTR_KIND_OPTIMIZE_NONE:
1283     return Attribute::OptimizeNone;
1284   case bitc::ATTR_KIND_READ_NONE:
1285     return Attribute::ReadNone;
1286   case bitc::ATTR_KIND_READ_ONLY:
1287     return Attribute::ReadOnly;
1288   case bitc::ATTR_KIND_RETURNED:
1289     return Attribute::Returned;
1290   case bitc::ATTR_KIND_RETURNS_TWICE:
1291     return Attribute::ReturnsTwice;
1292   case bitc::ATTR_KIND_S_EXT:
1293     return Attribute::SExt;
1294   case bitc::ATTR_KIND_STACK_ALIGNMENT:
1295     return Attribute::StackAlignment;
1296   case bitc::ATTR_KIND_STACK_PROTECT:
1297     return Attribute::StackProtect;
1298   case bitc::ATTR_KIND_STACK_PROTECT_REQ:
1299     return Attribute::StackProtectReq;
1300   case bitc::ATTR_KIND_STACK_PROTECT_STRONG:
1301     return Attribute::StackProtectStrong;
1302   case bitc::ATTR_KIND_SAFESTACK:
1303     return Attribute::SafeStack;
1304   case bitc::ATTR_KIND_STRUCT_RET:
1305     return Attribute::StructRet;
1306   case bitc::ATTR_KIND_SANITIZE_ADDRESS:
1307     return Attribute::SanitizeAddress;
1308   case bitc::ATTR_KIND_SANITIZE_THREAD:
1309     return Attribute::SanitizeThread;
1310   case bitc::ATTR_KIND_SANITIZE_MEMORY:
1311     return Attribute::SanitizeMemory;
1312   case bitc::ATTR_KIND_UW_TABLE:
1313     return Attribute::UWTable;
1314   case bitc::ATTR_KIND_Z_EXT:
1315     return Attribute::ZExt;
1316   }
1317 }
1318 
parseAlignmentValue(uint64_t Exponent,unsigned & Alignment)1319 std::error_code BitcodeReader::parseAlignmentValue(uint64_t Exponent,
1320                                                    unsigned &Alignment) {
1321   // Note: Alignment in bitcode files is incremented by 1, so that zero
1322   // can be used for default alignment.
1323   if (Exponent > Value::MaxAlignmentExponent + 1)
1324     return error("Invalid alignment value");
1325   Alignment = (1 << static_cast<unsigned>(Exponent)) >> 1;
1326   return std::error_code();
1327 }
1328 
parseAttrKind(uint64_t Code,Attribute::AttrKind * Kind)1329 std::error_code BitcodeReader::parseAttrKind(uint64_t Code,
1330                                              Attribute::AttrKind *Kind) {
1331   *Kind = getAttrFromCode(Code);
1332   if (*Kind == Attribute::None)
1333     return error(BitcodeError::CorruptedBitcode,
1334                  "Unknown attribute kind (" + Twine(Code) + ")");
1335   return std::error_code();
1336 }
1337 
parseAttributeGroupBlock()1338 std::error_code BitcodeReader::parseAttributeGroupBlock() {
1339   if (Stream.EnterSubBlock(bitc::PARAMATTR_GROUP_BLOCK_ID))
1340     return error("Invalid record");
1341 
1342   if (!MAttributeGroups.empty())
1343     return error("Invalid multiple blocks");
1344 
1345   SmallVector<uint64_t, 64> Record;
1346 
1347   // Read all the records.
1348   while (1) {
1349     BitstreamEntry Entry = Stream.advanceSkippingSubblocks();
1350 
1351     switch (Entry.Kind) {
1352     case BitstreamEntry::SubBlock: // Handled for us already.
1353     case BitstreamEntry::Error:
1354       return error("Malformed block");
1355     case BitstreamEntry::EndBlock:
1356       return std::error_code();
1357     case BitstreamEntry::Record:
1358       // The interesting case.
1359       break;
1360     }
1361 
1362     // Read a record.
1363     Record.clear();
1364     switch (Stream.readRecord(Entry.ID, Record)) {
1365     default:  // Default behavior: ignore.
1366       break;
1367     case bitc::PARAMATTR_GRP_CODE_ENTRY: { // ENTRY: [grpid, idx, a0, a1, ...]
1368       if (Record.size() < 3)
1369         return error("Invalid record");
1370 
1371       uint64_t GrpID = Record[0];
1372       uint64_t Idx = Record[1]; // Index of the object this attribute refers to.
1373 
1374       AttrBuilder B;
1375       for (unsigned i = 2, e = Record.size(); i != e; ++i) {
1376         if (Record[i] == 0) {        // Enum attribute
1377           Attribute::AttrKind Kind;
1378           if (std::error_code EC = parseAttrKind(Record[++i], &Kind))
1379             return EC;
1380 
1381           B.addAttribute(Kind);
1382         } else if (Record[i] == 1) { // Integer attribute
1383           Attribute::AttrKind Kind;
1384           if (std::error_code EC = parseAttrKind(Record[++i], &Kind))
1385             return EC;
1386           if (Kind == Attribute::Alignment)
1387             B.addAlignmentAttr(Record[++i]);
1388           else if (Kind == Attribute::StackAlignment)
1389             B.addStackAlignmentAttr(Record[++i]);
1390           else if (Kind == Attribute::Dereferenceable)
1391             B.addDereferenceableAttr(Record[++i]);
1392           else if (Kind == Attribute::DereferenceableOrNull)
1393             B.addDereferenceableOrNullAttr(Record[++i]);
1394         } else {                     // String attribute
1395           assert((Record[i] == 3 || Record[i] == 4) &&
1396                  "Invalid attribute group entry");
1397           bool HasValue = (Record[i++] == 4);
1398           SmallString<64> KindStr;
1399           SmallString<64> ValStr;
1400 
1401           while (Record[i] != 0 && i != e)
1402             KindStr += Record[i++];
1403           assert(Record[i] == 0 && "Kind string not null terminated");
1404 
1405           if (HasValue) {
1406             // Has a value associated with it.
1407             ++i; // Skip the '0' that terminates the "kind" string.
1408             while (Record[i] != 0 && i != e)
1409               ValStr += Record[i++];
1410             assert(Record[i] == 0 && "Value string not null terminated");
1411           }
1412 
1413           B.addAttribute(KindStr.str(), ValStr.str());
1414         }
1415       }
1416 
1417       MAttributeGroups[GrpID] = AttributeSet::get(Context, Idx, B);
1418       break;
1419     }
1420     }
1421   }
1422 }
1423 
parseTypeTable()1424 std::error_code BitcodeReader::parseTypeTable() {
1425   if (Stream.EnterSubBlock(bitc::TYPE_BLOCK_ID_NEW))
1426     return error("Invalid record");
1427 
1428   return parseTypeTableBody();
1429 }
1430 
parseTypeTableBody()1431 std::error_code BitcodeReader::parseTypeTableBody() {
1432   if (!TypeList.empty())
1433     return error("Invalid multiple blocks");
1434 
1435   SmallVector<uint64_t, 64> Record;
1436   unsigned NumRecords = 0;
1437 
1438   SmallString<64> TypeName;
1439 
1440   // Read all the records for this type table.
1441   while (1) {
1442     BitstreamEntry Entry = Stream.advanceSkippingSubblocks();
1443 
1444     switch (Entry.Kind) {
1445     case BitstreamEntry::SubBlock: // Handled for us already.
1446     case BitstreamEntry::Error:
1447       return error("Malformed block");
1448     case BitstreamEntry::EndBlock:
1449       if (NumRecords != TypeList.size())
1450         return error("Malformed block");
1451       return std::error_code();
1452     case BitstreamEntry::Record:
1453       // The interesting case.
1454       break;
1455     }
1456 
1457     // Read a record.
1458     Record.clear();
1459     Type *ResultTy = nullptr;
1460     switch (Stream.readRecord(Entry.ID, Record)) {
1461     default:
1462       return error("Invalid value");
1463     case bitc::TYPE_CODE_NUMENTRY: // TYPE_CODE_NUMENTRY: [numentries]
1464       // TYPE_CODE_NUMENTRY contains a count of the number of types in the
1465       // type list.  This allows us to reserve space.
1466       if (Record.size() < 1)
1467         return error("Invalid record");
1468       TypeList.resize(Record[0]);
1469       continue;
1470     case bitc::TYPE_CODE_VOID:      // VOID
1471       ResultTy = Type::getVoidTy(Context);
1472       break;
1473     case bitc::TYPE_CODE_HALF:     // HALF
1474       ResultTy = Type::getHalfTy(Context);
1475       break;
1476     case bitc::TYPE_CODE_FLOAT:     // FLOAT
1477       ResultTy = Type::getFloatTy(Context);
1478       break;
1479     case bitc::TYPE_CODE_DOUBLE:    // DOUBLE
1480       ResultTy = Type::getDoubleTy(Context);
1481       break;
1482     case bitc::TYPE_CODE_X86_FP80:  // X86_FP80
1483       ResultTy = Type::getX86_FP80Ty(Context);
1484       break;
1485     case bitc::TYPE_CODE_FP128:     // FP128
1486       ResultTy = Type::getFP128Ty(Context);
1487       break;
1488     case bitc::TYPE_CODE_PPC_FP128: // PPC_FP128
1489       ResultTy = Type::getPPC_FP128Ty(Context);
1490       break;
1491     case bitc::TYPE_CODE_LABEL:     // LABEL
1492       ResultTy = Type::getLabelTy(Context);
1493       break;
1494     case bitc::TYPE_CODE_METADATA:  // METADATA
1495       ResultTy = Type::getMetadataTy(Context);
1496       break;
1497     case bitc::TYPE_CODE_X86_MMX:   // X86_MMX
1498       ResultTy = Type::getX86_MMXTy(Context);
1499       break;
1500     case bitc::TYPE_CODE_TOKEN:     // TOKEN
1501       ResultTy = Type::getTokenTy(Context);
1502       break;
1503     case bitc::TYPE_CODE_INTEGER: { // INTEGER: [width]
1504       if (Record.size() < 1)
1505         return error("Invalid record");
1506 
1507       uint64_t NumBits = Record[0];
1508       if (NumBits < IntegerType::MIN_INT_BITS ||
1509           NumBits > IntegerType::MAX_INT_BITS)
1510         return error("Bitwidth for integer type out of range");
1511       ResultTy = IntegerType::get(Context, NumBits);
1512       break;
1513     }
1514     case bitc::TYPE_CODE_POINTER: { // POINTER: [pointee type] or
1515                                     //          [pointee type, address space]
1516       if (Record.size() < 1)
1517         return error("Invalid record");
1518       unsigned AddressSpace = 0;
1519       if (Record.size() == 2)
1520         AddressSpace = Record[1];
1521       ResultTy = getTypeByID(Record[0]);
1522       if (!ResultTy ||
1523           !PointerType::isValidElementType(ResultTy))
1524         return error("Invalid type");
1525       ResultTy = PointerType::get(ResultTy, AddressSpace);
1526       break;
1527     }
1528     case bitc::TYPE_CODE_FUNCTION_OLD: {
1529       // FIXME: attrid is dead, remove it in LLVM 4.0
1530       // FUNCTION: [vararg, attrid, retty, paramty x N]
1531       if (Record.size() < 3)
1532         return error("Invalid record");
1533       SmallVector<Type*, 8> ArgTys;
1534       for (unsigned i = 3, e = Record.size(); i != e; ++i) {
1535         if (Type *T = getTypeByID(Record[i]))
1536           ArgTys.push_back(T);
1537         else
1538           break;
1539       }
1540 
1541       ResultTy = getTypeByID(Record[2]);
1542       if (!ResultTy || ArgTys.size() < Record.size()-3)
1543         return error("Invalid type");
1544 
1545       ResultTy = FunctionType::get(ResultTy, ArgTys, Record[0]);
1546       break;
1547     }
1548     case bitc::TYPE_CODE_FUNCTION: {
1549       // FUNCTION: [vararg, retty, paramty x N]
1550       if (Record.size() < 2)
1551         return error("Invalid record");
1552       SmallVector<Type*, 8> ArgTys;
1553       for (unsigned i = 2, e = Record.size(); i != e; ++i) {
1554         if (Type *T = getTypeByID(Record[i])) {
1555           if (!FunctionType::isValidArgumentType(T))
1556             return error("Invalid function argument type");
1557           ArgTys.push_back(T);
1558         }
1559         else
1560           break;
1561       }
1562 
1563       ResultTy = getTypeByID(Record[1]);
1564       if (!ResultTy || ArgTys.size() < Record.size()-2)
1565         return error("Invalid type");
1566 
1567       ResultTy = FunctionType::get(ResultTy, ArgTys, Record[0]);
1568       break;
1569     }
1570     case bitc::TYPE_CODE_STRUCT_ANON: {  // STRUCT: [ispacked, eltty x N]
1571       if (Record.size() < 1)
1572         return error("Invalid record");
1573       SmallVector<Type*, 8> EltTys;
1574       for (unsigned i = 1, e = Record.size(); i != e; ++i) {
1575         if (Type *T = getTypeByID(Record[i]))
1576           EltTys.push_back(T);
1577         else
1578           break;
1579       }
1580       if (EltTys.size() != Record.size()-1)
1581         return error("Invalid type");
1582       ResultTy = StructType::get(Context, EltTys, Record[0]);
1583       break;
1584     }
1585     case bitc::TYPE_CODE_STRUCT_NAME:   // STRUCT_NAME: [strchr x N]
1586       if (convertToString(Record, 0, TypeName))
1587         return error("Invalid record");
1588       continue;
1589 
1590     case bitc::TYPE_CODE_STRUCT_NAMED: { // STRUCT: [ispacked, eltty x N]
1591       if (Record.size() < 1)
1592         return error("Invalid record");
1593 
1594       if (NumRecords >= TypeList.size())
1595         return error("Invalid TYPE table");
1596 
1597       // Check to see if this was forward referenced, if so fill in the temp.
1598       StructType *Res = cast_or_null<StructType>(TypeList[NumRecords]);
1599       if (Res) {
1600         Res->setName(TypeName);
1601         TypeList[NumRecords] = nullptr;
1602       } else  // Otherwise, create a new struct.
1603         Res = createIdentifiedStructType(Context, TypeName);
1604       TypeName.clear();
1605 
1606       SmallVector<Type*, 8> EltTys;
1607       for (unsigned i = 1, e = Record.size(); i != e; ++i) {
1608         if (Type *T = getTypeByID(Record[i]))
1609           EltTys.push_back(T);
1610         else
1611           break;
1612       }
1613       if (EltTys.size() != Record.size()-1)
1614         return error("Invalid record");
1615       Res->setBody(EltTys, Record[0]);
1616       ResultTy = Res;
1617       break;
1618     }
1619     case bitc::TYPE_CODE_OPAQUE: {       // OPAQUE: []
1620       if (Record.size() != 1)
1621         return error("Invalid record");
1622 
1623       if (NumRecords >= TypeList.size())
1624         return error("Invalid TYPE table");
1625 
1626       // Check to see if this was forward referenced, if so fill in the temp.
1627       StructType *Res = cast_or_null<StructType>(TypeList[NumRecords]);
1628       if (Res) {
1629         Res->setName(TypeName);
1630         TypeList[NumRecords] = nullptr;
1631       } else  // Otherwise, create a new struct with no body.
1632         Res = createIdentifiedStructType(Context, TypeName);
1633       TypeName.clear();
1634       ResultTy = Res;
1635       break;
1636     }
1637     case bitc::TYPE_CODE_ARRAY:     // ARRAY: [numelts, eltty]
1638       if (Record.size() < 2)
1639         return error("Invalid record");
1640       ResultTy = getTypeByID(Record[1]);
1641       if (!ResultTy || !ArrayType::isValidElementType(ResultTy))
1642         return error("Invalid type");
1643       ResultTy = ArrayType::get(ResultTy, Record[0]);
1644       break;
1645     case bitc::TYPE_CODE_VECTOR:    // VECTOR: [numelts, eltty]
1646       if (Record.size() < 2)
1647         return error("Invalid record");
1648       if (Record[0] == 0)
1649         return error("Invalid vector length");
1650       ResultTy = getTypeByID(Record[1]);
1651       if (!ResultTy || !StructType::isValidElementType(ResultTy))
1652         return error("Invalid type");
1653       ResultTy = VectorType::get(ResultTy, Record[0]);
1654       break;
1655     }
1656 
1657     if (NumRecords >= TypeList.size())
1658       return error("Invalid TYPE table");
1659     if (TypeList[NumRecords])
1660       return error(
1661           "Invalid TYPE table: Only named structs can be forward referenced");
1662     assert(ResultTy && "Didn't read a type?");
1663     TypeList[NumRecords++] = ResultTy;
1664   }
1665 }
1666 
parseOperandBundleTags()1667 std::error_code BitcodeReader::parseOperandBundleTags() {
1668   if (Stream.EnterSubBlock(bitc::OPERAND_BUNDLE_TAGS_BLOCK_ID))
1669     return error("Invalid record");
1670 
1671   if (!BundleTags.empty())
1672     return error("Invalid multiple blocks");
1673 
1674   SmallVector<uint64_t, 64> Record;
1675 
1676   while (1) {
1677     BitstreamEntry Entry = Stream.advanceSkippingSubblocks();
1678 
1679     switch (Entry.Kind) {
1680     case BitstreamEntry::SubBlock: // Handled for us already.
1681     case BitstreamEntry::Error:
1682       return error("Malformed block");
1683     case BitstreamEntry::EndBlock:
1684       return std::error_code();
1685     case BitstreamEntry::Record:
1686       // The interesting case.
1687       break;
1688     }
1689 
1690     // Tags are implicitly mapped to integers by their order.
1691 
1692     if (Stream.readRecord(Entry.ID, Record) != bitc::OPERAND_BUNDLE_TAG)
1693       return error("Invalid record");
1694 
1695     // OPERAND_BUNDLE_TAG: [strchr x N]
1696     BundleTags.emplace_back();
1697     if (convertToString(Record, 0, BundleTags.back()))
1698       return error("Invalid record");
1699     Record.clear();
1700   }
1701 }
1702 
1703 /// Associate a value with its name from the given index in the provided record.
recordValue(SmallVectorImpl<uint64_t> & Record,unsigned NameIndex,Triple & TT)1704 ErrorOr<Value *> BitcodeReader::recordValue(SmallVectorImpl<uint64_t> &Record,
1705                                             unsigned NameIndex, Triple &TT) {
1706   SmallString<128> ValueName;
1707   if (convertToString(Record, NameIndex, ValueName))
1708     return error("Invalid record");
1709   unsigned ValueID = Record[0];
1710   if (ValueID >= ValueList.size() || !ValueList[ValueID])
1711     return error("Invalid record");
1712   Value *V = ValueList[ValueID];
1713 
1714   StringRef NameStr(ValueName.data(), ValueName.size());
1715   if (NameStr.find_first_of(0) != StringRef::npos)
1716     return error("Invalid value name");
1717   V->setName(NameStr);
1718   auto *GO = dyn_cast<GlobalObject>(V);
1719   if (GO) {
1720     if (GO->getComdat() == reinterpret_cast<Comdat *>(1)) {
1721       if (TT.isOSBinFormatMachO())
1722         GO->setComdat(nullptr);
1723       else
1724         GO->setComdat(TheModule->getOrInsertComdat(V->getName()));
1725     }
1726   }
1727   return V;
1728 }
1729 
1730 /// Parse the value symbol table at either the current parsing location or
1731 /// at the given bit offset if provided.
parseValueSymbolTable(uint64_t Offset)1732 std::error_code BitcodeReader::parseValueSymbolTable(uint64_t Offset) {
1733   uint64_t CurrentBit;
1734   // Pass in the Offset to distinguish between calling for the module-level
1735   // VST (where we want to jump to the VST offset) and the function-level
1736   // VST (where we don't).
1737   if (Offset > 0) {
1738     // Save the current parsing location so we can jump back at the end
1739     // of the VST read.
1740     CurrentBit = Stream.GetCurrentBitNo();
1741     Stream.JumpToBit(Offset * 32);
1742 #ifndef NDEBUG
1743     // Do some checking if we are in debug mode.
1744     BitstreamEntry Entry = Stream.advance();
1745     assert(Entry.Kind == BitstreamEntry::SubBlock);
1746     assert(Entry.ID == bitc::VALUE_SYMTAB_BLOCK_ID);
1747 #else
1748     // In NDEBUG mode ignore the output so we don't get an unused variable
1749     // warning.
1750     Stream.advance();
1751 #endif
1752   }
1753 
1754   // Compute the delta between the bitcode indices in the VST (the word offset
1755   // to the word-aligned ENTER_SUBBLOCK for the function block, and that
1756   // expected by the lazy reader. The reader's EnterSubBlock expects to have
1757   // already read the ENTER_SUBBLOCK code (size getAbbrevIDWidth) and BlockID
1758   // (size BlockIDWidth). Note that we access the stream's AbbrevID width here
1759   // just before entering the VST subblock because: 1) the EnterSubBlock
1760   // changes the AbbrevID width; 2) the VST block is nested within the same
1761   // outer MODULE_BLOCK as the FUNCTION_BLOCKs and therefore have the same
1762   // AbbrevID width before calling EnterSubBlock; and 3) when we want to
1763   // jump to the FUNCTION_BLOCK using this offset later, we don't want
1764   // to rely on the stream's AbbrevID width being that of the MODULE_BLOCK.
1765   unsigned FuncBitcodeOffsetDelta =
1766       Stream.getAbbrevIDWidth() + bitc::BlockIDWidth;
1767 
1768   if (Stream.EnterSubBlock(bitc::VALUE_SYMTAB_BLOCK_ID))
1769     return error("Invalid record");
1770 
1771   SmallVector<uint64_t, 64> Record;
1772 
1773   Triple TT(TheModule->getTargetTriple());
1774 
1775   // Read all the records for this value table.
1776   SmallString<128> ValueName;
1777   while (1) {
1778     BitstreamEntry Entry = Stream.advanceSkippingSubblocks();
1779 
1780     switch (Entry.Kind) {
1781     case BitstreamEntry::SubBlock: // Handled for us already.
1782     case BitstreamEntry::Error:
1783       return error("Malformed block");
1784     case BitstreamEntry::EndBlock:
1785       if (Offset > 0)
1786         Stream.JumpToBit(CurrentBit);
1787       return std::error_code();
1788     case BitstreamEntry::Record:
1789       // The interesting case.
1790       break;
1791     }
1792 
1793     // Read a record.
1794     Record.clear();
1795     switch (Stream.readRecord(Entry.ID, Record)) {
1796     default:  // Default behavior: unknown type.
1797       break;
1798     case bitc::VST_CODE_ENTRY: {  // VST_ENTRY: [valueid, namechar x N]
1799       ErrorOr<Value *> ValOrErr = recordValue(Record, 1, TT);
1800       if (std::error_code EC = ValOrErr.getError())
1801         return EC;
1802       ValOrErr.get();
1803       break;
1804     }
1805     case bitc::VST_CODE_FNENTRY: {
1806       // VST_FNENTRY: [valueid, offset, namechar x N]
1807       ErrorOr<Value *> ValOrErr = recordValue(Record, 2, TT);
1808       if (std::error_code EC = ValOrErr.getError())
1809         return EC;
1810       Value *V = ValOrErr.get();
1811 
1812       auto *GO = dyn_cast<GlobalObject>(V);
1813       if (!GO) {
1814         // If this is an alias, need to get the actual Function object
1815         // it aliases, in order to set up the DeferredFunctionInfo entry below.
1816         auto *GA = dyn_cast<GlobalAlias>(V);
1817         if (GA)
1818           GO = GA->getBaseObject();
1819         assert(GO);
1820       }
1821 
1822       uint64_t FuncWordOffset = Record[1];
1823       Function *F = dyn_cast<Function>(GO);
1824       assert(F);
1825       uint64_t FuncBitOffset = FuncWordOffset * 32;
1826       DeferredFunctionInfo[F] = FuncBitOffset + FuncBitcodeOffsetDelta;
1827       // Set the LastFunctionBlockBit to point to the last function block.
1828       // Later when parsing is resumed after function materialization,
1829       // we can simply skip that last function block.
1830       if (FuncBitOffset > LastFunctionBlockBit)
1831         LastFunctionBlockBit = FuncBitOffset;
1832       break;
1833     }
1834     case bitc::VST_CODE_BBENTRY: {
1835       if (convertToString(Record, 1, ValueName))
1836         return error("Invalid record");
1837       BasicBlock *BB = getBasicBlock(Record[0]);
1838       if (!BB)
1839         return error("Invalid record");
1840 
1841       BB->setName(StringRef(ValueName.data(), ValueName.size()));
1842       ValueName.clear();
1843       break;
1844     }
1845     }
1846   }
1847 }
1848 
1849 /// Parse a single METADATA_KIND record, inserting result in MDKindMap.
1850 std::error_code
parseMetadataKindRecord(SmallVectorImpl<uint64_t> & Record)1851 BitcodeReader::parseMetadataKindRecord(SmallVectorImpl<uint64_t> &Record) {
1852   if (Record.size() < 2)
1853     return error("Invalid record");
1854 
1855   unsigned Kind = Record[0];
1856   SmallString<8> Name(Record.begin() + 1, Record.end());
1857 
1858   unsigned NewKind = TheModule->getMDKindID(Name.str());
1859   if (!MDKindMap.insert(std::make_pair(Kind, NewKind)).second)
1860     return error("Conflicting METADATA_KIND records");
1861   return std::error_code();
1862 }
1863 
unrotateSign(uint64_t U)1864 static int64_t unrotateSign(uint64_t U) { return U & 1 ? ~(U >> 1) : U >> 1; }
1865 
1866 /// Parse a METADATA_BLOCK. If ModuleLevel is true then we are parsing
1867 /// module level metadata.
parseMetadata(bool ModuleLevel)1868 std::error_code BitcodeReader::parseMetadata(bool ModuleLevel) {
1869   IsMetadataMaterialized = true;
1870   unsigned NextMDValueNo = MDValueList.size();
1871   if (ModuleLevel && SeenModuleValuesRecord) {
1872     // Now that we are parsing the module level metadata, we want to restart
1873     // the numbering of the MD values, and replace temp MD created earlier
1874     // with their real values. If we saw a METADATA_VALUE record then we
1875     // would have set the MDValueList size to the number specified in that
1876     // record, to support parsing function-level metadata first, and we need
1877     // to reset back to 0 to fill the MDValueList in with the parsed module
1878     // The function-level metadata parsing should have reset the MDValueList
1879     // size back to the value reported by the METADATA_VALUE record, saved in
1880     // NumModuleMDs.
1881     assert(NumModuleMDs == MDValueList.size() &&
1882            "Expected MDValueList to only contain module level values");
1883     NextMDValueNo = 0;
1884   }
1885 
1886   if (Stream.EnterSubBlock(bitc::METADATA_BLOCK_ID))
1887     return error("Invalid record");
1888 
1889   SmallVector<uint64_t, 64> Record;
1890 
1891   auto getMD =
1892       [&](unsigned ID) -> Metadata *{ return MDValueList.getValueFwdRef(ID); };
1893   auto getMDOrNull = [&](unsigned ID) -> Metadata *{
1894     if (ID)
1895       return getMD(ID - 1);
1896     return nullptr;
1897   };
1898   auto getMDString = [&](unsigned ID) -> MDString *{
1899     // This requires that the ID is not really a forward reference.  In
1900     // particular, the MDString must already have been resolved.
1901     return cast_or_null<MDString>(getMDOrNull(ID));
1902   };
1903 
1904 #define GET_OR_DISTINCT(CLASS, DISTINCT, ARGS)                                 \
1905   (DISTINCT ? CLASS::getDistinct ARGS : CLASS::get ARGS)
1906 
1907   // Read all the records.
1908   while (1) {
1909     BitstreamEntry Entry = Stream.advanceSkippingSubblocks();
1910 
1911     switch (Entry.Kind) {
1912     case BitstreamEntry::SubBlock: // Handled for us already.
1913     case BitstreamEntry::Error:
1914       return error("Malformed block");
1915     case BitstreamEntry::EndBlock:
1916       MDValueList.tryToResolveCycles();
1917       assert((!(ModuleLevel && SeenModuleValuesRecord) ||
1918               NumModuleMDs == MDValueList.size()) &&
1919              "Inconsistent bitcode: METADATA_VALUES mismatch");
1920       return std::error_code();
1921     case BitstreamEntry::Record:
1922       // The interesting case.
1923       break;
1924     }
1925 
1926     // Read a record.
1927     Record.clear();
1928     unsigned Code = Stream.readRecord(Entry.ID, Record);
1929     bool IsDistinct = false;
1930     switch (Code) {
1931     default:  // Default behavior: ignore.
1932       break;
1933     case bitc::METADATA_NAME: {
1934       // Read name of the named metadata.
1935       SmallString<8> Name(Record.begin(), Record.end());
1936       Record.clear();
1937       Code = Stream.ReadCode();
1938 
1939       unsigned NextBitCode = Stream.readRecord(Code, Record);
1940       if (NextBitCode != bitc::METADATA_NAMED_NODE)
1941         return error("METADATA_NAME not followed by METADATA_NAMED_NODE");
1942 
1943       // Read named metadata elements.
1944       unsigned Size = Record.size();
1945       NamedMDNode *NMD = TheModule->getOrInsertNamedMetadata(Name);
1946       for (unsigned i = 0; i != Size; ++i) {
1947         MDNode *MD = dyn_cast_or_null<MDNode>(MDValueList.getValueFwdRef(Record[i]));
1948         if (!MD)
1949           return error("Invalid record");
1950         NMD->addOperand(MD);
1951       }
1952       break;
1953     }
1954     case bitc::METADATA_OLD_FN_NODE: {
1955       // FIXME: Remove in 4.0.
1956       // This is a LocalAsMetadata record, the only type of function-local
1957       // metadata.
1958       if (Record.size() % 2 == 1)
1959         return error("Invalid record");
1960 
1961       // If this isn't a LocalAsMetadata record, we're dropping it.  This used
1962       // to be legal, but there's no upgrade path.
1963       auto dropRecord = [&] {
1964         MDValueList.assignValue(MDNode::get(Context, None), NextMDValueNo++);
1965       };
1966       if (Record.size() != 2) {
1967         dropRecord();
1968         break;
1969       }
1970 
1971       Type *Ty = getTypeByID(Record[0]);
1972       if (Ty->isMetadataTy() || Ty->isVoidTy()) {
1973         dropRecord();
1974         break;
1975       }
1976 
1977       MDValueList.assignValue(
1978           LocalAsMetadata::get(ValueList.getValueFwdRef(Record[1], Ty)),
1979           NextMDValueNo++);
1980       break;
1981     }
1982     case bitc::METADATA_OLD_NODE: {
1983       // FIXME: Remove in 4.0.
1984       if (Record.size() % 2 == 1)
1985         return error("Invalid record");
1986 
1987       unsigned Size = Record.size();
1988       SmallVector<Metadata *, 8> Elts;
1989       for (unsigned i = 0; i != Size; i += 2) {
1990         Type *Ty = getTypeByID(Record[i]);
1991         if (!Ty)
1992           return error("Invalid record");
1993         if (Ty->isMetadataTy())
1994           Elts.push_back(MDValueList.getValueFwdRef(Record[i+1]));
1995         else if (!Ty->isVoidTy()) {
1996           auto *MD =
1997               ValueAsMetadata::get(ValueList.getValueFwdRef(Record[i + 1], Ty));
1998           assert(isa<ConstantAsMetadata>(MD) &&
1999                  "Expected non-function-local metadata");
2000           Elts.push_back(MD);
2001         } else
2002           Elts.push_back(nullptr);
2003       }
2004       MDValueList.assignValue(MDNode::get(Context, Elts), NextMDValueNo++);
2005       break;
2006     }
2007     case bitc::METADATA_VALUE: {
2008       if (Record.size() != 2)
2009         return error("Invalid record");
2010 
2011       Type *Ty = getTypeByID(Record[0]);
2012       if (Ty->isMetadataTy() || Ty->isVoidTy())
2013         return error("Invalid record");
2014 
2015       MDValueList.assignValue(
2016           ValueAsMetadata::get(ValueList.getValueFwdRef(Record[1], Ty)),
2017           NextMDValueNo++);
2018       break;
2019     }
2020     case bitc::METADATA_DISTINCT_NODE:
2021       IsDistinct = true;
2022       // fallthrough...
2023     case bitc::METADATA_NODE: {
2024       SmallVector<Metadata *, 8> Elts;
2025       Elts.reserve(Record.size());
2026       for (unsigned ID : Record)
2027         Elts.push_back(ID ? MDValueList.getValueFwdRef(ID - 1) : nullptr);
2028       MDValueList.assignValue(IsDistinct ? MDNode::getDistinct(Context, Elts)
2029                                          : MDNode::get(Context, Elts),
2030                               NextMDValueNo++);
2031       break;
2032     }
2033     case bitc::METADATA_LOCATION: {
2034       if (Record.size() != 5)
2035         return error("Invalid record");
2036 
2037       unsigned Line = Record[1];
2038       unsigned Column = Record[2];
2039       MDNode *Scope = cast<MDNode>(MDValueList.getValueFwdRef(Record[3]));
2040       Metadata *InlinedAt =
2041           Record[4] ? MDValueList.getValueFwdRef(Record[4] - 1) : nullptr;
2042       MDValueList.assignValue(
2043           GET_OR_DISTINCT(DILocation, Record[0],
2044                           (Context, Line, Column, Scope, InlinedAt)),
2045           NextMDValueNo++);
2046       break;
2047     }
2048     case bitc::METADATA_GENERIC_DEBUG: {
2049       if (Record.size() < 4)
2050         return error("Invalid record");
2051 
2052       unsigned Tag = Record[1];
2053       unsigned Version = Record[2];
2054 
2055       if (Tag >= 1u << 16 || Version != 0)
2056         return error("Invalid record");
2057 
2058       auto *Header = getMDString(Record[3]);
2059       SmallVector<Metadata *, 8> DwarfOps;
2060       for (unsigned I = 4, E = Record.size(); I != E; ++I)
2061         DwarfOps.push_back(Record[I] ? MDValueList.getValueFwdRef(Record[I] - 1)
2062                                      : nullptr);
2063       MDValueList.assignValue(GET_OR_DISTINCT(GenericDINode, Record[0],
2064                                               (Context, Tag, Header, DwarfOps)),
2065                               NextMDValueNo++);
2066       break;
2067     }
2068     case bitc::METADATA_SUBRANGE: {
2069       if (Record.size() != 3)
2070         return error("Invalid record");
2071 
2072       MDValueList.assignValue(
2073           GET_OR_DISTINCT(DISubrange, Record[0],
2074                           (Context, Record[1], unrotateSign(Record[2]))),
2075           NextMDValueNo++);
2076       break;
2077     }
2078     case bitc::METADATA_ENUMERATOR: {
2079       if (Record.size() != 3)
2080         return error("Invalid record");
2081 
2082       MDValueList.assignValue(GET_OR_DISTINCT(DIEnumerator, Record[0],
2083                                               (Context, unrotateSign(Record[1]),
2084                                                getMDString(Record[2]))),
2085                               NextMDValueNo++);
2086       break;
2087     }
2088     case bitc::METADATA_BASIC_TYPE: {
2089       if (Record.size() != 6)
2090         return error("Invalid record");
2091 
2092       MDValueList.assignValue(
2093           GET_OR_DISTINCT(DIBasicType, Record[0],
2094                           (Context, Record[1], getMDString(Record[2]),
2095                            Record[3], Record[4], Record[5])),
2096           NextMDValueNo++);
2097       break;
2098     }
2099     case bitc::METADATA_DERIVED_TYPE: {
2100       if (Record.size() != 12)
2101         return error("Invalid record");
2102 
2103       MDValueList.assignValue(
2104           GET_OR_DISTINCT(DIDerivedType, Record[0],
2105                           (Context, Record[1], getMDString(Record[2]),
2106                            getMDOrNull(Record[3]), Record[4],
2107                            getMDOrNull(Record[5]), getMDOrNull(Record[6]),
2108                            Record[7], Record[8], Record[9], Record[10],
2109                            getMDOrNull(Record[11]))),
2110           NextMDValueNo++);
2111       break;
2112     }
2113     case bitc::METADATA_COMPOSITE_TYPE: {
2114       if (Record.size() != 16)
2115         return error("Invalid record");
2116 
2117       MDValueList.assignValue(
2118           GET_OR_DISTINCT(DICompositeType, Record[0],
2119                           (Context, Record[1], getMDString(Record[2]),
2120                            getMDOrNull(Record[3]), Record[4],
2121                            getMDOrNull(Record[5]), getMDOrNull(Record[6]),
2122                            Record[7], Record[8], Record[9], Record[10],
2123                            getMDOrNull(Record[11]), Record[12],
2124                            getMDOrNull(Record[13]), getMDOrNull(Record[14]),
2125                            getMDString(Record[15]))),
2126           NextMDValueNo++);
2127       break;
2128     }
2129     case bitc::METADATA_SUBROUTINE_TYPE: {
2130       if (Record.size() != 3)
2131         return error("Invalid record");
2132 
2133       MDValueList.assignValue(
2134           GET_OR_DISTINCT(DISubroutineType, Record[0],
2135                           (Context, Record[1], getMDOrNull(Record[2]))),
2136           NextMDValueNo++);
2137       break;
2138     }
2139 
2140     case bitc::METADATA_MODULE: {
2141       if (Record.size() != 6)
2142         return error("Invalid record");
2143 
2144       MDValueList.assignValue(
2145           GET_OR_DISTINCT(DIModule, Record[0],
2146                           (Context, getMDOrNull(Record[1]),
2147                           getMDString(Record[2]), getMDString(Record[3]),
2148                           getMDString(Record[4]), getMDString(Record[5]))),
2149           NextMDValueNo++);
2150       break;
2151     }
2152 
2153     case bitc::METADATA_FILE: {
2154       if (Record.size() != 3)
2155         return error("Invalid record");
2156 
2157       MDValueList.assignValue(
2158           GET_OR_DISTINCT(DIFile, Record[0], (Context, getMDString(Record[1]),
2159                                               getMDString(Record[2]))),
2160           NextMDValueNo++);
2161       break;
2162     }
2163     case bitc::METADATA_COMPILE_UNIT: {
2164       if (Record.size() < 14 || Record.size() > 16)
2165         return error("Invalid record");
2166 
2167       // Ignore Record[0], which indicates whether this compile unit is
2168       // distinct.  It's always distinct.
2169       MDValueList.assignValue(
2170           DICompileUnit::getDistinct(
2171               Context, Record[1], getMDOrNull(Record[2]),
2172               getMDString(Record[3]), Record[4], getMDString(Record[5]),
2173               Record[6], getMDString(Record[7]), Record[8],
2174               getMDOrNull(Record[9]), getMDOrNull(Record[10]),
2175               getMDOrNull(Record[11]), getMDOrNull(Record[12]),
2176               getMDOrNull(Record[13]),
2177               Record.size() <= 15 ? 0 : getMDOrNull(Record[15]),
2178               Record.size() <= 14 ? 0 : Record[14]),
2179           NextMDValueNo++);
2180       break;
2181     }
2182     case bitc::METADATA_SUBPROGRAM: {
2183       if (Record.size() != 18 && Record.size() != 19)
2184         return error("Invalid record");
2185 
2186       bool HasFn = Record.size() == 19;
2187       DISubprogram *SP = GET_OR_DISTINCT(
2188           DISubprogram,
2189           Record[0] || Record[8], // All definitions should be distinct.
2190           (Context, getMDOrNull(Record[1]), getMDString(Record[2]),
2191            getMDString(Record[3]), getMDOrNull(Record[4]), Record[5],
2192            getMDOrNull(Record[6]), Record[7], Record[8], Record[9],
2193            getMDOrNull(Record[10]), Record[11], Record[12], Record[13],
2194            Record[14], getMDOrNull(Record[15 + HasFn]),
2195            getMDOrNull(Record[16 + HasFn]), getMDOrNull(Record[17 + HasFn])));
2196       MDValueList.assignValue(SP, NextMDValueNo++);
2197 
2198       // Upgrade sp->function mapping to function->sp mapping.
2199       if (HasFn && Record[15]) {
2200         if (auto *CMD = dyn_cast<ConstantAsMetadata>(getMDOrNull(Record[15])))
2201           if (auto *F = dyn_cast<Function>(CMD->getValue())) {
2202             if (F->isMaterializable())
2203               // Defer until materialized; unmaterialized functions may not have
2204               // metadata.
2205               FunctionsWithSPs[F] = SP;
2206             else if (!F->empty())
2207               F->setSubprogram(SP);
2208           }
2209       }
2210       break;
2211     }
2212     case bitc::METADATA_LEXICAL_BLOCK: {
2213       if (Record.size() != 5)
2214         return error("Invalid record");
2215 
2216       MDValueList.assignValue(
2217           GET_OR_DISTINCT(DILexicalBlock, Record[0],
2218                           (Context, getMDOrNull(Record[1]),
2219                            getMDOrNull(Record[2]), Record[3], Record[4])),
2220           NextMDValueNo++);
2221       break;
2222     }
2223     case bitc::METADATA_LEXICAL_BLOCK_FILE: {
2224       if (Record.size() != 4)
2225         return error("Invalid record");
2226 
2227       MDValueList.assignValue(
2228           GET_OR_DISTINCT(DILexicalBlockFile, Record[0],
2229                           (Context, getMDOrNull(Record[1]),
2230                            getMDOrNull(Record[2]), Record[3])),
2231           NextMDValueNo++);
2232       break;
2233     }
2234     case bitc::METADATA_NAMESPACE: {
2235       if (Record.size() != 5)
2236         return error("Invalid record");
2237 
2238       MDValueList.assignValue(
2239           GET_OR_DISTINCT(DINamespace, Record[0],
2240                           (Context, getMDOrNull(Record[1]),
2241                            getMDOrNull(Record[2]), getMDString(Record[3]),
2242                            Record[4])),
2243           NextMDValueNo++);
2244       break;
2245     }
2246     case bitc::METADATA_MACRO: {
2247       if (Record.size() != 5)
2248         return error("Invalid record");
2249 
2250       MDValueList.assignValue(
2251           GET_OR_DISTINCT(DIMacro, Record[0],
2252                           (Context, Record[1], Record[2],
2253                            getMDString(Record[3]), getMDString(Record[4]))),
2254           NextMDValueNo++);
2255       break;
2256     }
2257     case bitc::METADATA_MACRO_FILE: {
2258       if (Record.size() != 5)
2259         return error("Invalid record");
2260 
2261       MDValueList.assignValue(
2262           GET_OR_DISTINCT(DIMacroFile, Record[0],
2263                           (Context, Record[1], Record[2],
2264                            getMDOrNull(Record[3]), getMDOrNull(Record[4]))),
2265           NextMDValueNo++);
2266       break;
2267     }
2268     case bitc::METADATA_TEMPLATE_TYPE: {
2269       if (Record.size() != 3)
2270         return error("Invalid record");
2271 
2272       MDValueList.assignValue(GET_OR_DISTINCT(DITemplateTypeParameter,
2273                                               Record[0],
2274                                               (Context, getMDString(Record[1]),
2275                                                getMDOrNull(Record[2]))),
2276                               NextMDValueNo++);
2277       break;
2278     }
2279     case bitc::METADATA_TEMPLATE_VALUE: {
2280       if (Record.size() != 5)
2281         return error("Invalid record");
2282 
2283       MDValueList.assignValue(
2284           GET_OR_DISTINCT(DITemplateValueParameter, Record[0],
2285                           (Context, Record[1], getMDString(Record[2]),
2286                            getMDOrNull(Record[3]), getMDOrNull(Record[4]))),
2287           NextMDValueNo++);
2288       break;
2289     }
2290     case bitc::METADATA_GLOBAL_VAR: {
2291       if (Record.size() != 11)
2292         return error("Invalid record");
2293 
2294       MDValueList.assignValue(
2295           GET_OR_DISTINCT(DIGlobalVariable, Record[0],
2296                           (Context, getMDOrNull(Record[1]),
2297                            getMDString(Record[2]), getMDString(Record[3]),
2298                            getMDOrNull(Record[4]), Record[5],
2299                            getMDOrNull(Record[6]), Record[7], Record[8],
2300                            getMDOrNull(Record[9]), getMDOrNull(Record[10]))),
2301           NextMDValueNo++);
2302       break;
2303     }
2304     case bitc::METADATA_LOCAL_VAR: {
2305       // 10th field is for the obseleted 'inlinedAt:' field.
2306       if (Record.size() < 8 || Record.size() > 10)
2307         return error("Invalid record");
2308 
2309       // 2nd field used to be an artificial tag, either DW_TAG_auto_variable or
2310       // DW_TAG_arg_variable.
2311       bool HasTag = Record.size() > 8;
2312       MDValueList.assignValue(
2313           GET_OR_DISTINCT(DILocalVariable, Record[0],
2314                           (Context, getMDOrNull(Record[1 + HasTag]),
2315                            getMDString(Record[2 + HasTag]),
2316                            getMDOrNull(Record[3 + HasTag]), Record[4 + HasTag],
2317                            getMDOrNull(Record[5 + HasTag]), Record[6 + HasTag],
2318                            Record[7 + HasTag])),
2319           NextMDValueNo++);
2320       break;
2321     }
2322     case bitc::METADATA_EXPRESSION: {
2323       if (Record.size() < 1)
2324         return error("Invalid record");
2325 
2326       MDValueList.assignValue(
2327           GET_OR_DISTINCT(DIExpression, Record[0],
2328                           (Context, makeArrayRef(Record).slice(1))),
2329           NextMDValueNo++);
2330       break;
2331     }
2332     case bitc::METADATA_OBJC_PROPERTY: {
2333       if (Record.size() != 8)
2334         return error("Invalid record");
2335 
2336       MDValueList.assignValue(
2337           GET_OR_DISTINCT(DIObjCProperty, Record[0],
2338                           (Context, getMDString(Record[1]),
2339                            getMDOrNull(Record[2]), Record[3],
2340                            getMDString(Record[4]), getMDString(Record[5]),
2341                            Record[6], getMDOrNull(Record[7]))),
2342           NextMDValueNo++);
2343       break;
2344     }
2345     case bitc::METADATA_IMPORTED_ENTITY: {
2346       if (Record.size() != 6)
2347         return error("Invalid record");
2348 
2349       MDValueList.assignValue(
2350           GET_OR_DISTINCT(DIImportedEntity, Record[0],
2351                           (Context, Record[1], getMDOrNull(Record[2]),
2352                            getMDOrNull(Record[3]), Record[4],
2353                            getMDString(Record[5]))),
2354           NextMDValueNo++);
2355       break;
2356     }
2357     case bitc::METADATA_STRING: {
2358       std::string String(Record.begin(), Record.end());
2359       llvm::UpgradeMDStringConstant(String);
2360       Metadata *MD = MDString::get(Context, String);
2361       MDValueList.assignValue(MD, NextMDValueNo++);
2362       break;
2363     }
2364     case bitc::METADATA_KIND: {
2365       // Support older bitcode files that had METADATA_KIND records in a
2366       // block with METADATA_BLOCK_ID.
2367       if (std::error_code EC = parseMetadataKindRecord(Record))
2368         return EC;
2369       break;
2370     }
2371     }
2372   }
2373 #undef GET_OR_DISTINCT
2374 }
2375 
2376 /// Parse the metadata kinds out of the METADATA_KIND_BLOCK.
parseMetadataKinds()2377 std::error_code BitcodeReader::parseMetadataKinds() {
2378   if (Stream.EnterSubBlock(bitc::METADATA_KIND_BLOCK_ID))
2379     return error("Invalid record");
2380 
2381   SmallVector<uint64_t, 64> Record;
2382 
2383   // Read all the records.
2384   while (1) {
2385     BitstreamEntry Entry = Stream.advanceSkippingSubblocks();
2386 
2387     switch (Entry.Kind) {
2388     case BitstreamEntry::SubBlock: // Handled for us already.
2389     case BitstreamEntry::Error:
2390       return error("Malformed block");
2391     case BitstreamEntry::EndBlock:
2392       return std::error_code();
2393     case BitstreamEntry::Record:
2394       // The interesting case.
2395       break;
2396     }
2397 
2398     // Read a record.
2399     Record.clear();
2400     unsigned Code = Stream.readRecord(Entry.ID, Record);
2401     switch (Code) {
2402     default: // Default behavior: ignore.
2403       break;
2404     case bitc::METADATA_KIND: {
2405       if (std::error_code EC = parseMetadataKindRecord(Record))
2406         return EC;
2407       break;
2408     }
2409     }
2410   }
2411 }
2412 
2413 /// Decode a signed value stored with the sign bit in the LSB for dense VBR
2414 /// encoding.
decodeSignRotatedValue(uint64_t V)2415 uint64_t BitcodeReader::decodeSignRotatedValue(uint64_t V) {
2416   if ((V & 1) == 0)
2417     return V >> 1;
2418   if (V != 1)
2419     return -(V >> 1);
2420   // There is no such thing as -0 with integers.  "-0" really means MININT.
2421   return 1ULL << 63;
2422 }
2423 
2424 /// Resolve all of the initializers for global values and aliases that we can.
resolveGlobalAndAliasInits()2425 std::error_code BitcodeReader::resolveGlobalAndAliasInits() {
2426   std::vector<std::pair<GlobalVariable*, unsigned> > GlobalInitWorklist;
2427   std::vector<std::pair<GlobalAlias*, unsigned> > AliasInitWorklist;
2428   std::vector<std::pair<Function*, unsigned> > FunctionPrefixWorklist;
2429   std::vector<std::pair<Function*, unsigned> > FunctionPrologueWorklist;
2430   std::vector<std::pair<Function*, unsigned> > FunctionPersonalityFnWorklist;
2431 
2432   GlobalInitWorklist.swap(GlobalInits);
2433   AliasInitWorklist.swap(AliasInits);
2434   FunctionPrefixWorklist.swap(FunctionPrefixes);
2435   FunctionPrologueWorklist.swap(FunctionPrologues);
2436   FunctionPersonalityFnWorklist.swap(FunctionPersonalityFns);
2437 
2438   while (!GlobalInitWorklist.empty()) {
2439     unsigned ValID = GlobalInitWorklist.back().second;
2440     if (ValID >= ValueList.size()) {
2441       // Not ready to resolve this yet, it requires something later in the file.
2442       GlobalInits.push_back(GlobalInitWorklist.back());
2443     } else {
2444       if (Constant *C = dyn_cast_or_null<Constant>(ValueList[ValID]))
2445         GlobalInitWorklist.back().first->setInitializer(C);
2446       else
2447         return error("Expected a constant");
2448     }
2449     GlobalInitWorklist.pop_back();
2450   }
2451 
2452   while (!AliasInitWorklist.empty()) {
2453     unsigned ValID = AliasInitWorklist.back().second;
2454     if (ValID >= ValueList.size()) {
2455       AliasInits.push_back(AliasInitWorklist.back());
2456     } else {
2457       Constant *C = dyn_cast_or_null<Constant>(ValueList[ValID]);
2458       if (!C)
2459         return error("Expected a constant");
2460       GlobalAlias *Alias = AliasInitWorklist.back().first;
2461       if (C->getType() != Alias->getType())
2462         return error("Alias and aliasee types don't match");
2463       Alias->setAliasee(C);
2464     }
2465     AliasInitWorklist.pop_back();
2466   }
2467 
2468   while (!FunctionPrefixWorklist.empty()) {
2469     unsigned ValID = FunctionPrefixWorklist.back().second;
2470     if (ValID >= ValueList.size()) {
2471       FunctionPrefixes.push_back(FunctionPrefixWorklist.back());
2472     } else {
2473       if (Constant *C = dyn_cast_or_null<Constant>(ValueList[ValID]))
2474         FunctionPrefixWorklist.back().first->setPrefixData(C);
2475       else
2476         return error("Expected a constant");
2477     }
2478     FunctionPrefixWorklist.pop_back();
2479   }
2480 
2481   while (!FunctionPrologueWorklist.empty()) {
2482     unsigned ValID = FunctionPrologueWorklist.back().second;
2483     if (ValID >= ValueList.size()) {
2484       FunctionPrologues.push_back(FunctionPrologueWorklist.back());
2485     } else {
2486       if (Constant *C = dyn_cast_or_null<Constant>(ValueList[ValID]))
2487         FunctionPrologueWorklist.back().first->setPrologueData(C);
2488       else
2489         return error("Expected a constant");
2490     }
2491     FunctionPrologueWorklist.pop_back();
2492   }
2493 
2494   while (!FunctionPersonalityFnWorklist.empty()) {
2495     unsigned ValID = FunctionPersonalityFnWorklist.back().second;
2496     if (ValID >= ValueList.size()) {
2497       FunctionPersonalityFns.push_back(FunctionPersonalityFnWorklist.back());
2498     } else {
2499       if (Constant *C = dyn_cast_or_null<Constant>(ValueList[ValID]))
2500         FunctionPersonalityFnWorklist.back().first->setPersonalityFn(C);
2501       else
2502         return error("Expected a constant");
2503     }
2504     FunctionPersonalityFnWorklist.pop_back();
2505   }
2506 
2507   return std::error_code();
2508 }
2509 
readWideAPInt(ArrayRef<uint64_t> Vals,unsigned TypeBits)2510 static APInt readWideAPInt(ArrayRef<uint64_t> Vals, unsigned TypeBits) {
2511   SmallVector<uint64_t, 8> Words(Vals.size());
2512   std::transform(Vals.begin(), Vals.end(), Words.begin(),
2513                  BitcodeReader::decodeSignRotatedValue);
2514 
2515   return APInt(TypeBits, Words);
2516 }
2517 
parseConstants()2518 std::error_code BitcodeReader::parseConstants() {
2519   if (Stream.EnterSubBlock(bitc::CONSTANTS_BLOCK_ID))
2520     return error("Invalid record");
2521 
2522   SmallVector<uint64_t, 64> Record;
2523 
2524   // Read all the records for this value table.
2525   Type *CurTy = Type::getInt32Ty(Context);
2526   unsigned NextCstNo = ValueList.size();
2527   while (1) {
2528     BitstreamEntry Entry = Stream.advanceSkippingSubblocks();
2529 
2530     switch (Entry.Kind) {
2531     case BitstreamEntry::SubBlock: // Handled for us already.
2532     case BitstreamEntry::Error:
2533       return error("Malformed block");
2534     case BitstreamEntry::EndBlock:
2535       if (NextCstNo != ValueList.size())
2536         return error("Invalid ronstant reference");
2537 
2538       // Once all the constants have been read, go through and resolve forward
2539       // references.
2540       ValueList.resolveConstantForwardRefs();
2541       return std::error_code();
2542     case BitstreamEntry::Record:
2543       // The interesting case.
2544       break;
2545     }
2546 
2547     // Read a record.
2548     Record.clear();
2549     Value *V = nullptr;
2550     unsigned BitCode = Stream.readRecord(Entry.ID, Record);
2551     switch (BitCode) {
2552     default:  // Default behavior: unknown constant
2553     case bitc::CST_CODE_UNDEF:     // UNDEF
2554       V = UndefValue::get(CurTy);
2555       break;
2556     case bitc::CST_CODE_SETTYPE:   // SETTYPE: [typeid]
2557       if (Record.empty())
2558         return error("Invalid record");
2559       if (Record[0] >= TypeList.size() || !TypeList[Record[0]])
2560         return error("Invalid record");
2561       CurTy = TypeList[Record[0]];
2562       continue;  // Skip the ValueList manipulation.
2563     case bitc::CST_CODE_NULL:      // NULL
2564       V = Constant::getNullValue(CurTy);
2565       break;
2566     case bitc::CST_CODE_INTEGER:   // INTEGER: [intval]
2567       if (!CurTy->isIntegerTy() || Record.empty())
2568         return error("Invalid record");
2569       V = ConstantInt::get(CurTy, decodeSignRotatedValue(Record[0]));
2570       break;
2571     case bitc::CST_CODE_WIDE_INTEGER: {// WIDE_INTEGER: [n x intval]
2572       if (!CurTy->isIntegerTy() || Record.empty())
2573         return error("Invalid record");
2574 
2575       APInt VInt =
2576           readWideAPInt(Record, cast<IntegerType>(CurTy)->getBitWidth());
2577       V = ConstantInt::get(Context, VInt);
2578 
2579       break;
2580     }
2581     case bitc::CST_CODE_FLOAT: {    // FLOAT: [fpval]
2582       if (Record.empty())
2583         return error("Invalid record");
2584       if (CurTy->isHalfTy())
2585         V = ConstantFP::get(Context, APFloat(APFloat::IEEEhalf,
2586                                              APInt(16, (uint16_t)Record[0])));
2587       else if (CurTy->isFloatTy())
2588         V = ConstantFP::get(Context, APFloat(APFloat::IEEEsingle,
2589                                              APInt(32, (uint32_t)Record[0])));
2590       else if (CurTy->isDoubleTy())
2591         V = ConstantFP::get(Context, APFloat(APFloat::IEEEdouble,
2592                                              APInt(64, Record[0])));
2593       else if (CurTy->isX86_FP80Ty()) {
2594         // Bits are not stored the same way as a normal i80 APInt, compensate.
2595         uint64_t Rearrange[2];
2596         Rearrange[0] = (Record[1] & 0xffffLL) | (Record[0] << 16);
2597         Rearrange[1] = Record[0] >> 48;
2598         V = ConstantFP::get(Context, APFloat(APFloat::x87DoubleExtended,
2599                                              APInt(80, Rearrange)));
2600       } else if (CurTy->isFP128Ty())
2601         V = ConstantFP::get(Context, APFloat(APFloat::IEEEquad,
2602                                              APInt(128, Record)));
2603       else if (CurTy->isPPC_FP128Ty())
2604         V = ConstantFP::get(Context, APFloat(APFloat::PPCDoubleDouble,
2605                                              APInt(128, Record)));
2606       else
2607         V = UndefValue::get(CurTy);
2608       break;
2609     }
2610 
2611     case bitc::CST_CODE_AGGREGATE: {// AGGREGATE: [n x value number]
2612       if (Record.empty())
2613         return error("Invalid record");
2614 
2615       unsigned Size = Record.size();
2616       SmallVector<Constant*, 16> Elts;
2617 
2618       if (StructType *STy = dyn_cast<StructType>(CurTy)) {
2619         for (unsigned i = 0; i != Size; ++i)
2620           Elts.push_back(ValueList.getConstantFwdRef(Record[i],
2621                                                      STy->getElementType(i)));
2622         V = ConstantStruct::get(STy, Elts);
2623       } else if (ArrayType *ATy = dyn_cast<ArrayType>(CurTy)) {
2624         Type *EltTy = ATy->getElementType();
2625         for (unsigned i = 0; i != Size; ++i)
2626           Elts.push_back(ValueList.getConstantFwdRef(Record[i], EltTy));
2627         V = ConstantArray::get(ATy, Elts);
2628       } else if (VectorType *VTy = dyn_cast<VectorType>(CurTy)) {
2629         Type *EltTy = VTy->getElementType();
2630         for (unsigned i = 0; i != Size; ++i)
2631           Elts.push_back(ValueList.getConstantFwdRef(Record[i], EltTy));
2632         V = ConstantVector::get(Elts);
2633       } else {
2634         V = UndefValue::get(CurTy);
2635       }
2636       break;
2637     }
2638     case bitc::CST_CODE_STRING:    // STRING: [values]
2639     case bitc::CST_CODE_CSTRING: { // CSTRING: [values]
2640       if (Record.empty())
2641         return error("Invalid record");
2642 
2643       SmallString<16> Elts(Record.begin(), Record.end());
2644       V = ConstantDataArray::getString(Context, Elts,
2645                                        BitCode == bitc::CST_CODE_CSTRING);
2646       break;
2647     }
2648     case bitc::CST_CODE_DATA: {// DATA: [n x value]
2649       if (Record.empty())
2650         return error("Invalid record");
2651 
2652       Type *EltTy = cast<SequentialType>(CurTy)->getElementType();
2653       if (EltTy->isIntegerTy(8)) {
2654         SmallVector<uint8_t, 16> Elts(Record.begin(), Record.end());
2655         if (isa<VectorType>(CurTy))
2656           V = ConstantDataVector::get(Context, Elts);
2657         else
2658           V = ConstantDataArray::get(Context, Elts);
2659       } else if (EltTy->isIntegerTy(16)) {
2660         SmallVector<uint16_t, 16> Elts(Record.begin(), Record.end());
2661         if (isa<VectorType>(CurTy))
2662           V = ConstantDataVector::get(Context, Elts);
2663         else
2664           V = ConstantDataArray::get(Context, Elts);
2665       } else if (EltTy->isIntegerTy(32)) {
2666         SmallVector<uint32_t, 16> Elts(Record.begin(), Record.end());
2667         if (isa<VectorType>(CurTy))
2668           V = ConstantDataVector::get(Context, Elts);
2669         else
2670           V = ConstantDataArray::get(Context, Elts);
2671       } else if (EltTy->isIntegerTy(64)) {
2672         SmallVector<uint64_t, 16> Elts(Record.begin(), Record.end());
2673         if (isa<VectorType>(CurTy))
2674           V = ConstantDataVector::get(Context, Elts);
2675         else
2676           V = ConstantDataArray::get(Context, Elts);
2677       } else if (EltTy->isHalfTy()) {
2678         SmallVector<uint16_t, 16> Elts(Record.begin(), Record.end());
2679         if (isa<VectorType>(CurTy))
2680           V = ConstantDataVector::getFP(Context, Elts);
2681         else
2682           V = ConstantDataArray::getFP(Context, Elts);
2683       } else if (EltTy->isFloatTy()) {
2684         SmallVector<uint32_t, 16> Elts(Record.begin(), Record.end());
2685         if (isa<VectorType>(CurTy))
2686           V = ConstantDataVector::getFP(Context, Elts);
2687         else
2688           V = ConstantDataArray::getFP(Context, Elts);
2689       } else if (EltTy->isDoubleTy()) {
2690         SmallVector<uint64_t, 16> Elts(Record.begin(), Record.end());
2691         if (isa<VectorType>(CurTy))
2692           V = ConstantDataVector::getFP(Context, Elts);
2693         else
2694           V = ConstantDataArray::getFP(Context, Elts);
2695       } else {
2696         return error("Invalid type for value");
2697       }
2698       break;
2699     }
2700 
2701     case bitc::CST_CODE_CE_BINOP: {  // CE_BINOP: [opcode, opval, opval]
2702       if (Record.size() < 3)
2703         return error("Invalid record");
2704       int Opc = getDecodedBinaryOpcode(Record[0], CurTy);
2705       if (Opc < 0) {
2706         V = UndefValue::get(CurTy);  // Unknown binop.
2707       } else {
2708         Constant *LHS = ValueList.getConstantFwdRef(Record[1], CurTy);
2709         Constant *RHS = ValueList.getConstantFwdRef(Record[2], CurTy);
2710         unsigned Flags = 0;
2711         if (Record.size() >= 4) {
2712           if (Opc == Instruction::Add ||
2713               Opc == Instruction::Sub ||
2714               Opc == Instruction::Mul ||
2715               Opc == Instruction::Shl) {
2716             if (Record[3] & (1 << bitc::OBO_NO_SIGNED_WRAP))
2717               Flags |= OverflowingBinaryOperator::NoSignedWrap;
2718             if (Record[3] & (1 << bitc::OBO_NO_UNSIGNED_WRAP))
2719               Flags |= OverflowingBinaryOperator::NoUnsignedWrap;
2720           } else if (Opc == Instruction::SDiv ||
2721                      Opc == Instruction::UDiv ||
2722                      Opc == Instruction::LShr ||
2723                      Opc == Instruction::AShr) {
2724             if (Record[3] & (1 << bitc::PEO_EXACT))
2725               Flags |= SDivOperator::IsExact;
2726           }
2727         }
2728         V = ConstantExpr::get(Opc, LHS, RHS, Flags);
2729       }
2730       break;
2731     }
2732     case bitc::CST_CODE_CE_CAST: {  // CE_CAST: [opcode, opty, opval]
2733       if (Record.size() < 3)
2734         return error("Invalid record");
2735       int Opc = getDecodedCastOpcode(Record[0]);
2736       if (Opc < 0) {
2737         V = UndefValue::get(CurTy);  // Unknown cast.
2738       } else {
2739         Type *OpTy = getTypeByID(Record[1]);
2740         if (!OpTy)
2741           return error("Invalid record");
2742         Constant *Op = ValueList.getConstantFwdRef(Record[2], OpTy);
2743         V = UpgradeBitCastExpr(Opc, Op, CurTy);
2744         if (!V) V = ConstantExpr::getCast(Opc, Op, CurTy);
2745       }
2746       break;
2747     }
2748     case bitc::CST_CODE_CE_INBOUNDS_GEP:
2749     case bitc::CST_CODE_CE_GEP: {  // CE_GEP:        [n x operands]
2750       unsigned OpNum = 0;
2751       Type *PointeeType = nullptr;
2752       if (Record.size() % 2)
2753         PointeeType = getTypeByID(Record[OpNum++]);
2754       SmallVector<Constant*, 16> Elts;
2755       while (OpNum != Record.size()) {
2756         Type *ElTy = getTypeByID(Record[OpNum++]);
2757         if (!ElTy)
2758           return error("Invalid record");
2759         Elts.push_back(ValueList.getConstantFwdRef(Record[OpNum++], ElTy));
2760       }
2761 
2762       if (PointeeType &&
2763           PointeeType !=
2764               cast<SequentialType>(Elts[0]->getType()->getScalarType())
2765                   ->getElementType())
2766         return error("Explicit gep operator type does not match pointee type "
2767                      "of pointer operand");
2768 
2769       ArrayRef<Constant *> Indices(Elts.begin() + 1, Elts.end());
2770       V = ConstantExpr::getGetElementPtr(PointeeType, Elts[0], Indices,
2771                                          BitCode ==
2772                                              bitc::CST_CODE_CE_INBOUNDS_GEP);
2773       break;
2774     }
2775     case bitc::CST_CODE_CE_SELECT: {  // CE_SELECT: [opval#, opval#, opval#]
2776       if (Record.size() < 3)
2777         return error("Invalid record");
2778 
2779       Type *SelectorTy = Type::getInt1Ty(Context);
2780 
2781       // The selector might be an i1 or an <n x i1>
2782       // Get the type from the ValueList before getting a forward ref.
2783       if (VectorType *VTy = dyn_cast<VectorType>(CurTy))
2784         if (Value *V = ValueList[Record[0]])
2785           if (SelectorTy != V->getType())
2786             SelectorTy = VectorType::get(SelectorTy, VTy->getNumElements());
2787 
2788       V = ConstantExpr::getSelect(ValueList.getConstantFwdRef(Record[0],
2789                                                               SelectorTy),
2790                                   ValueList.getConstantFwdRef(Record[1],CurTy),
2791                                   ValueList.getConstantFwdRef(Record[2],CurTy));
2792       break;
2793     }
2794     case bitc::CST_CODE_CE_EXTRACTELT
2795         : { // CE_EXTRACTELT: [opty, opval, opty, opval]
2796       if (Record.size() < 3)
2797         return error("Invalid record");
2798       VectorType *OpTy =
2799         dyn_cast_or_null<VectorType>(getTypeByID(Record[0]));
2800       if (!OpTy)
2801         return error("Invalid record");
2802       Constant *Op0 = ValueList.getConstantFwdRef(Record[1], OpTy);
2803       Constant *Op1 = nullptr;
2804       if (Record.size() == 4) {
2805         Type *IdxTy = getTypeByID(Record[2]);
2806         if (!IdxTy)
2807           return error("Invalid record");
2808         Op1 = ValueList.getConstantFwdRef(Record[3], IdxTy);
2809       } else // TODO: Remove with llvm 4.0
2810         Op1 = ValueList.getConstantFwdRef(Record[2], Type::getInt32Ty(Context));
2811       if (!Op1)
2812         return error("Invalid record");
2813       V = ConstantExpr::getExtractElement(Op0, Op1);
2814       break;
2815     }
2816     case bitc::CST_CODE_CE_INSERTELT
2817         : { // CE_INSERTELT: [opval, opval, opty, opval]
2818       VectorType *OpTy = dyn_cast<VectorType>(CurTy);
2819       if (Record.size() < 3 || !OpTy)
2820         return error("Invalid record");
2821       Constant *Op0 = ValueList.getConstantFwdRef(Record[0], OpTy);
2822       Constant *Op1 = ValueList.getConstantFwdRef(Record[1],
2823                                                   OpTy->getElementType());
2824       Constant *Op2 = nullptr;
2825       if (Record.size() == 4) {
2826         Type *IdxTy = getTypeByID(Record[2]);
2827         if (!IdxTy)
2828           return error("Invalid record");
2829         Op2 = ValueList.getConstantFwdRef(Record[3], IdxTy);
2830       } else // TODO: Remove with llvm 4.0
2831         Op2 = ValueList.getConstantFwdRef(Record[2], Type::getInt32Ty(Context));
2832       if (!Op2)
2833         return error("Invalid record");
2834       V = ConstantExpr::getInsertElement(Op0, Op1, Op2);
2835       break;
2836     }
2837     case bitc::CST_CODE_CE_SHUFFLEVEC: { // CE_SHUFFLEVEC: [opval, opval, opval]
2838       VectorType *OpTy = dyn_cast<VectorType>(CurTy);
2839       if (Record.size() < 3 || !OpTy)
2840         return error("Invalid record");
2841       Constant *Op0 = ValueList.getConstantFwdRef(Record[0], OpTy);
2842       Constant *Op1 = ValueList.getConstantFwdRef(Record[1], OpTy);
2843       Type *ShufTy = VectorType::get(Type::getInt32Ty(Context),
2844                                                  OpTy->getNumElements());
2845       Constant *Op2 = ValueList.getConstantFwdRef(Record[2], ShufTy);
2846       V = ConstantExpr::getShuffleVector(Op0, Op1, Op2);
2847       break;
2848     }
2849     case bitc::CST_CODE_CE_SHUFVEC_EX: { // [opty, opval, opval, opval]
2850       VectorType *RTy = dyn_cast<VectorType>(CurTy);
2851       VectorType *OpTy =
2852         dyn_cast_or_null<VectorType>(getTypeByID(Record[0]));
2853       if (Record.size() < 4 || !RTy || !OpTy)
2854         return error("Invalid record");
2855       Constant *Op0 = ValueList.getConstantFwdRef(Record[1], OpTy);
2856       Constant *Op1 = ValueList.getConstantFwdRef(Record[2], OpTy);
2857       Type *ShufTy = VectorType::get(Type::getInt32Ty(Context),
2858                                                  RTy->getNumElements());
2859       Constant *Op2 = ValueList.getConstantFwdRef(Record[3], ShufTy);
2860       V = ConstantExpr::getShuffleVector(Op0, Op1, Op2);
2861       break;
2862     }
2863     case bitc::CST_CODE_CE_CMP: {     // CE_CMP: [opty, opval, opval, pred]
2864       if (Record.size() < 4)
2865         return error("Invalid record");
2866       Type *OpTy = getTypeByID(Record[0]);
2867       if (!OpTy)
2868         return error("Invalid record");
2869       Constant *Op0 = ValueList.getConstantFwdRef(Record[1], OpTy);
2870       Constant *Op1 = ValueList.getConstantFwdRef(Record[2], OpTy);
2871 
2872       if (OpTy->isFPOrFPVectorTy())
2873         V = ConstantExpr::getFCmp(Record[3], Op0, Op1);
2874       else
2875         V = ConstantExpr::getICmp(Record[3], Op0, Op1);
2876       break;
2877     }
2878     // This maintains backward compatibility, pre-asm dialect keywords.
2879     // FIXME: Remove with the 4.0 release.
2880     case bitc::CST_CODE_INLINEASM_OLD: {
2881       if (Record.size() < 2)
2882         return error("Invalid record");
2883       std::string AsmStr, ConstrStr;
2884       bool HasSideEffects = Record[0] & 1;
2885       bool IsAlignStack = Record[0] >> 1;
2886       unsigned AsmStrSize = Record[1];
2887       if (2+AsmStrSize >= Record.size())
2888         return error("Invalid record");
2889       unsigned ConstStrSize = Record[2+AsmStrSize];
2890       if (3+AsmStrSize+ConstStrSize > Record.size())
2891         return error("Invalid record");
2892 
2893       for (unsigned i = 0; i != AsmStrSize; ++i)
2894         AsmStr += (char)Record[2+i];
2895       for (unsigned i = 0; i != ConstStrSize; ++i)
2896         ConstrStr += (char)Record[3+AsmStrSize+i];
2897       PointerType *PTy = cast<PointerType>(CurTy);
2898       V = InlineAsm::get(cast<FunctionType>(PTy->getElementType()),
2899                          AsmStr, ConstrStr, HasSideEffects, IsAlignStack);
2900       break;
2901     }
2902     // This version adds support for the asm dialect keywords (e.g.,
2903     // inteldialect).
2904     case bitc::CST_CODE_INLINEASM: {
2905       if (Record.size() < 2)
2906         return error("Invalid record");
2907       std::string AsmStr, ConstrStr;
2908       bool HasSideEffects = Record[0] & 1;
2909       bool IsAlignStack = (Record[0] >> 1) & 1;
2910       unsigned AsmDialect = Record[0] >> 2;
2911       unsigned AsmStrSize = Record[1];
2912       if (2+AsmStrSize >= Record.size())
2913         return error("Invalid record");
2914       unsigned ConstStrSize = Record[2+AsmStrSize];
2915       if (3+AsmStrSize+ConstStrSize > Record.size())
2916         return error("Invalid record");
2917 
2918       for (unsigned i = 0; i != AsmStrSize; ++i)
2919         AsmStr += (char)Record[2+i];
2920       for (unsigned i = 0; i != ConstStrSize; ++i)
2921         ConstrStr += (char)Record[3+AsmStrSize+i];
2922       PointerType *PTy = cast<PointerType>(CurTy);
2923       V = InlineAsm::get(cast<FunctionType>(PTy->getElementType()),
2924                          AsmStr, ConstrStr, HasSideEffects, IsAlignStack,
2925                          InlineAsm::AsmDialect(AsmDialect));
2926       break;
2927     }
2928     case bitc::CST_CODE_BLOCKADDRESS:{
2929       if (Record.size() < 3)
2930         return error("Invalid record");
2931       Type *FnTy = getTypeByID(Record[0]);
2932       if (!FnTy)
2933         return error("Invalid record");
2934       Function *Fn =
2935         dyn_cast_or_null<Function>(ValueList.getConstantFwdRef(Record[1],FnTy));
2936       if (!Fn)
2937         return error("Invalid record");
2938 
2939       // If the function is already parsed we can insert the block address right
2940       // away.
2941       BasicBlock *BB;
2942       unsigned BBID = Record[2];
2943       if (!BBID)
2944         // Invalid reference to entry block.
2945         return error("Invalid ID");
2946       if (!Fn->empty()) {
2947         Function::iterator BBI = Fn->begin(), BBE = Fn->end();
2948         for (size_t I = 0, E = BBID; I != E; ++I) {
2949           if (BBI == BBE)
2950             return error("Invalid ID");
2951           ++BBI;
2952         }
2953         BB = &*BBI;
2954       } else {
2955         // Otherwise insert a placeholder and remember it so it can be inserted
2956         // when the function is parsed.
2957         auto &FwdBBs = BasicBlockFwdRefs[Fn];
2958         if (FwdBBs.empty())
2959           BasicBlockFwdRefQueue.push_back(Fn);
2960         if (FwdBBs.size() < BBID + 1)
2961           FwdBBs.resize(BBID + 1);
2962         if (!FwdBBs[BBID])
2963           FwdBBs[BBID] = BasicBlock::Create(Context);
2964         BB = FwdBBs[BBID];
2965       }
2966       V = BlockAddress::get(Fn, BB);
2967       break;
2968     }
2969     }
2970 
2971     ValueList.assignValue(V, NextCstNo);
2972     ++NextCstNo;
2973   }
2974 }
2975 
parseUseLists()2976 std::error_code BitcodeReader::parseUseLists() {
2977   if (Stream.EnterSubBlock(bitc::USELIST_BLOCK_ID))
2978     return error("Invalid record");
2979 
2980   // Read all the records.
2981   SmallVector<uint64_t, 64> Record;
2982   while (1) {
2983     BitstreamEntry Entry = Stream.advanceSkippingSubblocks();
2984 
2985     switch (Entry.Kind) {
2986     case BitstreamEntry::SubBlock: // Handled for us already.
2987     case BitstreamEntry::Error:
2988       return error("Malformed block");
2989     case BitstreamEntry::EndBlock:
2990       return std::error_code();
2991     case BitstreamEntry::Record:
2992       // The interesting case.
2993       break;
2994     }
2995 
2996     // Read a use list record.
2997     Record.clear();
2998     bool IsBB = false;
2999     switch (Stream.readRecord(Entry.ID, Record)) {
3000     default:  // Default behavior: unknown type.
3001       break;
3002     case bitc::USELIST_CODE_BB:
3003       IsBB = true;
3004       // fallthrough
3005     case bitc::USELIST_CODE_DEFAULT: {
3006       unsigned RecordLength = Record.size();
3007       if (RecordLength < 3)
3008         // Records should have at least an ID and two indexes.
3009         return error("Invalid record");
3010       unsigned ID = Record.back();
3011       Record.pop_back();
3012 
3013       Value *V;
3014       if (IsBB) {
3015         assert(ID < FunctionBBs.size() && "Basic block not found");
3016         V = FunctionBBs[ID];
3017       } else
3018         V = ValueList[ID];
3019       unsigned NumUses = 0;
3020       SmallDenseMap<const Use *, unsigned, 16> Order;
3021       for (const Use &U : V->materialized_uses()) {
3022         if (++NumUses > Record.size())
3023           break;
3024         Order[&U] = Record[NumUses - 1];
3025       }
3026       if (Order.size() != Record.size() || NumUses > Record.size())
3027         // Mismatches can happen if the functions are being materialized lazily
3028         // (out-of-order), or a value has been upgraded.
3029         break;
3030 
3031       V->sortUseList([&](const Use &L, const Use &R) {
3032         return Order.lookup(&L) < Order.lookup(&R);
3033       });
3034       break;
3035     }
3036     }
3037   }
3038 }
3039 
3040 /// When we see the block for metadata, remember where it is and then skip it.
3041 /// This lets us lazily deserialize the metadata.
rememberAndSkipMetadata()3042 std::error_code BitcodeReader::rememberAndSkipMetadata() {
3043   // Save the current stream state.
3044   uint64_t CurBit = Stream.GetCurrentBitNo();
3045   DeferredMetadataInfo.push_back(CurBit);
3046 
3047   // Skip over the block for now.
3048   if (Stream.SkipBlock())
3049     return error("Invalid record");
3050   return std::error_code();
3051 }
3052 
materializeMetadata()3053 std::error_code BitcodeReader::materializeMetadata() {
3054   for (uint64_t BitPos : DeferredMetadataInfo) {
3055     // Move the bit stream to the saved position.
3056     Stream.JumpToBit(BitPos);
3057     if (std::error_code EC = parseMetadata(true))
3058       return EC;
3059   }
3060   DeferredMetadataInfo.clear();
3061   return std::error_code();
3062 }
3063 
setStripDebugInfo()3064 void BitcodeReader::setStripDebugInfo() { StripDebugInfo = true; }
3065 
saveMDValueList(DenseMap<const Metadata *,unsigned> & MDValueToValIDMap,bool OnlyTempMD)3066 void BitcodeReader::saveMDValueList(
3067     DenseMap<const Metadata *, unsigned> &MDValueToValIDMap, bool OnlyTempMD) {
3068   for (unsigned ValID = 0; ValID < MDValueList.size(); ++ValID) {
3069     Metadata *MD = MDValueList[ValID];
3070     auto *N = dyn_cast_or_null<MDNode>(MD);
3071     // Save all values if !OnlyTempMD, otherwise just the temporary metadata.
3072     if (!OnlyTempMD || (N && N->isTemporary())) {
3073       // Will call this after materializing each function, in order to
3074       // handle remapping of the function's instructions/metadata.
3075       // See if we already have an entry in that case.
3076       if (OnlyTempMD && MDValueToValIDMap.count(MD)) {
3077         assert(MDValueToValIDMap[MD] == ValID &&
3078                "Inconsistent metadata value id");
3079         continue;
3080       }
3081       MDValueToValIDMap[MD] = ValID;
3082     }
3083   }
3084 }
3085 
3086 /// When we see the block for a function body, remember where it is and then
3087 /// skip it.  This lets us lazily deserialize the functions.
rememberAndSkipFunctionBody()3088 std::error_code BitcodeReader::rememberAndSkipFunctionBody() {
3089   // Get the function we are talking about.
3090   if (FunctionsWithBodies.empty())
3091     return error("Insufficient function protos");
3092 
3093   Function *Fn = FunctionsWithBodies.back();
3094   FunctionsWithBodies.pop_back();
3095 
3096   // Save the current stream state.
3097   uint64_t CurBit = Stream.GetCurrentBitNo();
3098   assert(
3099       (DeferredFunctionInfo[Fn] == 0 || DeferredFunctionInfo[Fn] == CurBit) &&
3100       "Mismatch between VST and scanned function offsets");
3101   DeferredFunctionInfo[Fn] = CurBit;
3102 
3103   // Skip over the function block for now.
3104   if (Stream.SkipBlock())
3105     return error("Invalid record");
3106   return std::error_code();
3107 }
3108 
globalCleanup()3109 std::error_code BitcodeReader::globalCleanup() {
3110   // Patch the initializers for globals and aliases up.
3111   resolveGlobalAndAliasInits();
3112   if (!GlobalInits.empty() || !AliasInits.empty())
3113     return error("Malformed global initializer set");
3114 
3115   // Look for intrinsic functions which need to be upgraded at some point
3116   for (Function &F : *TheModule) {
3117     Function *NewFn;
3118     if (UpgradeIntrinsicFunction(&F, NewFn))
3119       UpgradedIntrinsics[&F] = NewFn;
3120   }
3121 
3122   // Look for global variables which need to be renamed.
3123   for (GlobalVariable &GV : TheModule->globals())
3124     UpgradeGlobalVariable(&GV);
3125 
3126   // Force deallocation of memory for these vectors to favor the client that
3127   // want lazy deserialization.
3128   std::vector<std::pair<GlobalVariable*, unsigned> >().swap(GlobalInits);
3129   std::vector<std::pair<GlobalAlias*, unsigned> >().swap(AliasInits);
3130   return std::error_code();
3131 }
3132 
3133 /// Support for lazy parsing of function bodies. This is required if we
3134 /// either have an old bitcode file without a VST forward declaration record,
3135 /// or if we have an anonymous function being materialized, since anonymous
3136 /// functions do not have a name and are therefore not in the VST.
rememberAndSkipFunctionBodies()3137 std::error_code BitcodeReader::rememberAndSkipFunctionBodies() {
3138   Stream.JumpToBit(NextUnreadBit);
3139 
3140   if (Stream.AtEndOfStream())
3141     return error("Could not find function in stream");
3142 
3143   if (!SeenFirstFunctionBody)
3144     return error("Trying to materialize functions before seeing function blocks");
3145 
3146   // An old bitcode file with the symbol table at the end would have
3147   // finished the parse greedily.
3148   assert(SeenValueSymbolTable);
3149 
3150   SmallVector<uint64_t, 64> Record;
3151 
3152   while (1) {
3153     BitstreamEntry Entry = Stream.advance();
3154     switch (Entry.Kind) {
3155     default:
3156       return error("Expect SubBlock");
3157     case BitstreamEntry::SubBlock:
3158       switch (Entry.ID) {
3159       default:
3160         return error("Expect function block");
3161       case bitc::FUNCTION_BLOCK_ID:
3162         if (std::error_code EC = rememberAndSkipFunctionBody())
3163           return EC;
3164         NextUnreadBit = Stream.GetCurrentBitNo();
3165         return std::error_code();
3166       }
3167     }
3168   }
3169 }
3170 
parseBitcodeVersion()3171 std::error_code BitcodeReader::parseBitcodeVersion() {
3172   if (Stream.EnterSubBlock(bitc::IDENTIFICATION_BLOCK_ID))
3173     return error("Invalid record");
3174 
3175   // Read all the records.
3176   SmallVector<uint64_t, 64> Record;
3177   while (1) {
3178     BitstreamEntry Entry = Stream.advance();
3179 
3180     switch (Entry.Kind) {
3181     default:
3182     case BitstreamEntry::Error:
3183       return error("Malformed block");
3184     case BitstreamEntry::EndBlock:
3185       return std::error_code();
3186     case BitstreamEntry::Record:
3187       // The interesting case.
3188       break;
3189     }
3190 
3191     // Read a record.
3192     Record.clear();
3193     unsigned BitCode = Stream.readRecord(Entry.ID, Record);
3194     switch (BitCode) {
3195     default: // Default behavior: reject
3196       return error("Invalid value");
3197     case bitc::IDENTIFICATION_CODE_STRING: { // IDENTIFICATION:      [strchr x
3198                                              // N]
3199       convertToString(Record, 0, ProducerIdentification);
3200       break;
3201     }
3202     case bitc::IDENTIFICATION_CODE_EPOCH: { // EPOCH:      [epoch#]
3203       unsigned epoch = (unsigned)Record[0];
3204       if (epoch != bitc::BITCODE_CURRENT_EPOCH) {
3205         return error(
3206           Twine("Incompatible epoch: Bitcode '") + Twine(epoch) +
3207           "' vs current: '" + Twine(bitc::BITCODE_CURRENT_EPOCH) + "'");
3208       }
3209     }
3210     }
3211   }
3212 }
3213 
parseModule(uint64_t ResumeBit,bool ShouldLazyLoadMetadata)3214 std::error_code BitcodeReader::parseModule(uint64_t ResumeBit,
3215                                            bool ShouldLazyLoadMetadata) {
3216   if (ResumeBit)
3217     Stream.JumpToBit(ResumeBit);
3218   else if (Stream.EnterSubBlock(bitc::MODULE_BLOCK_ID))
3219     return error("Invalid record");
3220 
3221   SmallVector<uint64_t, 64> Record;
3222   std::vector<std::string> SectionTable;
3223   std::vector<std::string> GCTable;
3224 
3225   // Read all the records for this module.
3226   while (1) {
3227     BitstreamEntry Entry = Stream.advance();
3228 
3229     switch (Entry.Kind) {
3230     case BitstreamEntry::Error:
3231       return error("Malformed block");
3232     case BitstreamEntry::EndBlock:
3233       return globalCleanup();
3234 
3235     case BitstreamEntry::SubBlock:
3236       switch (Entry.ID) {
3237       default:  // Skip unknown content.
3238         if (Stream.SkipBlock())
3239           return error("Invalid record");
3240         break;
3241       case bitc::BLOCKINFO_BLOCK_ID:
3242         if (Stream.ReadBlockInfoBlock())
3243           return error("Malformed block");
3244         break;
3245       case bitc::PARAMATTR_BLOCK_ID:
3246         if (std::error_code EC = parseAttributeBlock())
3247           return EC;
3248         break;
3249       case bitc::PARAMATTR_GROUP_BLOCK_ID:
3250         if (std::error_code EC = parseAttributeGroupBlock())
3251           return EC;
3252         break;
3253       case bitc::TYPE_BLOCK_ID_NEW:
3254         if (std::error_code EC = parseTypeTable())
3255           return EC;
3256         break;
3257       case bitc::VALUE_SYMTAB_BLOCK_ID:
3258         if (!SeenValueSymbolTable) {
3259           // Either this is an old form VST without function index and an
3260           // associated VST forward declaration record (which would have caused
3261           // the VST to be jumped to and parsed before it was encountered
3262           // normally in the stream), or there were no function blocks to
3263           // trigger an earlier parsing of the VST.
3264           assert(VSTOffset == 0 || FunctionsWithBodies.empty());
3265           if (std::error_code EC = parseValueSymbolTable())
3266             return EC;
3267           SeenValueSymbolTable = true;
3268         } else {
3269           // We must have had a VST forward declaration record, which caused
3270           // the parser to jump to and parse the VST earlier.
3271           assert(VSTOffset > 0);
3272           if (Stream.SkipBlock())
3273             return error("Invalid record");
3274         }
3275         break;
3276       case bitc::CONSTANTS_BLOCK_ID:
3277         if (std::error_code EC = parseConstants())
3278           return EC;
3279         if (std::error_code EC = resolveGlobalAndAliasInits())
3280           return EC;
3281         break;
3282       case bitc::METADATA_BLOCK_ID:
3283         if (ShouldLazyLoadMetadata && !IsMetadataMaterialized) {
3284           if (std::error_code EC = rememberAndSkipMetadata())
3285             return EC;
3286           break;
3287         }
3288         assert(DeferredMetadataInfo.empty() && "Unexpected deferred metadata");
3289         if (std::error_code EC = parseMetadata(true))
3290           return EC;
3291         break;
3292       case bitc::METADATA_KIND_BLOCK_ID:
3293         if (std::error_code EC = parseMetadataKinds())
3294           return EC;
3295         break;
3296       case bitc::FUNCTION_BLOCK_ID:
3297         // If this is the first function body we've seen, reverse the
3298         // FunctionsWithBodies list.
3299         if (!SeenFirstFunctionBody) {
3300           std::reverse(FunctionsWithBodies.begin(), FunctionsWithBodies.end());
3301           if (std::error_code EC = globalCleanup())
3302             return EC;
3303           SeenFirstFunctionBody = true;
3304         }
3305 
3306         if (VSTOffset > 0) {
3307           // If we have a VST forward declaration record, make sure we
3308           // parse the VST now if we haven't already. It is needed to
3309           // set up the DeferredFunctionInfo vector for lazy reading.
3310           if (!SeenValueSymbolTable) {
3311             if (std::error_code EC =
3312                     BitcodeReader::parseValueSymbolTable(VSTOffset))
3313               return EC;
3314             SeenValueSymbolTable = true;
3315             // Fall through so that we record the NextUnreadBit below.
3316             // This is necessary in case we have an anonymous function that
3317             // is later materialized. Since it will not have a VST entry we
3318             // need to fall back to the lazy parse to find its offset.
3319           } else {
3320             // If we have a VST forward declaration record, but have already
3321             // parsed the VST (just above, when the first function body was
3322             // encountered here), then we are resuming the parse after
3323             // materializing functions. The ResumeBit points to the
3324             // start of the last function block recorded in the
3325             // DeferredFunctionInfo map. Skip it.
3326             if (Stream.SkipBlock())
3327               return error("Invalid record");
3328             continue;
3329           }
3330         }
3331 
3332         // Support older bitcode files that did not have the function
3333         // index in the VST, nor a VST forward declaration record, as
3334         // well as anonymous functions that do not have VST entries.
3335         // Build the DeferredFunctionInfo vector on the fly.
3336         if (std::error_code EC = rememberAndSkipFunctionBody())
3337           return EC;
3338 
3339         // Suspend parsing when we reach the function bodies. Subsequent
3340         // materialization calls will resume it when necessary. If the bitcode
3341         // file is old, the symbol table will be at the end instead and will not
3342         // have been seen yet. In this case, just finish the parse now.
3343         if (SeenValueSymbolTable) {
3344           NextUnreadBit = Stream.GetCurrentBitNo();
3345           return std::error_code();
3346         }
3347         break;
3348       case bitc::USELIST_BLOCK_ID:
3349         if (std::error_code EC = parseUseLists())
3350           return EC;
3351         break;
3352       case bitc::OPERAND_BUNDLE_TAGS_BLOCK_ID:
3353         if (std::error_code EC = parseOperandBundleTags())
3354           return EC;
3355         break;
3356       }
3357       continue;
3358 
3359     case BitstreamEntry::Record:
3360       // The interesting case.
3361       break;
3362     }
3363 
3364 
3365     // Read a record.
3366     auto BitCode = Stream.readRecord(Entry.ID, Record);
3367     switch (BitCode) {
3368     default: break;  // Default behavior, ignore unknown content.
3369     case bitc::MODULE_CODE_VERSION: {  // VERSION: [version#]
3370       if (Record.size() < 1)
3371         return error("Invalid record");
3372       // Only version #0 and #1 are supported so far.
3373       unsigned module_version = Record[0];
3374       switch (module_version) {
3375         default:
3376           return error("Invalid value");
3377         case 0:
3378           UseRelativeIDs = false;
3379           break;
3380         case 1:
3381           UseRelativeIDs = true;
3382           break;
3383       }
3384       break;
3385     }
3386     case bitc::MODULE_CODE_TRIPLE: {  // TRIPLE: [strchr x N]
3387       std::string S;
3388       if (convertToString(Record, 0, S))
3389         return error("Invalid record");
3390       TheModule->setTargetTriple(S);
3391       break;
3392     }
3393     case bitc::MODULE_CODE_DATALAYOUT: {  // DATALAYOUT: [strchr x N]
3394       std::string S;
3395       if (convertToString(Record, 0, S))
3396         return error("Invalid record");
3397       TheModule->setDataLayout(S);
3398       break;
3399     }
3400     case bitc::MODULE_CODE_ASM: {  // ASM: [strchr x N]
3401       std::string S;
3402       if (convertToString(Record, 0, S))
3403         return error("Invalid record");
3404       TheModule->setModuleInlineAsm(S);
3405       break;
3406     }
3407     case bitc::MODULE_CODE_DEPLIB: {  // DEPLIB: [strchr x N]
3408       // FIXME: Remove in 4.0.
3409       std::string S;
3410       if (convertToString(Record, 0, S))
3411         return error("Invalid record");
3412       // Ignore value.
3413       break;
3414     }
3415     case bitc::MODULE_CODE_SECTIONNAME: {  // SECTIONNAME: [strchr x N]
3416       std::string S;
3417       if (convertToString(Record, 0, S))
3418         return error("Invalid record");
3419       SectionTable.push_back(S);
3420       break;
3421     }
3422     case bitc::MODULE_CODE_GCNAME: {  // SECTIONNAME: [strchr x N]
3423       std::string S;
3424       if (convertToString(Record, 0, S))
3425         return error("Invalid record");
3426       GCTable.push_back(S);
3427       break;
3428     }
3429     case bitc::MODULE_CODE_COMDAT: { // COMDAT: [selection_kind, name]
3430       if (Record.size() < 2)
3431         return error("Invalid record");
3432       Comdat::SelectionKind SK = getDecodedComdatSelectionKind(Record[0]);
3433       unsigned ComdatNameSize = Record[1];
3434       std::string ComdatName;
3435       ComdatName.reserve(ComdatNameSize);
3436       for (unsigned i = 0; i != ComdatNameSize; ++i)
3437         ComdatName += (char)Record[2 + i];
3438       Comdat *C = TheModule->getOrInsertComdat(ComdatName);
3439       C->setSelectionKind(SK);
3440       ComdatList.push_back(C);
3441       break;
3442     }
3443     // GLOBALVAR: [pointer type, isconst, initid,
3444     //             linkage, alignment, section, visibility, threadlocal,
3445     //             unnamed_addr, externally_initialized, dllstorageclass,
3446     //             comdat]
3447     case bitc::MODULE_CODE_GLOBALVAR: {
3448       if (Record.size() < 6)
3449         return error("Invalid record");
3450       Type *Ty = getTypeByID(Record[0]);
3451       if (!Ty)
3452         return error("Invalid record");
3453       bool isConstant = Record[1] & 1;
3454       bool explicitType = Record[1] & 2;
3455       unsigned AddressSpace;
3456       if (explicitType) {
3457         AddressSpace = Record[1] >> 2;
3458       } else {
3459         if (!Ty->isPointerTy())
3460           return error("Invalid type for value");
3461         AddressSpace = cast<PointerType>(Ty)->getAddressSpace();
3462         Ty = cast<PointerType>(Ty)->getElementType();
3463       }
3464 
3465       uint64_t RawLinkage = Record[3];
3466       GlobalValue::LinkageTypes Linkage = getDecodedLinkage(RawLinkage);
3467       unsigned Alignment;
3468       if (std::error_code EC = parseAlignmentValue(Record[4], Alignment))
3469         return EC;
3470       std::string Section;
3471       if (Record[5]) {
3472         if (Record[5]-1 >= SectionTable.size())
3473           return error("Invalid ID");
3474         Section = SectionTable[Record[5]-1];
3475       }
3476       GlobalValue::VisibilityTypes Visibility = GlobalValue::DefaultVisibility;
3477       // Local linkage must have default visibility.
3478       if (Record.size() > 6 && !GlobalValue::isLocalLinkage(Linkage))
3479         // FIXME: Change to an error if non-default in 4.0.
3480         Visibility = getDecodedVisibility(Record[6]);
3481 
3482       GlobalVariable::ThreadLocalMode TLM = GlobalVariable::NotThreadLocal;
3483       if (Record.size() > 7)
3484         TLM = getDecodedThreadLocalMode(Record[7]);
3485 
3486       bool UnnamedAddr = false;
3487       if (Record.size() > 8)
3488         UnnamedAddr = Record[8];
3489 
3490       bool ExternallyInitialized = false;
3491       if (Record.size() > 9)
3492         ExternallyInitialized = Record[9];
3493 
3494       GlobalVariable *NewGV =
3495         new GlobalVariable(*TheModule, Ty, isConstant, Linkage, nullptr, "", nullptr,
3496                            TLM, AddressSpace, ExternallyInitialized);
3497       NewGV->setAlignment(Alignment);
3498       if (!Section.empty())
3499         NewGV->setSection(Section);
3500       NewGV->setVisibility(Visibility);
3501       NewGV->setUnnamedAddr(UnnamedAddr);
3502 
3503       if (Record.size() > 10)
3504         NewGV->setDLLStorageClass(getDecodedDLLStorageClass(Record[10]));
3505       else
3506         upgradeDLLImportExportLinkage(NewGV, RawLinkage);
3507 
3508       ValueList.push_back(NewGV);
3509 
3510       // Remember which value to use for the global initializer.
3511       if (unsigned InitID = Record[2])
3512         GlobalInits.push_back(std::make_pair(NewGV, InitID-1));
3513 
3514       if (Record.size() > 11) {
3515         if (unsigned ComdatID = Record[11]) {
3516           if (ComdatID > ComdatList.size())
3517             return error("Invalid global variable comdat ID");
3518           NewGV->setComdat(ComdatList[ComdatID - 1]);
3519         }
3520       } else if (hasImplicitComdat(RawLinkage)) {
3521         NewGV->setComdat(reinterpret_cast<Comdat *>(1));
3522       }
3523       break;
3524     }
3525     // FUNCTION:  [type, callingconv, isproto, linkage, paramattr,
3526     //             alignment, section, visibility, gc, unnamed_addr,
3527     //             prologuedata, dllstorageclass, comdat, prefixdata]
3528     case bitc::MODULE_CODE_FUNCTION: {
3529       if (Record.size() < 8)
3530         return error("Invalid record");
3531       Type *Ty = getTypeByID(Record[0]);
3532       if (!Ty)
3533         return error("Invalid record");
3534       if (auto *PTy = dyn_cast<PointerType>(Ty))
3535         Ty = PTy->getElementType();
3536       auto *FTy = dyn_cast<FunctionType>(Ty);
3537       if (!FTy)
3538         return error("Invalid type for value");
3539       auto CC = static_cast<CallingConv::ID>(Record[1]);
3540       if (CC & ~CallingConv::MaxID)
3541         return error("Invalid calling convention ID");
3542 
3543       Function *Func = Function::Create(FTy, GlobalValue::ExternalLinkage,
3544                                         "", TheModule);
3545 
3546       Func->setCallingConv(CC);
3547       bool isProto = Record[2];
3548       uint64_t RawLinkage = Record[3];
3549       Func->setLinkage(getDecodedLinkage(RawLinkage));
3550       Func->setAttributes(getAttributes(Record[4]));
3551 
3552       unsigned Alignment;
3553       if (std::error_code EC = parseAlignmentValue(Record[5], Alignment))
3554         return EC;
3555       Func->setAlignment(Alignment);
3556       if (Record[6]) {
3557         if (Record[6]-1 >= SectionTable.size())
3558           return error("Invalid ID");
3559         Func->setSection(SectionTable[Record[6]-1]);
3560       }
3561       // Local linkage must have default visibility.
3562       if (!Func->hasLocalLinkage())
3563         // FIXME: Change to an error if non-default in 4.0.
3564         Func->setVisibility(getDecodedVisibility(Record[7]));
3565       if (Record.size() > 8 && Record[8]) {
3566         if (Record[8]-1 >= GCTable.size())
3567           return error("Invalid ID");
3568         Func->setGC(GCTable[Record[8]-1].c_str());
3569       }
3570       bool UnnamedAddr = false;
3571       if (Record.size() > 9)
3572         UnnamedAddr = Record[9];
3573       Func->setUnnamedAddr(UnnamedAddr);
3574       if (Record.size() > 10 && Record[10] != 0)
3575         FunctionPrologues.push_back(std::make_pair(Func, Record[10]-1));
3576 
3577       if (Record.size() > 11)
3578         Func->setDLLStorageClass(getDecodedDLLStorageClass(Record[11]));
3579       else
3580         upgradeDLLImportExportLinkage(Func, RawLinkage);
3581 
3582       if (Record.size() > 12) {
3583         if (unsigned ComdatID = Record[12]) {
3584           if (ComdatID > ComdatList.size())
3585             return error("Invalid function comdat ID");
3586           Func->setComdat(ComdatList[ComdatID - 1]);
3587         }
3588       } else if (hasImplicitComdat(RawLinkage)) {
3589         Func->setComdat(reinterpret_cast<Comdat *>(1));
3590       }
3591 
3592       if (Record.size() > 13 && Record[13] != 0)
3593         FunctionPrefixes.push_back(std::make_pair(Func, Record[13]-1));
3594 
3595       if (Record.size() > 14 && Record[14] != 0)
3596         FunctionPersonalityFns.push_back(std::make_pair(Func, Record[14] - 1));
3597 
3598       ValueList.push_back(Func);
3599 
3600       // If this is a function with a body, remember the prototype we are
3601       // creating now, so that we can match up the body with them later.
3602       if (!isProto) {
3603         Func->setIsMaterializable(true);
3604         FunctionsWithBodies.push_back(Func);
3605         DeferredFunctionInfo[Func] = 0;
3606       }
3607       break;
3608     }
3609     // ALIAS: [alias type, addrspace, aliasee val#, linkage]
3610     // ALIAS: [alias type, addrspace, aliasee val#, linkage, visibility, dllstorageclass]
3611     case bitc::MODULE_CODE_ALIAS:
3612     case bitc::MODULE_CODE_ALIAS_OLD: {
3613       bool NewRecord = BitCode == bitc::MODULE_CODE_ALIAS;
3614       if (Record.size() < (3 + (unsigned)NewRecord))
3615         return error("Invalid record");
3616       unsigned OpNum = 0;
3617       Type *Ty = getTypeByID(Record[OpNum++]);
3618       if (!Ty)
3619         return error("Invalid record");
3620 
3621       unsigned AddrSpace;
3622       if (!NewRecord) {
3623         auto *PTy = dyn_cast<PointerType>(Ty);
3624         if (!PTy)
3625           return error("Invalid type for value");
3626         Ty = PTy->getElementType();
3627         AddrSpace = PTy->getAddressSpace();
3628       } else {
3629         AddrSpace = Record[OpNum++];
3630       }
3631 
3632       auto Val = Record[OpNum++];
3633       auto Linkage = Record[OpNum++];
3634       auto *NewGA = GlobalAlias::create(
3635           Ty, AddrSpace, getDecodedLinkage(Linkage), "", TheModule);
3636       // Old bitcode files didn't have visibility field.
3637       // Local linkage must have default visibility.
3638       if (OpNum != Record.size()) {
3639         auto VisInd = OpNum++;
3640         if (!NewGA->hasLocalLinkage())
3641           // FIXME: Change to an error if non-default in 4.0.
3642           NewGA->setVisibility(getDecodedVisibility(Record[VisInd]));
3643       }
3644       if (OpNum != Record.size())
3645         NewGA->setDLLStorageClass(getDecodedDLLStorageClass(Record[OpNum++]));
3646       else
3647         upgradeDLLImportExportLinkage(NewGA, Linkage);
3648       if (OpNum != Record.size())
3649         NewGA->setThreadLocalMode(getDecodedThreadLocalMode(Record[OpNum++]));
3650       if (OpNum != Record.size())
3651         NewGA->setUnnamedAddr(Record[OpNum++]);
3652       ValueList.push_back(NewGA);
3653       AliasInits.push_back(std::make_pair(NewGA, Val));
3654       break;
3655     }
3656     /// MODULE_CODE_PURGEVALS: [numvals]
3657     case bitc::MODULE_CODE_PURGEVALS:
3658       // Trim down the value list to the specified size.
3659       if (Record.size() < 1 || Record[0] > ValueList.size())
3660         return error("Invalid record");
3661       ValueList.shrinkTo(Record[0]);
3662       break;
3663     /// MODULE_CODE_VSTOFFSET: [offset]
3664     case bitc::MODULE_CODE_VSTOFFSET:
3665       if (Record.size() < 1)
3666         return error("Invalid record");
3667       VSTOffset = Record[0];
3668       break;
3669     /// MODULE_CODE_METADATA_VALUES: [numvals]
3670     case bitc::MODULE_CODE_METADATA_VALUES:
3671       if (Record.size() < 1)
3672         return error("Invalid record");
3673       assert(!IsMetadataMaterialized);
3674       // This record contains the number of metadata values in the module-level
3675       // METADATA_BLOCK. It is used to support lazy parsing of metadata as
3676       // a postpass, where we will parse function-level metadata first.
3677       // This is needed because the ids of metadata are assigned implicitly
3678       // based on their ordering in the bitcode, with the function-level
3679       // metadata ids starting after the module-level metadata ids. Otherwise,
3680       // we would have to parse the module-level metadata block to prime the
3681       // MDValueList when we are lazy loading metadata during function
3682       // importing. Initialize the MDValueList size here based on the
3683       // record value, regardless of whether we are doing lazy metadata
3684       // loading, so that we have consistent handling and assertion
3685       // checking in parseMetadata for module-level metadata.
3686       NumModuleMDs = Record[0];
3687       SeenModuleValuesRecord = true;
3688       assert(MDValueList.size() == 0);
3689       MDValueList.resize(NumModuleMDs);
3690       break;
3691     }
3692     Record.clear();
3693   }
3694 }
3695 
3696 /// Helper to read the header common to all bitcode files.
hasValidBitcodeHeader(BitstreamCursor & Stream)3697 static bool hasValidBitcodeHeader(BitstreamCursor &Stream) {
3698   // Sniff for the signature.
3699   if (Stream.Read(8) != 'B' ||
3700       Stream.Read(8) != 'C' ||
3701       Stream.Read(4) != 0x0 ||
3702       Stream.Read(4) != 0xC ||
3703       Stream.Read(4) != 0xE ||
3704       Stream.Read(4) != 0xD)
3705     return false;
3706   return true;
3707 }
3708 
3709 std::error_code
parseBitcodeInto(std::unique_ptr<DataStreamer> Streamer,Module * M,bool ShouldLazyLoadMetadata)3710 BitcodeReader::parseBitcodeInto(std::unique_ptr<DataStreamer> Streamer,
3711                                 Module *M, bool ShouldLazyLoadMetadata) {
3712   TheModule = M;
3713 
3714   if (std::error_code EC = initStream(std::move(Streamer)))
3715     return EC;
3716 
3717   // Sniff for the signature.
3718   if (!hasValidBitcodeHeader(Stream))
3719     return error("Invalid bitcode signature");
3720 
3721   // We expect a number of well-defined blocks, though we don't necessarily
3722   // need to understand them all.
3723   while (1) {
3724     if (Stream.AtEndOfStream()) {
3725       // We didn't really read a proper Module.
3726       return error("Malformed IR file");
3727     }
3728 
3729     BitstreamEntry Entry =
3730       Stream.advance(BitstreamCursor::AF_DontAutoprocessAbbrevs);
3731 
3732     if (Entry.Kind != BitstreamEntry::SubBlock)
3733       return error("Malformed block");
3734 
3735     if (Entry.ID == bitc::IDENTIFICATION_BLOCK_ID) {
3736       parseBitcodeVersion();
3737       continue;
3738     }
3739 
3740     if (Entry.ID == bitc::MODULE_BLOCK_ID)
3741       return parseModule(0, ShouldLazyLoadMetadata);
3742 
3743     if (Stream.SkipBlock())
3744       return error("Invalid record");
3745   }
3746 }
3747 
parseModuleTriple()3748 ErrorOr<std::string> BitcodeReader::parseModuleTriple() {
3749   if (Stream.EnterSubBlock(bitc::MODULE_BLOCK_ID))
3750     return error("Invalid record");
3751 
3752   SmallVector<uint64_t, 64> Record;
3753 
3754   std::string Triple;
3755   // Read all the records for this module.
3756   while (1) {
3757     BitstreamEntry Entry = Stream.advanceSkippingSubblocks();
3758 
3759     switch (Entry.Kind) {
3760     case BitstreamEntry::SubBlock: // Handled for us already.
3761     case BitstreamEntry::Error:
3762       return error("Malformed block");
3763     case BitstreamEntry::EndBlock:
3764       return Triple;
3765     case BitstreamEntry::Record:
3766       // The interesting case.
3767       break;
3768     }
3769 
3770     // Read a record.
3771     switch (Stream.readRecord(Entry.ID, Record)) {
3772     default: break;  // Default behavior, ignore unknown content.
3773     case bitc::MODULE_CODE_TRIPLE: {  // TRIPLE: [strchr x N]
3774       std::string S;
3775       if (convertToString(Record, 0, S))
3776         return error("Invalid record");
3777       Triple = S;
3778       break;
3779     }
3780     }
3781     Record.clear();
3782   }
3783   llvm_unreachable("Exit infinite loop");
3784 }
3785 
parseTriple()3786 ErrorOr<std::string> BitcodeReader::parseTriple() {
3787   if (std::error_code EC = initStream(nullptr))
3788     return EC;
3789 
3790   // Sniff for the signature.
3791   if (!hasValidBitcodeHeader(Stream))
3792     return error("Invalid bitcode signature");
3793 
3794   // We expect a number of well-defined blocks, though we don't necessarily
3795   // need to understand them all.
3796   while (1) {
3797     BitstreamEntry Entry = Stream.advance();
3798 
3799     switch (Entry.Kind) {
3800     case BitstreamEntry::Error:
3801       return error("Malformed block");
3802     case BitstreamEntry::EndBlock:
3803       return std::error_code();
3804 
3805     case BitstreamEntry::SubBlock:
3806       if (Entry.ID == bitc::MODULE_BLOCK_ID)
3807         return parseModuleTriple();
3808 
3809       // Ignore other sub-blocks.
3810       if (Stream.SkipBlock())
3811         return error("Malformed block");
3812       continue;
3813 
3814     case BitstreamEntry::Record:
3815       Stream.skipRecord(Entry.ID);
3816       continue;
3817     }
3818   }
3819 }
3820 
parseIdentificationBlock()3821 ErrorOr<std::string> BitcodeReader::parseIdentificationBlock() {
3822   if (std::error_code EC = initStream(nullptr))
3823     return EC;
3824 
3825   // Sniff for the signature.
3826   if (!hasValidBitcodeHeader(Stream))
3827     return error("Invalid bitcode signature");
3828 
3829   // We expect a number of well-defined blocks, though we don't necessarily
3830   // need to understand them all.
3831   while (1) {
3832     BitstreamEntry Entry = Stream.advance();
3833     switch (Entry.Kind) {
3834     case BitstreamEntry::Error:
3835       return error("Malformed block");
3836     case BitstreamEntry::EndBlock:
3837       return std::error_code();
3838 
3839     case BitstreamEntry::SubBlock:
3840       if (Entry.ID == bitc::IDENTIFICATION_BLOCK_ID) {
3841         if (std::error_code EC = parseBitcodeVersion())
3842           return EC;
3843         return ProducerIdentification;
3844       }
3845       // Ignore other sub-blocks.
3846       if (Stream.SkipBlock())
3847         return error("Malformed block");
3848       continue;
3849     case BitstreamEntry::Record:
3850       Stream.skipRecord(Entry.ID);
3851       continue;
3852     }
3853   }
3854 }
3855 
3856 /// Parse metadata attachments.
parseMetadataAttachment(Function & F)3857 std::error_code BitcodeReader::parseMetadataAttachment(Function &F) {
3858   if (Stream.EnterSubBlock(bitc::METADATA_ATTACHMENT_ID))
3859     return error("Invalid record");
3860 
3861   SmallVector<uint64_t, 64> Record;
3862   while (1) {
3863     BitstreamEntry Entry = Stream.advanceSkippingSubblocks();
3864 
3865     switch (Entry.Kind) {
3866     case BitstreamEntry::SubBlock: // Handled for us already.
3867     case BitstreamEntry::Error:
3868       return error("Malformed block");
3869     case BitstreamEntry::EndBlock:
3870       return std::error_code();
3871     case BitstreamEntry::Record:
3872       // The interesting case.
3873       break;
3874     }
3875 
3876     // Read a metadata attachment record.
3877     Record.clear();
3878     switch (Stream.readRecord(Entry.ID, Record)) {
3879     default:  // Default behavior: ignore.
3880       break;
3881     case bitc::METADATA_ATTACHMENT: {
3882       unsigned RecordLength = Record.size();
3883       if (Record.empty())
3884         return error("Invalid record");
3885       if (RecordLength % 2 == 0) {
3886         // A function attachment.
3887         for (unsigned I = 0; I != RecordLength; I += 2) {
3888           auto K = MDKindMap.find(Record[I]);
3889           if (K == MDKindMap.end())
3890             return error("Invalid ID");
3891           Metadata *MD = MDValueList.getValueFwdRef(Record[I + 1]);
3892           F.setMetadata(K->second, cast<MDNode>(MD));
3893         }
3894         continue;
3895       }
3896 
3897       // An instruction attachment.
3898       Instruction *Inst = InstructionList[Record[0]];
3899       for (unsigned i = 1; i != RecordLength; i = i+2) {
3900         unsigned Kind = Record[i];
3901         DenseMap<unsigned, unsigned>::iterator I =
3902           MDKindMap.find(Kind);
3903         if (I == MDKindMap.end())
3904           return error("Invalid ID");
3905         Metadata *Node = MDValueList.getValueFwdRef(Record[i + 1]);
3906         if (isa<LocalAsMetadata>(Node))
3907           // Drop the attachment.  This used to be legal, but there's no
3908           // upgrade path.
3909           break;
3910         Inst->setMetadata(I->second, cast<MDNode>(Node));
3911         if (I->second == LLVMContext::MD_tbaa)
3912           InstsWithTBAATag.push_back(Inst);
3913       }
3914       break;
3915     }
3916     }
3917   }
3918 }
3919 
typeCheckLoadStoreInst(Type * ValType,Type * PtrType)3920 static std::error_code typeCheckLoadStoreInst(Type *ValType, Type *PtrType) {
3921   LLVMContext &Context = PtrType->getContext();
3922   if (!isa<PointerType>(PtrType))
3923     return error(Context, "Load/Store operand is not a pointer type");
3924   Type *ElemType = cast<PointerType>(PtrType)->getElementType();
3925 
3926   if (ValType && ValType != ElemType)
3927     return error(Context, "Explicit load/store type does not match pointee "
3928                           "type of pointer operand");
3929   if (!PointerType::isLoadableOrStorableType(ElemType))
3930     return error(Context, "Cannot load/store from pointer");
3931   return std::error_code();
3932 }
3933 
3934 /// Lazily parse the specified function body block.
parseFunctionBody(Function * F)3935 std::error_code BitcodeReader::parseFunctionBody(Function *F) {
3936   if (Stream.EnterSubBlock(bitc::FUNCTION_BLOCK_ID))
3937     return error("Invalid record");
3938 
3939   InstructionList.clear();
3940   unsigned ModuleValueListSize = ValueList.size();
3941   unsigned ModuleMDValueListSize = MDValueList.size();
3942 
3943   // Add all the function arguments to the value table.
3944   for (Argument &I : F->args())
3945     ValueList.push_back(&I);
3946 
3947   unsigned NextValueNo = ValueList.size();
3948   BasicBlock *CurBB = nullptr;
3949   unsigned CurBBNo = 0;
3950 
3951   DebugLoc LastLoc;
3952   auto getLastInstruction = [&]() -> Instruction * {
3953     if (CurBB && !CurBB->empty())
3954       return &CurBB->back();
3955     else if (CurBBNo && FunctionBBs[CurBBNo - 1] &&
3956              !FunctionBBs[CurBBNo - 1]->empty())
3957       return &FunctionBBs[CurBBNo - 1]->back();
3958     return nullptr;
3959   };
3960 
3961   std::vector<OperandBundleDef> OperandBundles;
3962 
3963   // Read all the records.
3964   SmallVector<uint64_t, 64> Record;
3965   while (1) {
3966     BitstreamEntry Entry = Stream.advance();
3967 
3968     switch (Entry.Kind) {
3969     case BitstreamEntry::Error:
3970       return error("Malformed block");
3971     case BitstreamEntry::EndBlock:
3972       goto OutOfRecordLoop;
3973 
3974     case BitstreamEntry::SubBlock:
3975       switch (Entry.ID) {
3976       default:  // Skip unknown content.
3977         if (Stream.SkipBlock())
3978           return error("Invalid record");
3979         break;
3980       case bitc::CONSTANTS_BLOCK_ID:
3981         if (std::error_code EC = parseConstants())
3982           return EC;
3983         NextValueNo = ValueList.size();
3984         break;
3985       case bitc::VALUE_SYMTAB_BLOCK_ID:
3986         if (std::error_code EC = parseValueSymbolTable())
3987           return EC;
3988         break;
3989       case bitc::METADATA_ATTACHMENT_ID:
3990         if (std::error_code EC = parseMetadataAttachment(*F))
3991           return EC;
3992         break;
3993       case bitc::METADATA_BLOCK_ID:
3994         if (std::error_code EC = parseMetadata())
3995           return EC;
3996         break;
3997       case bitc::USELIST_BLOCK_ID:
3998         if (std::error_code EC = parseUseLists())
3999           return EC;
4000         break;
4001       }
4002       continue;
4003 
4004     case BitstreamEntry::Record:
4005       // The interesting case.
4006       break;
4007     }
4008 
4009     // Read a record.
4010     Record.clear();
4011     Instruction *I = nullptr;
4012     unsigned BitCode = Stream.readRecord(Entry.ID, Record);
4013     switch (BitCode) {
4014     default: // Default behavior: reject
4015       return error("Invalid value");
4016     case bitc::FUNC_CODE_DECLAREBLOCKS: {   // DECLAREBLOCKS: [nblocks]
4017       if (Record.size() < 1 || Record[0] == 0)
4018         return error("Invalid record");
4019       // Create all the basic blocks for the function.
4020       FunctionBBs.resize(Record[0]);
4021 
4022       // See if anything took the address of blocks in this function.
4023       auto BBFRI = BasicBlockFwdRefs.find(F);
4024       if (BBFRI == BasicBlockFwdRefs.end()) {
4025         for (unsigned i = 0, e = FunctionBBs.size(); i != e; ++i)
4026           FunctionBBs[i] = BasicBlock::Create(Context, "", F);
4027       } else {
4028         auto &BBRefs = BBFRI->second;
4029         // Check for invalid basic block references.
4030         if (BBRefs.size() > FunctionBBs.size())
4031           return error("Invalid ID");
4032         assert(!BBRefs.empty() && "Unexpected empty array");
4033         assert(!BBRefs.front() && "Invalid reference to entry block");
4034         for (unsigned I = 0, E = FunctionBBs.size(), RE = BBRefs.size(); I != E;
4035              ++I)
4036           if (I < RE && BBRefs[I]) {
4037             BBRefs[I]->insertInto(F);
4038             FunctionBBs[I] = BBRefs[I];
4039           } else {
4040             FunctionBBs[I] = BasicBlock::Create(Context, "", F);
4041           }
4042 
4043         // Erase from the table.
4044         BasicBlockFwdRefs.erase(BBFRI);
4045       }
4046 
4047       CurBB = FunctionBBs[0];
4048       continue;
4049     }
4050 
4051     case bitc::FUNC_CODE_DEBUG_LOC_AGAIN:  // DEBUG_LOC_AGAIN
4052       // This record indicates that the last instruction is at the same
4053       // location as the previous instruction with a location.
4054       I = getLastInstruction();
4055 
4056       if (!I)
4057         return error("Invalid record");
4058       I->setDebugLoc(LastLoc);
4059       I = nullptr;
4060       continue;
4061 
4062     case bitc::FUNC_CODE_DEBUG_LOC: {      // DEBUG_LOC: [line, col, scope, ia]
4063       I = getLastInstruction();
4064       if (!I || Record.size() < 4)
4065         return error("Invalid record");
4066 
4067       unsigned Line = Record[0], Col = Record[1];
4068       unsigned ScopeID = Record[2], IAID = Record[3];
4069 
4070       MDNode *Scope = nullptr, *IA = nullptr;
4071       if (ScopeID) Scope = cast<MDNode>(MDValueList.getValueFwdRef(ScopeID-1));
4072       if (IAID)    IA = cast<MDNode>(MDValueList.getValueFwdRef(IAID-1));
4073       LastLoc = DebugLoc::get(Line, Col, Scope, IA);
4074       I->setDebugLoc(LastLoc);
4075       I = nullptr;
4076       continue;
4077     }
4078 
4079     case bitc::FUNC_CODE_INST_BINOP: {    // BINOP: [opval, ty, opval, opcode]
4080       unsigned OpNum = 0;
4081       Value *LHS, *RHS;
4082       if (getValueTypePair(Record, OpNum, NextValueNo, LHS) ||
4083           popValue(Record, OpNum, NextValueNo, LHS->getType(), RHS) ||
4084           OpNum+1 > Record.size())
4085         return error("Invalid record");
4086 
4087       int Opc = getDecodedBinaryOpcode(Record[OpNum++], LHS->getType());
4088       if (Opc == -1)
4089         return error("Invalid record");
4090       I = BinaryOperator::Create((Instruction::BinaryOps)Opc, LHS, RHS);
4091       InstructionList.push_back(I);
4092       if (OpNum < Record.size()) {
4093         if (Opc == Instruction::Add ||
4094             Opc == Instruction::Sub ||
4095             Opc == Instruction::Mul ||
4096             Opc == Instruction::Shl) {
4097           if (Record[OpNum] & (1 << bitc::OBO_NO_SIGNED_WRAP))
4098             cast<BinaryOperator>(I)->setHasNoSignedWrap(true);
4099           if (Record[OpNum] & (1 << bitc::OBO_NO_UNSIGNED_WRAP))
4100             cast<BinaryOperator>(I)->setHasNoUnsignedWrap(true);
4101         } else if (Opc == Instruction::SDiv ||
4102                    Opc == Instruction::UDiv ||
4103                    Opc == Instruction::LShr ||
4104                    Opc == Instruction::AShr) {
4105           if (Record[OpNum] & (1 << bitc::PEO_EXACT))
4106             cast<BinaryOperator>(I)->setIsExact(true);
4107         } else if (isa<FPMathOperator>(I)) {
4108           FastMathFlags FMF = getDecodedFastMathFlags(Record[OpNum]);
4109           if (FMF.any())
4110             I->setFastMathFlags(FMF);
4111         }
4112 
4113       }
4114       break;
4115     }
4116     case bitc::FUNC_CODE_INST_CAST: {    // CAST: [opval, opty, destty, castopc]
4117       unsigned OpNum = 0;
4118       Value *Op;
4119       if (getValueTypePair(Record, OpNum, NextValueNo, Op) ||
4120           OpNum+2 != Record.size())
4121         return error("Invalid record");
4122 
4123       Type *ResTy = getTypeByID(Record[OpNum]);
4124       int Opc = getDecodedCastOpcode(Record[OpNum + 1]);
4125       if (Opc == -1 || !ResTy)
4126         return error("Invalid record");
4127       Instruction *Temp = nullptr;
4128       if ((I = UpgradeBitCastInst(Opc, Op, ResTy, Temp))) {
4129         if (Temp) {
4130           InstructionList.push_back(Temp);
4131           CurBB->getInstList().push_back(Temp);
4132         }
4133       } else {
4134         auto CastOp = (Instruction::CastOps)Opc;
4135         if (!CastInst::castIsValid(CastOp, Op, ResTy))
4136           return error("Invalid cast");
4137         I = CastInst::Create(CastOp, Op, ResTy);
4138       }
4139       InstructionList.push_back(I);
4140       break;
4141     }
4142     case bitc::FUNC_CODE_INST_INBOUNDS_GEP_OLD:
4143     case bitc::FUNC_CODE_INST_GEP_OLD:
4144     case bitc::FUNC_CODE_INST_GEP: { // GEP: type, [n x operands]
4145       unsigned OpNum = 0;
4146 
4147       Type *Ty;
4148       bool InBounds;
4149 
4150       if (BitCode == bitc::FUNC_CODE_INST_GEP) {
4151         InBounds = Record[OpNum++];
4152         Ty = getTypeByID(Record[OpNum++]);
4153       } else {
4154         InBounds = BitCode == bitc::FUNC_CODE_INST_INBOUNDS_GEP_OLD;
4155         Ty = nullptr;
4156       }
4157 
4158       Value *BasePtr;
4159       if (getValueTypePair(Record, OpNum, NextValueNo, BasePtr))
4160         return error("Invalid record");
4161 
4162       if (!Ty)
4163         Ty = cast<SequentialType>(BasePtr->getType()->getScalarType())
4164                  ->getElementType();
4165       else if (Ty !=
4166                cast<SequentialType>(BasePtr->getType()->getScalarType())
4167                    ->getElementType())
4168         return error(
4169             "Explicit gep type does not match pointee type of pointer operand");
4170 
4171       SmallVector<Value*, 16> GEPIdx;
4172       while (OpNum != Record.size()) {
4173         Value *Op;
4174         if (getValueTypePair(Record, OpNum, NextValueNo, Op))
4175           return error("Invalid record");
4176         GEPIdx.push_back(Op);
4177       }
4178 
4179       I = GetElementPtrInst::Create(Ty, BasePtr, GEPIdx);
4180 
4181       InstructionList.push_back(I);
4182       if (InBounds)
4183         cast<GetElementPtrInst>(I)->setIsInBounds(true);
4184       break;
4185     }
4186 
4187     case bitc::FUNC_CODE_INST_EXTRACTVAL: {
4188                                        // EXTRACTVAL: [opty, opval, n x indices]
4189       unsigned OpNum = 0;
4190       Value *Agg;
4191       if (getValueTypePair(Record, OpNum, NextValueNo, Agg))
4192         return error("Invalid record");
4193 
4194       unsigned RecSize = Record.size();
4195       if (OpNum == RecSize)
4196         return error("EXTRACTVAL: Invalid instruction with 0 indices");
4197 
4198       SmallVector<unsigned, 4> EXTRACTVALIdx;
4199       Type *CurTy = Agg->getType();
4200       for (; OpNum != RecSize; ++OpNum) {
4201         bool IsArray = CurTy->isArrayTy();
4202         bool IsStruct = CurTy->isStructTy();
4203         uint64_t Index = Record[OpNum];
4204 
4205         if (!IsStruct && !IsArray)
4206           return error("EXTRACTVAL: Invalid type");
4207         if ((unsigned)Index != Index)
4208           return error("Invalid value");
4209         if (IsStruct && Index >= CurTy->subtypes().size())
4210           return error("EXTRACTVAL: Invalid struct index");
4211         if (IsArray && Index >= CurTy->getArrayNumElements())
4212           return error("EXTRACTVAL: Invalid array index");
4213         EXTRACTVALIdx.push_back((unsigned)Index);
4214 
4215         if (IsStruct)
4216           CurTy = CurTy->subtypes()[Index];
4217         else
4218           CurTy = CurTy->subtypes()[0];
4219       }
4220 
4221       I = ExtractValueInst::Create(Agg, EXTRACTVALIdx);
4222       InstructionList.push_back(I);
4223       break;
4224     }
4225 
4226     case bitc::FUNC_CODE_INST_INSERTVAL: {
4227                            // INSERTVAL: [opty, opval, opty, opval, n x indices]
4228       unsigned OpNum = 0;
4229       Value *Agg;
4230       if (getValueTypePair(Record, OpNum, NextValueNo, Agg))
4231         return error("Invalid record");
4232       Value *Val;
4233       if (getValueTypePair(Record, OpNum, NextValueNo, Val))
4234         return error("Invalid record");
4235 
4236       unsigned RecSize = Record.size();
4237       if (OpNum == RecSize)
4238         return error("INSERTVAL: Invalid instruction with 0 indices");
4239 
4240       SmallVector<unsigned, 4> INSERTVALIdx;
4241       Type *CurTy = Agg->getType();
4242       for (; OpNum != RecSize; ++OpNum) {
4243         bool IsArray = CurTy->isArrayTy();
4244         bool IsStruct = CurTy->isStructTy();
4245         uint64_t Index = Record[OpNum];
4246 
4247         if (!IsStruct && !IsArray)
4248           return error("INSERTVAL: Invalid type");
4249         if ((unsigned)Index != Index)
4250           return error("Invalid value");
4251         if (IsStruct && Index >= CurTy->subtypes().size())
4252           return error("INSERTVAL: Invalid struct index");
4253         if (IsArray && Index >= CurTy->getArrayNumElements())
4254           return error("INSERTVAL: Invalid array index");
4255 
4256         INSERTVALIdx.push_back((unsigned)Index);
4257         if (IsStruct)
4258           CurTy = CurTy->subtypes()[Index];
4259         else
4260           CurTy = CurTy->subtypes()[0];
4261       }
4262 
4263       if (CurTy != Val->getType())
4264         return error("Inserted value type doesn't match aggregate type");
4265 
4266       I = InsertValueInst::Create(Agg, Val, INSERTVALIdx);
4267       InstructionList.push_back(I);
4268       break;
4269     }
4270 
4271     case bitc::FUNC_CODE_INST_SELECT: { // SELECT: [opval, ty, opval, opval]
4272       // obsolete form of select
4273       // handles select i1 ... in old bitcode
4274       unsigned OpNum = 0;
4275       Value *TrueVal, *FalseVal, *Cond;
4276       if (getValueTypePair(Record, OpNum, NextValueNo, TrueVal) ||
4277           popValue(Record, OpNum, NextValueNo, TrueVal->getType(), FalseVal) ||
4278           popValue(Record, OpNum, NextValueNo, Type::getInt1Ty(Context), Cond))
4279         return error("Invalid record");
4280 
4281       I = SelectInst::Create(Cond, TrueVal, FalseVal);
4282       InstructionList.push_back(I);
4283       break;
4284     }
4285 
4286     case bitc::FUNC_CODE_INST_VSELECT: {// VSELECT: [ty,opval,opval,predty,pred]
4287       // new form of select
4288       // handles select i1 or select [N x i1]
4289       unsigned OpNum = 0;
4290       Value *TrueVal, *FalseVal, *Cond;
4291       if (getValueTypePair(Record, OpNum, NextValueNo, TrueVal) ||
4292           popValue(Record, OpNum, NextValueNo, TrueVal->getType(), FalseVal) ||
4293           getValueTypePair(Record, OpNum, NextValueNo, Cond))
4294         return error("Invalid record");
4295 
4296       // select condition can be either i1 or [N x i1]
4297       if (VectorType* vector_type =
4298           dyn_cast<VectorType>(Cond->getType())) {
4299         // expect <n x i1>
4300         if (vector_type->getElementType() != Type::getInt1Ty(Context))
4301           return error("Invalid type for value");
4302       } else {
4303         // expect i1
4304         if (Cond->getType() != Type::getInt1Ty(Context))
4305           return error("Invalid type for value");
4306       }
4307 
4308       I = SelectInst::Create(Cond, TrueVal, FalseVal);
4309       InstructionList.push_back(I);
4310       break;
4311     }
4312 
4313     case bitc::FUNC_CODE_INST_EXTRACTELT: { // EXTRACTELT: [opty, opval, opval]
4314       unsigned OpNum = 0;
4315       Value *Vec, *Idx;
4316       if (getValueTypePair(Record, OpNum, NextValueNo, Vec) ||
4317           getValueTypePair(Record, OpNum, NextValueNo, Idx))
4318         return error("Invalid record");
4319       if (!Vec->getType()->isVectorTy())
4320         return error("Invalid type for value");
4321       I = ExtractElementInst::Create(Vec, Idx);
4322       InstructionList.push_back(I);
4323       break;
4324     }
4325 
4326     case bitc::FUNC_CODE_INST_INSERTELT: { // INSERTELT: [ty, opval,opval,opval]
4327       unsigned OpNum = 0;
4328       Value *Vec, *Elt, *Idx;
4329       if (getValueTypePair(Record, OpNum, NextValueNo, Vec))
4330         return error("Invalid record");
4331       if (!Vec->getType()->isVectorTy())
4332         return error("Invalid type for value");
4333       if (popValue(Record, OpNum, NextValueNo,
4334                    cast<VectorType>(Vec->getType())->getElementType(), Elt) ||
4335           getValueTypePair(Record, OpNum, NextValueNo, Idx))
4336         return error("Invalid record");
4337       I = InsertElementInst::Create(Vec, Elt, Idx);
4338       InstructionList.push_back(I);
4339       break;
4340     }
4341 
4342     case bitc::FUNC_CODE_INST_SHUFFLEVEC: {// SHUFFLEVEC: [opval,ty,opval,opval]
4343       unsigned OpNum = 0;
4344       Value *Vec1, *Vec2, *Mask;
4345       if (getValueTypePair(Record, OpNum, NextValueNo, Vec1) ||
4346           popValue(Record, OpNum, NextValueNo, Vec1->getType(), Vec2))
4347         return error("Invalid record");
4348 
4349       if (getValueTypePair(Record, OpNum, NextValueNo, Mask))
4350         return error("Invalid record");
4351       if (!Vec1->getType()->isVectorTy() || !Vec2->getType()->isVectorTy())
4352         return error("Invalid type for value");
4353       I = new ShuffleVectorInst(Vec1, Vec2, Mask);
4354       InstructionList.push_back(I);
4355       break;
4356     }
4357 
4358     case bitc::FUNC_CODE_INST_CMP:   // CMP: [opty, opval, opval, pred]
4359       // Old form of ICmp/FCmp returning bool
4360       // Existed to differentiate between icmp/fcmp and vicmp/vfcmp which were
4361       // both legal on vectors but had different behaviour.
4362     case bitc::FUNC_CODE_INST_CMP2: { // CMP2: [opty, opval, opval, pred]
4363       // FCmp/ICmp returning bool or vector of bool
4364 
4365       unsigned OpNum = 0;
4366       Value *LHS, *RHS;
4367       if (getValueTypePair(Record, OpNum, NextValueNo, LHS) ||
4368           popValue(Record, OpNum, NextValueNo, LHS->getType(), RHS))
4369         return error("Invalid record");
4370 
4371       unsigned PredVal = Record[OpNum];
4372       bool IsFP = LHS->getType()->isFPOrFPVectorTy();
4373       FastMathFlags FMF;
4374       if (IsFP && Record.size() > OpNum+1)
4375         FMF = getDecodedFastMathFlags(Record[++OpNum]);
4376 
4377       if (OpNum+1 != Record.size())
4378         return error("Invalid record");
4379 
4380       if (LHS->getType()->isFPOrFPVectorTy())
4381         I = new FCmpInst((FCmpInst::Predicate)PredVal, LHS, RHS);
4382       else
4383         I = new ICmpInst((ICmpInst::Predicate)PredVal, LHS, RHS);
4384 
4385       if (FMF.any())
4386         I->setFastMathFlags(FMF);
4387       InstructionList.push_back(I);
4388       break;
4389     }
4390 
4391     case bitc::FUNC_CODE_INST_RET: // RET: [opty,opval<optional>]
4392       {
4393         unsigned Size = Record.size();
4394         if (Size == 0) {
4395           I = ReturnInst::Create(Context);
4396           InstructionList.push_back(I);
4397           break;
4398         }
4399 
4400         unsigned OpNum = 0;
4401         Value *Op = nullptr;
4402         if (getValueTypePair(Record, OpNum, NextValueNo, Op))
4403           return error("Invalid record");
4404         if (OpNum != Record.size())
4405           return error("Invalid record");
4406 
4407         I = ReturnInst::Create(Context, Op);
4408         InstructionList.push_back(I);
4409         break;
4410       }
4411     case bitc::FUNC_CODE_INST_BR: { // BR: [bb#, bb#, opval] or [bb#]
4412       if (Record.size() != 1 && Record.size() != 3)
4413         return error("Invalid record");
4414       BasicBlock *TrueDest = getBasicBlock(Record[0]);
4415       if (!TrueDest)
4416         return error("Invalid record");
4417 
4418       if (Record.size() == 1) {
4419         I = BranchInst::Create(TrueDest);
4420         InstructionList.push_back(I);
4421       }
4422       else {
4423         BasicBlock *FalseDest = getBasicBlock(Record[1]);
4424         Value *Cond = getValue(Record, 2, NextValueNo,
4425                                Type::getInt1Ty(Context));
4426         if (!FalseDest || !Cond)
4427           return error("Invalid record");
4428         I = BranchInst::Create(TrueDest, FalseDest, Cond);
4429         InstructionList.push_back(I);
4430       }
4431       break;
4432     }
4433     case bitc::FUNC_CODE_INST_CLEANUPRET: { // CLEANUPRET: [val] or [val,bb#]
4434       if (Record.size() != 1 && Record.size() != 2)
4435         return error("Invalid record");
4436       unsigned Idx = 0;
4437       Value *CleanupPad =
4438           getValue(Record, Idx++, NextValueNo, Type::getTokenTy(Context));
4439       if (!CleanupPad)
4440         return error("Invalid record");
4441       BasicBlock *UnwindDest = nullptr;
4442       if (Record.size() == 2) {
4443         UnwindDest = getBasicBlock(Record[Idx++]);
4444         if (!UnwindDest)
4445           return error("Invalid record");
4446       }
4447 
4448       I = CleanupReturnInst::Create(CleanupPad, UnwindDest);
4449       InstructionList.push_back(I);
4450       break;
4451     }
4452     case bitc::FUNC_CODE_INST_CATCHRET: { // CATCHRET: [val,bb#]
4453       if (Record.size() != 2)
4454         return error("Invalid record");
4455       unsigned Idx = 0;
4456       Value *CatchPad =
4457           getValue(Record, Idx++, NextValueNo, Type::getTokenTy(Context));
4458       if (!CatchPad)
4459         return error("Invalid record");
4460       BasicBlock *BB = getBasicBlock(Record[Idx++]);
4461       if (!BB)
4462         return error("Invalid record");
4463 
4464       I = CatchReturnInst::Create(CatchPad, BB);
4465       InstructionList.push_back(I);
4466       break;
4467     }
4468     case bitc::FUNC_CODE_INST_CATCHSWITCH: { // CATCHSWITCH: [tok,num,(bb)*,bb?]
4469       // We must have, at minimum, the outer scope and the number of arguments.
4470       if (Record.size() < 2)
4471         return error("Invalid record");
4472 
4473       unsigned Idx = 0;
4474 
4475       Value *ParentPad =
4476           getValue(Record, Idx++, NextValueNo, Type::getTokenTy(Context));
4477 
4478       unsigned NumHandlers = Record[Idx++];
4479 
4480       SmallVector<BasicBlock *, 2> Handlers;
4481       for (unsigned Op = 0; Op != NumHandlers; ++Op) {
4482         BasicBlock *BB = getBasicBlock(Record[Idx++]);
4483         if (!BB)
4484           return error("Invalid record");
4485         Handlers.push_back(BB);
4486       }
4487 
4488       BasicBlock *UnwindDest = nullptr;
4489       if (Idx + 1 == Record.size()) {
4490         UnwindDest = getBasicBlock(Record[Idx++]);
4491         if (!UnwindDest)
4492           return error("Invalid record");
4493       }
4494 
4495       if (Record.size() != Idx)
4496         return error("Invalid record");
4497 
4498       auto *CatchSwitch =
4499           CatchSwitchInst::Create(ParentPad, UnwindDest, NumHandlers);
4500       for (BasicBlock *Handler : Handlers)
4501         CatchSwitch->addHandler(Handler);
4502       I = CatchSwitch;
4503       InstructionList.push_back(I);
4504       break;
4505     }
4506     case bitc::FUNC_CODE_INST_CATCHPAD:
4507     case bitc::FUNC_CODE_INST_CLEANUPPAD: { // [tok,num,(ty,val)*]
4508       // We must have, at minimum, the outer scope and the number of arguments.
4509       if (Record.size() < 2)
4510         return error("Invalid record");
4511 
4512       unsigned Idx = 0;
4513 
4514       Value *ParentPad =
4515           getValue(Record, Idx++, NextValueNo, Type::getTokenTy(Context));
4516 
4517       unsigned NumArgOperands = Record[Idx++];
4518 
4519       SmallVector<Value *, 2> Args;
4520       for (unsigned Op = 0; Op != NumArgOperands; ++Op) {
4521         Value *Val;
4522         if (getValueTypePair(Record, Idx, NextValueNo, Val))
4523           return error("Invalid record");
4524         Args.push_back(Val);
4525       }
4526 
4527       if (Record.size() != Idx)
4528         return error("Invalid record");
4529 
4530       if (BitCode == bitc::FUNC_CODE_INST_CLEANUPPAD)
4531         I = CleanupPadInst::Create(ParentPad, Args);
4532       else
4533         I = CatchPadInst::Create(ParentPad, Args);
4534       InstructionList.push_back(I);
4535       break;
4536     }
4537     case bitc::FUNC_CODE_INST_SWITCH: { // SWITCH: [opty, op0, op1, ...]
4538       // Check magic
4539       if ((Record[0] >> 16) == SWITCH_INST_MAGIC) {
4540         // "New" SwitchInst format with case ranges. The changes to write this
4541         // format were reverted but we still recognize bitcode that uses it.
4542         // Hopefully someday we will have support for case ranges and can use
4543         // this format again.
4544 
4545         Type *OpTy = getTypeByID(Record[1]);
4546         unsigned ValueBitWidth = cast<IntegerType>(OpTy)->getBitWidth();
4547 
4548         Value *Cond = getValue(Record, 2, NextValueNo, OpTy);
4549         BasicBlock *Default = getBasicBlock(Record[3]);
4550         if (!OpTy || !Cond || !Default)
4551           return error("Invalid record");
4552 
4553         unsigned NumCases = Record[4];
4554 
4555         SwitchInst *SI = SwitchInst::Create(Cond, Default, NumCases);
4556         InstructionList.push_back(SI);
4557 
4558         unsigned CurIdx = 5;
4559         for (unsigned i = 0; i != NumCases; ++i) {
4560           SmallVector<ConstantInt*, 1> CaseVals;
4561           unsigned NumItems = Record[CurIdx++];
4562           for (unsigned ci = 0; ci != NumItems; ++ci) {
4563             bool isSingleNumber = Record[CurIdx++];
4564 
4565             APInt Low;
4566             unsigned ActiveWords = 1;
4567             if (ValueBitWidth > 64)
4568               ActiveWords = Record[CurIdx++];
4569             Low = readWideAPInt(makeArrayRef(&Record[CurIdx], ActiveWords),
4570                                 ValueBitWidth);
4571             CurIdx += ActiveWords;
4572 
4573             if (!isSingleNumber) {
4574               ActiveWords = 1;
4575               if (ValueBitWidth > 64)
4576                 ActiveWords = Record[CurIdx++];
4577               APInt High = readWideAPInt(
4578                   makeArrayRef(&Record[CurIdx], ActiveWords), ValueBitWidth);
4579               CurIdx += ActiveWords;
4580 
4581               // FIXME: It is not clear whether values in the range should be
4582               // compared as signed or unsigned values. The partially
4583               // implemented changes that used this format in the past used
4584               // unsigned comparisons.
4585               for ( ; Low.ule(High); ++Low)
4586                 CaseVals.push_back(ConstantInt::get(Context, Low));
4587             } else
4588               CaseVals.push_back(ConstantInt::get(Context, Low));
4589           }
4590           BasicBlock *DestBB = getBasicBlock(Record[CurIdx++]);
4591           for (SmallVector<ConstantInt*, 1>::iterator cvi = CaseVals.begin(),
4592                  cve = CaseVals.end(); cvi != cve; ++cvi)
4593             SI->addCase(*cvi, DestBB);
4594         }
4595         I = SI;
4596         break;
4597       }
4598 
4599       // Old SwitchInst format without case ranges.
4600 
4601       if (Record.size() < 3 || (Record.size() & 1) == 0)
4602         return error("Invalid record");
4603       Type *OpTy = getTypeByID(Record[0]);
4604       Value *Cond = getValue(Record, 1, NextValueNo, OpTy);
4605       BasicBlock *Default = getBasicBlock(Record[2]);
4606       if (!OpTy || !Cond || !Default)
4607         return error("Invalid record");
4608       unsigned NumCases = (Record.size()-3)/2;
4609       SwitchInst *SI = SwitchInst::Create(Cond, Default, NumCases);
4610       InstructionList.push_back(SI);
4611       for (unsigned i = 0, e = NumCases; i != e; ++i) {
4612         ConstantInt *CaseVal =
4613           dyn_cast_or_null<ConstantInt>(getFnValueByID(Record[3+i*2], OpTy));
4614         BasicBlock *DestBB = getBasicBlock(Record[1+3+i*2]);
4615         if (!CaseVal || !DestBB) {
4616           delete SI;
4617           return error("Invalid record");
4618         }
4619         SI->addCase(CaseVal, DestBB);
4620       }
4621       I = SI;
4622       break;
4623     }
4624     case bitc::FUNC_CODE_INST_INDIRECTBR: { // INDIRECTBR: [opty, op0, op1, ...]
4625       if (Record.size() < 2)
4626         return error("Invalid record");
4627       Type *OpTy = getTypeByID(Record[0]);
4628       Value *Address = getValue(Record, 1, NextValueNo, OpTy);
4629       if (!OpTy || !Address)
4630         return error("Invalid record");
4631       unsigned NumDests = Record.size()-2;
4632       IndirectBrInst *IBI = IndirectBrInst::Create(Address, NumDests);
4633       InstructionList.push_back(IBI);
4634       for (unsigned i = 0, e = NumDests; i != e; ++i) {
4635         if (BasicBlock *DestBB = getBasicBlock(Record[2+i])) {
4636           IBI->addDestination(DestBB);
4637         } else {
4638           delete IBI;
4639           return error("Invalid record");
4640         }
4641       }
4642       I = IBI;
4643       break;
4644     }
4645 
4646     case bitc::FUNC_CODE_INST_INVOKE: {
4647       // INVOKE: [attrs, cc, normBB, unwindBB, fnty, op0,op1,op2, ...]
4648       if (Record.size() < 4)
4649         return error("Invalid record");
4650       unsigned OpNum = 0;
4651       AttributeSet PAL = getAttributes(Record[OpNum++]);
4652       unsigned CCInfo = Record[OpNum++];
4653       BasicBlock *NormalBB = getBasicBlock(Record[OpNum++]);
4654       BasicBlock *UnwindBB = getBasicBlock(Record[OpNum++]);
4655 
4656       FunctionType *FTy = nullptr;
4657       if (CCInfo >> 13 & 1 &&
4658           !(FTy = dyn_cast<FunctionType>(getTypeByID(Record[OpNum++]))))
4659         return error("Explicit invoke type is not a function type");
4660 
4661       Value *Callee;
4662       if (getValueTypePair(Record, OpNum, NextValueNo, Callee))
4663         return error("Invalid record");
4664 
4665       PointerType *CalleeTy = dyn_cast<PointerType>(Callee->getType());
4666       if (!CalleeTy)
4667         return error("Callee is not a pointer");
4668       if (!FTy) {
4669         FTy = dyn_cast<FunctionType>(CalleeTy->getElementType());
4670         if (!FTy)
4671           return error("Callee is not of pointer to function type");
4672       } else if (CalleeTy->getElementType() != FTy)
4673         return error("Explicit invoke type does not match pointee type of "
4674                      "callee operand");
4675       if (Record.size() < FTy->getNumParams() + OpNum)
4676         return error("Insufficient operands to call");
4677 
4678       SmallVector<Value*, 16> Ops;
4679       for (unsigned i = 0, e = FTy->getNumParams(); i != e; ++i, ++OpNum) {
4680         Ops.push_back(getValue(Record, OpNum, NextValueNo,
4681                                FTy->getParamType(i)));
4682         if (!Ops.back())
4683           return error("Invalid record");
4684       }
4685 
4686       if (!FTy->isVarArg()) {
4687         if (Record.size() != OpNum)
4688           return error("Invalid record");
4689       } else {
4690         // Read type/value pairs for varargs params.
4691         while (OpNum != Record.size()) {
4692           Value *Op;
4693           if (getValueTypePair(Record, OpNum, NextValueNo, Op))
4694             return error("Invalid record");
4695           Ops.push_back(Op);
4696         }
4697       }
4698 
4699       I = InvokeInst::Create(Callee, NormalBB, UnwindBB, Ops, OperandBundles);
4700       OperandBundles.clear();
4701       InstructionList.push_back(I);
4702       cast<InvokeInst>(I)->setCallingConv(
4703           static_cast<CallingConv::ID>(CallingConv::MaxID & CCInfo));
4704       cast<InvokeInst>(I)->setAttributes(PAL);
4705       break;
4706     }
4707     case bitc::FUNC_CODE_INST_RESUME: { // RESUME: [opval]
4708       unsigned Idx = 0;
4709       Value *Val = nullptr;
4710       if (getValueTypePair(Record, Idx, NextValueNo, Val))
4711         return error("Invalid record");
4712       I = ResumeInst::Create(Val);
4713       InstructionList.push_back(I);
4714       break;
4715     }
4716     case bitc::FUNC_CODE_INST_UNREACHABLE: // UNREACHABLE
4717       I = new UnreachableInst(Context);
4718       InstructionList.push_back(I);
4719       break;
4720     case bitc::FUNC_CODE_INST_PHI: { // PHI: [ty, val0,bb0, ...]
4721       if (Record.size() < 1 || ((Record.size()-1)&1))
4722         return error("Invalid record");
4723       Type *Ty = getTypeByID(Record[0]);
4724       if (!Ty)
4725         return error("Invalid record");
4726 
4727       PHINode *PN = PHINode::Create(Ty, (Record.size()-1)/2);
4728       InstructionList.push_back(PN);
4729 
4730       for (unsigned i = 0, e = Record.size()-1; i != e; i += 2) {
4731         Value *V;
4732         // With the new function encoding, it is possible that operands have
4733         // negative IDs (for forward references).  Use a signed VBR
4734         // representation to keep the encoding small.
4735         if (UseRelativeIDs)
4736           V = getValueSigned(Record, 1+i, NextValueNo, Ty);
4737         else
4738           V = getValue(Record, 1+i, NextValueNo, Ty);
4739         BasicBlock *BB = getBasicBlock(Record[2+i]);
4740         if (!V || !BB)
4741           return error("Invalid record");
4742         PN->addIncoming(V, BB);
4743       }
4744       I = PN;
4745       break;
4746     }
4747 
4748     case bitc::FUNC_CODE_INST_LANDINGPAD:
4749     case bitc::FUNC_CODE_INST_LANDINGPAD_OLD: {
4750       // LANDINGPAD: [ty, val, val, num, (id0,val0 ...)?]
4751       unsigned Idx = 0;
4752       if (BitCode == bitc::FUNC_CODE_INST_LANDINGPAD) {
4753         if (Record.size() < 3)
4754           return error("Invalid record");
4755       } else {
4756         assert(BitCode == bitc::FUNC_CODE_INST_LANDINGPAD_OLD);
4757         if (Record.size() < 4)
4758           return error("Invalid record");
4759       }
4760       Type *Ty = getTypeByID(Record[Idx++]);
4761       if (!Ty)
4762         return error("Invalid record");
4763       if (BitCode == bitc::FUNC_CODE_INST_LANDINGPAD_OLD) {
4764         Value *PersFn = nullptr;
4765         if (getValueTypePair(Record, Idx, NextValueNo, PersFn))
4766           return error("Invalid record");
4767 
4768         if (!F->hasPersonalityFn())
4769           F->setPersonalityFn(cast<Constant>(PersFn));
4770         else if (F->getPersonalityFn() != cast<Constant>(PersFn))
4771           return error("Personality function mismatch");
4772       }
4773 
4774       bool IsCleanup = !!Record[Idx++];
4775       unsigned NumClauses = Record[Idx++];
4776       LandingPadInst *LP = LandingPadInst::Create(Ty, NumClauses);
4777       LP->setCleanup(IsCleanup);
4778       for (unsigned J = 0; J != NumClauses; ++J) {
4779         LandingPadInst::ClauseType CT =
4780           LandingPadInst::ClauseType(Record[Idx++]); (void)CT;
4781         Value *Val;
4782 
4783         if (getValueTypePair(Record, Idx, NextValueNo, Val)) {
4784           delete LP;
4785           return error("Invalid record");
4786         }
4787 
4788         assert((CT != LandingPadInst::Catch ||
4789                 !isa<ArrayType>(Val->getType())) &&
4790                "Catch clause has a invalid type!");
4791         assert((CT != LandingPadInst::Filter ||
4792                 isa<ArrayType>(Val->getType())) &&
4793                "Filter clause has invalid type!");
4794         LP->addClause(cast<Constant>(Val));
4795       }
4796 
4797       I = LP;
4798       InstructionList.push_back(I);
4799       break;
4800     }
4801 
4802     case bitc::FUNC_CODE_INST_ALLOCA: { // ALLOCA: [instty, opty, op, align]
4803       if (Record.size() != 4)
4804         return error("Invalid record");
4805       uint64_t AlignRecord = Record[3];
4806       const uint64_t InAllocaMask = uint64_t(1) << 5;
4807       const uint64_t ExplicitTypeMask = uint64_t(1) << 6;
4808       // Reserve bit 7 for SwiftError flag.
4809       // const uint64_t SwiftErrorMask = uint64_t(1) << 7;
4810       const uint64_t FlagMask = InAllocaMask | ExplicitTypeMask;
4811       bool InAlloca = AlignRecord & InAllocaMask;
4812       Type *Ty = getTypeByID(Record[0]);
4813       if ((AlignRecord & ExplicitTypeMask) == 0) {
4814         auto *PTy = dyn_cast_or_null<PointerType>(Ty);
4815         if (!PTy)
4816           return error("Old-style alloca with a non-pointer type");
4817         Ty = PTy->getElementType();
4818       }
4819       Type *OpTy = getTypeByID(Record[1]);
4820       Value *Size = getFnValueByID(Record[2], OpTy);
4821       unsigned Align;
4822       if (std::error_code EC =
4823               parseAlignmentValue(AlignRecord & ~FlagMask, Align)) {
4824         return EC;
4825       }
4826       if (!Ty || !Size)
4827         return error("Invalid record");
4828       AllocaInst *AI = new AllocaInst(Ty, Size, Align);
4829       AI->setUsedWithInAlloca(InAlloca);
4830       I = AI;
4831       InstructionList.push_back(I);
4832       break;
4833     }
4834     case bitc::FUNC_CODE_INST_LOAD: { // LOAD: [opty, op, align, vol]
4835       unsigned OpNum = 0;
4836       Value *Op;
4837       if (getValueTypePair(Record, OpNum, NextValueNo, Op) ||
4838           (OpNum + 2 != Record.size() && OpNum + 3 != Record.size()))
4839         return error("Invalid record");
4840 
4841       Type *Ty = nullptr;
4842       if (OpNum + 3 == Record.size())
4843         Ty = getTypeByID(Record[OpNum++]);
4844       if (std::error_code EC = typeCheckLoadStoreInst(Ty, Op->getType()))
4845         return EC;
4846       if (!Ty)
4847         Ty = cast<PointerType>(Op->getType())->getElementType();
4848 
4849       unsigned Align;
4850       if (std::error_code EC = parseAlignmentValue(Record[OpNum], Align))
4851         return EC;
4852       I = new LoadInst(Ty, Op, "", Record[OpNum + 1], Align);
4853 
4854       InstructionList.push_back(I);
4855       break;
4856     }
4857     case bitc::FUNC_CODE_INST_LOADATOMIC: {
4858        // LOADATOMIC: [opty, op, align, vol, ordering, synchscope]
4859       unsigned OpNum = 0;
4860       Value *Op;
4861       if (getValueTypePair(Record, OpNum, NextValueNo, Op) ||
4862           (OpNum + 4 != Record.size() && OpNum + 5 != Record.size()))
4863         return error("Invalid record");
4864 
4865       Type *Ty = nullptr;
4866       if (OpNum + 5 == Record.size())
4867         Ty = getTypeByID(Record[OpNum++]);
4868       if (std::error_code EC = typeCheckLoadStoreInst(Ty, Op->getType()))
4869         return EC;
4870       if (!Ty)
4871         Ty = cast<PointerType>(Op->getType())->getElementType();
4872 
4873       AtomicOrdering Ordering = getDecodedOrdering(Record[OpNum + 2]);
4874       if (Ordering == NotAtomic || Ordering == Release ||
4875           Ordering == AcquireRelease)
4876         return error("Invalid record");
4877       if (Ordering != NotAtomic && Record[OpNum] == 0)
4878         return error("Invalid record");
4879       SynchronizationScope SynchScope = getDecodedSynchScope(Record[OpNum + 3]);
4880 
4881       unsigned Align;
4882       if (std::error_code EC = parseAlignmentValue(Record[OpNum], Align))
4883         return EC;
4884       I = new LoadInst(Op, "", Record[OpNum+1], Align, Ordering, SynchScope);
4885 
4886       InstructionList.push_back(I);
4887       break;
4888     }
4889     case bitc::FUNC_CODE_INST_STORE:
4890     case bitc::FUNC_CODE_INST_STORE_OLD: { // STORE2:[ptrty, ptr, val, align, vol]
4891       unsigned OpNum = 0;
4892       Value *Val, *Ptr;
4893       if (getValueTypePair(Record, OpNum, NextValueNo, Ptr) ||
4894           (BitCode == bitc::FUNC_CODE_INST_STORE
4895                ? getValueTypePair(Record, OpNum, NextValueNo, Val)
4896                : popValue(Record, OpNum, NextValueNo,
4897                           cast<PointerType>(Ptr->getType())->getElementType(),
4898                           Val)) ||
4899           OpNum + 2 != Record.size())
4900         return error("Invalid record");
4901 
4902       if (std::error_code EC =
4903               typeCheckLoadStoreInst(Val->getType(), Ptr->getType()))
4904         return EC;
4905       unsigned Align;
4906       if (std::error_code EC = parseAlignmentValue(Record[OpNum], Align))
4907         return EC;
4908       I = new StoreInst(Val, Ptr, Record[OpNum+1], Align);
4909       InstructionList.push_back(I);
4910       break;
4911     }
4912     case bitc::FUNC_CODE_INST_STOREATOMIC:
4913     case bitc::FUNC_CODE_INST_STOREATOMIC_OLD: {
4914       // STOREATOMIC: [ptrty, ptr, val, align, vol, ordering, synchscope]
4915       unsigned OpNum = 0;
4916       Value *Val, *Ptr;
4917       if (getValueTypePair(Record, OpNum, NextValueNo, Ptr) ||
4918           (BitCode == bitc::FUNC_CODE_INST_STOREATOMIC
4919                ? getValueTypePair(Record, OpNum, NextValueNo, Val)
4920                : popValue(Record, OpNum, NextValueNo,
4921                           cast<PointerType>(Ptr->getType())->getElementType(),
4922                           Val)) ||
4923           OpNum + 4 != Record.size())
4924         return error("Invalid record");
4925 
4926       if (std::error_code EC =
4927               typeCheckLoadStoreInst(Val->getType(), Ptr->getType()))
4928         return EC;
4929       AtomicOrdering Ordering = getDecodedOrdering(Record[OpNum + 2]);
4930       if (Ordering == NotAtomic || Ordering == Acquire ||
4931           Ordering == AcquireRelease)
4932         return error("Invalid record");
4933       SynchronizationScope SynchScope = getDecodedSynchScope(Record[OpNum + 3]);
4934       if (Ordering != NotAtomic && Record[OpNum] == 0)
4935         return error("Invalid record");
4936 
4937       unsigned Align;
4938       if (std::error_code EC = parseAlignmentValue(Record[OpNum], Align))
4939         return EC;
4940       I = new StoreInst(Val, Ptr, Record[OpNum+1], Align, Ordering, SynchScope);
4941       InstructionList.push_back(I);
4942       break;
4943     }
4944     case bitc::FUNC_CODE_INST_CMPXCHG_OLD:
4945     case bitc::FUNC_CODE_INST_CMPXCHG: {
4946       // CMPXCHG:[ptrty, ptr, cmp, new, vol, successordering, synchscope,
4947       //          failureordering?, isweak?]
4948       unsigned OpNum = 0;
4949       Value *Ptr, *Cmp, *New;
4950       if (getValueTypePair(Record, OpNum, NextValueNo, Ptr) ||
4951           (BitCode == bitc::FUNC_CODE_INST_CMPXCHG
4952                ? getValueTypePair(Record, OpNum, NextValueNo, Cmp)
4953                : popValue(Record, OpNum, NextValueNo,
4954                           cast<PointerType>(Ptr->getType())->getElementType(),
4955                           Cmp)) ||
4956           popValue(Record, OpNum, NextValueNo, Cmp->getType(), New) ||
4957           Record.size() < OpNum + 3 || Record.size() > OpNum + 5)
4958         return error("Invalid record");
4959       AtomicOrdering SuccessOrdering = getDecodedOrdering(Record[OpNum + 1]);
4960       if (SuccessOrdering == NotAtomic || SuccessOrdering == Unordered)
4961         return error("Invalid record");
4962       SynchronizationScope SynchScope = getDecodedSynchScope(Record[OpNum + 2]);
4963 
4964       if (std::error_code EC =
4965               typeCheckLoadStoreInst(Cmp->getType(), Ptr->getType()))
4966         return EC;
4967       AtomicOrdering FailureOrdering;
4968       if (Record.size() < 7)
4969         FailureOrdering =
4970             AtomicCmpXchgInst::getStrongestFailureOrdering(SuccessOrdering);
4971       else
4972         FailureOrdering = getDecodedOrdering(Record[OpNum + 3]);
4973 
4974       I = new AtomicCmpXchgInst(Ptr, Cmp, New, SuccessOrdering, FailureOrdering,
4975                                 SynchScope);
4976       cast<AtomicCmpXchgInst>(I)->setVolatile(Record[OpNum]);
4977 
4978       if (Record.size() < 8) {
4979         // Before weak cmpxchgs existed, the instruction simply returned the
4980         // value loaded from memory, so bitcode files from that era will be
4981         // expecting the first component of a modern cmpxchg.
4982         CurBB->getInstList().push_back(I);
4983         I = ExtractValueInst::Create(I, 0);
4984       } else {
4985         cast<AtomicCmpXchgInst>(I)->setWeak(Record[OpNum+4]);
4986       }
4987 
4988       InstructionList.push_back(I);
4989       break;
4990     }
4991     case bitc::FUNC_CODE_INST_ATOMICRMW: {
4992       // ATOMICRMW:[ptrty, ptr, val, op, vol, ordering, synchscope]
4993       unsigned OpNum = 0;
4994       Value *Ptr, *Val;
4995       if (getValueTypePair(Record, OpNum, NextValueNo, Ptr) ||
4996           popValue(Record, OpNum, NextValueNo,
4997                     cast<PointerType>(Ptr->getType())->getElementType(), Val) ||
4998           OpNum+4 != Record.size())
4999         return error("Invalid record");
5000       AtomicRMWInst::BinOp Operation = getDecodedRMWOperation(Record[OpNum]);
5001       if (Operation < AtomicRMWInst::FIRST_BINOP ||
5002           Operation > AtomicRMWInst::LAST_BINOP)
5003         return error("Invalid record");
5004       AtomicOrdering Ordering = getDecodedOrdering(Record[OpNum + 2]);
5005       if (Ordering == NotAtomic || Ordering == Unordered)
5006         return error("Invalid record");
5007       SynchronizationScope SynchScope = getDecodedSynchScope(Record[OpNum + 3]);
5008       I = new AtomicRMWInst(Operation, Ptr, Val, Ordering, SynchScope);
5009       cast<AtomicRMWInst>(I)->setVolatile(Record[OpNum+1]);
5010       InstructionList.push_back(I);
5011       break;
5012     }
5013     case bitc::FUNC_CODE_INST_FENCE: { // FENCE:[ordering, synchscope]
5014       if (2 != Record.size())
5015         return error("Invalid record");
5016       AtomicOrdering Ordering = getDecodedOrdering(Record[0]);
5017       if (Ordering == NotAtomic || Ordering == Unordered ||
5018           Ordering == Monotonic)
5019         return error("Invalid record");
5020       SynchronizationScope SynchScope = getDecodedSynchScope(Record[1]);
5021       I = new FenceInst(Context, Ordering, SynchScope);
5022       InstructionList.push_back(I);
5023       break;
5024     }
5025     case bitc::FUNC_CODE_INST_CALL: {
5026       // CALL: [paramattrs, cc, fmf, fnty, fnid, arg0, arg1...]
5027       if (Record.size() < 3)
5028         return error("Invalid record");
5029 
5030       unsigned OpNum = 0;
5031       AttributeSet PAL = getAttributes(Record[OpNum++]);
5032       unsigned CCInfo = Record[OpNum++];
5033 
5034       FastMathFlags FMF;
5035       if ((CCInfo >> bitc::CALL_FMF) & 1) {
5036         FMF = getDecodedFastMathFlags(Record[OpNum++]);
5037         if (!FMF.any())
5038           return error("Fast math flags indicator set for call with no FMF");
5039       }
5040 
5041       FunctionType *FTy = nullptr;
5042       if (CCInfo >> bitc::CALL_EXPLICIT_TYPE & 1 &&
5043           !(FTy = dyn_cast<FunctionType>(getTypeByID(Record[OpNum++]))))
5044         return error("Explicit call type is not a function type");
5045 
5046       Value *Callee;
5047       if (getValueTypePair(Record, OpNum, NextValueNo, Callee))
5048         return error("Invalid record");
5049 
5050       PointerType *OpTy = dyn_cast<PointerType>(Callee->getType());
5051       if (!OpTy)
5052         return error("Callee is not a pointer type");
5053       if (!FTy) {
5054         FTy = dyn_cast<FunctionType>(OpTy->getElementType());
5055         if (!FTy)
5056           return error("Callee is not of pointer to function type");
5057       } else if (OpTy->getElementType() != FTy)
5058         return error("Explicit call type does not match pointee type of "
5059                      "callee operand");
5060       if (Record.size() < FTy->getNumParams() + OpNum)
5061         return error("Insufficient operands to call");
5062 
5063       SmallVector<Value*, 16> Args;
5064       // Read the fixed params.
5065       for (unsigned i = 0, e = FTy->getNumParams(); i != e; ++i, ++OpNum) {
5066         if (FTy->getParamType(i)->isLabelTy())
5067           Args.push_back(getBasicBlock(Record[OpNum]));
5068         else
5069           Args.push_back(getValue(Record, OpNum, NextValueNo,
5070                                   FTy->getParamType(i)));
5071         if (!Args.back())
5072           return error("Invalid record");
5073       }
5074 
5075       // Read type/value pairs for varargs params.
5076       if (!FTy->isVarArg()) {
5077         if (OpNum != Record.size())
5078           return error("Invalid record");
5079       } else {
5080         while (OpNum != Record.size()) {
5081           Value *Op;
5082           if (getValueTypePair(Record, OpNum, NextValueNo, Op))
5083             return error("Invalid record");
5084           Args.push_back(Op);
5085         }
5086       }
5087 
5088       I = CallInst::Create(FTy, Callee, Args, OperandBundles);
5089       OperandBundles.clear();
5090       InstructionList.push_back(I);
5091       cast<CallInst>(I)->setCallingConv(
5092           static_cast<CallingConv::ID>((0x7ff & CCInfo) >> bitc::CALL_CCONV));
5093       CallInst::TailCallKind TCK = CallInst::TCK_None;
5094       if (CCInfo & 1 << bitc::CALL_TAIL)
5095         TCK = CallInst::TCK_Tail;
5096       if (CCInfo & (1 << bitc::CALL_MUSTTAIL))
5097         TCK = CallInst::TCK_MustTail;
5098       if (CCInfo & (1 << bitc::CALL_NOTAIL))
5099         TCK = CallInst::TCK_NoTail;
5100       cast<CallInst>(I)->setTailCallKind(TCK);
5101       cast<CallInst>(I)->setAttributes(PAL);
5102       if (FMF.any()) {
5103         if (!isa<FPMathOperator>(I))
5104           return error("Fast-math-flags specified for call without "
5105                        "floating-point scalar or vector return type");
5106         I->setFastMathFlags(FMF);
5107       }
5108       break;
5109     }
5110     case bitc::FUNC_CODE_INST_VAARG: { // VAARG: [valistty, valist, instty]
5111       if (Record.size() < 3)
5112         return error("Invalid record");
5113       Type *OpTy = getTypeByID(Record[0]);
5114       Value *Op = getValue(Record, 1, NextValueNo, OpTy);
5115       Type *ResTy = getTypeByID(Record[2]);
5116       if (!OpTy || !Op || !ResTy)
5117         return error("Invalid record");
5118       I = new VAArgInst(Op, ResTy);
5119       InstructionList.push_back(I);
5120       break;
5121     }
5122 
5123     case bitc::FUNC_CODE_OPERAND_BUNDLE: {
5124       // A call or an invoke can be optionally prefixed with some variable
5125       // number of operand bundle blocks.  These blocks are read into
5126       // OperandBundles and consumed at the next call or invoke instruction.
5127 
5128       if (Record.size() < 1 || Record[0] >= BundleTags.size())
5129         return error("Invalid record");
5130 
5131       std::vector<Value *> Inputs;
5132 
5133       unsigned OpNum = 1;
5134       while (OpNum != Record.size()) {
5135         Value *Op;
5136         if (getValueTypePair(Record, OpNum, NextValueNo, Op))
5137           return error("Invalid record");
5138         Inputs.push_back(Op);
5139       }
5140 
5141       OperandBundles.emplace_back(BundleTags[Record[0]], std::move(Inputs));
5142       continue;
5143     }
5144     }
5145 
5146     // Add instruction to end of current BB.  If there is no current BB, reject
5147     // this file.
5148     if (!CurBB) {
5149       delete I;
5150       return error("Invalid instruction with no BB");
5151     }
5152     if (!OperandBundles.empty()) {
5153       delete I;
5154       return error("Operand bundles found with no consumer");
5155     }
5156     CurBB->getInstList().push_back(I);
5157 
5158     // If this was a terminator instruction, move to the next block.
5159     if (isa<TerminatorInst>(I)) {
5160       ++CurBBNo;
5161       CurBB = CurBBNo < FunctionBBs.size() ? FunctionBBs[CurBBNo] : nullptr;
5162     }
5163 
5164     // Non-void values get registered in the value table for future use.
5165     if (I && !I->getType()->isVoidTy())
5166       ValueList.assignValue(I, NextValueNo++);
5167   }
5168 
5169 OutOfRecordLoop:
5170 
5171   if (!OperandBundles.empty())
5172     return error("Operand bundles found with no consumer");
5173 
5174   // Check the function list for unresolved values.
5175   if (Argument *A = dyn_cast<Argument>(ValueList.back())) {
5176     if (!A->getParent()) {
5177       // We found at least one unresolved value.  Nuke them all to avoid leaks.
5178       for (unsigned i = ModuleValueListSize, e = ValueList.size(); i != e; ++i){
5179         if ((A = dyn_cast_or_null<Argument>(ValueList[i])) && !A->getParent()) {
5180           A->replaceAllUsesWith(UndefValue::get(A->getType()));
5181           delete A;
5182         }
5183       }
5184       return error("Never resolved value found in function");
5185     }
5186   }
5187 
5188   // FIXME: Check for unresolved forward-declared metadata references
5189   // and clean up leaks.
5190 
5191   // Trim the value list down to the size it was before we parsed this function.
5192   ValueList.shrinkTo(ModuleValueListSize);
5193   MDValueList.shrinkTo(ModuleMDValueListSize);
5194   std::vector<BasicBlock*>().swap(FunctionBBs);
5195   return std::error_code();
5196 }
5197 
5198 /// Find the function body in the bitcode stream
findFunctionInStream(Function * F,DenseMap<Function *,uint64_t>::iterator DeferredFunctionInfoIterator)5199 std::error_code BitcodeReader::findFunctionInStream(
5200     Function *F,
5201     DenseMap<Function *, uint64_t>::iterator DeferredFunctionInfoIterator) {
5202   while (DeferredFunctionInfoIterator->second == 0) {
5203     // This is the fallback handling for the old format bitcode that
5204     // didn't contain the function index in the VST, or when we have
5205     // an anonymous function which would not have a VST entry.
5206     // Assert that we have one of those two cases.
5207     assert(VSTOffset == 0 || !F->hasName());
5208     // Parse the next body in the stream and set its position in the
5209     // DeferredFunctionInfo map.
5210     if (std::error_code EC = rememberAndSkipFunctionBodies())
5211       return EC;
5212   }
5213   return std::error_code();
5214 }
5215 
5216 //===----------------------------------------------------------------------===//
5217 // GVMaterializer implementation
5218 //===----------------------------------------------------------------------===//
5219 
releaseBuffer()5220 void BitcodeReader::releaseBuffer() { Buffer.release(); }
5221 
materialize(GlobalValue * GV)5222 std::error_code BitcodeReader::materialize(GlobalValue *GV) {
5223   // In older bitcode we must materialize the metadata before parsing
5224   // any functions, in order to set up the MDValueList properly.
5225   if (!SeenModuleValuesRecord) {
5226     if (std::error_code EC = materializeMetadata())
5227       return EC;
5228   }
5229 
5230   Function *F = dyn_cast<Function>(GV);
5231   // If it's not a function or is already material, ignore the request.
5232   if (!F || !F->isMaterializable())
5233     return std::error_code();
5234 
5235   DenseMap<Function*, uint64_t>::iterator DFII = DeferredFunctionInfo.find(F);
5236   assert(DFII != DeferredFunctionInfo.end() && "Deferred function not found!");
5237   // If its position is recorded as 0, its body is somewhere in the stream
5238   // but we haven't seen it yet.
5239   if (DFII->second == 0)
5240     if (std::error_code EC = findFunctionInStream(F, DFII))
5241       return EC;
5242 
5243   // Move the bit stream to the saved position of the deferred function body.
5244   Stream.JumpToBit(DFII->second);
5245 
5246   if (std::error_code EC = parseFunctionBody(F))
5247     return EC;
5248   F->setIsMaterializable(false);
5249 
5250   if (StripDebugInfo)
5251     stripDebugInfo(*F);
5252 
5253   // Upgrade any old intrinsic calls in the function.
5254   for (auto &I : UpgradedIntrinsics) {
5255     for (auto UI = I.first->materialized_user_begin(), UE = I.first->user_end();
5256          UI != UE;) {
5257       User *U = *UI;
5258       ++UI;
5259       if (CallInst *CI = dyn_cast<CallInst>(U))
5260         UpgradeIntrinsicCall(CI, I.second);
5261     }
5262   }
5263 
5264   // Finish fn->subprogram upgrade for materialized functions.
5265   if (DISubprogram *SP = FunctionsWithSPs.lookup(F))
5266     F->setSubprogram(SP);
5267 
5268   // Bring in any functions that this function forward-referenced via
5269   // blockaddresses.
5270   return materializeForwardReferencedFunctions();
5271 }
5272 
materializeModule()5273 std::error_code BitcodeReader::materializeModule() {
5274   if (std::error_code EC = materializeMetadata())
5275     return EC;
5276 
5277   // Promise to materialize all forward references.
5278   WillMaterializeAllForwardRefs = true;
5279 
5280   // Iterate over the module, deserializing any functions that are still on
5281   // disk.
5282   for (Function &F : *TheModule) {
5283     if (std::error_code EC = materialize(&F))
5284       return EC;
5285   }
5286   // At this point, if there are any function bodies, parse the rest of
5287   // the bits in the module past the last function block we have recorded
5288   // through either lazy scanning or the VST.
5289   if (LastFunctionBlockBit || NextUnreadBit)
5290     parseModule(LastFunctionBlockBit > NextUnreadBit ? LastFunctionBlockBit
5291                                                      : NextUnreadBit);
5292 
5293   // Check that all block address forward references got resolved (as we
5294   // promised above).
5295   if (!BasicBlockFwdRefs.empty())
5296     return error("Never resolved function from blockaddress");
5297 
5298   // Upgrade any intrinsic calls that slipped through (should not happen!) and
5299   // delete the old functions to clean up. We can't do this unless the entire
5300   // module is materialized because there could always be another function body
5301   // with calls to the old function.
5302   for (auto &I : UpgradedIntrinsics) {
5303     for (auto *U : I.first->users()) {
5304       if (CallInst *CI = dyn_cast<CallInst>(U))
5305         UpgradeIntrinsicCall(CI, I.second);
5306     }
5307     if (!I.first->use_empty())
5308       I.first->replaceAllUsesWith(I.second);
5309     I.first->eraseFromParent();
5310   }
5311   UpgradedIntrinsics.clear();
5312 
5313   for (unsigned I = 0, E = InstsWithTBAATag.size(); I < E; I++)
5314     UpgradeInstWithTBAATag(InstsWithTBAATag[I]);
5315 
5316   UpgradeDebugInfo(*TheModule);
5317   return std::error_code();
5318 }
5319 
getIdentifiedStructTypes() const5320 std::vector<StructType *> BitcodeReader::getIdentifiedStructTypes() const {
5321   return IdentifiedStructTypes;
5322 }
5323 
5324 std::error_code
initStream(std::unique_ptr<DataStreamer> Streamer)5325 BitcodeReader::initStream(std::unique_ptr<DataStreamer> Streamer) {
5326   if (Streamer)
5327     return initLazyStream(std::move(Streamer));
5328   return initStreamFromBuffer();
5329 }
5330 
initStreamFromBuffer()5331 std::error_code BitcodeReader::initStreamFromBuffer() {
5332   const unsigned char *BufPtr = (const unsigned char*)Buffer->getBufferStart();
5333   const unsigned char *BufEnd = BufPtr+Buffer->getBufferSize();
5334 
5335   if (Buffer->getBufferSize() & 3)
5336     return error("Invalid bitcode signature");
5337 
5338   // If we have a wrapper header, parse it and ignore the non-bc file contents.
5339   // The magic number is 0x0B17C0DE stored in little endian.
5340   if (isBitcodeWrapper(BufPtr, BufEnd))
5341     if (SkipBitcodeWrapperHeader(BufPtr, BufEnd, true))
5342       return error("Invalid bitcode wrapper header");
5343 
5344   StreamFile.reset(new BitstreamReader(BufPtr, BufEnd));
5345   Stream.init(&*StreamFile);
5346 
5347   return std::error_code();
5348 }
5349 
5350 std::error_code
initLazyStream(std::unique_ptr<DataStreamer> Streamer)5351 BitcodeReader::initLazyStream(std::unique_ptr<DataStreamer> Streamer) {
5352   // Check and strip off the bitcode wrapper; BitstreamReader expects never to
5353   // see it.
5354   auto OwnedBytes =
5355       llvm::make_unique<StreamingMemoryObject>(std::move(Streamer));
5356   StreamingMemoryObject &Bytes = *OwnedBytes;
5357   StreamFile = llvm::make_unique<BitstreamReader>(std::move(OwnedBytes));
5358   Stream.init(&*StreamFile);
5359 
5360   unsigned char buf[16];
5361   if (Bytes.readBytes(buf, 16, 0) != 16)
5362     return error("Invalid bitcode signature");
5363 
5364   if (!isBitcode(buf, buf + 16))
5365     return error("Invalid bitcode signature");
5366 
5367   if (isBitcodeWrapper(buf, buf + 4)) {
5368     const unsigned char *bitcodeStart = buf;
5369     const unsigned char *bitcodeEnd = buf + 16;
5370     SkipBitcodeWrapperHeader(bitcodeStart, bitcodeEnd, false);
5371     Bytes.dropLeadingBytes(bitcodeStart - buf);
5372     Bytes.setKnownObjectSize(bitcodeEnd - bitcodeStart);
5373   }
5374   return std::error_code();
5375 }
5376 
error(BitcodeError E,const Twine & Message)5377 std::error_code FunctionIndexBitcodeReader::error(BitcodeError E,
5378                                                   const Twine &Message) {
5379   return ::error(DiagnosticHandler, make_error_code(E), Message);
5380 }
5381 
error(const Twine & Message)5382 std::error_code FunctionIndexBitcodeReader::error(const Twine &Message) {
5383   return ::error(DiagnosticHandler,
5384                  make_error_code(BitcodeError::CorruptedBitcode), Message);
5385 }
5386 
error(BitcodeError E)5387 std::error_code FunctionIndexBitcodeReader::error(BitcodeError E) {
5388   return ::error(DiagnosticHandler, make_error_code(E));
5389 }
5390 
FunctionIndexBitcodeReader(MemoryBuffer * Buffer,DiagnosticHandlerFunction DiagnosticHandler,bool IsLazy,bool CheckFuncSummaryPresenceOnly)5391 FunctionIndexBitcodeReader::FunctionIndexBitcodeReader(
5392     MemoryBuffer *Buffer, DiagnosticHandlerFunction DiagnosticHandler,
5393     bool IsLazy, bool CheckFuncSummaryPresenceOnly)
5394     : DiagnosticHandler(DiagnosticHandler), Buffer(Buffer), IsLazy(IsLazy),
5395       CheckFuncSummaryPresenceOnly(CheckFuncSummaryPresenceOnly) {}
5396 
FunctionIndexBitcodeReader(DiagnosticHandlerFunction DiagnosticHandler,bool IsLazy,bool CheckFuncSummaryPresenceOnly)5397 FunctionIndexBitcodeReader::FunctionIndexBitcodeReader(
5398     DiagnosticHandlerFunction DiagnosticHandler, bool IsLazy,
5399     bool CheckFuncSummaryPresenceOnly)
5400     : DiagnosticHandler(DiagnosticHandler), Buffer(nullptr), IsLazy(IsLazy),
5401       CheckFuncSummaryPresenceOnly(CheckFuncSummaryPresenceOnly) {}
5402 
freeState()5403 void FunctionIndexBitcodeReader::freeState() { Buffer = nullptr; }
5404 
releaseBuffer()5405 void FunctionIndexBitcodeReader::releaseBuffer() { Buffer.release(); }
5406 
5407 // Specialized value symbol table parser used when reading function index
5408 // blocks where we don't actually create global values.
5409 // At the end of this routine the function index is populated with a map
5410 // from function name to FunctionInfo. The function info contains
5411 // the function block's bitcode offset as well as the offset into the
5412 // function summary section.
parseValueSymbolTable()5413 std::error_code FunctionIndexBitcodeReader::parseValueSymbolTable() {
5414   if (Stream.EnterSubBlock(bitc::VALUE_SYMTAB_BLOCK_ID))
5415     return error("Invalid record");
5416 
5417   SmallVector<uint64_t, 64> Record;
5418 
5419   // Read all the records for this value table.
5420   SmallString<128> ValueName;
5421   while (1) {
5422     BitstreamEntry Entry = Stream.advanceSkippingSubblocks();
5423 
5424     switch (Entry.Kind) {
5425     case BitstreamEntry::SubBlock: // Handled for us already.
5426     case BitstreamEntry::Error:
5427       return error("Malformed block");
5428     case BitstreamEntry::EndBlock:
5429       return std::error_code();
5430     case BitstreamEntry::Record:
5431       // The interesting case.
5432       break;
5433     }
5434 
5435     // Read a record.
5436     Record.clear();
5437     switch (Stream.readRecord(Entry.ID, Record)) {
5438     default: // Default behavior: ignore (e.g. VST_CODE_BBENTRY records).
5439       break;
5440     case bitc::VST_CODE_FNENTRY: {
5441       // VST_FNENTRY: [valueid, offset, namechar x N]
5442       if (convertToString(Record, 2, ValueName))
5443         return error("Invalid record");
5444       unsigned ValueID = Record[0];
5445       uint64_t FuncOffset = Record[1];
5446       std::unique_ptr<FunctionInfo> FuncInfo =
5447           llvm::make_unique<FunctionInfo>(FuncOffset);
5448       if (foundFuncSummary() && !IsLazy) {
5449         DenseMap<uint64_t, std::unique_ptr<FunctionSummary>>::iterator SMI =
5450             SummaryMap.find(ValueID);
5451         assert(SMI != SummaryMap.end() && "Summary info not found");
5452         FuncInfo->setFunctionSummary(std::move(SMI->second));
5453       }
5454       TheIndex->addFunctionInfo(ValueName, std::move(FuncInfo));
5455 
5456       ValueName.clear();
5457       break;
5458     }
5459     case bitc::VST_CODE_COMBINED_FNENTRY: {
5460       // VST_FNENTRY: [offset, namechar x N]
5461       if (convertToString(Record, 1, ValueName))
5462         return error("Invalid record");
5463       uint64_t FuncSummaryOffset = Record[0];
5464       std::unique_ptr<FunctionInfo> FuncInfo =
5465           llvm::make_unique<FunctionInfo>(FuncSummaryOffset);
5466       if (foundFuncSummary() && !IsLazy) {
5467         DenseMap<uint64_t, std::unique_ptr<FunctionSummary>>::iterator SMI =
5468             SummaryMap.find(FuncSummaryOffset);
5469         assert(SMI != SummaryMap.end() && "Summary info not found");
5470         FuncInfo->setFunctionSummary(std::move(SMI->second));
5471       }
5472       TheIndex->addFunctionInfo(ValueName, std::move(FuncInfo));
5473 
5474       ValueName.clear();
5475       break;
5476     }
5477     }
5478   }
5479 }
5480 
5481 // Parse just the blocks needed for function index building out of the module.
5482 // At the end of this routine the function Index is populated with a map
5483 // from function name to FunctionInfo. The function info contains
5484 // either the parsed function summary information (when parsing summaries
5485 // eagerly), or just to the function summary record's offset
5486 // if parsing lazily (IsLazy).
parseModule()5487 std::error_code FunctionIndexBitcodeReader::parseModule() {
5488   if (Stream.EnterSubBlock(bitc::MODULE_BLOCK_ID))
5489     return error("Invalid record");
5490 
5491   // Read the function index for this module.
5492   while (1) {
5493     BitstreamEntry Entry = Stream.advance();
5494 
5495     switch (Entry.Kind) {
5496     case BitstreamEntry::Error:
5497       return error("Malformed block");
5498     case BitstreamEntry::EndBlock:
5499       return std::error_code();
5500 
5501     case BitstreamEntry::SubBlock:
5502       if (CheckFuncSummaryPresenceOnly) {
5503         if (Entry.ID == bitc::FUNCTION_SUMMARY_BLOCK_ID) {
5504           SeenFuncSummary = true;
5505           // No need to parse the rest since we found the summary.
5506           return std::error_code();
5507         }
5508         if (Stream.SkipBlock())
5509           return error("Invalid record");
5510         continue;
5511       }
5512       switch (Entry.ID) {
5513       default: // Skip unknown content.
5514         if (Stream.SkipBlock())
5515           return error("Invalid record");
5516         break;
5517       case bitc::BLOCKINFO_BLOCK_ID:
5518         // Need to parse these to get abbrev ids (e.g. for VST)
5519         if (Stream.ReadBlockInfoBlock())
5520           return error("Malformed block");
5521         break;
5522       case bitc::VALUE_SYMTAB_BLOCK_ID:
5523         if (std::error_code EC = parseValueSymbolTable())
5524           return EC;
5525         break;
5526       case bitc::FUNCTION_SUMMARY_BLOCK_ID:
5527         SeenFuncSummary = true;
5528         if (IsLazy) {
5529           // Lazy parsing of summary info, skip it.
5530           if (Stream.SkipBlock())
5531             return error("Invalid record");
5532         } else if (std::error_code EC = parseEntireSummary())
5533           return EC;
5534         break;
5535       case bitc::MODULE_STRTAB_BLOCK_ID:
5536         if (std::error_code EC = parseModuleStringTable())
5537           return EC;
5538         break;
5539       }
5540       continue;
5541 
5542     case BitstreamEntry::Record:
5543       Stream.skipRecord(Entry.ID);
5544       continue;
5545     }
5546   }
5547 }
5548 
5549 // Eagerly parse the entire function summary block (i.e. for all functions
5550 // in the index). This populates the FunctionSummary objects in
5551 // the index.
parseEntireSummary()5552 std::error_code FunctionIndexBitcodeReader::parseEntireSummary() {
5553   if (Stream.EnterSubBlock(bitc::FUNCTION_SUMMARY_BLOCK_ID))
5554     return error("Invalid record");
5555 
5556   SmallVector<uint64_t, 64> Record;
5557 
5558   while (1) {
5559     BitstreamEntry Entry = Stream.advanceSkippingSubblocks();
5560 
5561     switch (Entry.Kind) {
5562     case BitstreamEntry::SubBlock: // Handled for us already.
5563     case BitstreamEntry::Error:
5564       return error("Malformed block");
5565     case BitstreamEntry::EndBlock:
5566       return std::error_code();
5567     case BitstreamEntry::Record:
5568       // The interesting case.
5569       break;
5570     }
5571 
5572     // Read a record. The record format depends on whether this
5573     // is a per-module index or a combined index file. In the per-module
5574     // case the records contain the associated value's ID for correlation
5575     // with VST entries. In the combined index the correlation is done
5576     // via the bitcode offset of the summary records (which were saved
5577     // in the combined index VST entries). The records also contain
5578     // information used for ThinLTO renaming and importing.
5579     Record.clear();
5580     uint64_t CurRecordBit = Stream.GetCurrentBitNo();
5581     switch (Stream.readRecord(Entry.ID, Record)) {
5582     default: // Default behavior: ignore.
5583       break;
5584     // FS_PERMODULE_ENTRY: [valueid, islocal, instcount]
5585     case bitc::FS_CODE_PERMODULE_ENTRY: {
5586       unsigned ValueID = Record[0];
5587       bool IsLocal = Record[1];
5588       unsigned InstCount = Record[2];
5589       std::unique_ptr<FunctionSummary> FS =
5590           llvm::make_unique<FunctionSummary>(InstCount);
5591       FS->setLocalFunction(IsLocal);
5592       // The module path string ref set in the summary must be owned by the
5593       // index's module string table. Since we don't have a module path
5594       // string table section in the per-module index, we create a single
5595       // module path string table entry with an empty (0) ID to take
5596       // ownership.
5597       FS->setModulePath(
5598           TheIndex->addModulePath(Buffer->getBufferIdentifier(), 0));
5599       SummaryMap[ValueID] = std::move(FS);
5600     }
5601     // FS_COMBINED_ENTRY: [modid, instcount]
5602     case bitc::FS_CODE_COMBINED_ENTRY: {
5603       uint64_t ModuleId = Record[0];
5604       unsigned InstCount = Record[1];
5605       std::unique_ptr<FunctionSummary> FS =
5606           llvm::make_unique<FunctionSummary>(InstCount);
5607       FS->setModulePath(ModuleIdMap[ModuleId]);
5608       SummaryMap[CurRecordBit] = std::move(FS);
5609     }
5610     }
5611   }
5612   llvm_unreachable("Exit infinite loop");
5613 }
5614 
5615 // Parse the  module string table block into the Index.
5616 // This populates the ModulePathStringTable map in the index.
parseModuleStringTable()5617 std::error_code FunctionIndexBitcodeReader::parseModuleStringTable() {
5618   if (Stream.EnterSubBlock(bitc::MODULE_STRTAB_BLOCK_ID))
5619     return error("Invalid record");
5620 
5621   SmallVector<uint64_t, 64> Record;
5622 
5623   SmallString<128> ModulePath;
5624   while (1) {
5625     BitstreamEntry Entry = Stream.advanceSkippingSubblocks();
5626 
5627     switch (Entry.Kind) {
5628     case BitstreamEntry::SubBlock: // Handled for us already.
5629     case BitstreamEntry::Error:
5630       return error("Malformed block");
5631     case BitstreamEntry::EndBlock:
5632       return std::error_code();
5633     case BitstreamEntry::Record:
5634       // The interesting case.
5635       break;
5636     }
5637 
5638     Record.clear();
5639     switch (Stream.readRecord(Entry.ID, Record)) {
5640     default: // Default behavior: ignore.
5641       break;
5642     case bitc::MST_CODE_ENTRY: {
5643       // MST_ENTRY: [modid, namechar x N]
5644       if (convertToString(Record, 1, ModulePath))
5645         return error("Invalid record");
5646       uint64_t ModuleId = Record[0];
5647       StringRef ModulePathInMap = TheIndex->addModulePath(ModulePath, ModuleId);
5648       ModuleIdMap[ModuleId] = ModulePathInMap;
5649       ModulePath.clear();
5650       break;
5651     }
5652     }
5653   }
5654   llvm_unreachable("Exit infinite loop");
5655 }
5656 
5657 // Parse the function info index from the bitcode streamer into the given index.
parseSummaryIndexInto(std::unique_ptr<DataStreamer> Streamer,FunctionInfoIndex * I)5658 std::error_code FunctionIndexBitcodeReader::parseSummaryIndexInto(
5659     std::unique_ptr<DataStreamer> Streamer, FunctionInfoIndex *I) {
5660   TheIndex = I;
5661 
5662   if (std::error_code EC = initStream(std::move(Streamer)))
5663     return EC;
5664 
5665   // Sniff for the signature.
5666   if (!hasValidBitcodeHeader(Stream))
5667     return error("Invalid bitcode signature");
5668 
5669   // We expect a number of well-defined blocks, though we don't necessarily
5670   // need to understand them all.
5671   while (1) {
5672     if (Stream.AtEndOfStream()) {
5673       // We didn't really read a proper Module block.
5674       return error("Malformed block");
5675     }
5676 
5677     BitstreamEntry Entry =
5678         Stream.advance(BitstreamCursor::AF_DontAutoprocessAbbrevs);
5679 
5680     if (Entry.Kind != BitstreamEntry::SubBlock)
5681       return error("Malformed block");
5682 
5683     // If we see a MODULE_BLOCK, parse it to find the blocks needed for
5684     // building the function summary index.
5685     if (Entry.ID == bitc::MODULE_BLOCK_ID)
5686       return parseModule();
5687 
5688     if (Stream.SkipBlock())
5689       return error("Invalid record");
5690   }
5691 }
5692 
5693 // Parse the function information at the given offset in the buffer into
5694 // the index. Used to support lazy parsing of function summaries from the
5695 // combined index during importing.
5696 // TODO: This function is not yet complete as it won't have a consumer
5697 // until ThinLTO function importing is added.
parseFunctionSummary(std::unique_ptr<DataStreamer> Streamer,FunctionInfoIndex * I,size_t FunctionSummaryOffset)5698 std::error_code FunctionIndexBitcodeReader::parseFunctionSummary(
5699     std::unique_ptr<DataStreamer> Streamer, FunctionInfoIndex *I,
5700     size_t FunctionSummaryOffset) {
5701   TheIndex = I;
5702 
5703   if (std::error_code EC = initStream(std::move(Streamer)))
5704     return EC;
5705 
5706   // Sniff for the signature.
5707   if (!hasValidBitcodeHeader(Stream))
5708     return error("Invalid bitcode signature");
5709 
5710   Stream.JumpToBit(FunctionSummaryOffset);
5711 
5712   BitstreamEntry Entry = Stream.advanceSkippingSubblocks();
5713 
5714   switch (Entry.Kind) {
5715   default:
5716     return error("Malformed block");
5717   case BitstreamEntry::Record:
5718     // The expected case.
5719     break;
5720   }
5721 
5722   // TODO: Read a record. This interface will be completed when ThinLTO
5723   // importing is added so that it can be tested.
5724   SmallVector<uint64_t, 64> Record;
5725   switch (Stream.readRecord(Entry.ID, Record)) {
5726   case bitc::FS_CODE_COMBINED_ENTRY:
5727   default:
5728     return error("Invalid record");
5729   }
5730 
5731   return std::error_code();
5732 }
5733 
5734 std::error_code
initStream(std::unique_ptr<DataStreamer> Streamer)5735 FunctionIndexBitcodeReader::initStream(std::unique_ptr<DataStreamer> Streamer) {
5736   if (Streamer)
5737     return initLazyStream(std::move(Streamer));
5738   return initStreamFromBuffer();
5739 }
5740 
initStreamFromBuffer()5741 std::error_code FunctionIndexBitcodeReader::initStreamFromBuffer() {
5742   const unsigned char *BufPtr = (const unsigned char *)Buffer->getBufferStart();
5743   const unsigned char *BufEnd = BufPtr + Buffer->getBufferSize();
5744 
5745   if (Buffer->getBufferSize() & 3)
5746     return error("Invalid bitcode signature");
5747 
5748   // If we have a wrapper header, parse it and ignore the non-bc file contents.
5749   // The magic number is 0x0B17C0DE stored in little endian.
5750   if (isBitcodeWrapper(BufPtr, BufEnd))
5751     if (SkipBitcodeWrapperHeader(BufPtr, BufEnd, true))
5752       return error("Invalid bitcode wrapper header");
5753 
5754   StreamFile.reset(new BitstreamReader(BufPtr, BufEnd));
5755   Stream.init(&*StreamFile);
5756 
5757   return std::error_code();
5758 }
5759 
initLazyStream(std::unique_ptr<DataStreamer> Streamer)5760 std::error_code FunctionIndexBitcodeReader::initLazyStream(
5761     std::unique_ptr<DataStreamer> Streamer) {
5762   // Check and strip off the bitcode wrapper; BitstreamReader expects never to
5763   // see it.
5764   auto OwnedBytes =
5765       llvm::make_unique<StreamingMemoryObject>(std::move(Streamer));
5766   StreamingMemoryObject &Bytes = *OwnedBytes;
5767   StreamFile = llvm::make_unique<BitstreamReader>(std::move(OwnedBytes));
5768   Stream.init(&*StreamFile);
5769 
5770   unsigned char buf[16];
5771   if (Bytes.readBytes(buf, 16, 0) != 16)
5772     return error("Invalid bitcode signature");
5773 
5774   if (!isBitcode(buf, buf + 16))
5775     return error("Invalid bitcode signature");
5776 
5777   if (isBitcodeWrapper(buf, buf + 4)) {
5778     const unsigned char *bitcodeStart = buf;
5779     const unsigned char *bitcodeEnd = buf + 16;
5780     SkipBitcodeWrapperHeader(bitcodeStart, bitcodeEnd, false);
5781     Bytes.dropLeadingBytes(bitcodeStart - buf);
5782     Bytes.setKnownObjectSize(bitcodeEnd - bitcodeStart);
5783   }
5784   return std::error_code();
5785 }
5786 
5787 namespace {
5788 class BitcodeErrorCategoryType : public std::error_category {
name() const5789   const char *name() const LLVM_NOEXCEPT override {
5790     return "llvm.bitcode";
5791   }
message(int IE) const5792   std::string message(int IE) const override {
5793     BitcodeError E = static_cast<BitcodeError>(IE);
5794     switch (E) {
5795     case BitcodeError::InvalidBitcodeSignature:
5796       return "Invalid bitcode signature";
5797     case BitcodeError::CorruptedBitcode:
5798       return "Corrupted bitcode";
5799     }
5800     llvm_unreachable("Unknown error type!");
5801   }
5802 };
5803 }
5804 
5805 static ManagedStatic<BitcodeErrorCategoryType> ErrorCategory;
5806 
BitcodeErrorCategory()5807 const std::error_category &llvm::BitcodeErrorCategory() {
5808   return *ErrorCategory;
5809 }
5810 
5811 //===----------------------------------------------------------------------===//
5812 // External interface
5813 //===----------------------------------------------------------------------===//
5814 
5815 static ErrorOr<std::unique_ptr<Module>>
getBitcodeModuleImpl(std::unique_ptr<DataStreamer> Streamer,StringRef Name,BitcodeReader * R,LLVMContext & Context,bool MaterializeAll,bool ShouldLazyLoadMetadata)5816 getBitcodeModuleImpl(std::unique_ptr<DataStreamer> Streamer, StringRef Name,
5817                      BitcodeReader *R, LLVMContext &Context,
5818                      bool MaterializeAll, bool ShouldLazyLoadMetadata) {
5819   std::unique_ptr<Module> M = make_unique<Module>(Name, Context);
5820   M->setMaterializer(R);
5821 
5822   auto cleanupOnError = [&](std::error_code EC) {
5823     R->releaseBuffer(); // Never take ownership on error.
5824     return EC;
5825   };
5826 
5827   // Delay parsing Metadata if ShouldLazyLoadMetadata is true.
5828   if (std::error_code EC = R->parseBitcodeInto(std::move(Streamer), M.get(),
5829                                                ShouldLazyLoadMetadata))
5830     return cleanupOnError(EC);
5831 
5832   if (MaterializeAll) {
5833     // Read in the entire module, and destroy the BitcodeReader.
5834     if (std::error_code EC = M->materializeAll())
5835       return cleanupOnError(EC);
5836   } else {
5837     // Resolve forward references from blockaddresses.
5838     if (std::error_code EC = R->materializeForwardReferencedFunctions())
5839       return cleanupOnError(EC);
5840   }
5841   return std::move(M);
5842 }
5843 
5844 /// \brief Get a lazy one-at-time loading module from bitcode.
5845 ///
5846 /// This isn't always used in a lazy context.  In particular, it's also used by
5847 /// \a parseBitcodeFile().  If this is truly lazy, then we need to eagerly pull
5848 /// in forward-referenced functions from block address references.
5849 ///
5850 /// \param[in] MaterializeAll Set to \c true if we should materialize
5851 /// everything.
5852 static ErrorOr<std::unique_ptr<Module>>
getLazyBitcodeModuleImpl(std::unique_ptr<MemoryBuffer> && Buffer,LLVMContext & Context,bool MaterializeAll,bool ShouldLazyLoadMetadata=false)5853 getLazyBitcodeModuleImpl(std::unique_ptr<MemoryBuffer> &&Buffer,
5854                          LLVMContext &Context, bool MaterializeAll,
5855                          bool ShouldLazyLoadMetadata = false) {
5856   BitcodeReader *R = new BitcodeReader(Buffer.get(), Context);
5857 
5858   ErrorOr<std::unique_ptr<Module>> Ret =
5859       getBitcodeModuleImpl(nullptr, Buffer->getBufferIdentifier(), R, Context,
5860                            MaterializeAll, ShouldLazyLoadMetadata);
5861   if (!Ret)
5862     return Ret;
5863 
5864   Buffer.release(); // The BitcodeReader owns it now.
5865   return Ret;
5866 }
5867 
5868 ErrorOr<std::unique_ptr<Module>>
getLazyBitcodeModule(std::unique_ptr<MemoryBuffer> && Buffer,LLVMContext & Context,bool ShouldLazyLoadMetadata)5869 llvm::getLazyBitcodeModule(std::unique_ptr<MemoryBuffer> &&Buffer,
5870                            LLVMContext &Context, bool ShouldLazyLoadMetadata) {
5871   return getLazyBitcodeModuleImpl(std::move(Buffer), Context, false,
5872                                   ShouldLazyLoadMetadata);
5873 }
5874 
5875 ErrorOr<std::unique_ptr<Module>>
getStreamedBitcodeModule(StringRef Name,std::unique_ptr<DataStreamer> Streamer,LLVMContext & Context)5876 llvm::getStreamedBitcodeModule(StringRef Name,
5877                                std::unique_ptr<DataStreamer> Streamer,
5878                                LLVMContext &Context) {
5879   std::unique_ptr<Module> M = make_unique<Module>(Name, Context);
5880   BitcodeReader *R = new BitcodeReader(Context);
5881 
5882   return getBitcodeModuleImpl(std::move(Streamer), Name, R, Context, false,
5883                               false);
5884 }
5885 
parseBitcodeFile(MemoryBufferRef Buffer,LLVMContext & Context)5886 ErrorOr<std::unique_ptr<Module>> llvm::parseBitcodeFile(MemoryBufferRef Buffer,
5887                                                         LLVMContext &Context) {
5888   std::unique_ptr<MemoryBuffer> Buf = MemoryBuffer::getMemBuffer(Buffer, false);
5889   return getLazyBitcodeModuleImpl(std::move(Buf), Context, true);
5890   // TODO: Restore the use-lists to the in-memory state when the bitcode was
5891   // written.  We must defer until the Module has been fully materialized.
5892 }
5893 
getBitcodeTargetTriple(MemoryBufferRef Buffer,LLVMContext & Context)5894 std::string llvm::getBitcodeTargetTriple(MemoryBufferRef Buffer,
5895                                          LLVMContext &Context) {
5896   std::unique_ptr<MemoryBuffer> Buf = MemoryBuffer::getMemBuffer(Buffer, false);
5897   auto R = llvm::make_unique<BitcodeReader>(Buf.release(), Context);
5898   ErrorOr<std::string> Triple = R->parseTriple();
5899   if (Triple.getError())
5900     return "";
5901   return Triple.get();
5902 }
5903 
getBitcodeProducerString(MemoryBufferRef Buffer,LLVMContext & Context)5904 std::string llvm::getBitcodeProducerString(MemoryBufferRef Buffer,
5905                                            LLVMContext &Context) {
5906   std::unique_ptr<MemoryBuffer> Buf = MemoryBuffer::getMemBuffer(Buffer, false);
5907   BitcodeReader R(Buf.release(), Context);
5908   ErrorOr<std::string> ProducerString = R.parseIdentificationBlock();
5909   if (ProducerString.getError())
5910     return "";
5911   return ProducerString.get();
5912 }
5913 
5914 // Parse the specified bitcode buffer, returning the function info index.
5915 // If IsLazy is false, parse the entire function summary into
5916 // the index. Otherwise skip the function summary section, and only create
5917 // an index object with a map from function name to function summary offset.
5918 // The index is used to perform lazy function summary reading later.
5919 ErrorOr<std::unique_ptr<FunctionInfoIndex>>
getFunctionInfoIndex(MemoryBufferRef Buffer,DiagnosticHandlerFunction DiagnosticHandler,bool IsLazy)5920 llvm::getFunctionInfoIndex(MemoryBufferRef Buffer,
5921                            DiagnosticHandlerFunction DiagnosticHandler,
5922                            bool IsLazy) {
5923   std::unique_ptr<MemoryBuffer> Buf = MemoryBuffer::getMemBuffer(Buffer, false);
5924   FunctionIndexBitcodeReader R(Buf.get(), DiagnosticHandler, IsLazy);
5925 
5926   auto Index = llvm::make_unique<FunctionInfoIndex>();
5927 
5928   auto cleanupOnError = [&](std::error_code EC) {
5929     R.releaseBuffer(); // Never take ownership on error.
5930     return EC;
5931   };
5932 
5933   if (std::error_code EC = R.parseSummaryIndexInto(nullptr, Index.get()))
5934     return cleanupOnError(EC);
5935 
5936   Buf.release(); // The FunctionIndexBitcodeReader owns it now.
5937   return std::move(Index);
5938 }
5939 
5940 // Check if the given bitcode buffer contains a function summary block.
hasFunctionSummary(MemoryBufferRef Buffer,DiagnosticHandlerFunction DiagnosticHandler)5941 bool llvm::hasFunctionSummary(MemoryBufferRef Buffer,
5942                               DiagnosticHandlerFunction DiagnosticHandler) {
5943   std::unique_ptr<MemoryBuffer> Buf = MemoryBuffer::getMemBuffer(Buffer, false);
5944   FunctionIndexBitcodeReader R(Buf.get(), DiagnosticHandler, false, true);
5945 
5946   auto cleanupOnError = [&](std::error_code EC) {
5947     R.releaseBuffer(); // Never take ownership on error.
5948     return false;
5949   };
5950 
5951   if (std::error_code EC = R.parseSummaryIndexInto(nullptr, nullptr))
5952     return cleanupOnError(EC);
5953 
5954   Buf.release(); // The FunctionIndexBitcodeReader owns it now.
5955   return R.foundFuncSummary();
5956 }
5957 
5958 // This method supports lazy reading of function summary data from the combined
5959 // index during ThinLTO function importing. When reading the combined index
5960 // file, getFunctionInfoIndex is first invoked with IsLazy=true.
5961 // Then this method is called for each function considered for importing,
5962 // to parse the summary information for the given function name into
5963 // the index.
readFunctionSummary(MemoryBufferRef Buffer,DiagnosticHandlerFunction DiagnosticHandler,StringRef FunctionName,std::unique_ptr<FunctionInfoIndex> Index)5964 std::error_code llvm::readFunctionSummary(
5965     MemoryBufferRef Buffer, DiagnosticHandlerFunction DiagnosticHandler,
5966     StringRef FunctionName, std::unique_ptr<FunctionInfoIndex> Index) {
5967   std::unique_ptr<MemoryBuffer> Buf = MemoryBuffer::getMemBuffer(Buffer, false);
5968   FunctionIndexBitcodeReader R(Buf.get(), DiagnosticHandler);
5969 
5970   auto cleanupOnError = [&](std::error_code EC) {
5971     R.releaseBuffer(); // Never take ownership on error.
5972     return EC;
5973   };
5974 
5975   // Lookup the given function name in the FunctionMap, which may
5976   // contain a list of function infos in the case of a COMDAT. Walk through
5977   // and parse each function summary info at the function summary offset
5978   // recorded when parsing the value symbol table.
5979   for (const auto &FI : Index->getFunctionInfoList(FunctionName)) {
5980     size_t FunctionSummaryOffset = FI->bitcodeIndex();
5981     if (std::error_code EC =
5982             R.parseFunctionSummary(nullptr, Index.get(), FunctionSummaryOffset))
5983       return cleanupOnError(EC);
5984   }
5985 
5986   Buf.release(); // The FunctionIndexBitcodeReader owns it now.
5987   return std::error_code();
5988 }
5989