1 //===- BitcodeReader.h - Internal BitcodeReader impl ------------*- C++ -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This header defines the BitcodeReader class.
12 //===----------------------------------------------------------------------===//
14 #ifndef LLVM_LIB_BITCODE_READER_BITCODEREADER_H
15 #define LLVM_LIB_BITCODE_READER_BITCODEREADER_H
17 #include "llvm/ADT/DenseMap.h"
18 #include "llvm/Bitcode/BitstreamReader.h"
19 #include "llvm/Bitcode/LLVMBitCodes.h"
20 #include "llvm/IR/Attributes.h"
21 #include "llvm/IR/GVMaterializer.h"
22 #include "llvm/IR/Metadata.h"
23 #include "llvm/IR/OperandTraits.h"
24 #include "llvm/IR/TrackingMDRef.h"
25 #include "llvm/IR/Type.h"
26 #include "llvm/IR/ValueHandle.h"
28 #include <system_error>
36 //===----------------------------------------------------------------------===//
37 // BitcodeReaderValueList Class
38 //===----------------------------------------------------------------------===//
40 class BitcodeReaderValueList {
41 std::vector<WeakVH> ValuePtrs;
43 /// ResolveConstants - As we resolve forward-referenced constants, we add
44 /// information about them to this vector. This allows us to resolve them in
45 /// bulk instead of resolving each reference at a time. See the code in
46 /// ResolveConstantForwardRefs for more information about this.
48 /// The key of this vector is the placeholder constant, the value is the slot
49 /// number that holds the resolved value.
50 typedef std::vector<std::pair<Constant*, unsigned> > ResolveConstantsTy;
51 ResolveConstantsTy ResolveConstants;
54 BitcodeReaderValueList(LLVMContext &C) : Context(C) {}
55 ~BitcodeReaderValueList() {
56 assert(ResolveConstants.empty() && "Constants not resolved?");
59 // vector compatibility methods
60 unsigned size() const { return ValuePtrs.size(); }
61 void resize(unsigned N) { ValuePtrs.resize(N); }
62 void push_back(Value *V) {
63 ValuePtrs.push_back(V);
67 assert(ResolveConstants.empty() && "Constants not resolved?");
71 Value *operator[](unsigned i) const {
72 assert(i < ValuePtrs.size());
76 Value *back() const { return ValuePtrs.back(); }
77 void pop_back() { ValuePtrs.pop_back(); }
78 bool empty() const { return ValuePtrs.empty(); }
79 void shrinkTo(unsigned N) {
80 assert(N <= size() && "Invalid shrinkTo request!");
84 Constant *getConstantFwdRef(unsigned Idx, Type *Ty);
85 Value *getValueFwdRef(unsigned Idx, Type *Ty);
87 void AssignValue(Value *V, unsigned Idx);
89 /// ResolveConstantForwardRefs - Once all constants are read, this method bulk
90 /// resolves any forward references.
91 void ResolveConstantForwardRefs();
95 //===----------------------------------------------------------------------===//
96 // BitcodeReaderMDValueList Class
97 //===----------------------------------------------------------------------===//
99 class BitcodeReaderMDValueList {
104 std::vector<TrackingMDRef> MDValuePtrs;
106 LLVMContext &Context;
108 BitcodeReaderMDValueList(LLVMContext &C)
109 : NumFwdRefs(0), AnyFwdRefs(false), Context(C) {}
111 // vector compatibility methods
112 unsigned size() const { return MDValuePtrs.size(); }
113 void resize(unsigned N) { MDValuePtrs.resize(N); }
114 void push_back(Metadata *MD) { MDValuePtrs.emplace_back(MD); }
115 void clear() { MDValuePtrs.clear(); }
116 Metadata *back() const { return MDValuePtrs.back(); }
117 void pop_back() { MDValuePtrs.pop_back(); }
118 bool empty() const { return MDValuePtrs.empty(); }
120 Metadata *operator[](unsigned i) const {
121 assert(i < MDValuePtrs.size());
122 return MDValuePtrs[i];
125 void shrinkTo(unsigned N) {
126 assert(N <= size() && "Invalid shrinkTo request!");
127 MDValuePtrs.resize(N);
130 Metadata *getValueFwdRef(unsigned Idx);
131 void AssignValue(Metadata *MD, unsigned Idx);
132 void tryToResolveCycles();
135 class BitcodeReader : public GVMaterializer {
136 LLVMContext &Context;
137 DiagnosticHandlerFunction DiagnosticHandler;
139 std::unique_ptr<MemoryBuffer> Buffer;
140 std::unique_ptr<BitstreamReader> StreamFile;
141 BitstreamCursor Stream;
142 DataStreamer *LazyStreamer;
143 uint64_t NextUnreadBit;
144 bool SeenValueSymbolTable;
146 std::vector<Type*> TypeList;
147 BitcodeReaderValueList ValueList;
148 BitcodeReaderMDValueList MDValueList;
149 std::vector<Comdat *> ComdatList;
150 SmallVector<Instruction *, 64> InstructionList;
152 std::vector<std::pair<GlobalVariable*, unsigned> > GlobalInits;
153 std::vector<std::pair<GlobalAlias*, unsigned> > AliasInits;
154 std::vector<std::pair<Function*, unsigned> > FunctionPrefixes;
155 std::vector<std::pair<Function*, unsigned> > FunctionPrologues;
157 SmallVector<Instruction*, 64> InstsWithTBAATag;
159 /// MAttributes - The set of attributes by index. Index zero in the
160 /// file is for null, and is thus not represented here. As such all indices
162 std::vector<AttributeSet> MAttributes;
164 /// \brief The set of attribute groups.
165 std::map<unsigned, AttributeSet> MAttributeGroups;
167 /// FunctionBBs - While parsing a function body, this is a list of the basic
168 /// blocks for the function.
169 std::vector<BasicBlock*> FunctionBBs;
171 // When reading the module header, this list is populated with functions that
172 // have bodies later in the file.
173 std::vector<Function*> FunctionsWithBodies;
175 // When intrinsic functions are encountered which require upgrading they are
176 // stored here with their replacement function.
177 typedef std::vector<std::pair<Function*, Function*> > UpgradedIntrinsicMap;
178 UpgradedIntrinsicMap UpgradedIntrinsics;
180 // Map the bitcode's custom MDKind ID to the Module's MDKind ID.
181 DenseMap<unsigned, unsigned> MDKindMap;
183 // Several operations happen after the module header has been read, but
184 // before function bodies are processed. This keeps track of whether
185 // we've done this yet.
186 bool SeenFirstFunctionBody;
188 /// DeferredFunctionInfo - When function bodies are initially scanned, this
189 /// map contains info about where to find deferred function body in the
191 DenseMap<Function*, uint64_t> DeferredFunctionInfo;
193 /// These are basic blocks forward-referenced by block addresses. They are
194 /// inserted lazily into functions when they're loaded. The basic block ID is
195 /// its index into the vector.
196 DenseMap<Function *, std::vector<BasicBlock *>> BasicBlockFwdRefs;
197 std::deque<Function *> BasicBlockFwdRefQueue;
199 /// UseRelativeIDs - Indicates that we are using a new encoding for
200 /// instruction operands where most operands in the current
201 /// FUNCTION_BLOCK are encoded relative to the instruction number,
202 /// for a more compact encoding. Some instruction operands are not
203 /// relative to the instruction ID: basic block numbers, and types.
204 /// Once the old style function blocks have been phased out, we would
205 /// not need this flag.
208 /// True if all functions will be materialized, negating the need to process
209 /// (e.g.) blockaddress forward references.
210 bool WillMaterializeAllForwardRefs;
212 /// Functions that have block addresses taken. This is usually empty.
213 SmallPtrSet<const Function *, 4> BlockAddressesTaken;
216 std::error_code Error(BitcodeError E, const Twine &Message);
217 std::error_code Error(BitcodeError E);
218 std::error_code Error(const Twine &Message);
220 explicit BitcodeReader(MemoryBuffer *buffer, LLVMContext &C,
221 DiagnosticHandlerFunction DiagnosticHandler);
222 explicit BitcodeReader(DataStreamer *streamer, LLVMContext &C,
223 DiagnosticHandlerFunction DiagnosticHandler);
224 ~BitcodeReader() { FreeState(); }
226 std::error_code materializeForwardReferencedFunctions();
230 void releaseBuffer();
232 bool isDematerializable(const GlobalValue *GV) const override;
233 std::error_code materialize(GlobalValue *GV) override;
234 std::error_code MaterializeModule(Module *M) override;
235 std::vector<StructType *> getIdentifiedStructTypes() const override;
236 void Dematerialize(GlobalValue *GV) override;
238 /// @brief Main interface to parsing a bitcode buffer.
239 /// @returns true if an error occurred.
240 std::error_code ParseBitcodeInto(Module *M);
242 /// @brief Cheap mechanism to just extract module triple
243 /// @returns true if an error occurred.
244 ErrorOr<std::string> parseTriple();
246 static uint64_t decodeSignRotatedValue(uint64_t V);
249 std::vector<StructType *> IdentifiedStructTypes;
250 StructType *createIdentifiedStructType(LLVMContext &Context, StringRef Name);
251 StructType *createIdentifiedStructType(LLVMContext &Context);
253 Type *getTypeByID(unsigned ID);
254 Value *getFnValueByID(unsigned ID, Type *Ty) {
255 if (Ty && Ty->isMetadataTy())
256 return MetadataAsValue::get(Ty->getContext(), getFnMetadataByID(ID));
257 return ValueList.getValueFwdRef(ID, Ty);
259 Metadata *getFnMetadataByID(unsigned ID) {
260 return MDValueList.getValueFwdRef(ID);
262 BasicBlock *getBasicBlock(unsigned ID) const {
263 if (ID >= FunctionBBs.size()) return nullptr; // Invalid ID
264 return FunctionBBs[ID];
266 AttributeSet getAttributes(unsigned i) const {
267 if (i-1 < MAttributes.size())
268 return MAttributes[i-1];
269 return AttributeSet();
272 /// getValueTypePair - Read a value/type pair out of the specified record from
273 /// slot 'Slot'. Increment Slot past the number of slots used in the record.
274 /// Return true on failure.
275 bool getValueTypePair(SmallVectorImpl<uint64_t> &Record, unsigned &Slot,
276 unsigned InstNum, Value *&ResVal) {
277 if (Slot == Record.size()) return true;
278 unsigned ValNo = (unsigned)Record[Slot++];
279 // Adjust the ValNo, if it was encoded relative to the InstNum.
281 ValNo = InstNum - ValNo;
282 if (ValNo < InstNum) {
283 // If this is not a forward reference, just return the value we already
285 ResVal = getFnValueByID(ValNo, nullptr);
286 return ResVal == nullptr;
287 } else if (Slot == Record.size()) {
291 unsigned TypeNo = (unsigned)Record[Slot++];
292 ResVal = getFnValueByID(ValNo, getTypeByID(TypeNo));
293 return ResVal == nullptr;
296 /// popValue - Read a value out of the specified record from slot 'Slot'.
297 /// Increment Slot past the number of slots used by the value in the record.
298 /// Return true if there is an error.
299 bool popValue(SmallVectorImpl<uint64_t> &Record, unsigned &Slot,
300 unsigned InstNum, Type *Ty, Value *&ResVal) {
301 if (getValue(Record, Slot, InstNum, Ty, ResVal))
303 // All values currently take a single record slot.
308 /// getValue -- Like popValue, but does not increment the Slot number.
309 bool getValue(SmallVectorImpl<uint64_t> &Record, unsigned Slot,
310 unsigned InstNum, Type *Ty, Value *&ResVal) {
311 ResVal = getValue(Record, Slot, InstNum, Ty);
312 return ResVal == nullptr;
315 /// getValue -- Version of getValue that returns ResVal directly,
316 /// or 0 if there is an error.
317 Value *getValue(SmallVectorImpl<uint64_t> &Record, unsigned Slot,
318 unsigned InstNum, Type *Ty) {
319 if (Slot == Record.size()) return nullptr;
320 unsigned ValNo = (unsigned)Record[Slot];
321 // Adjust the ValNo, if it was encoded relative to the InstNum.
323 ValNo = InstNum - ValNo;
324 return getFnValueByID(ValNo, Ty);
327 /// getValueSigned -- Like getValue, but decodes signed VBRs.
328 Value *getValueSigned(SmallVectorImpl<uint64_t> &Record, unsigned Slot,
329 unsigned InstNum, Type *Ty) {
330 if (Slot == Record.size()) return nullptr;
331 unsigned ValNo = (unsigned)decodeSignRotatedValue(Record[Slot]);
332 // Adjust the ValNo, if it was encoded relative to the InstNum.
334 ValNo = InstNum - ValNo;
335 return getFnValueByID(ValNo, Ty);
338 /// Converts alignment exponent (i.e. power of two (or zero)) to the
339 /// corresponding alignment to use. If alignment is too large, returns
340 /// a corresponding error code.
341 std::error_code parseAlignmentValue(uint64_t Exponent, unsigned &Alignment);
342 std::error_code ParseAttrKind(uint64_t Code, Attribute::AttrKind *Kind);
343 std::error_code ParseModule(bool Resume);
344 std::error_code ParseAttributeBlock();
345 std::error_code ParseAttributeGroupBlock();
346 std::error_code ParseTypeTable();
347 std::error_code ParseTypeTableBody();
349 std::error_code ParseValueSymbolTable();
350 std::error_code ParseConstants();
351 std::error_code RememberAndSkipFunctionBody();
352 std::error_code ParseFunctionBody(Function *F);
353 std::error_code GlobalCleanup();
354 std::error_code ResolveGlobalAndAliasInits();
355 std::error_code ParseMetadata();
356 std::error_code ParseMetadataAttachment();
357 ErrorOr<std::string> parseModuleTriple();
358 std::error_code ParseUseLists();
359 std::error_code InitStream();
360 std::error_code InitStreamFromBuffer();
361 std::error_code InitLazyStream();
362 std::error_code FindFunctionInStream(
364 DenseMap<Function *, uint64_t>::iterator DeferredFunctionInfoIterator);
367 } // End llvm namespace