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 BITCODE_READER_H
15 #define BITCODE_READER_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/OperandTraits.h"
23 #include "llvm/IR/Type.h"
24 #include "llvm/IR/ValueHandle.h"
25 #include <system_error>
32 //===----------------------------------------------------------------------===//
33 // BitcodeReaderValueList Class
34 //===----------------------------------------------------------------------===//
36 class BitcodeReaderValueList {
37 std::vector<WeakVH> ValuePtrs;
39 /// ResolveConstants - As we resolve forward-referenced constants, we add
40 /// information about them to this vector. This allows us to resolve them in
41 /// bulk instead of resolving each reference at a time. See the code in
42 /// ResolveConstantForwardRefs for more information about this.
44 /// The key of this vector is the placeholder constant, the value is the slot
45 /// number that holds the resolved value.
46 typedef std::vector<std::pair<Constant*, unsigned> > ResolveConstantsTy;
47 ResolveConstantsTy ResolveConstants;
50 BitcodeReaderValueList(LLVMContext &C) : Context(C) {}
51 ~BitcodeReaderValueList() {
52 assert(ResolveConstants.empty() && "Constants not resolved?");
55 // vector compatibility methods
56 unsigned size() const { return ValuePtrs.size(); }
57 void resize(unsigned N) { ValuePtrs.resize(N); }
58 void push_back(Value *V) {
59 ValuePtrs.push_back(V);
63 assert(ResolveConstants.empty() && "Constants not resolved?");
67 Value *operator[](unsigned i) const {
68 assert(i < ValuePtrs.size());
72 Value *back() const { return ValuePtrs.back(); }
73 void pop_back() { ValuePtrs.pop_back(); }
74 bool empty() const { return ValuePtrs.empty(); }
75 void shrinkTo(unsigned N) {
76 assert(N <= size() && "Invalid shrinkTo request!");
80 Constant *getConstantFwdRef(unsigned Idx, Type *Ty);
81 Value *getValueFwdRef(unsigned Idx, Type *Ty);
83 void AssignValue(Value *V, unsigned Idx);
85 /// ResolveConstantForwardRefs - Once all constants are read, this method bulk
86 /// resolves any forward references.
87 void ResolveConstantForwardRefs();
91 //===----------------------------------------------------------------------===//
92 // BitcodeReaderMDValueList Class
93 //===----------------------------------------------------------------------===//
95 class BitcodeReaderMDValueList {
96 std::vector<WeakVH> MDValuePtrs;
100 BitcodeReaderMDValueList(LLVMContext& C) : Context(C) {}
102 // vector compatibility methods
103 unsigned size() const { return MDValuePtrs.size(); }
104 void resize(unsigned N) { MDValuePtrs.resize(N); }
105 void push_back(Value *V) { MDValuePtrs.push_back(V); }
106 void clear() { MDValuePtrs.clear(); }
107 Value *back() const { return MDValuePtrs.back(); }
108 void pop_back() { MDValuePtrs.pop_back(); }
109 bool empty() const { return MDValuePtrs.empty(); }
111 Value *operator[](unsigned i) const {
112 assert(i < MDValuePtrs.size());
113 return MDValuePtrs[i];
116 void shrinkTo(unsigned N) {
117 assert(N <= size() && "Invalid shrinkTo request!");
118 MDValuePtrs.resize(N);
121 Value *getValueFwdRef(unsigned Idx);
122 void AssignValue(Value *V, unsigned Idx);
125 class BitcodeReader : public GVMaterializer {
126 LLVMContext &Context;
128 std::unique_ptr<MemoryBuffer> Buffer;
129 std::unique_ptr<BitstreamReader> StreamFile;
130 BitstreamCursor Stream;
131 DataStreamer *LazyStreamer;
132 uint64_t NextUnreadBit;
133 bool SeenValueSymbolTable;
135 std::vector<Type*> TypeList;
136 BitcodeReaderValueList ValueList;
137 BitcodeReaderMDValueList MDValueList;
138 SmallVector<Instruction *, 64> InstructionList;
139 SmallVector<SmallVector<uint64_t, 64>, 64> UseListRecords;
141 std::vector<std::pair<GlobalVariable*, unsigned> > GlobalInits;
142 std::vector<std::pair<GlobalAlias*, unsigned> > AliasInits;
143 std::vector<std::pair<Function*, unsigned> > FunctionPrefixes;
145 SmallVector<Instruction*, 64> InstsWithTBAATag;
147 /// MAttributes - The set of attributes by index. Index zero in the
148 /// file is for null, and is thus not represented here. As such all indices
150 std::vector<AttributeSet> MAttributes;
152 /// \brief The set of attribute groups.
153 std::map<unsigned, AttributeSet> MAttributeGroups;
155 /// FunctionBBs - While parsing a function body, this is a list of the basic
156 /// blocks for the function.
157 std::vector<BasicBlock*> FunctionBBs;
159 // When reading the module header, this list is populated with functions that
160 // have bodies later in the file.
161 std::vector<Function*> FunctionsWithBodies;
163 // When intrinsic functions are encountered which require upgrading they are
164 // stored here with their replacement function.
165 typedef std::vector<std::pair<Function*, Function*> > UpgradedIntrinsicMap;
166 UpgradedIntrinsicMap UpgradedIntrinsics;
168 // Map the bitcode's custom MDKind ID to the Module's MDKind ID.
169 DenseMap<unsigned, unsigned> MDKindMap;
171 // Several operations happen after the module header has been read, but
172 // before function bodies are processed. This keeps track of whether
173 // we've done this yet.
174 bool SeenFirstFunctionBody;
176 /// DeferredFunctionInfo - When function bodies are initially scanned, this
177 /// map contains info about where to find deferred function body in the
179 DenseMap<Function*, uint64_t> DeferredFunctionInfo;
181 /// BlockAddrFwdRefs - These are blockaddr references to basic blocks. These
182 /// are resolved lazily when functions are loaded.
183 typedef std::pair<unsigned, GlobalVariable*> BlockAddrRefTy;
184 DenseMap<Function*, std::vector<BlockAddrRefTy> > BlockAddrFwdRefs;
186 /// UseRelativeIDs - Indicates that we are using a new encoding for
187 /// instruction operands where most operands in the current
188 /// FUNCTION_BLOCK are encoded relative to the instruction number,
189 /// for a more compact encoding. Some instruction operands are not
190 /// relative to the instruction ID: basic block numbers, and types.
191 /// Once the old style function blocks have been phased out, we would
192 /// not need this flag.
195 static const std::error_category &BitcodeErrorCategory();
199 BitcodeStreamInvalidSize,
200 ConflictingMETADATA_KINDRecords,
201 CouldNotFindFunctionInStream,
203 InsufficientFunctionProtos,
204 InvalidBitcodeSignature,
205 InvalidBitcodeWrapperHeader,
206 InvalidConstantReference,
207 InvalidID, // A read identifier is not found in the table it should be in.
208 InvalidInstructionWithNoBB,
209 InvalidRecord, // A read record doesn't have the expected size or structure
210 InvalidTypeForValue, // Type read OK, but is invalid for its use
212 InvalidType, // We were unable to read a type
213 MalformedBlock, // We are unable to advance in the stream.
214 MalformedGlobalInitializerSet,
215 InvalidMultipleBlocks, // We found multiple blocks of a kind that should
217 NeverResolvedValueFoundInFunction,
218 InvalidValue // Invalid version, inst number, attr number, etc
221 std::error_code Error(ErrorType E) {
222 return std::error_code(E, BitcodeErrorCategory());
225 explicit BitcodeReader(MemoryBuffer *buffer, LLVMContext &C)
226 : Context(C), TheModule(nullptr), Buffer(buffer), LazyStreamer(nullptr),
227 NextUnreadBit(0), SeenValueSymbolTable(false), ValueList(C),
228 MDValueList(C), SeenFirstFunctionBody(false), UseRelativeIDs(false) {}
229 explicit BitcodeReader(DataStreamer *streamer, LLVMContext &C)
230 : Context(C), TheModule(nullptr), Buffer(nullptr), LazyStreamer(streamer),
231 NextUnreadBit(0), SeenValueSymbolTable(false), ValueList(C),
232 MDValueList(C), SeenFirstFunctionBody(false), UseRelativeIDs(false) {}
233 ~BitcodeReader() { FreeState(); }
235 void materializeForwardReferencedFunctions();
239 void releaseBuffer() override;
241 bool isMaterializable(const GlobalValue *GV) const override;
242 bool isDematerializable(const GlobalValue *GV) const override;
243 std::error_code Materialize(GlobalValue *GV) override;
244 std::error_code MaterializeModule(Module *M) override;
245 void Dematerialize(GlobalValue *GV) override;
247 /// @brief Main interface to parsing a bitcode buffer.
248 /// @returns true if an error occurred.
249 std::error_code ParseBitcodeInto(Module *M);
251 /// @brief Cheap mechanism to just extract module triple
252 /// @returns true if an error occurred.
253 std::error_code ParseTriple(std::string &Triple);
255 static uint64_t decodeSignRotatedValue(uint64_t V);
258 Type *getTypeByID(unsigned ID);
259 Value *getFnValueByID(unsigned ID, Type *Ty) {
260 if (Ty && Ty->isMetadataTy())
261 return MDValueList.getValueFwdRef(ID);
262 return ValueList.getValueFwdRef(ID, Ty);
264 BasicBlock *getBasicBlock(unsigned ID) const {
265 if (ID >= FunctionBBs.size()) return nullptr; // Invalid ID
266 return FunctionBBs[ID];
268 AttributeSet getAttributes(unsigned i) const {
269 if (i-1 < MAttributes.size())
270 return MAttributes[i-1];
271 return AttributeSet();
274 /// getValueTypePair - Read a value/type pair out of the specified record from
275 /// slot 'Slot'. Increment Slot past the number of slots used in the record.
276 /// Return true on failure.
277 bool getValueTypePair(SmallVectorImpl<uint64_t> &Record, unsigned &Slot,
278 unsigned InstNum, Value *&ResVal) {
279 if (Slot == Record.size()) return true;
280 unsigned ValNo = (unsigned)Record[Slot++];
281 // Adjust the ValNo, if it was encoded relative to the InstNum.
283 ValNo = InstNum - ValNo;
284 if (ValNo < InstNum) {
285 // If this is not a forward reference, just return the value we already
287 ResVal = getFnValueByID(ValNo, nullptr);
288 return ResVal == nullptr;
289 } else if (Slot == Record.size()) {
293 unsigned TypeNo = (unsigned)Record[Slot++];
294 ResVal = getFnValueByID(ValNo, getTypeByID(TypeNo));
295 return ResVal == nullptr;
298 /// popValue - Read a value out of the specified record from slot 'Slot'.
299 /// Increment Slot past the number of slots used by the value in the record.
300 /// Return true if there is an error.
301 bool popValue(SmallVectorImpl<uint64_t> &Record, unsigned &Slot,
302 unsigned InstNum, Type *Ty, Value *&ResVal) {
303 if (getValue(Record, Slot, InstNum, Ty, ResVal))
305 // All values currently take a single record slot.
310 /// getValue -- Like popValue, but does not increment the Slot number.
311 bool getValue(SmallVectorImpl<uint64_t> &Record, unsigned Slot,
312 unsigned InstNum, Type *Ty, Value *&ResVal) {
313 ResVal = getValue(Record, Slot, InstNum, Ty);
314 return ResVal == nullptr;
317 /// getValue -- Version of getValue that returns ResVal directly,
318 /// or 0 if there is an error.
319 Value *getValue(SmallVectorImpl<uint64_t> &Record, unsigned Slot,
320 unsigned InstNum, Type *Ty) {
321 if (Slot == Record.size()) return nullptr;
322 unsigned ValNo = (unsigned)Record[Slot];
323 // Adjust the ValNo, if it was encoded relative to the InstNum.
325 ValNo = InstNum - ValNo;
326 return getFnValueByID(ValNo, Ty);
329 /// getValueSigned -- Like getValue, but decodes signed VBRs.
330 Value *getValueSigned(SmallVectorImpl<uint64_t> &Record, unsigned Slot,
331 unsigned InstNum, Type *Ty) {
332 if (Slot == Record.size()) return nullptr;
333 unsigned ValNo = (unsigned)decodeSignRotatedValue(Record[Slot]);
334 // Adjust the ValNo, if it was encoded relative to the InstNum.
336 ValNo = InstNum - ValNo;
337 return getFnValueByID(ValNo, Ty);
340 std::error_code ParseAttrKind(uint64_t Code, Attribute::AttrKind *Kind);
341 std::error_code ParseModule(bool Resume);
342 std::error_code ParseAttributeBlock();
343 std::error_code ParseAttributeGroupBlock();
344 std::error_code ParseTypeTable();
345 std::error_code ParseTypeTableBody();
347 std::error_code ParseValueSymbolTable();
348 std::error_code ParseConstants();
349 std::error_code RememberAndSkipFunctionBody();
350 std::error_code ParseFunctionBody(Function *F);
351 std::error_code GlobalCleanup();
352 std::error_code ResolveGlobalAndAliasInits();
353 std::error_code ParseMetadata();
354 std::error_code ParseMetadataAttachment();
355 std::error_code ParseModuleTriple(std::string &Triple);
356 std::error_code ParseUseLists();
357 std::error_code InitStream();
358 std::error_code InitStreamFromBuffer();
359 std::error_code InitLazyStream();
360 std::error_code FindFunctionInStream(
362 DenseMap<Function *, uint64_t>::iterator DeferredFunctionInfoIterator);
365 } // End llvm namespace