1//===- BitcodeReader.h - Internal BitcodeReader impl ------------*- C++ -*-===//
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// This header defines the BitcodeReader class.
11//
12//===----------------------------------------------------------------------===//
13
14#ifndef BITCODE_READER_H
15#define BITCODE_READER_H
16
17#include "llvm/ADT/DenseMap.h"
18#include "llvm/Bitcode/BitstreamReader.h"
19#include "llvm/Bitcode/LLVMBitCodes.h"
20#include "llvm/GVMaterializer.h"
21#include "llvm/IR/Attributes.h"
22#include "llvm/IR/OperandTraits.h"
23#include "llvm/IR/Type.h"
24#include "llvm/Support/ValueHandle.h"
25#include <vector>
26
27namespace llvm {
28  class MemoryBuffer;
29  class LLVMContext;
30
31//===----------------------------------------------------------------------===//
32//                          BitcodeReaderValueList Class
33//===----------------------------------------------------------------------===//
34
35class BitcodeReaderValueList {
36  std::vector<WeakVH> ValuePtrs;
37
38  /// ResolveConstants - As we resolve forward-referenced constants, we add
39  /// information about them to this vector.  This allows us to resolve them in
40  /// bulk instead of resolving each reference at a time.  See the code in
41  /// ResolveConstantForwardRefs for more information about this.
42  ///
43  /// The key of this vector is the placeholder constant, the value is the slot
44  /// number that holds the resolved value.
45  typedef std::vector<std::pair<Constant*, unsigned> > ResolveConstantsTy;
46  ResolveConstantsTy ResolveConstants;
47  LLVMContext &Context;
48public:
49  BitcodeReaderValueList(LLVMContext &C) : Context(C) {}
50  ~BitcodeReaderValueList() {
51    assert(ResolveConstants.empty() && "Constants not resolved?");
52  }
53
54  // vector compatibility methods
55  unsigned size() const { return ValuePtrs.size(); }
56  void resize(unsigned N) { ValuePtrs.resize(N); }
57  void push_back(Value *V) {
58    ValuePtrs.push_back(V);
59  }
60
61  void clear() {
62    assert(ResolveConstants.empty() && "Constants not resolved?");
63    ValuePtrs.clear();
64  }
65
66  Value *operator[](unsigned i) const {
67    assert(i < ValuePtrs.size());
68    return ValuePtrs[i];
69  }
70
71  Value *back() const { return ValuePtrs.back(); }
72    void pop_back() { ValuePtrs.pop_back(); }
73  bool empty() const { return ValuePtrs.empty(); }
74  void shrinkTo(unsigned N) {
75    assert(N <= size() && "Invalid shrinkTo request!");
76    ValuePtrs.resize(N);
77  }
78
79  Constant *getConstantFwdRef(unsigned Idx, Type *Ty);
80  Value *getValueFwdRef(unsigned Idx, Type *Ty);
81
82  void AssignValue(Value *V, unsigned Idx);
83
84  /// ResolveConstantForwardRefs - Once all constants are read, this method bulk
85  /// resolves any forward references.
86  void ResolveConstantForwardRefs();
87};
88
89
90//===----------------------------------------------------------------------===//
91//                          BitcodeReaderMDValueList Class
92//===----------------------------------------------------------------------===//
93
94class BitcodeReaderMDValueList {
95  std::vector<WeakVH> MDValuePtrs;
96
97  LLVMContext &Context;
98public:
99  BitcodeReaderMDValueList(LLVMContext& C) : Context(C) {}
100
101  // vector compatibility methods
102  unsigned size() const       { return MDValuePtrs.size(); }
103  void resize(unsigned N)     { MDValuePtrs.resize(N); }
104  void push_back(Value *V)    { MDValuePtrs.push_back(V);  }
105  void clear()                { MDValuePtrs.clear();  }
106  Value *back() const         { return MDValuePtrs.back(); }
107  void pop_back()             { MDValuePtrs.pop_back(); }
108  bool empty() const          { return MDValuePtrs.empty(); }
109
110  Value *operator[](unsigned i) const {
111    assert(i < MDValuePtrs.size());
112    return MDValuePtrs[i];
113  }
114
115  void shrinkTo(unsigned N) {
116    assert(N <= size() && "Invalid shrinkTo request!");
117    MDValuePtrs.resize(N);
118  }
119
120  Value *getValueFwdRef(unsigned Idx);
121  void AssignValue(Value *V, unsigned Idx);
122};
123
124class BitcodeReader : public GVMaterializer {
125  LLVMContext &Context;
126  Module *TheModule;
127  MemoryBuffer *Buffer;
128  bool BufferOwned;
129  OwningPtr<BitstreamReader> StreamFile;
130  BitstreamCursor Stream;
131  DataStreamer *LazyStreamer;
132  uint64_t NextUnreadBit;
133  bool SeenValueSymbolTable;
134
135  const char *ErrorString;
136
137  std::vector<Type*> TypeList;
138  BitcodeReaderValueList ValueList;
139  BitcodeReaderMDValueList MDValueList;
140  SmallVector<Instruction *, 64> InstructionList;
141  SmallVector<SmallVector<uint64_t, 64>, 64> UseListRecords;
142
143  std::vector<std::pair<GlobalVariable*, unsigned> > GlobalInits;
144  std::vector<std::pair<GlobalAlias*, unsigned> > AliasInits;
145
146  /// MAttributes - The set of attributes by index.  Index zero in the
147  /// file is for null, and is thus not represented here.  As such all indices
148  /// are off by one.
149  std::vector<AttributeSet> MAttributes;
150
151  /// \brief The set of attribute groups.
152  std::map<unsigned, AttributeSet> MAttributeGroups;
153
154  /// FunctionBBs - While parsing a function body, this is a list of the basic
155  /// blocks for the function.
156  std::vector<BasicBlock*> FunctionBBs;
157
158  // When reading the module header, this list is populated with functions that
159  // have bodies later in the file.
160  std::vector<Function*> FunctionsWithBodies;
161
162  // When intrinsic functions are encountered which require upgrading they are
163  // stored here with their replacement function.
164  typedef std::vector<std::pair<Function*, Function*> > UpgradedIntrinsicMap;
165  UpgradedIntrinsicMap UpgradedIntrinsics;
166
167  // Map the bitcode's custom MDKind ID to the Module's MDKind ID.
168  DenseMap<unsigned, unsigned> MDKindMap;
169
170  // Several operations happen after the module header has been read, but
171  // before function bodies are processed. This keeps track of whether
172  // we've done this yet.
173  bool SeenFirstFunctionBody;
174
175  /// DeferredFunctionInfo - When function bodies are initially scanned, this
176  /// map contains info about where to find deferred function body in the
177  /// stream.
178  DenseMap<Function*, uint64_t> DeferredFunctionInfo;
179
180  /// BlockAddrFwdRefs - These are blockaddr references to basic blocks.  These
181  /// are resolved lazily when functions are loaded.
182  typedef std::pair<unsigned, GlobalVariable*> BlockAddrRefTy;
183  DenseMap<Function*, std::vector<BlockAddrRefTy> > BlockAddrFwdRefs;
184
185  /// UseRelativeIDs - Indicates that we are using a new encoding for
186  /// instruction operands where most operands in the current
187  /// FUNCTION_BLOCK are encoded relative to the instruction number,
188  /// for a more compact encoding.  Some instruction operands are not
189  /// relative to the instruction ID: basic block numbers, and types.
190  /// Once the old style function blocks have been phased out, we would
191  /// not need this flag.
192  bool UseRelativeIDs;
193
194public:
195  explicit BitcodeReader(MemoryBuffer *buffer, LLVMContext &C)
196    : Context(C), TheModule(0), Buffer(buffer), BufferOwned(false),
197      LazyStreamer(0), NextUnreadBit(0), SeenValueSymbolTable(false),
198      ErrorString(0), ValueList(C), MDValueList(C),
199      SeenFirstFunctionBody(false), UseRelativeIDs(false) {
200  }
201  explicit BitcodeReader(DataStreamer *streamer, LLVMContext &C)
202    : Context(C), TheModule(0), Buffer(0), BufferOwned(false),
203      LazyStreamer(streamer), NextUnreadBit(0), SeenValueSymbolTable(false),
204      ErrorString(0), ValueList(C), MDValueList(C),
205      SeenFirstFunctionBody(false), UseRelativeIDs(false) {
206  }
207  ~BitcodeReader() {
208    FreeState();
209  }
210
211  void materializeForwardReferencedFunctions();
212
213  void FreeState();
214
215  /// setBufferOwned - If this is true, the reader will destroy the MemoryBuffer
216  /// when the reader is destroyed.
217  void setBufferOwned(bool Owned) { BufferOwned = Owned; }
218
219  virtual bool isMaterializable(const GlobalValue *GV) const;
220  virtual bool isDematerializable(const GlobalValue *GV) const;
221  virtual bool Materialize(GlobalValue *GV, std::string *ErrInfo = 0);
222  virtual bool MaterializeModule(Module *M, std::string *ErrInfo = 0);
223  virtual void Dematerialize(GlobalValue *GV);
224
225  bool Error(const char *Str) {
226    ErrorString = Str;
227    return true;
228  }
229  const char *getErrorString() const { return ErrorString; }
230
231  /// @brief Main interface to parsing a bitcode buffer.
232  /// @returns true if an error occurred.
233  bool ParseBitcodeInto(Module *M);
234
235  /// @brief Cheap mechanism to just extract module triple
236  /// @returns true if an error occurred.
237  bool ParseTriple(std::string &Triple);
238
239  static uint64_t decodeSignRotatedValue(uint64_t V);
240
241private:
242  Type *getTypeByID(unsigned ID);
243  Value *getFnValueByID(unsigned ID, Type *Ty) {
244    if (Ty && Ty->isMetadataTy())
245      return MDValueList.getValueFwdRef(ID);
246    return ValueList.getValueFwdRef(ID, Ty);
247  }
248  BasicBlock *getBasicBlock(unsigned ID) const {
249    if (ID >= FunctionBBs.size()) return 0; // Invalid ID
250    return FunctionBBs[ID];
251  }
252  AttributeSet getAttributes(unsigned i) const {
253    if (i-1 < MAttributes.size())
254      return MAttributes[i-1];
255    return AttributeSet();
256  }
257
258  /// getValueTypePair - Read a value/type pair out of the specified record from
259  /// slot 'Slot'.  Increment Slot past the number of slots used in the record.
260  /// Return true on failure.
261  bool getValueTypePair(SmallVectorImpl<uint64_t> &Record, unsigned &Slot,
262                        unsigned InstNum, Value *&ResVal) {
263    if (Slot == Record.size()) return true;
264    unsigned ValNo = (unsigned)Record[Slot++];
265    // Adjust the ValNo, if it was encoded relative to the InstNum.
266    if (UseRelativeIDs)
267      ValNo = InstNum - ValNo;
268    if (ValNo < InstNum) {
269      // If this is not a forward reference, just return the value we already
270      // have.
271      ResVal = getFnValueByID(ValNo, 0);
272      return ResVal == 0;
273    } else if (Slot == Record.size()) {
274      return true;
275    }
276
277    unsigned TypeNo = (unsigned)Record[Slot++];
278    ResVal = getFnValueByID(ValNo, getTypeByID(TypeNo));
279    return ResVal == 0;
280  }
281
282  /// popValue - Read a value out of the specified record from slot 'Slot'.
283  /// Increment Slot past the number of slots used by the value in the record.
284  /// Return true if there is an error.
285  bool popValue(SmallVectorImpl<uint64_t> &Record, unsigned &Slot,
286                unsigned InstNum, Type *Ty, Value *&ResVal) {
287    if (getValue(Record, Slot, InstNum, Ty, ResVal))
288      return true;
289    // All values currently take a single record slot.
290    ++Slot;
291    return false;
292  }
293
294  /// getValue -- Like popValue, but does not increment the Slot number.
295  bool getValue(SmallVectorImpl<uint64_t> &Record, unsigned Slot,
296                unsigned InstNum, Type *Ty, Value *&ResVal) {
297    ResVal = getValue(Record, Slot, InstNum, Ty);
298    return ResVal == 0;
299  }
300
301  /// getValue -- Version of getValue that returns ResVal directly,
302  /// or 0 if there is an error.
303  Value *getValue(SmallVectorImpl<uint64_t> &Record, unsigned Slot,
304                  unsigned InstNum, Type *Ty) {
305    if (Slot == Record.size()) return 0;
306    unsigned ValNo = (unsigned)Record[Slot];
307    // Adjust the ValNo, if it was encoded relative to the InstNum.
308    if (UseRelativeIDs)
309      ValNo = InstNum - ValNo;
310    return getFnValueByID(ValNo, Ty);
311  }
312
313  /// getValueSigned -- Like getValue, but decodes signed VBRs.
314  Value *getValueSigned(SmallVectorImpl<uint64_t> &Record, unsigned Slot,
315                        unsigned InstNum, Type *Ty) {
316    if (Slot == Record.size()) return 0;
317    unsigned ValNo = (unsigned)decodeSignRotatedValue(Record[Slot]);
318    // Adjust the ValNo, if it was encoded relative to the InstNum.
319    if (UseRelativeIDs)
320      ValNo = InstNum - ValNo;
321    return getFnValueByID(ValNo, Ty);
322  }
323
324  bool ParseAttrKind(uint64_t Code, Attribute::AttrKind *Kind);
325  bool ParseModule(bool Resume);
326  bool ParseAttributeBlock();
327  bool ParseAttributeGroupBlock();
328  bool ParseTypeTable();
329  bool ParseTypeTableBody();
330
331  bool ParseValueSymbolTable();
332  bool ParseConstants();
333  bool RememberAndSkipFunctionBody();
334  bool ParseFunctionBody(Function *F);
335  bool GlobalCleanup();
336  bool ResolveGlobalAndAliasInits();
337  bool ParseMetadata();
338  bool ParseMetadataAttachment();
339  bool ParseModuleTriple(std::string &Triple);
340  bool ParseUseLists();
341  bool InitStream();
342  bool InitStreamFromBuffer();
343  bool InitLazyStream();
344  bool FindFunctionInStream(Function *F,
345         DenseMap<Function*, uint64_t>::iterator DeferredFunctionInfoIterator);
346};
347
348} // End llvm namespace
349
350#endif
351