1 //===-- Reader.h - Interface To Bytecode Reading ----------------*- C++ -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file was developed by Reid Spencer and is distributed under the
6 // University of Illinois Open Source License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This header file defines the interface to the Bytecode Reader which is
11 // responsible for correctly interpreting bytecode files (backwards compatible)
12 // and materializing a module from the bytecode read.
14 //===----------------------------------------------------------------------===//
16 #ifndef BYTECODE_PARSER_H
17 #define BYTECODE_PARSER_H
19 #include "llvm/Constants.h"
20 #include "llvm/DerivedTypes.h"
21 #include "llvm/ModuleProvider.h"
22 #include "llvm/Bytecode/Analyzer.h"
23 #include "llvm/ADT/SmallVector.h"
29 // Forward declarations
30 class BytecodeHandler;
31 class TypeSymbolTable;
32 class ValueSymbolTable;
34 /// This class defines the interface for parsing a buffer of bytecode. The
35 /// parser itself takes no action except to call the various functions of
36 /// the handler interface. The parser's sole responsibility is the correct
37 /// interpretation of the bytecode buffer. The handler is responsible for
38 /// instantiating and keeping track of all values. As a convenience, the parser
39 /// is responsible for materializing types and will pass them through the
40 /// handler interface as necessary.
41 /// @see BytecodeHandler
42 /// @brief Bytecode Reader interface
43 class BytecodeReader : public ModuleProvider {
45 /// @name Constructors
48 /// @brief Default constructor. By default, no handler is used.
49 BytecodeReader(BytecodeHandler* h = 0) {
50 decompressedBlock = 0;
56 if (decompressedBlock) {
57 ::free(decompressedBlock);
58 decompressedBlock = 0;
67 /// @brief A convenience type for the buffer pointer
68 typedef const unsigned char* BufPtr;
70 /// @brief The type used for a vector of potentially abstract types
71 typedef std::vector<PATypeHolder> TypeListTy;
73 /// This type provides a vector of Value* via the User class for
74 /// storage of Values that have been constructed when reading the
75 /// bytecode. Because of forward referencing, constant replacement
76 /// can occur so we ensure that our list of Value* is updated
77 /// properly through those transitions. This ensures that the
78 /// correct Value* is in our list when it comes time to associate
79 /// constants with global variables at the end of reading the
81 /// @brief A list of values as a User of those Values.
82 class ValueList : public User {
83 std::vector<Use> Uses;
85 ValueList() : User(Type::VoidTy, Value::ArgumentVal, 0, 0) {}
87 // vector compatibility methods
88 unsigned size() const { return getNumOperands(); }
89 void push_back(Value *V) {
90 Uses.push_back(Use(V, this));
91 OperandList = &Uses[0];
94 Value *back() const { return Uses.back(); }
95 void pop_back() { Uses.pop_back(); --NumOperands; }
96 bool empty() const { return NumOperands == 0; }
97 virtual void print(std::ostream& os) const {
98 for (unsigned i = 0; i < size(); ++i) {
100 getOperand(i)->print(os);
106 /// @brief A 2 dimensional table of values
107 typedef std::vector<ValueList*> ValueTable;
109 /// This map is needed so that forward references to constants can be looked
110 /// up by Type and slot number when resolving those references.
111 /// @brief A mapping of a Type/slot pair to a Constant*.
112 typedef std::map<std::pair<unsigned,unsigned>, Constant*> ConstantRefsType;
114 /// For lazy read-in of functions, we need to save the location in the
115 /// data stream where the function is located. This structure provides that
116 /// information. Lazy read-in is used mostly by the JIT which only wants to
117 /// resolve functions as it needs them.
118 /// @brief Keeps pointers to function contents for later use.
119 struct LazyFunctionInfo {
120 const unsigned char *Buf, *EndBuf;
121 LazyFunctionInfo(const unsigned char *B = 0, const unsigned char *EB = 0)
122 : Buf(B), EndBuf(EB) {}
125 /// @brief A mapping of functions to their LazyFunctionInfo for lazy reading.
126 typedef std::map<Function*, LazyFunctionInfo> LazyFunctionMap;
128 /// @brief A list of global variables and the slot number that initializes
130 typedef std::vector<std::pair<GlobalVariable*, unsigned> > GlobalInitsList;
132 /// This type maps a typeslot/valueslot pair to the corresponding Value*.
133 /// It is used for dealing with forward references as values are read in.
134 /// @brief A map for dealing with forward references of values.
135 typedef std::map<std::pair<unsigned,unsigned>,Value*> ForwardReferenceMap;
141 typedef size_t BCDecompressor_t(const char *, size_t, char*&, std::string*);
143 /// @returns true if an error occurred
144 /// @brief Main interface to parsing a bytecode buffer.
146 volatile BufPtr Buf, ///< Beginning of the bytecode buffer
147 unsigned Length, ///< Length of the bytecode buffer
148 const std::string &ModuleID, ///< An identifier for the module constructed.
149 BCDecompressor_t *Decompressor = 0, ///< Optional decompressor.
150 std::string* ErrMsg = 0 ///< Optional place for error message
153 /// @brief Parse all function bodies
154 bool ParseAllFunctionBodies(std::string* ErrMsg);
156 /// @brief Parse the next function of specific type
157 bool ParseFunction(Function* Func, std::string* ErrMsg) ;
159 /// This method is abstract in the parent ModuleProvider class. Its
160 /// implementation is identical to the ParseFunction method.
161 /// @see ParseFunction
162 /// @brief Make a specific function materialize.
163 virtual bool materializeFunction(Function *F, std::string *ErrMsg = 0) {
164 LazyFunctionMap::iterator Fi = LazyFunctionLoadMap.find(F);
165 if (Fi == LazyFunctionLoadMap.end())
167 if (ParseFunction(F,ErrMsg))
172 /// This method is abstract in the parent ModuleProvider class. Its
173 /// implementation is identical to ParseAllFunctionBodies.
174 /// @see ParseAllFunctionBodies
175 /// @brief Make the whole module materialize
176 virtual Module* materializeModule(std::string *ErrMsg = 0) {
177 if (ParseAllFunctionBodies(ErrMsg))
182 /// This method is provided by the parent ModuleProvde class and overriden
183 /// here. It simply releases the module from its provided and frees up our
185 /// @brief Release our hold on the generated module
186 Module* releaseModule(std::string *ErrInfo = 0) {
187 // Since we're losing control of this Module, we must hand it back complete
188 Module *M = ModuleProvider::releaseModule(ErrInfo);
194 /// @name Parsing Units For Subclasses
197 /// @brief Parse whole module scope
200 /// @brief Parse the version information block
201 void ParseVersionInfo();
203 /// @brief Parse the ModuleGlobalInfo block
204 void ParseModuleGlobalInfo();
206 /// @brief Parse a value symbol table
207 void ParseTypeSymbolTable(TypeSymbolTable *ST);
209 /// @brief Parse a value symbol table
210 void ParseValueSymbolTable(Function* Func, ValueSymbolTable *ST);
212 /// @brief Parse functions lazily.
213 void ParseFunctionLazily();
215 /// @brief Parse a function body
216 void ParseFunctionBody(Function* Func);
218 /// @brief Parse global types
219 void ParseGlobalTypes();
221 /// @brief Parse a basic block (for LLVM 1.0 basic block blocks)
222 BasicBlock* ParseBasicBlock(unsigned BlockNo);
224 /// @brief parse an instruction list (for post LLVM 1.0 instruction lists
225 /// with blocks differentiated by terminating instructions.
226 unsigned ParseInstructionList(
227 Function* F ///< The function into which BBs will be inserted
230 /// @brief Parse a single instruction.
231 void ParseInstruction(
232 SmallVector <unsigned, 8>& Args, ///< The arguments to be filled in
233 BasicBlock* BB ///< The BB the instruction goes in
236 /// @brief Parse the whole constant pool
237 void ParseConstantPool(ValueTable& Values, TypeListTy& Types,
240 /// @brief Parse a single constant pool value
241 Value *ParseConstantPoolValue(unsigned TypeID);
243 /// @brief Parse a block of types constants
244 void ParseTypes(TypeListTy &Tab, unsigned NumEntries);
246 /// @brief Parse a single type constant
247 const Type *ParseType();
249 /// @brief Parse a string constants block
250 void ParseStringConstants(unsigned NumEntries, ValueTable &Tab);
252 /// @brief Release our memory.
254 freeTable(FunctionValues);
255 freeTable(ModuleValues);
262 std::string ErrorMsg; ///< A place to hold an error message through longjmp
263 jmp_buf context; ///< Where to return to if an error occurs.
264 char* decompressedBlock; ///< Result of decompression
265 BufPtr MemStart; ///< Start of the memory buffer
266 BufPtr MemEnd; ///< End of the memory buffer
267 BufPtr BlockStart; ///< Start of current block being parsed
268 BufPtr BlockEnd; ///< End of current block being parsed
269 BufPtr At; ///< Where we're currently parsing at
271 /// Information about the module, extracted from the bytecode revision number.
273 unsigned char RevisionNum; // The rev # itself
275 /// @brief This vector is used to deal with forward references to types in
277 TypeListTy ModuleTypes;
279 /// @brief This is an inverse mapping of ModuleTypes from the type to an
280 /// index. Because refining types causes the index of this map to be
281 /// invalidated, any time we refine a type, we clear this cache and recompute
282 /// it next time we need it. These entries are ordered by the pointer value.
283 std::vector<std::pair<const Type*, unsigned> > ModuleTypeIDCache;
285 /// @brief This vector is used to deal with forward references to types in
287 TypeListTy FunctionTypes;
289 /// When the ModuleGlobalInfo section is read, we create a Function object
290 /// for each function in the module. When the function is loaded, after the
291 /// module global info is read, this Function is populated. Until then, the
292 /// functions in this vector just hold the function signature.
293 std::vector<Function*> FunctionSignatureList;
295 /// @brief This is the table of values belonging to the current function
296 ValueTable FunctionValues;
298 /// @brief This is the table of values belonging to the module (global)
299 ValueTable ModuleValues;
301 /// @brief This keeps track of function level forward references.
302 ForwardReferenceMap ForwardReferences;
304 /// @brief The basic blocks we've parsed, while parsing a function.
305 std::vector<BasicBlock*> ParsedBasicBlocks;
307 /// This maintains a mapping between <Type, Slot #>'s and forward references
308 /// to constants. Such values may be referenced before they are defined, and
309 /// if so, the temporary object that they represent is held here. @brief
310 /// Temporary place for forward references to constants.
311 ConstantRefsType ConstantFwdRefs;
313 /// Constant values are read in after global variables. Because of this, we
314 /// must defer setting the initializers on global variables until after module
315 /// level constants have been read. In the mean time, this list keeps track
316 /// of what we must do.
317 GlobalInitsList GlobalInits;
319 // For lazy reading-in of functions, we need to save away several pieces of
320 // information about each function: its begin and end pointer in the buffer
321 // and its FunctionSlot.
322 LazyFunctionMap LazyFunctionLoadMap;
324 /// This stores the parser's handler which is used for handling tasks other
325 /// just than reading bytecode into the IR. If this is non-null, calls on
326 /// the (polymorphic) BytecodeHandler interface (see llvm/Bytecode/Handler.h)
327 /// will be made to report the logical structure of the bytecode file. What
328 /// the handler does with the events it receives is completely orthogonal to
329 /// the business of parsing the bytecode and building the IR. This is used,
330 /// for example, by the llvm-abcd tool for analysis of byte code.
331 /// @brief Handler for parsing events.
332 BytecodeHandler* Handler;
336 /// @name Implementation Details
339 /// @brief Determines if this module has a function or not.
340 bool hasFunctions() { return ! FunctionSignatureList.empty(); }
342 /// @brief Determines if the type id has an implicit null value.
343 bool hasImplicitNull(unsigned TyID );
345 /// @brief Converts a type slot number to its Type*
346 const Type *getType(unsigned ID);
348 /// @brief Read in a type id and turn it into a Type*
349 inline const Type* readType();
351 /// @brief Converts a Type* to its type slot number
352 unsigned getTypeSlot(const Type *Ty);
354 /// @brief Gets the global type corresponding to the TypeId
355 const Type *getGlobalTableType(unsigned TypeId);
357 /// @brief Get a value from its typeid and slot number
358 Value* getValue(unsigned TypeID, unsigned num, bool Create = true);
360 /// @brief Get a basic block for current function
361 BasicBlock *getBasicBlock(unsigned ID);
363 /// @brief Get a constant value from its typeid and value slot.
364 Constant* getConstantValue(unsigned typeSlot, unsigned valSlot);
366 /// @brief Convenience function for getting a constant value when
367 /// the Type has already been resolved.
368 Constant* getConstantValue(const Type *Ty, unsigned valSlot) {
369 return getConstantValue(getTypeSlot(Ty), valSlot);
372 /// @brief Insert a newly created value
373 unsigned insertValue(Value *V, unsigned Type, ValueTable &Table);
375 /// @brief Insert the arguments of a function.
376 void insertArguments(Function* F );
378 /// @brief Resolve all references to the placeholder (if any) for the
380 void ResolveReferencesToConstant(Constant *C, unsigned Typ, unsigned Slot);
382 /// @brief Free a table, making sure to free the ValueList in the table.
383 void freeTable(ValueTable &Tab) {
384 while (!Tab.empty()) {
390 inline void error(const std::string& errmsg);
392 BytecodeReader(const BytecodeReader &); // DO NOT IMPLEMENT
393 void operator=(const BytecodeReader &); // DO NOT IMPLEMENT
395 // This enum provides the values of the well-known type slots that are always
396 // emitted as the first few types in the table by the BytecodeWriter class.
397 enum WellKnownTypeSlots {
398 VoidTypeSlot = 0, ///< TypeID == VoidTyID
399 FloatTySlot = 1, ///< TypeID == FloatTyID
400 DoubleTySlot = 2, ///< TypeID == DoubleTyID
401 LabelTySlot = 3, ///< TypeID == LabelTyID
402 BoolTySlot = 4, ///< TypeID == IntegerTyID, width = 1
403 Int8TySlot = 5, ///< TypeID == IntegerTyID, width = 8
404 Int16TySlot = 6, ///< TypeID == IntegerTyID, width = 16
405 Int32TySlot = 7, ///< TypeID == IntegerTyID, width = 32
406 Int64TySlot = 8 ///< TypeID == IntegerTyID, width = 64
410 /// @name Reader Primitives
414 /// @brief Is there more to parse in the current block?
415 inline bool moreInBlock();
417 /// @brief Have we read past the end of the block
418 inline void checkPastBlockEnd(const char * block_name);
420 /// @brief Align to 32 bits
421 inline void align32();
423 /// @brief Read an unsigned integer as 32-bits
424 inline unsigned read_uint();
426 /// @brief Read an unsigned integer with variable bit rate encoding
427 inline unsigned read_vbr_uint();
429 /// @brief Read an unsigned integer of no more than 24-bits with variable
430 /// bit rate encoding.
431 inline unsigned read_vbr_uint24();
433 /// @brief Read an unsigned 64-bit integer with variable bit rate encoding.
434 inline uint64_t read_vbr_uint64();
436 /// @brief Read a signed 64-bit integer with variable bit rate encoding.
437 inline int64_t read_vbr_int64();
439 /// @brief Read a string
440 inline std::string read_str();
441 inline void read_str(SmallVectorImpl<char> &StrData);
443 /// @brief Read a float value
444 inline void read_float(float& FloatVal);
446 /// @brief Read a double value
447 inline void read_double(double& DoubleVal);
449 /// @brief Read an arbitrary data chunk of fixed length
450 inline void read_data(void *Ptr, void *End);
452 /// @brief Read a bytecode block header
453 inline void read_block(unsigned &Type, unsigned &Size);
457 } // End llvm namespace