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/GlobalValue.h"
22 #include "llvm/Function.h"
23 #include "llvm/ModuleProvider.h"
24 #include "llvm/Bytecode/Analyzer.h"
31 class BytecodeHandler; ///< Forward declare the handler interface
32 class TypeSymbolTable; ///< Forward declare
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 /// @returns true if an error occurred
142 /// @brief Main interface to parsing a bytecode buffer.
144 volatile BufPtr Buf, ///< Beginning of the bytecode buffer
145 unsigned Length, ///< Length of the bytecode buffer
146 const std::string &ModuleID, ///< An identifier for the module constructed.
147 std::string* ErrMsg = 0 ///< Optional place for error message
150 /// @brief Parse all function bodies
151 bool ParseAllFunctionBodies(std::string* ErrMsg);
153 /// @brief Parse the next function of specific type
154 bool ParseFunction(Function* Func, std::string* ErrMsg) ;
156 /// This method is abstract in the parent ModuleProvider class. Its
157 /// implementation is identical to the ParseFunction method.
158 /// @see ParseFunction
159 /// @brief Make a specific function materialize.
160 virtual bool materializeFunction(Function *F, std::string *ErrMsg = 0) {
161 LazyFunctionMap::iterator Fi = LazyFunctionLoadMap.find(F);
162 if (Fi == LazyFunctionLoadMap.end())
164 if (ParseFunction(F,ErrMsg))
169 /// This method is abstract in the parent ModuleProvider class. Its
170 /// implementation is identical to ParseAllFunctionBodies.
171 /// @see ParseAllFunctionBodies
172 /// @brief Make the whole module materialize
173 virtual Module* materializeModule(std::string *ErrMsg = 0) {
174 if (ParseAllFunctionBodies(ErrMsg))
179 /// This method is provided by the parent ModuleProvde class and overriden
180 /// here. It simply releases the module from its provided and frees up our
182 /// @brief Release our hold on the generated module
183 Module* releaseModule(std::string *ErrInfo = 0) {
184 // Since we're losing control of this Module, we must hand it back complete
185 Module *M = ModuleProvider::releaseModule(ErrInfo);
191 /// @name Parsing Units For Subclasses
194 /// @brief Parse whole module scope
197 /// @brief Parse the version information block
198 void ParseVersionInfo();
200 /// @brief Parse the ModuleGlobalInfo block
201 void ParseModuleGlobalInfo();
203 /// @brief Parse a value symbol table
204 void ParseTypeSymbolTable(TypeSymbolTable *ST);
206 /// @brief Parse a value symbol table
207 void ParseValueSymbolTable(Function* Func, SymbolTable *ST);
209 /// @brief Parse functions lazily.
210 void ParseFunctionLazily();
212 /// @brief Parse a function body
213 void ParseFunctionBody(Function* Func);
215 /// @brief Parse global types
216 void ParseGlobalTypes();
218 /// @brief Parse a basic block (for LLVM 1.0 basic block blocks)
219 BasicBlock* ParseBasicBlock(unsigned BlockNo);
221 /// @brief parse an instruction list (for post LLVM 1.0 instruction lists
222 /// with blocks differentiated by terminating instructions.
223 unsigned ParseInstructionList(
224 Function* F ///< The function into which BBs will be inserted
227 /// @brief Parse a single instruction.
228 void ParseInstruction(
229 std::vector<unsigned>& Args, ///< The arguments to be filled in
230 BasicBlock* BB ///< The BB the instruction goes in
233 /// @brief Parse the whole constant pool
234 void ParseConstantPool(ValueTable& Values, TypeListTy& Types,
237 /// @brief Parse a single constant pool value
238 Value *ParseConstantPoolValue(unsigned TypeID);
240 /// @brief Parse a block of types constants
241 void ParseTypes(TypeListTy &Tab, unsigned NumEntries);
243 /// @brief Parse a single type constant
244 const Type *ParseType();
246 /// @brief Parse a string constants block
247 void ParseStringConstants(unsigned NumEntries, ValueTable &Tab);
249 /// @brief Release our memory.
251 freeTable(FunctionValues);
252 freeTable(ModuleValues);
259 std::string ErrorMsg; ///< A place to hold an error message through longjmp
260 jmp_buf context; ///< Where to return to if an error occurs.
261 char* decompressedBlock; ///< Result of decompression
262 BufPtr MemStart; ///< Start of the memory buffer
263 BufPtr MemEnd; ///< End of the memory buffer
264 BufPtr BlockStart; ///< Start of current block being parsed
265 BufPtr BlockEnd; ///< End of current block being parsed
266 BufPtr At; ///< Where we're currently parsing at
268 /// Information about the module, extracted from the bytecode revision number.
270 unsigned char RevisionNum; // The rev # itself
272 /// @brief This vector is used to deal with forward references to types in
274 TypeListTy ModuleTypes;
276 /// @brief This is an inverse mapping of ModuleTypes from the type to an
277 /// index. Because refining types causes the index of this map to be
278 /// invalidated, any time we refine a type, we clear this cache and recompute
279 /// it next time we need it. These entries are ordered by the pointer value.
280 std::vector<std::pair<const Type*, unsigned> > ModuleTypeIDCache;
282 /// @brief This vector is used to deal with forward references to types in
284 TypeListTy FunctionTypes;
286 /// When the ModuleGlobalInfo section is read, we create a Function object
287 /// for each function in the module. When the function is loaded, after the
288 /// module global info is read, this Function is populated. Until then, the
289 /// functions in this vector just hold the function signature.
290 std::vector<Function*> FunctionSignatureList;
292 /// @brief This is the table of values belonging to the current function
293 ValueTable FunctionValues;
295 /// @brief This is the table of values belonging to the module (global)
296 ValueTable ModuleValues;
298 /// @brief This keeps track of function level forward references.
299 ForwardReferenceMap ForwardReferences;
301 /// @brief The basic blocks we've parsed, while parsing a function.
302 std::vector<BasicBlock*> ParsedBasicBlocks;
304 /// This maintains a mapping between <Type, Slot #>'s and forward references
305 /// to constants. Such values may be referenced before they are defined, and
306 /// if so, the temporary object that they represent is held here. @brief
307 /// Temporary place for forward references to constants.
308 ConstantRefsType ConstantFwdRefs;
310 /// Constant values are read in after global variables. Because of this, we
311 /// must defer setting the initializers on global variables until after module
312 /// level constants have been read. In the mean time, this list keeps track
313 /// of what we must do.
314 GlobalInitsList GlobalInits;
316 // For lazy reading-in of functions, we need to save away several pieces of
317 // information about each function: its begin and end pointer in the buffer
318 // and its FunctionSlot.
319 LazyFunctionMap LazyFunctionLoadMap;
321 /// This stores the parser's handler which is used for handling tasks other
322 /// just than reading bytecode into the IR. If this is non-null, calls on
323 /// the (polymorphic) BytecodeHandler interface (see llvm/Bytecode/Handler.h)
324 /// will be made to report the logical structure of the bytecode file. What
325 /// the handler does with the events it receives is completely orthogonal to
326 /// the business of parsing the bytecode and building the IR. This is used,
327 /// for example, by the llvm-abcd tool for analysis of byte code.
328 /// @brief Handler for parsing events.
329 BytecodeHandler* Handler;
333 /// @name Implementation Details
336 /// @brief Determines if this module has a function or not.
337 bool hasFunctions() { return ! FunctionSignatureList.empty(); }
339 /// @brief Determines if the type id has an implicit null value.
340 bool hasImplicitNull(unsigned TyID );
342 /// @brief Converts a type slot number to its Type*
343 const Type *getType(unsigned ID);
345 /// @brief Read in a type id and turn it into a Type*
346 inline const Type* readType();
348 /// @brief Converts a Type* to its type slot number
349 unsigned getTypeSlot(const Type *Ty);
351 /// @brief Gets the global type corresponding to the TypeId
352 const Type *getGlobalTableType(unsigned TypeId);
354 /// @brief Get a value from its typeid and slot number
355 Value* getValue(unsigned TypeID, unsigned num, bool Create = true);
357 /// @brief Get a basic block for current function
358 BasicBlock *getBasicBlock(unsigned ID);
360 /// @brief Get a constant value from its typeid and value slot.
361 Constant* getConstantValue(unsigned typeSlot, unsigned valSlot);
363 /// @brief Convenience function for getting a constant value when
364 /// the Type has already been resolved.
365 Constant* getConstantValue(const Type *Ty, unsigned valSlot) {
366 return getConstantValue(getTypeSlot(Ty), valSlot);
369 /// @brief Insert a newly created value
370 unsigned insertValue(Value *V, unsigned Type, ValueTable &Table);
372 /// @brief Insert the arguments of a function.
373 void insertArguments(Function* F );
375 /// @brief Resolve all references to the placeholder (if any) for the
377 void ResolveReferencesToConstant(Constant *C, unsigned Typ, unsigned Slot);
379 /// @brief Free a table, making sure to free the ValueList in the table.
380 void freeTable(ValueTable &Tab) {
381 while (!Tab.empty()) {
387 inline void error(const std::string& errmsg);
389 BytecodeReader(const BytecodeReader &); // DO NOT IMPLEMENT
390 void operator=(const BytecodeReader &); // DO NOT IMPLEMENT
392 // This enum provides the values of the well-known type slots that are always
393 // emitted as the first few types in the table by the BytecodeWriter class.
394 enum WellKnownTypeSlots {
395 VoidTypeSlot = 0, ///< TypeID == VoidTyID
396 FloatTySlot = 1, ///< TypeID == FloatTyID
397 DoubleTySlot = 2, ///< TypeID == DoubleTyID
398 LabelTySlot = 3, ///< TypeID == LabelTyID
399 BoolTySlot = 4, ///< TypeID == IntegerTyID, width = 1
400 Int8TySlot = 5, ///< TypeID == IntegerTyID, width = 8
401 Int16TySlot = 6, ///< TypeID == IntegerTyID, width = 16
402 Int32TySlot = 7, ///< TypeID == IntegerTyID, width = 32
403 Int64TySlot = 8 ///< TypeID == IntegerTyID, width = 64
407 /// @name Reader Primitives
411 /// @brief Is there more to parse in the current block?
412 inline bool moreInBlock();
414 /// @brief Have we read past the end of the block
415 inline void checkPastBlockEnd(const char * block_name);
417 /// @brief Align to 32 bits
418 inline void align32();
420 /// @brief Read an unsigned integer as 32-bits
421 inline unsigned read_uint();
423 /// @brief Read an unsigned integer with variable bit rate encoding
424 inline unsigned read_vbr_uint();
426 /// @brief Read an unsigned integer of no more than 24-bits with variable
427 /// bit rate encoding.
428 inline unsigned read_vbr_uint24();
430 /// @brief Read an unsigned 64-bit integer with variable bit rate encoding.
431 inline uint64_t read_vbr_uint64();
433 /// @brief Read a signed 64-bit integer with variable bit rate encoding.
434 inline int64_t read_vbr_int64();
436 /// @brief Read a string
437 inline std::string read_str();
439 /// @brief Read a float value
440 inline void read_float(float& FloatVal);
442 /// @brief Read a double value
443 inline void read_double(double& DoubleVal);
445 /// @brief Read an arbitrary data chunk of fixed length
446 inline void read_data(void *Ptr, void *End);
448 /// @brief Read a bytecode block header
449 inline void read_block(unsigned &Type, unsigned &Size);
453 /// @brief A function for creating a BytecodeAnalzer as a handler
454 /// for the Bytecode reader.
455 BytecodeHandler* createBytecodeAnalyzerHandler(BytecodeAnalysis& bca,
456 std::ostream* output );
459 } // End llvm namespace