1 //===- Reader.cpp - Code to read bytecode files ---------------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file was developed by the LLVM research group and is distributed under
6 // the University of Illinois Open Source License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This library implements the functionality defined in llvm/Bytecode/Reader.h
12 // Note that this library should be as fast as possible, reentrant, and
15 // TODO: Allow passing in an option to ignore the symbol table
17 //===----------------------------------------------------------------------===//
20 #include "llvm/Bytecode/BytecodeHandler.h"
21 #include "llvm/BasicBlock.h"
22 #include "llvm/Constants.h"
23 #include "llvm/Instructions.h"
24 #include "llvm/SymbolTable.h"
25 #include "llvm/Bytecode/Format.h"
26 #include "llvm/Support/GetElementPtrTypeIterator.h"
27 #include "Support/StringExtras.h"
34 /// @brief A class for maintaining the slot number definition
35 /// as a placeholder for the actual definition for forward constants defs.
36 class ConstantPlaceHolder : public ConstantExpr {
38 ConstantPlaceHolder(); // DO NOT IMPLEMENT
39 void operator=(const ConstantPlaceHolder &); // DO NOT IMPLEMENT
41 ConstantPlaceHolder(const Type *Ty, unsigned id)
42 : ConstantExpr(Instruction::UserOp1, Constant::getNullValue(Ty), Ty),
44 unsigned getID() { return ID; }
49 // Provide some details on error
50 inline void BytecodeReader::error(std::string err) {
52 err += itostr(RevisionNum) ;
54 err += itostr(At-MemStart);
59 //===----------------------------------------------------------------------===//
60 // Bytecode Reading Methods
61 //===----------------------------------------------------------------------===//
63 /// Determine if the current block being read contains any more data.
64 inline bool BytecodeReader::moreInBlock() {
68 /// Throw an error if we've read past the end of the current block
69 inline void BytecodeReader::checkPastBlockEnd(const char * block_name) {
71 error(std::string("Attempt to read past the end of ") + block_name + " block.");
74 /// Align the buffer position to a 32 bit boundary
75 inline void BytecodeReader::align32() {
77 At = (const unsigned char *)((unsigned long)(At+3) & (~3UL));
79 if (Handler) Handler->handleAlignment(At - Save);
81 error("Ran out of data while aligning!");
84 /// Read a whole unsigned integer
85 inline unsigned BytecodeReader::read_uint() {
87 error("Ran out of data reading uint!");
89 return At[-4] | (At[-3] << 8) | (At[-2] << 16) | (At[-1] << 24);
92 /// Read a variable-bit-rate encoded unsigned integer
93 inline unsigned BytecodeReader::read_vbr_uint() {
100 error("Ran out of data reading vbr_uint!");
101 Result |= (unsigned)((*At++) & 0x7F) << Shift;
103 } while (At[-1] & 0x80);
104 if (Handler) Handler->handleVBR32(At-Save);
108 /// Read a variable-bit-rate encoded unsigned 64-bit integer.
109 inline uint64_t BytecodeReader::read_vbr_uint64() {
116 error("Ran out of data reading vbr_uint64!");
117 Result |= (uint64_t)((*At++) & 0x7F) << Shift;
119 } while (At[-1] & 0x80);
120 if (Handler) Handler->handleVBR64(At-Save);
124 /// Read a variable-bit-rate encoded signed 64-bit integer.
125 inline int64_t BytecodeReader::read_vbr_int64() {
126 uint64_t R = read_vbr_uint64();
129 return -(int64_t)(R >> 1);
130 else // There is no such thing as -0 with integers. "-0" really means
131 // 0x8000000000000000.
134 return (int64_t)(R >> 1);
137 /// Read a pascal-style string (length followed by text)
138 inline std::string BytecodeReader::read_str() {
139 unsigned Size = read_vbr_uint();
140 const unsigned char *OldAt = At;
142 if (At > BlockEnd) // Size invalid?
143 error("Ran out of data reading a string!");
144 return std::string((char*)OldAt, Size);
147 /// Read an arbitrary block of data
148 inline void BytecodeReader::read_data(void *Ptr, void *End) {
149 unsigned char *Start = (unsigned char *)Ptr;
150 unsigned Amount = (unsigned char *)End - Start;
151 if (At+Amount > BlockEnd)
152 error("Ran out of data!");
153 std::copy(At, At+Amount, Start);
157 /// Read a float value in little-endian order
158 inline void BytecodeReader::read_float(float& FloatVal) {
159 /// FIXME: This isn't optimal, it has size problems on some platforms
160 /// where FP is not IEEE.
165 FloatUnion.i = At[0] | (At[1] << 8) | (At[2] << 16) | (At[3] << 24);
166 At+=sizeof(uint32_t);
167 FloatVal = FloatUnion.f;
170 /// Read a double value in little-endian order
171 inline void BytecodeReader::read_double(double& DoubleVal) {
172 /// FIXME: This isn't optimal, it has size problems on some platforms
173 /// where FP is not IEEE.
178 DoubleUnion.i = At[0] | (At[1] << 8) | (At[2] << 16) | (At[3] << 24) |
179 (uint64_t(At[4]) << 32) | (uint64_t(At[5]) << 40) |
180 (uint64_t(At[6]) << 48) | (uint64_t(At[7]) << 56);
181 At+=sizeof(uint64_t);
182 DoubleVal = DoubleUnion.d;
185 /// Read a block header and obtain its type and size
186 inline void BytecodeReader::read_block(unsigned &Type, unsigned &Size) {
187 if ( hasLongBlockHeaders ) {
191 case BytecodeFormat::Reserved_DoNotUse :
192 error("Reserved_DoNotUse used as Module Type?");
193 Type = BytecodeFormat::Module; break;
194 case BytecodeFormat::Module:
195 Type = BytecodeFormat::ModuleBlockID; break;
196 case BytecodeFormat::Function:
197 Type = BytecodeFormat::FunctionBlockID; break;
198 case BytecodeFormat::ConstantPool:
199 Type = BytecodeFormat::ConstantPoolBlockID; break;
200 case BytecodeFormat::SymbolTable:
201 Type = BytecodeFormat::SymbolTableBlockID; break;
202 case BytecodeFormat::ModuleGlobalInfo:
203 Type = BytecodeFormat::ModuleGlobalInfoBlockID; break;
204 case BytecodeFormat::GlobalTypePlane:
205 Type = BytecodeFormat::GlobalTypePlaneBlockID; break;
206 case BytecodeFormat::InstructionList:
207 Type = BytecodeFormat::InstructionListBlockID; break;
208 case BytecodeFormat::CompactionTable:
209 Type = BytecodeFormat::CompactionTableBlockID; break;
210 case BytecodeFormat::BasicBlock:
211 /// This block type isn't used after version 1.1. However, we have to
212 /// still allow the value in case this is an old bc format file.
213 /// We just let its value creep thru.
216 error("Invalid module type found: " + utostr(Type));
221 Type = Size & 0x1F; // mask low order five bits
222 Size >>= 5; // get rid of five low order bits, leaving high 27
225 if (At + Size > BlockEnd)
226 error("Attempt to size a block past end of memory");
227 BlockEnd = At + Size;
228 if (Handler) Handler->handleBlock(Type, BlockStart, Size);
232 /// In LLVM 1.2 and before, Types were derived from Value and so they were
233 /// written as part of the type planes along with any other Value. In LLVM
234 /// 1.3 this changed so that Type does not derive from Value. Consequently,
235 /// the BytecodeReader's containers for Values can't contain Types because
236 /// there's no inheritance relationship. This means that the "Type Type"
237 /// plane is defunct along with the Type::TypeTyID TypeID. In LLVM 1.3
238 /// whenever a bytecode construct must have both types and values together,
239 /// the types are always read/written first and then the Values. Furthermore
240 /// since Type::TypeTyID no longer exists, its value (12) now corresponds to
241 /// Type::LabelTyID. In order to overcome this we must "sanitize" all the
242 /// type TypeIDs we encounter. For LLVM 1.3 bytecode files, there's no change.
243 /// For LLVM 1.2 and before, this function will decrement the type id by
244 /// one to account for the missing Type::TypeTyID enumerator if the value is
245 /// larger than 12 (Type::LabelTyID). If the value is exactly 12, then this
246 /// function returns true, otherwise false. This helps detect situations
247 /// where the pre 1.3 bytecode is indicating that what follows is a type.
248 /// @returns true iff type id corresponds to pre 1.3 "type type"
249 inline bool BytecodeReader::sanitizeTypeId(unsigned &TypeId) {
250 if (hasTypeDerivedFromValue) { /// do nothing if 1.3 or later
251 if (TypeId == Type::LabelTyID) {
252 TypeId = Type::VoidTyID; // sanitize it
253 return true; // indicate we got TypeTyID in pre 1.3 bytecode
254 } else if (TypeId > Type::LabelTyID)
255 --TypeId; // shift all planes down because type type plane is missing
260 /// Reads a vbr uint to read in a type id and does the necessary
261 /// conversion on it by calling sanitizeTypeId.
262 /// @returns true iff \p TypeId read corresponds to a pre 1.3 "type type"
263 /// @see sanitizeTypeId
264 inline bool BytecodeReader::read_typeid(unsigned &TypeId) {
265 TypeId = read_vbr_uint();
266 if ( !has32BitTypes )
267 if ( TypeId == 0x00FFFFFF )
268 TypeId = read_vbr_uint();
269 return sanitizeTypeId(TypeId);
272 //===----------------------------------------------------------------------===//
274 //===----------------------------------------------------------------------===//
276 /// Determine if a type id has an implicit null value
277 inline bool BytecodeReader::hasImplicitNull(unsigned TyID) {
278 if (!hasExplicitPrimitiveZeros)
279 return TyID != Type::LabelTyID && TyID != Type::VoidTyID;
280 return TyID >= Type::FirstDerivedTyID;
283 /// Obtain a type given a typeid and account for things like compaction tables,
284 /// function level vs module level, and the offsetting for the primitive types.
285 const Type *BytecodeReader::getType(unsigned ID) {
286 if (ID < Type::FirstDerivedTyID)
287 if (const Type *T = Type::getPrimitiveType((Type::TypeID)ID))
288 return T; // Asked for a primitive type...
290 // Otherwise, derived types need offset...
291 ID -= Type::FirstDerivedTyID;
293 if (!CompactionTypes.empty()) {
294 if (ID >= CompactionTypes.size())
295 error("Type ID out of range for compaction table!");
296 return CompactionTypes[ID];
299 // Is it a module-level type?
300 if (ID < ModuleTypes.size())
301 return ModuleTypes[ID].get();
303 // Nope, is it a function-level type?
304 ID -= ModuleTypes.size();
305 if (ID < FunctionTypes.size())
306 return FunctionTypes[ID].get();
308 error("Illegal type reference!");
312 /// Get a sanitized type id. This just makes sure that the \p ID
313 /// is both sanitized and not the "type type" of pre-1.3 bytecode.
314 /// @see sanitizeTypeId
315 inline const Type* BytecodeReader::getSanitizedType(unsigned& ID) {
316 if (sanitizeTypeId(ID))
317 error("Invalid type id encountered");
321 /// This method just saves some coding. It uses read_typeid to read
322 /// in a sanitized type id, errors that its not the type type, and
323 /// then calls getType to return the type value.
324 inline const Type* BytecodeReader::readSanitizedType() {
327 error("Invalid type id encountered");
331 /// Get the slot number associated with a type accounting for primitive
332 /// types, compaction tables, and function level vs module level.
333 unsigned BytecodeReader::getTypeSlot(const Type *Ty) {
334 if (Ty->isPrimitiveType())
335 return Ty->getTypeID();
337 // Scan the compaction table for the type if needed.
338 if (!CompactionTypes.empty()) {
339 std::vector<const Type*>::const_iterator I =
340 find(CompactionTypes.begin(), CompactionTypes.end(), Ty);
342 if (I == CompactionTypes.end())
343 error("Couldn't find type specified in compaction table!");
344 return Type::FirstDerivedTyID + (&*I - &CompactionTypes[0]);
347 // Check the function level types first...
348 TypeListTy::iterator I = find(FunctionTypes.begin(), FunctionTypes.end(), Ty);
350 if (I != FunctionTypes.end())
351 return Type::FirstDerivedTyID + ModuleTypes.size() +
352 (&*I - &FunctionTypes[0]);
354 // Check the module level types now...
355 I = find(ModuleTypes.begin(), ModuleTypes.end(), Ty);
356 if (I == ModuleTypes.end())
357 error("Didn't find type in ModuleTypes.");
358 return Type::FirstDerivedTyID + (&*I - &ModuleTypes[0]);
361 /// This is just like getType, but when a compaction table is in use, it is
362 /// ignored. It also ignores function level types.
364 const Type *BytecodeReader::getGlobalTableType(unsigned Slot) {
365 if (Slot < Type::FirstDerivedTyID) {
366 const Type *Ty = Type::getPrimitiveType((Type::TypeID)Slot);
368 error("Not a primitive type ID?");
371 Slot -= Type::FirstDerivedTyID;
372 if (Slot >= ModuleTypes.size())
373 error("Illegal compaction table type reference!");
374 return ModuleTypes[Slot];
377 /// This is just like getTypeSlot, but when a compaction table is in use, it
378 /// is ignored. It also ignores function level types.
379 unsigned BytecodeReader::getGlobalTableTypeSlot(const Type *Ty) {
380 if (Ty->isPrimitiveType())
381 return Ty->getTypeID();
382 TypeListTy::iterator I = find(ModuleTypes.begin(),
383 ModuleTypes.end(), Ty);
384 if (I == ModuleTypes.end())
385 error("Didn't find type in ModuleTypes.");
386 return Type::FirstDerivedTyID + (&*I - &ModuleTypes[0]);
389 /// Retrieve a value of a given type and slot number, possibly creating
390 /// it if it doesn't already exist.
391 Value * BytecodeReader::getValue(unsigned type, unsigned oNum, bool Create) {
392 assert(type != Type::LabelTyID && "getValue() cannot get blocks!");
395 // If there is a compaction table active, it defines the low-level numbers.
396 // If not, the module values define the low-level numbers.
397 if (CompactionValues.size() > type && !CompactionValues[type].empty()) {
398 if (Num < CompactionValues[type].size())
399 return CompactionValues[type][Num];
400 Num -= CompactionValues[type].size();
402 // By default, the global type id is the type id passed in
403 unsigned GlobalTyID = type;
405 // If the type plane was compactified, figure out the global type ID
406 // by adding the derived type ids and the distance.
407 if (!CompactionTypes.empty() && type >= Type::FirstDerivedTyID) {
408 const Type *Ty = CompactionTypes[type-Type::FirstDerivedTyID];
409 TypeListTy::iterator I =
410 find(ModuleTypes.begin(), ModuleTypes.end(), Ty);
411 assert(I != ModuleTypes.end());
412 GlobalTyID = Type::FirstDerivedTyID + (&*I - &ModuleTypes[0]);
415 if (hasImplicitNull(GlobalTyID)) {
417 return Constant::getNullValue(getType(type));
421 if (GlobalTyID < ModuleValues.size() && ModuleValues[GlobalTyID]) {
422 if (Num < ModuleValues[GlobalTyID]->size())
423 return ModuleValues[GlobalTyID]->getOperand(Num);
424 Num -= ModuleValues[GlobalTyID]->size();
428 if (FunctionValues.size() > type &&
429 FunctionValues[type] &&
430 Num < FunctionValues[type]->size())
431 return FunctionValues[type]->getOperand(Num);
433 if (!Create) return 0; // Do not create a placeholder?
435 std::pair<unsigned,unsigned> KeyValue(type, oNum);
436 ForwardReferenceMap::iterator I = ForwardReferences.lower_bound(KeyValue);
437 if (I != ForwardReferences.end() && I->first == KeyValue)
438 return I->second; // We have already created this placeholder
440 Value *Val = new Argument(getType(type));
441 ForwardReferences.insert(I, std::make_pair(KeyValue, Val));
445 /// This is just like getValue, but when a compaction table is in use, it
446 /// is ignored. Also, no forward references or other fancy features are
448 Value* BytecodeReader::getGlobalTableValue(const Type *Ty, unsigned SlotNo) {
449 // FIXME: getTypeSlot is inefficient!
450 unsigned TyID = getGlobalTableTypeSlot(Ty);
452 if (TyID != Type::LabelTyID) {
454 return Constant::getNullValue(Ty);
458 if (TyID >= ModuleValues.size() || ModuleValues[TyID] == 0 ||
459 SlotNo >= ModuleValues[TyID]->size()) {
460 error("Corrupt compaction table entry!"
461 + utostr(TyID) + ", " + utostr(SlotNo) + ": "
462 + utostr(ModuleValues.size()) + ", "
463 + utohexstr(intptr_t((void*)ModuleValues[TyID])) + ", "
464 + utostr(ModuleValues[TyID]->size()));
466 return ModuleValues[TyID]->getOperand(SlotNo);
469 /// Just like getValue, except that it returns a null pointer
470 /// only on error. It always returns a constant (meaning that if the value is
471 /// defined, but is not a constant, that is an error). If the specified
472 /// constant hasn't been parsed yet, a placeholder is defined and used.
473 /// Later, after the real value is parsed, the placeholder is eliminated.
474 Constant* BytecodeReader::getConstantValue(unsigned TypeSlot, unsigned Slot) {
475 if (Value *V = getValue(TypeSlot, Slot, false))
476 if (Constant *C = dyn_cast<Constant>(V))
477 return C; // If we already have the value parsed, just return it
479 error("Value for slot " + utostr(Slot) +
480 " is expected to be a constant!");
482 const Type *Ty = getType(TypeSlot);
483 std::pair<const Type*, unsigned> Key(Ty, Slot);
484 ConstantRefsType::iterator I = ConstantFwdRefs.lower_bound(Key);
486 if (I != ConstantFwdRefs.end() && I->first == Key) {
489 // Create a placeholder for the constant reference and
490 // keep track of the fact that we have a forward ref to recycle it
491 Constant *C = new ConstantPlaceHolder(Ty, Slot);
493 // Keep track of the fact that we have a forward ref to recycle it
494 ConstantFwdRefs.insert(I, std::make_pair(Key, C));
499 //===----------------------------------------------------------------------===//
500 // IR Construction Methods
501 //===----------------------------------------------------------------------===//
503 /// As values are created, they are inserted into the appropriate place
504 /// with this method. The ValueTable argument must be one of ModuleValues
505 /// or FunctionValues data members of this class.
506 unsigned BytecodeReader::insertValue(Value *Val, unsigned type,
507 ValueTable &ValueTab) {
508 assert((!isa<Constant>(Val) || !cast<Constant>(Val)->isNullValue()) ||
509 !hasImplicitNull(type) &&
510 "Cannot read null values from bytecode!");
512 if (ValueTab.size() <= type)
513 ValueTab.resize(type+1);
515 if (!ValueTab[type]) ValueTab[type] = new ValueList();
517 ValueTab[type]->push_back(Val);
519 bool HasOffset = hasImplicitNull(type);
520 return ValueTab[type]->size()-1 + HasOffset;
523 /// Insert the arguments of a function as new values in the reader.
524 void BytecodeReader::insertArguments(Function* F) {
525 const FunctionType *FT = F->getFunctionType();
526 Function::aiterator AI = F->abegin();
527 for (FunctionType::param_iterator It = FT->param_begin();
528 It != FT->param_end(); ++It, ++AI)
529 insertValue(AI, getTypeSlot(AI->getType()), FunctionValues);
532 //===----------------------------------------------------------------------===//
533 // Bytecode Parsing Methods
534 //===----------------------------------------------------------------------===//
536 /// This method parses a single instruction. The instruction is
537 /// inserted at the end of the \p BB provided. The arguments of
538 /// the instruction are provided in the \p Args vector.
539 void BytecodeReader::ParseInstruction(std::vector<unsigned> &Oprnds,
543 // Clear instruction data
547 unsigned Op = read_uint();
549 // bits Instruction format: Common to all formats
550 // --------------------------
551 // 01-00: Opcode type, fixed to 1.
553 Opcode = (Op >> 2) & 63;
554 Oprnds.resize((Op >> 0) & 03);
556 // Extract the operands
557 switch (Oprnds.size()) {
559 // bits Instruction format:
560 // --------------------------
561 // 19-08: Resulting type plane
562 // 31-20: Operand #1 (if set to (2^12-1), then zero operands)
564 iType = (Op >> 8) & 4095;
565 Oprnds[0] = (Op >> 20) & 4095;
566 if (Oprnds[0] == 4095) // Handle special encoding for 0 operands...
570 // bits Instruction format:
571 // --------------------------
572 // 15-08: Resulting type plane
576 iType = (Op >> 8) & 255;
577 Oprnds[0] = (Op >> 16) & 255;
578 Oprnds[1] = (Op >> 24) & 255;
581 // bits Instruction format:
582 // --------------------------
583 // 13-08: Resulting type plane
588 iType = (Op >> 8) & 63;
589 Oprnds[0] = (Op >> 14) & 63;
590 Oprnds[1] = (Op >> 20) & 63;
591 Oprnds[2] = (Op >> 26) & 63;
594 At -= 4; // Hrm, try this again...
595 Opcode = read_vbr_uint();
597 iType = read_vbr_uint();
599 unsigned NumOprnds = read_vbr_uint();
600 Oprnds.resize(NumOprnds);
603 error("Zero-argument instruction found; this is invalid.");
605 for (unsigned i = 0; i != NumOprnds; ++i)
606 Oprnds[i] = read_vbr_uint();
611 const Type *InstTy = getSanitizedType(iType);
613 // We have enough info to inform the handler now.
614 if (Handler) Handler->handleInstruction(Opcode, InstTy, Oprnds, At-SaveAt);
616 // Declare the resulting instruction we'll build.
617 Instruction *Result = 0;
619 // Handle binary operators
620 if (Opcode >= Instruction::BinaryOpsBegin &&
621 Opcode < Instruction::BinaryOpsEnd && Oprnds.size() == 2)
622 Result = BinaryOperator::create((Instruction::BinaryOps)Opcode,
623 getValue(iType, Oprnds[0]),
624 getValue(iType, Oprnds[1]));
629 error("Illegal instruction read!");
631 case Instruction::VAArg:
632 Result = new VAArgInst(getValue(iType, Oprnds[0]),
633 getSanitizedType(Oprnds[1]));
635 case Instruction::VANext:
636 Result = new VANextInst(getValue(iType, Oprnds[0]),
637 getSanitizedType(Oprnds[1]));
639 case Instruction::Cast:
640 Result = new CastInst(getValue(iType, Oprnds[0]),
641 getSanitizedType(Oprnds[1]));
643 case Instruction::Select:
644 Result = new SelectInst(getValue(Type::BoolTyID, Oprnds[0]),
645 getValue(iType, Oprnds[1]),
646 getValue(iType, Oprnds[2]));
648 case Instruction::PHI: {
649 if (Oprnds.size() == 0 || (Oprnds.size() & 1))
650 error("Invalid phi node encountered!");
652 PHINode *PN = new PHINode(InstTy);
653 PN->op_reserve(Oprnds.size());
654 for (unsigned i = 0, e = Oprnds.size(); i != e; i += 2)
655 PN->addIncoming(getValue(iType, Oprnds[i]), getBasicBlock(Oprnds[i+1]));
660 case Instruction::Shl:
661 case Instruction::Shr:
662 Result = new ShiftInst((Instruction::OtherOps)Opcode,
663 getValue(iType, Oprnds[0]),
664 getValue(Type::UByteTyID, Oprnds[1]));
666 case Instruction::Ret:
667 if (Oprnds.size() == 0)
668 Result = new ReturnInst();
669 else if (Oprnds.size() == 1)
670 Result = new ReturnInst(getValue(iType, Oprnds[0]));
672 error("Unrecognized instruction!");
675 case Instruction::Br:
676 if (Oprnds.size() == 1)
677 Result = new BranchInst(getBasicBlock(Oprnds[0]));
678 else if (Oprnds.size() == 3)
679 Result = new BranchInst(getBasicBlock(Oprnds[0]),
680 getBasicBlock(Oprnds[1]), getValue(Type::BoolTyID , Oprnds[2]));
682 error("Invalid number of operands for a 'br' instruction!");
684 case Instruction::Switch: {
685 if (Oprnds.size() & 1)
686 error("Switch statement with odd number of arguments!");
688 SwitchInst *I = new SwitchInst(getValue(iType, Oprnds[0]),
689 getBasicBlock(Oprnds[1]));
690 for (unsigned i = 2, e = Oprnds.size(); i != e; i += 2)
691 I->addCase(cast<Constant>(getValue(iType, Oprnds[i])),
692 getBasicBlock(Oprnds[i+1]));
697 case Instruction::Call: {
698 if (Oprnds.size() == 0)
699 error("Invalid call instruction encountered!");
701 Value *F = getValue(iType, Oprnds[0]);
703 // Check to make sure we have a pointer to function type
704 const PointerType *PTy = dyn_cast<PointerType>(F->getType());
705 if (PTy == 0) error("Call to non function pointer value!");
706 const FunctionType *FTy = dyn_cast<FunctionType>(PTy->getElementType());
707 if (FTy == 0) error("Call to non function pointer value!");
709 std::vector<Value *> Params;
710 if (!FTy->isVarArg()) {
711 FunctionType::param_iterator It = FTy->param_begin();
713 for (unsigned i = 1, e = Oprnds.size(); i != e; ++i) {
714 if (It == FTy->param_end())
715 error("Invalid call instruction!");
716 Params.push_back(getValue(getTypeSlot(*It++), Oprnds[i]));
718 if (It != FTy->param_end())
719 error("Invalid call instruction!");
721 Oprnds.erase(Oprnds.begin(), Oprnds.begin()+1);
723 unsigned FirstVariableOperand;
724 if (Oprnds.size() < FTy->getNumParams())
725 error("Call instruction missing operands!");
727 // Read all of the fixed arguments
728 for (unsigned i = 0, e = FTy->getNumParams(); i != e; ++i)
729 Params.push_back(getValue(getTypeSlot(FTy->getParamType(i)),Oprnds[i]));
731 FirstVariableOperand = FTy->getNumParams();
733 if ((Oprnds.size()-FirstVariableOperand) & 1) // Must be pairs of type/value
734 error("Invalid call instruction!");
736 for (unsigned i = FirstVariableOperand, e = Oprnds.size();
738 Params.push_back(getValue(Oprnds[i], Oprnds[i+1]));
741 Result = new CallInst(F, Params);
744 case Instruction::Invoke: {
745 if (Oprnds.size() < 3)
746 error("Invalid invoke instruction!");
747 Value *F = getValue(iType, Oprnds[0]);
749 // Check to make sure we have a pointer to function type
750 const PointerType *PTy = dyn_cast<PointerType>(F->getType());
752 error("Invoke to non function pointer value!");
753 const FunctionType *FTy = dyn_cast<FunctionType>(PTy->getElementType());
755 error("Invoke to non function pointer value!");
757 std::vector<Value *> Params;
758 BasicBlock *Normal, *Except;
760 if (!FTy->isVarArg()) {
761 Normal = getBasicBlock(Oprnds[1]);
762 Except = getBasicBlock(Oprnds[2]);
764 FunctionType::param_iterator It = FTy->param_begin();
765 for (unsigned i = 3, e = Oprnds.size(); i != e; ++i) {
766 if (It == FTy->param_end())
767 error("Invalid invoke instruction!");
768 Params.push_back(getValue(getTypeSlot(*It++), Oprnds[i]));
770 if (It != FTy->param_end())
771 error("Invalid invoke instruction!");
773 Oprnds.erase(Oprnds.begin(), Oprnds.begin()+1);
775 Normal = getBasicBlock(Oprnds[0]);
776 Except = getBasicBlock(Oprnds[1]);
778 unsigned FirstVariableArgument = FTy->getNumParams()+2;
779 for (unsigned i = 2; i != FirstVariableArgument; ++i)
780 Params.push_back(getValue(getTypeSlot(FTy->getParamType(i-2)),
783 if (Oprnds.size()-FirstVariableArgument & 1) // Must be type/value pairs
784 error("Invalid invoke instruction!");
786 for (unsigned i = FirstVariableArgument; i < Oprnds.size(); i += 2)
787 Params.push_back(getValue(Oprnds[i], Oprnds[i+1]));
790 Result = new InvokeInst(F, Normal, Except, Params);
793 case Instruction::Malloc:
794 if (Oprnds.size() > 2)
795 error("Invalid malloc instruction!");
796 if (!isa<PointerType>(InstTy))
797 error("Invalid malloc instruction!");
799 Result = new MallocInst(cast<PointerType>(InstTy)->getElementType(),
800 Oprnds.size() ? getValue(Type::UIntTyID,
804 case Instruction::Alloca:
805 if (Oprnds.size() > 2)
806 error("Invalid alloca instruction!");
807 if (!isa<PointerType>(InstTy))
808 error("Invalid alloca instruction!");
810 Result = new AllocaInst(cast<PointerType>(InstTy)->getElementType(),
811 Oprnds.size() ? getValue(Type::UIntTyID,
814 case Instruction::Free:
815 if (!isa<PointerType>(InstTy))
816 error("Invalid free instruction!");
817 Result = new FreeInst(getValue(iType, Oprnds[0]));
819 case Instruction::GetElementPtr: {
820 if (Oprnds.size() == 0 || !isa<PointerType>(InstTy))
821 error("Invalid getelementptr instruction!");
823 std::vector<Value*> Idx;
825 const Type *NextTy = InstTy;
826 for (unsigned i = 1, e = Oprnds.size(); i != e; ++i) {
827 const CompositeType *TopTy = dyn_cast_or_null<CompositeType>(NextTy);
829 error("Invalid getelementptr instruction!");
831 unsigned ValIdx = Oprnds[i];
833 if (!hasRestrictedGEPTypes) {
834 // Struct indices are always uints, sequential type indices can be any
835 // of the 32 or 64-bit integer types. The actual choice of type is
836 // encoded in the low two bits of the slot number.
837 if (isa<StructType>(TopTy))
838 IdxTy = Type::UIntTyID;
840 switch (ValIdx & 3) {
842 case 0: IdxTy = Type::UIntTyID; break;
843 case 1: IdxTy = Type::IntTyID; break;
844 case 2: IdxTy = Type::ULongTyID; break;
845 case 3: IdxTy = Type::LongTyID; break;
850 IdxTy = isa<StructType>(TopTy) ? Type::UByteTyID : Type::LongTyID;
853 Idx.push_back(getValue(IdxTy, ValIdx));
855 // Convert ubyte struct indices into uint struct indices.
856 if (isa<StructType>(TopTy) && hasRestrictedGEPTypes)
857 if (ConstantUInt *C = dyn_cast<ConstantUInt>(Idx.back()))
858 Idx[Idx.size()-1] = ConstantExpr::getCast(C, Type::UIntTy);
860 NextTy = GetElementPtrInst::getIndexedType(InstTy, Idx, true);
863 Result = new GetElementPtrInst(getValue(iType, Oprnds[0]), Idx);
867 case 62: // volatile load
868 case Instruction::Load:
869 if (Oprnds.size() != 1 || !isa<PointerType>(InstTy))
870 error("Invalid load instruction!");
871 Result = new LoadInst(getValue(iType, Oprnds[0]), "", Opcode == 62);
874 case 63: // volatile store
875 case Instruction::Store: {
876 if (!isa<PointerType>(InstTy) || Oprnds.size() != 2)
877 error("Invalid store instruction!");
879 Value *Ptr = getValue(iType, Oprnds[1]);
880 const Type *ValTy = cast<PointerType>(Ptr->getType())->getElementType();
881 Result = new StoreInst(getValue(getTypeSlot(ValTy), Oprnds[0]), Ptr,
885 case Instruction::Unwind:
886 if (Oprnds.size() != 0)
887 error("Invalid unwind instruction!");
888 Result = new UnwindInst();
890 } // end switch(Opcode)
893 if (Result->getType() == InstTy)
896 TypeSlot = getTypeSlot(Result->getType());
898 insertValue(Result, TypeSlot, FunctionValues);
899 BB->getInstList().push_back(Result);
902 /// Get a particular numbered basic block, which might be a forward reference.
903 /// This works together with ParseBasicBlock to handle these forward references
904 /// in a clean manner. This function is used when constructing phi, br, switch,
905 /// and other instructions that reference basic blocks. Blocks are numbered
906 /// sequentially as they appear in the function.
907 BasicBlock *BytecodeReader::getBasicBlock(unsigned ID) {
908 // Make sure there is room in the table...
909 if (ParsedBasicBlocks.size() <= ID) ParsedBasicBlocks.resize(ID+1);
911 // First check to see if this is a backwards reference, i.e., ParseBasicBlock
912 // has already created this block, or if the forward reference has already
914 if (ParsedBasicBlocks[ID])
915 return ParsedBasicBlocks[ID];
917 // Otherwise, the basic block has not yet been created. Do so and add it to
918 // the ParsedBasicBlocks list.
919 return ParsedBasicBlocks[ID] = new BasicBlock();
922 /// In LLVM 1.0 bytecode files, we used to output one basicblock at a time.
923 /// This method reads in one of the basicblock packets. This method is not used
924 /// for bytecode files after LLVM 1.0
925 /// @returns The basic block constructed.
926 BasicBlock *BytecodeReader::ParseBasicBlock(unsigned BlockNo) {
927 if (Handler) Handler->handleBasicBlockBegin(BlockNo);
931 if (ParsedBasicBlocks.size() == BlockNo)
932 ParsedBasicBlocks.push_back(BB = new BasicBlock());
933 else if (ParsedBasicBlocks[BlockNo] == 0)
934 BB = ParsedBasicBlocks[BlockNo] = new BasicBlock();
936 BB = ParsedBasicBlocks[BlockNo];
938 std::vector<unsigned> Operands;
939 while (moreInBlock())
940 ParseInstruction(Operands, BB);
942 if (Handler) Handler->handleBasicBlockEnd(BlockNo);
946 /// Parse all of the BasicBlock's & Instruction's in the body of a function.
947 /// In post 1.0 bytecode files, we no longer emit basic block individually,
948 /// in order to avoid per-basic-block overhead.
949 /// @returns Rhe number of basic blocks encountered.
950 unsigned BytecodeReader::ParseInstructionList(Function* F) {
951 unsigned BlockNo = 0;
952 std::vector<unsigned> Args;
954 while (moreInBlock()) {
955 if (Handler) Handler->handleBasicBlockBegin(BlockNo);
957 if (ParsedBasicBlocks.size() == BlockNo)
958 ParsedBasicBlocks.push_back(BB = new BasicBlock());
959 else if (ParsedBasicBlocks[BlockNo] == 0)
960 BB = ParsedBasicBlocks[BlockNo] = new BasicBlock();
962 BB = ParsedBasicBlocks[BlockNo];
964 F->getBasicBlockList().push_back(BB);
966 // Read instructions into this basic block until we get to a terminator
967 while (moreInBlock() && !BB->getTerminator())
968 ParseInstruction(Args, BB);
970 if (!BB->getTerminator())
971 error("Non-terminated basic block found!");
973 if (Handler) Handler->handleBasicBlockEnd(BlockNo-1);
979 /// Parse a symbol table. This works for both module level and function
980 /// level symbol tables. For function level symbol tables, the CurrentFunction
981 /// parameter must be non-zero and the ST parameter must correspond to
982 /// CurrentFunction's symbol table. For Module level symbol tables, the
983 /// CurrentFunction argument must be zero.
984 void BytecodeReader::ParseSymbolTable(Function *CurrentFunction,
986 if (Handler) Handler->handleSymbolTableBegin(CurrentFunction,ST);
988 // Allow efficient basic block lookup by number.
989 std::vector<BasicBlock*> BBMap;
991 for (Function::iterator I = CurrentFunction->begin(),
992 E = CurrentFunction->end(); I != E; ++I)
995 /// In LLVM 1.3 we write types separately from values so
996 /// The types are always first in the symbol table. This is
997 /// because Type no longer derives from Value.
998 if (!hasTypeDerivedFromValue) {
999 // Symtab block header: [num entries]
1000 unsigned NumEntries = read_vbr_uint();
1001 for (unsigned i = 0; i < NumEntries; ++i) {
1002 // Symtab entry: [def slot #][name]
1003 unsigned slot = read_vbr_uint();
1004 std::string Name = read_str();
1005 const Type* T = getType(slot);
1006 ST->insert(Name, T);
1010 while (moreInBlock()) {
1011 // Symtab block header: [num entries][type id number]
1012 unsigned NumEntries = read_vbr_uint();
1014 bool isTypeType = read_typeid(Typ);
1015 const Type *Ty = getType(Typ);
1017 for (unsigned i = 0; i != NumEntries; ++i) {
1018 // Symtab entry: [def slot #][name]
1019 unsigned slot = read_vbr_uint();
1020 std::string Name = read_str();
1022 // if we're reading a pre 1.3 bytecode file and the type plane
1023 // is the "type type", handle it here
1025 const Type* T = getType(slot);
1027 error("Failed type look-up for name '" + Name + "'");
1028 ST->insert(Name, T);
1029 continue; // code below must be short circuited
1032 if (Typ == Type::LabelTyID) {
1033 if (slot < BBMap.size())
1036 V = getValue(Typ, slot, false); // Find mapping...
1039 error("Failed value look-up for name '" + Name + "'");
1040 V->setName(Name, ST);
1044 checkPastBlockEnd("Symbol Table");
1045 if (Handler) Handler->handleSymbolTableEnd();
1048 /// Read in the types portion of a compaction table.
1049 void BytecodeReader::ParseCompactionTypes(unsigned NumEntries) {
1050 for (unsigned i = 0; i != NumEntries; ++i) {
1051 unsigned TypeSlot = 0;
1052 if (read_typeid(TypeSlot))
1053 error("Invalid type in compaction table: type type");
1054 const Type *Typ = getGlobalTableType(TypeSlot);
1055 CompactionTypes.push_back(Typ);
1056 if (Handler) Handler->handleCompactionTableType(i, TypeSlot, Typ);
1060 /// Parse a compaction table.
1061 void BytecodeReader::ParseCompactionTable() {
1063 // Notify handler that we're beginning a compaction table.
1064 if (Handler) Handler->handleCompactionTableBegin();
1066 // In LLVM 1.3 Type no longer derives from Value. So,
1067 // we always write them first in the compaction table
1068 // because they can't occupy a "type plane" where the
1070 if (! hasTypeDerivedFromValue) {
1071 unsigned NumEntries = read_vbr_uint();
1072 ParseCompactionTypes(NumEntries);
1075 // Compaction tables live in separate blocks so we have to loop
1076 // until we've read the whole thing.
1077 while (moreInBlock()) {
1078 // Read the number of Value* entries in the compaction table
1079 unsigned NumEntries = read_vbr_uint();
1081 unsigned isTypeType = false;
1083 // Decode the type from value read in. Most compaction table
1084 // planes will have one or two entries in them. If that's the
1085 // case then the length is encoded in the bottom two bits and
1086 // the higher bits encode the type. This saves another VBR value.
1087 if ((NumEntries & 3) == 3) {
1088 // In this case, both low-order bits are set (value 3). This
1089 // is a signal that the typeid follows.
1091 isTypeType = read_typeid(Ty);
1093 // In this case, the low-order bits specify the number of entries
1094 // and the high order bits specify the type.
1095 Ty = NumEntries >> 2;
1096 isTypeType = sanitizeTypeId(Ty);
1100 // if we're reading a pre 1.3 bytecode file and the type plane
1101 // is the "type type", handle it here
1103 ParseCompactionTypes(NumEntries);
1105 // Make sure we have enough room for the plane
1106 if (Ty >= CompactionValues.size())
1107 CompactionValues.resize(Ty+1);
1109 // Make sure the plane is empty or we have some kind of error
1110 if (!CompactionValues[Ty].empty())
1111 error("Compaction table plane contains multiple entries!");
1113 // Notify handler about the plane
1114 if (Handler) Handler->handleCompactionTablePlane(Ty, NumEntries);
1116 // Convert the type slot to a type
1117 const Type *Typ = getType(Ty);
1119 // Push the implicit zero
1120 CompactionValues[Ty].push_back(Constant::getNullValue(Typ));
1122 // Read in each of the entries, put them in the compaction table
1123 // and notify the handler that we have a new compaction table value.
1124 for (unsigned i = 0; i != NumEntries; ++i) {
1125 unsigned ValSlot = read_vbr_uint();
1126 Value *V = getGlobalTableValue(Typ, ValSlot);
1127 CompactionValues[Ty].push_back(V);
1128 if (Handler) Handler->handleCompactionTableValue(i, Ty, ValSlot, Typ);
1132 // Notify handler that the compaction table is done.
1133 if (Handler) Handler->handleCompactionTableEnd();
1136 // Parse a single type. The typeid is read in first. If its a primitive type
1137 // then nothing else needs to be read, we know how to instantiate it. If its
1138 // a derived type, then additional data is read to fill out the type
1140 const Type *BytecodeReader::ParseType() {
1141 unsigned PrimType = 0;
1142 if (read_typeid(PrimType))
1143 error("Invalid type (type type) in type constants!");
1145 const Type *Result = 0;
1146 if ((Result = Type::getPrimitiveType((Type::TypeID)PrimType)))
1150 case Type::FunctionTyID: {
1151 const Type *RetType = readSanitizedType();
1153 unsigned NumParams = read_vbr_uint();
1155 std::vector<const Type*> Params;
1157 Params.push_back(readSanitizedType());
1159 bool isVarArg = Params.size() && Params.back() == Type::VoidTy;
1160 if (isVarArg) Params.pop_back();
1162 Result = FunctionType::get(RetType, Params, isVarArg);
1165 case Type::ArrayTyID: {
1166 const Type *ElementType = readSanitizedType();
1167 unsigned NumElements = read_vbr_uint();
1168 Result = ArrayType::get(ElementType, NumElements);
1171 case Type::StructTyID: {
1172 std::vector<const Type*> Elements;
1174 if (read_typeid(Typ))
1175 error("Invalid element type (type type) for structure!");
1177 while (Typ) { // List is terminated by void/0 typeid
1178 Elements.push_back(getType(Typ));
1179 if (read_typeid(Typ))
1180 error("Invalid element type (type type) for structure!");
1183 Result = StructType::get(Elements);
1186 case Type::PointerTyID: {
1187 Result = PointerType::get(readSanitizedType());
1191 case Type::OpaqueTyID: {
1192 Result = OpaqueType::get();
1197 error("Don't know how to deserialize primitive type " + utostr(PrimType));
1200 if (Handler) Handler->handleType(Result);
1204 // ParseType - We have to use this weird code to handle recursive
1205 // types. We know that recursive types will only reference the current slab of
1206 // values in the type plane, but they can forward reference types before they
1207 // have been read. For example, Type #0 might be '{ Ty#1 }' and Type #1 might
1208 // be 'Ty#0*'. When reading Type #0, type number one doesn't exist. To fix
1209 // this ugly problem, we pessimistically insert an opaque type for each type we
1210 // are about to read. This means that forward references will resolve to
1211 // something and when we reread the type later, we can replace the opaque type
1212 // with a new resolved concrete type.
1214 void BytecodeReader::ParseTypes(TypeListTy &Tab, unsigned NumEntries){
1215 assert(Tab.size() == 0 && "should not have read type constants in before!");
1217 // Insert a bunch of opaque types to be resolved later...
1218 Tab.reserve(NumEntries);
1219 for (unsigned i = 0; i != NumEntries; ++i)
1220 Tab.push_back(OpaqueType::get());
1222 // Loop through reading all of the types. Forward types will make use of the
1223 // opaque types just inserted.
1225 for (unsigned i = 0; i != NumEntries; ++i) {
1226 const Type* NewTy = ParseType();
1227 const Type* OldTy = Tab[i].get();
1229 error("Couldn't parse type!");
1231 // Don't directly push the new type on the Tab. Instead we want to replace
1232 // the opaque type we previously inserted with the new concrete value. This
1233 // approach helps with forward references to types. The refinement from the
1234 // abstract (opaque) type to the new type causes all uses of the abstract
1235 // type to use the concrete type (NewTy). This will also cause the opaque
1236 // type to be deleted.
1237 cast<DerivedType>(const_cast<Type*>(OldTy))->refineAbstractTypeTo(NewTy);
1239 // This should have replaced the old opaque type with the new type in the
1240 // value table... or with a preexisting type that was already in the system.
1241 // Let's just make sure it did.
1242 assert(Tab[i] != OldTy && "refineAbstractType didn't work!");
1246 /// Parse a single constant value
1247 Constant *BytecodeReader::ParseConstantValue(unsigned TypeID) {
1248 // We must check for a ConstantExpr before switching by type because
1249 // a ConstantExpr can be of any type, and has no explicit value.
1251 // 0 if not expr; numArgs if is expr
1252 unsigned isExprNumArgs = read_vbr_uint();
1254 if (isExprNumArgs) {
1255 // FIXME: Encoding of constant exprs could be much more compact!
1256 std::vector<Constant*> ArgVec;
1257 ArgVec.reserve(isExprNumArgs);
1258 unsigned Opcode = read_vbr_uint();
1260 // Read the slot number and types of each of the arguments
1261 for (unsigned i = 0; i != isExprNumArgs; ++i) {
1262 unsigned ArgValSlot = read_vbr_uint();
1263 unsigned ArgTypeSlot = 0;
1264 if (read_typeid(ArgTypeSlot))
1265 error("Invalid argument type (type type) for constant value");
1267 // Get the arg value from its slot if it exists, otherwise a placeholder
1268 ArgVec.push_back(getConstantValue(ArgTypeSlot, ArgValSlot));
1271 // Construct a ConstantExpr of the appropriate kind
1272 if (isExprNumArgs == 1) { // All one-operand expressions
1273 if (Opcode != Instruction::Cast)
1274 error("Only Cast instruction has one argument for ConstantExpr");
1276 Constant* Result = ConstantExpr::getCast(ArgVec[0], getType(TypeID));
1277 if (Handler) Handler->handleConstantExpression(Opcode, ArgVec, Result);
1279 } else if (Opcode == Instruction::GetElementPtr) { // GetElementPtr
1280 std::vector<Constant*> IdxList(ArgVec.begin()+1, ArgVec.end());
1282 if (hasRestrictedGEPTypes) {
1283 const Type *BaseTy = ArgVec[0]->getType();
1284 generic_gep_type_iterator<std::vector<Constant*>::iterator>
1285 GTI = gep_type_begin(BaseTy, IdxList.begin(), IdxList.end()),
1286 E = gep_type_end(BaseTy, IdxList.begin(), IdxList.end());
1287 for (unsigned i = 0; GTI != E; ++GTI, ++i)
1288 if (isa<StructType>(*GTI)) {
1289 if (IdxList[i]->getType() != Type::UByteTy)
1290 error("Invalid index for getelementptr!");
1291 IdxList[i] = ConstantExpr::getCast(IdxList[i], Type::UIntTy);
1295 Constant* Result = ConstantExpr::getGetElementPtr(ArgVec[0], IdxList);
1296 if (Handler) Handler->handleConstantExpression(Opcode, ArgVec, Result);
1298 } else if (Opcode == Instruction::Select) {
1299 if (ArgVec.size() != 3)
1300 error("Select instruction must have three arguments.");
1301 Constant* Result = ConstantExpr::getSelect(ArgVec[0], ArgVec[1],
1303 if (Handler) Handler->handleConstantExpression(Opcode, ArgVec, Result);
1305 } else { // All other 2-operand expressions
1306 Constant* Result = ConstantExpr::get(Opcode, ArgVec[0], ArgVec[1]);
1307 if (Handler) Handler->handleConstantExpression(Opcode, ArgVec, Result);
1312 // Ok, not an ConstantExpr. We now know how to read the given type...
1313 const Type *Ty = getType(TypeID);
1314 switch (Ty->getTypeID()) {
1315 case Type::BoolTyID: {
1316 unsigned Val = read_vbr_uint();
1317 if (Val != 0 && Val != 1)
1318 error("Invalid boolean value read.");
1319 Constant* Result = ConstantBool::get(Val == 1);
1320 if (Handler) Handler->handleConstantValue(Result);
1324 case Type::UByteTyID: // Unsigned integer types...
1325 case Type::UShortTyID:
1326 case Type::UIntTyID: {
1327 unsigned Val = read_vbr_uint();
1328 if (!ConstantUInt::isValueValidForType(Ty, Val))
1329 error("Invalid unsigned byte/short/int read.");
1330 Constant* Result = ConstantUInt::get(Ty, Val);
1331 if (Handler) Handler->handleConstantValue(Result);
1335 case Type::ULongTyID: {
1336 Constant* Result = ConstantUInt::get(Ty, read_vbr_uint64());
1337 if (Handler) Handler->handleConstantValue(Result);
1341 case Type::SByteTyID: // Signed integer types...
1342 case Type::ShortTyID:
1343 case Type::IntTyID: {
1344 case Type::LongTyID:
1345 int64_t Val = read_vbr_int64();
1346 if (!ConstantSInt::isValueValidForType(Ty, Val))
1347 error("Invalid signed byte/short/int/long read.");
1348 Constant* Result = ConstantSInt::get(Ty, Val);
1349 if (Handler) Handler->handleConstantValue(Result);
1353 case Type::FloatTyID: {
1356 Constant* Result = ConstantFP::get(Ty, Val);
1357 if (Handler) Handler->handleConstantValue(Result);
1361 case Type::DoubleTyID: {
1364 Constant* Result = ConstantFP::get(Ty, Val);
1365 if (Handler) Handler->handleConstantValue(Result);
1369 case Type::ArrayTyID: {
1370 const ArrayType *AT = cast<ArrayType>(Ty);
1371 unsigned NumElements = AT->getNumElements();
1372 unsigned TypeSlot = getTypeSlot(AT->getElementType());
1373 std::vector<Constant*> Elements;
1374 Elements.reserve(NumElements);
1375 while (NumElements--) // Read all of the elements of the constant.
1376 Elements.push_back(getConstantValue(TypeSlot,
1378 Constant* Result = ConstantArray::get(AT, Elements);
1379 if (Handler) Handler->handleConstantArray(AT, Elements, TypeSlot, Result);
1383 case Type::StructTyID: {
1384 const StructType *ST = cast<StructType>(Ty);
1386 std::vector<Constant *> Elements;
1387 Elements.reserve(ST->getNumElements());
1388 for (unsigned i = 0; i != ST->getNumElements(); ++i)
1389 Elements.push_back(getConstantValue(ST->getElementType(i),
1392 Constant* Result = ConstantStruct::get(ST, Elements);
1393 if (Handler) Handler->handleConstantStruct(ST, Elements, Result);
1397 case Type::PointerTyID: { // ConstantPointerRef value...
1398 const PointerType *PT = cast<PointerType>(Ty);
1399 unsigned Slot = read_vbr_uint();
1401 // Check to see if we have already read this global variable...
1402 Value *Val = getValue(TypeID, Slot, false);
1405 if (!(GV = dyn_cast<GlobalValue>(Val)))
1406 error("GlobalValue not in ValueTable!");
1408 error("Forward references are not allowed here.");
1411 if (Handler) Handler->handleConstantPointer(PT, Slot, GV );
1416 error("Don't know how to deserialize constant value of type '" +
1417 Ty->getDescription());
1423 /// Resolve references for constants. This function resolves the forward
1424 /// referenced constants in the ConstantFwdRefs map. It uses the
1425 /// replaceAllUsesWith method of Value class to substitute the placeholder
1426 /// instance with the actual instance.
1427 void BytecodeReader::ResolveReferencesToConstant(Constant *NewV, unsigned Slot){
1428 ConstantRefsType::iterator I =
1429 ConstantFwdRefs.find(std::make_pair(NewV->getType(), Slot));
1430 if (I == ConstantFwdRefs.end()) return; // Never forward referenced?
1432 Value *PH = I->second; // Get the placeholder...
1433 PH->replaceAllUsesWith(NewV);
1434 delete PH; // Delete the old placeholder
1435 ConstantFwdRefs.erase(I); // Remove the map entry for it
1438 /// Parse the constant strings section.
1439 void BytecodeReader::ParseStringConstants(unsigned NumEntries, ValueTable &Tab){
1440 for (; NumEntries; --NumEntries) {
1442 if (read_typeid(Typ))
1443 error("Invalid type (type type) for string constant");
1444 const Type *Ty = getType(Typ);
1445 if (!isa<ArrayType>(Ty))
1446 error("String constant data invalid!");
1448 const ArrayType *ATy = cast<ArrayType>(Ty);
1449 if (ATy->getElementType() != Type::SByteTy &&
1450 ATy->getElementType() != Type::UByteTy)
1451 error("String constant data invalid!");
1453 // Read character data. The type tells us how long the string is.
1454 char Data[ATy->getNumElements()];
1455 read_data(Data, Data+ATy->getNumElements());
1457 std::vector<Constant*> Elements(ATy->getNumElements());
1458 if (ATy->getElementType() == Type::SByteTy)
1459 for (unsigned i = 0, e = ATy->getNumElements(); i != e; ++i)
1460 Elements[i] = ConstantSInt::get(Type::SByteTy, (signed char)Data[i]);
1462 for (unsigned i = 0, e = ATy->getNumElements(); i != e; ++i)
1463 Elements[i] = ConstantUInt::get(Type::UByteTy, (unsigned char)Data[i]);
1465 // Create the constant, inserting it as needed.
1466 Constant *C = ConstantArray::get(ATy, Elements);
1467 unsigned Slot = insertValue(C, Typ, Tab);
1468 ResolveReferencesToConstant(C, Slot);
1469 if (Handler) Handler->handleConstantString(cast<ConstantArray>(C));
1473 /// Parse the constant pool.
1474 void BytecodeReader::ParseConstantPool(ValueTable &Tab,
1475 TypeListTy &TypeTab,
1477 if (Handler) Handler->handleGlobalConstantsBegin();
1479 /// In LLVM 1.3 Type does not derive from Value so the types
1480 /// do not occupy a plane. Consequently, we read the types
1481 /// first in the constant pool.
1482 if (isFunction && !hasTypeDerivedFromValue) {
1483 unsigned NumEntries = read_vbr_uint();
1484 ParseTypes(TypeTab, NumEntries);
1487 while (moreInBlock()) {
1488 unsigned NumEntries = read_vbr_uint();
1490 bool isTypeType = read_typeid(Typ);
1492 /// In LLVM 1.2 and before, Types were written to the
1493 /// bytecode file in the "Type Type" plane (#12).
1494 /// In 1.3 plane 12 is now the label plane. Handle this here.
1496 ParseTypes(TypeTab, NumEntries);
1497 } else if (Typ == Type::VoidTyID) {
1498 /// Use of Type::VoidTyID is a misnomer. It actually means
1499 /// that the following plane is constant strings
1500 assert(&Tab == &ModuleValues && "Cannot read strings in functions!");
1501 ParseStringConstants(NumEntries, Tab);
1503 for (unsigned i = 0; i < NumEntries; ++i) {
1504 Constant *C = ParseConstantValue(Typ);
1505 assert(C && "ParseConstantValue returned NULL!");
1506 unsigned Slot = insertValue(C, Typ, Tab);
1508 // If we are reading a function constant table, make sure that we adjust
1509 // the slot number to be the real global constant number.
1511 if (&Tab != &ModuleValues && Typ < ModuleValues.size() &&
1513 Slot += ModuleValues[Typ]->size();
1514 ResolveReferencesToConstant(C, Slot);
1518 checkPastBlockEnd("Constant Pool");
1519 if (Handler) Handler->handleGlobalConstantsEnd();
1522 /// Parse the contents of a function. Note that this function can be
1523 /// called lazily by materializeFunction
1524 /// @see materializeFunction
1525 void BytecodeReader::ParseFunctionBody(Function* F) {
1527 unsigned FuncSize = BlockEnd - At;
1528 GlobalValue::LinkageTypes Linkage = GlobalValue::ExternalLinkage;
1530 unsigned LinkageType = read_vbr_uint();
1531 switch (LinkageType) {
1532 case 0: Linkage = GlobalValue::ExternalLinkage; break;
1533 case 1: Linkage = GlobalValue::WeakLinkage; break;
1534 case 2: Linkage = GlobalValue::AppendingLinkage; break;
1535 case 3: Linkage = GlobalValue::InternalLinkage; break;
1536 case 4: Linkage = GlobalValue::LinkOnceLinkage; break;
1538 error("Invalid linkage type for Function.");
1539 Linkage = GlobalValue::InternalLinkage;
1543 F->setLinkage(Linkage);
1544 if (Handler) Handler->handleFunctionBegin(F,FuncSize);
1546 // Keep track of how many basic blocks we have read in...
1547 unsigned BlockNum = 0;
1548 bool InsertedArguments = false;
1550 BufPtr MyEnd = BlockEnd;
1551 while (At < MyEnd) {
1552 unsigned Type, Size;
1554 read_block(Type, Size);
1557 case BytecodeFormat::ConstantPoolBlockID:
1558 if (!InsertedArguments) {
1559 // Insert arguments into the value table before we parse the first basic
1560 // block in the function, but after we potentially read in the
1561 // compaction table.
1563 InsertedArguments = true;
1566 ParseConstantPool(FunctionValues, FunctionTypes, true);
1569 case BytecodeFormat::CompactionTableBlockID:
1570 ParseCompactionTable();
1573 case BytecodeFormat::BasicBlock: {
1574 if (!InsertedArguments) {
1575 // Insert arguments into the value table before we parse the first basic
1576 // block in the function, but after we potentially read in the
1577 // compaction table.
1579 InsertedArguments = true;
1582 BasicBlock *BB = ParseBasicBlock(BlockNum++);
1583 F->getBasicBlockList().push_back(BB);
1587 case BytecodeFormat::InstructionListBlockID: {
1588 // Insert arguments into the value table before we parse the instruction
1589 // list for the function, but after we potentially read in the compaction
1591 if (!InsertedArguments) {
1593 InsertedArguments = true;
1597 error("Already parsed basic blocks!");
1598 BlockNum = ParseInstructionList(F);
1602 case BytecodeFormat::SymbolTableBlockID:
1603 ParseSymbolTable(F, &F->getSymbolTable());
1609 error("Wrapped around reading bytecode.");
1614 // Malformed bc file if read past end of block.
1618 // Make sure there were no references to non-existant basic blocks.
1619 if (BlockNum != ParsedBasicBlocks.size())
1620 error("Illegal basic block operand reference");
1622 ParsedBasicBlocks.clear();
1624 // Resolve forward references. Replace any uses of a forward reference value
1625 // with the real value.
1627 // replaceAllUsesWith is very inefficient for instructions which have a LARGE
1628 // number of operands. PHI nodes often have forward references, and can also
1629 // often have a very large number of operands.
1631 // FIXME: REEVALUATE. replaceAllUsesWith is _much_ faster now, and this code
1632 // should be simplified back to using it!
1634 std::map<Value*, Value*> ForwardRefMapping;
1635 for (std::map<std::pair<unsigned,unsigned>, Value*>::iterator
1636 I = ForwardReferences.begin(), E = ForwardReferences.end();
1638 ForwardRefMapping[I->second] = getValue(I->first.first, I->first.second,
1641 for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
1642 for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
1643 for (unsigned i = 0, e = I->getNumOperands(); i != e; ++i)
1644 if (Argument *A = dyn_cast<Argument>(I->getOperand(i))) {
1645 std::map<Value*, Value*>::iterator It = ForwardRefMapping.find(A);
1646 if (It != ForwardRefMapping.end()) I->setOperand(i, It->second);
1649 while (!ForwardReferences.empty()) {
1650 std::map<std::pair<unsigned,unsigned>, Value*>::iterator I =
1651 ForwardReferences.begin();
1652 Value *PlaceHolder = I->second;
1653 ForwardReferences.erase(I);
1655 // Now that all the uses are gone, delete the placeholder...
1656 // If we couldn't find a def (error case), then leak a little
1657 // memory, because otherwise we can't remove all uses!
1661 // Clear out function-level types...
1662 FunctionTypes.clear();
1663 CompactionTypes.clear();
1664 CompactionValues.clear();
1665 freeTable(FunctionValues);
1667 if (Handler) Handler->handleFunctionEnd(F);
1670 /// This function parses LLVM functions lazily. It obtains the type of the
1671 /// function and records where the body of the function is in the bytecode
1672 /// buffer. The caller can then use the ParseNextFunction and
1673 /// ParseAllFunctionBodies to get handler events for the functions.
1674 void BytecodeReader::ParseFunctionLazily() {
1675 if (FunctionSignatureList.empty())
1676 error("FunctionSignatureList empty!");
1678 Function *Func = FunctionSignatureList.back();
1679 FunctionSignatureList.pop_back();
1681 // Save the information for future reading of the function
1682 LazyFunctionLoadMap[Func] = LazyFunctionInfo(BlockStart, BlockEnd);
1684 // Pretend we've `parsed' this function
1688 /// The ParserFunction method lazily parses one function. Use this method to
1689 /// casue the parser to parse a specific function in the module. Note that
1690 /// this will remove the function from what is to be included by
1691 /// ParseAllFunctionBodies.
1692 /// @see ParseAllFunctionBodies
1693 /// @see ParseBytecode
1694 void BytecodeReader::ParseFunction(Function* Func) {
1695 // Find {start, end} pointers and slot in the map. If not there, we're done.
1696 LazyFunctionMap::iterator Fi = LazyFunctionLoadMap.find(Func);
1698 // Make sure we found it
1699 if (Fi == LazyFunctionLoadMap.end()) {
1700 error("Unrecognized function of type " + Func->getType()->getDescription());
1704 BlockStart = At = Fi->second.Buf;
1705 BlockEnd = Fi->second.EndBuf;
1706 assert(Fi->first == Func && "Found wrong function?");
1708 LazyFunctionLoadMap.erase(Fi);
1710 this->ParseFunctionBody(Func);
1713 /// The ParseAllFunctionBodies method parses through all the previously
1714 /// unparsed functions in the bytecode file. If you want to completely parse
1715 /// a bytecode file, this method should be called after Parsebytecode because
1716 /// Parsebytecode only records the locations in the bytecode file of where
1717 /// the function definitions are located. This function uses that information
1718 /// to materialize the functions.
1719 /// @see ParseBytecode
1720 void BytecodeReader::ParseAllFunctionBodies() {
1721 LazyFunctionMap::iterator Fi = LazyFunctionLoadMap.begin();
1722 LazyFunctionMap::iterator Fe = LazyFunctionLoadMap.end();
1725 Function* Func = Fi->first;
1726 BlockStart = At = Fi->second.Buf;
1727 BlockEnd = Fi->second.EndBuf;
1728 this->ParseFunctionBody(Func);
1733 /// Parse the global type list
1734 void BytecodeReader::ParseGlobalTypes() {
1735 // Read the number of types
1736 unsigned NumEntries = read_vbr_uint();
1738 // Ignore the type plane identifier for types if the bc file is pre 1.3
1739 if (hasTypeDerivedFromValue)
1742 ParseTypes(ModuleTypes, NumEntries);
1745 /// Parse the Global info (types, global vars, constants)
1746 void BytecodeReader::ParseModuleGlobalInfo() {
1748 if (Handler) Handler->handleModuleGlobalsBegin();
1750 // Read global variables...
1751 unsigned VarType = read_vbr_uint();
1752 while (VarType != Type::VoidTyID) { // List is terminated by Void
1753 // VarType Fields: bit0 = isConstant, bit1 = hasInitializer, bit2,3,4 =
1754 // Linkage, bit4+ = slot#
1755 unsigned SlotNo = VarType >> 5;
1756 if (sanitizeTypeId(SlotNo))
1757 error("Invalid type (type type) for global var!");
1758 unsigned LinkageID = (VarType >> 2) & 7;
1759 bool isConstant = VarType & 1;
1760 bool hasInitializer = VarType & 2;
1761 GlobalValue::LinkageTypes Linkage;
1763 switch (LinkageID) {
1764 case 0: Linkage = GlobalValue::ExternalLinkage; break;
1765 case 1: Linkage = GlobalValue::WeakLinkage; break;
1766 case 2: Linkage = GlobalValue::AppendingLinkage; break;
1767 case 3: Linkage = GlobalValue::InternalLinkage; break;
1768 case 4: Linkage = GlobalValue::LinkOnceLinkage; break;
1770 error("Unknown linkage type: " + utostr(LinkageID));
1771 Linkage = GlobalValue::InternalLinkage;
1775 const Type *Ty = getType(SlotNo);
1777 error("Global has no type! SlotNo=" + utostr(SlotNo));
1780 if (!isa<PointerType>(Ty)) {
1781 error("Global not a pointer type! Ty= " + Ty->getDescription());
1784 const Type *ElTy = cast<PointerType>(Ty)->getElementType();
1786 // Create the global variable...
1787 GlobalVariable *GV = new GlobalVariable(ElTy, isConstant, Linkage,
1789 insertValue(GV, SlotNo, ModuleValues);
1791 unsigned initSlot = 0;
1792 if (hasInitializer) {
1793 initSlot = read_vbr_uint();
1794 GlobalInits.push_back(std::make_pair(GV, initSlot));
1797 // Notify handler about the global value.
1798 if (Handler) Handler->handleGlobalVariable(ElTy, isConstant, Linkage, SlotNo, initSlot);
1801 VarType = read_vbr_uint();
1804 // Read the function objects for all of the functions that are coming
1805 unsigned FnSignature = 0;
1806 if (read_typeid(FnSignature))
1807 error("Invalid function type (type type) found");
1809 while (FnSignature != Type::VoidTyID) { // List is terminated by Void
1810 const Type *Ty = getType(FnSignature);
1811 if (!isa<PointerType>(Ty) ||
1812 !isa<FunctionType>(cast<PointerType>(Ty)->getElementType())) {
1813 error("Function not a pointer to function type! Ty = " +
1814 Ty->getDescription());
1815 // FIXME: what should Ty be if handler continues?
1818 // We create functions by passing the underlying FunctionType to create...
1819 const FunctionType* FTy =
1820 cast<FunctionType>(cast<PointerType>(Ty)->getElementType());
1822 // Insert the place hodler
1823 Function* Func = new Function(FTy, GlobalValue::InternalLinkage,
1825 insertValue(Func, FnSignature, ModuleValues);
1827 // Save this for later so we know type of lazily instantiated functions
1828 FunctionSignatureList.push_back(Func);
1830 if (Handler) Handler->handleFunctionDeclaration(Func);
1832 // Get Next function signature
1833 if (read_typeid(FnSignature))
1834 error("Invalid function type (type type) found");
1837 // Now that the function signature list is set up, reverse it so that we can
1838 // remove elements efficiently from the back of the vector.
1839 std::reverse(FunctionSignatureList.begin(), FunctionSignatureList.end());
1841 // If this bytecode format has dependent library information in it ..
1842 if (!hasNoDependentLibraries) {
1843 // Read in the number of dependent library items that follow
1844 unsigned num_dep_libs = read_vbr_uint();
1845 std::string dep_lib;
1846 while( num_dep_libs-- ) {
1847 dep_lib = read_str();
1848 TheModule->addLibrary(dep_lib);
1851 // Read target triple and place into the module
1852 std::string triple = read_str();
1853 TheModule->setTargetTriple(triple);
1856 if (hasInconsistentModuleGlobalInfo)
1859 // This is for future proofing... in the future extra fields may be added that
1860 // we don't understand, so we transparently ignore them.
1864 if (Handler) Handler->handleModuleGlobalsEnd();
1867 /// Parse the version information and decode it by setting flags on the
1868 /// Reader that enable backward compatibility of the reader.
1869 void BytecodeReader::ParseVersionInfo() {
1870 unsigned Version = read_vbr_uint();
1872 // Unpack version number: low four bits are for flags, top bits = version
1873 Module::Endianness Endianness;
1874 Module::PointerSize PointerSize;
1875 Endianness = (Version & 1) ? Module::BigEndian : Module::LittleEndian;
1876 PointerSize = (Version & 2) ? Module::Pointer64 : Module::Pointer32;
1878 bool hasNoEndianness = Version & 4;
1879 bool hasNoPointerSize = Version & 8;
1881 RevisionNum = Version >> 4;
1883 // Default values for the current bytecode version
1884 hasInconsistentModuleGlobalInfo = false;
1885 hasExplicitPrimitiveZeros = false;
1886 hasRestrictedGEPTypes = false;
1887 hasTypeDerivedFromValue = false;
1888 hasLongBlockHeaders = false;
1889 has32BitTypes = false;
1890 hasNoDependentLibraries = false;
1892 switch (RevisionNum) {
1893 case 0: // LLVM 1.0, 1.1 release version
1894 // Base LLVM 1.0 bytecode format.
1895 hasInconsistentModuleGlobalInfo = true;
1896 hasExplicitPrimitiveZeros = true;
1900 case 1: // LLVM 1.2 release version
1901 // LLVM 1.2 added explicit support for emitting strings efficiently.
1903 // Also, it fixed the problem where the size of the ModuleGlobalInfo block
1904 // included the size for the alignment at the end, where the rest of the
1907 // LLVM 1.2 and before required that GEP indices be ubyte constants for
1908 // structures and longs for sequential types.
1909 hasRestrictedGEPTypes = true;
1911 // LLVM 1.2 and before had the Type class derive from Value class. This
1912 // changed in release 1.3 and consequently LLVM 1.3 bytecode files are
1913 // written differently because Types can no longer be part of the
1914 // type planes for Values.
1915 hasTypeDerivedFromValue = true;
1919 case 2: /// 1.2.5 (mid-release) version
1921 /// LLVM 1.2 and earlier had two-word block headers. This is a bit wasteful,
1922 /// especially for small files where the 8 bytes per block is a large fraction
1923 /// of the total block size. In LLVM 1.3, the block type and length are
1924 /// compressed into a single 32-bit unsigned integer. 27 bits for length, 5
1925 /// bits for block type.
1926 hasLongBlockHeaders = true;
1928 /// LLVM 1.2 and earlier wrote type slot numbers as vbr_uint32. In LLVM 1.3
1929 /// this has been reduced to vbr_uint24. It shouldn't make much difference
1930 /// since we haven't run into a module with > 24 million types, but for safety
1931 /// the 24-bit restriction has been enforced in 1.3 to free some bits in
1932 /// various places and to ensure consistency.
1933 has32BitTypes = true;
1935 /// LLVM 1.2 and earlier did not provide a target triple nor a list of
1936 /// libraries on which the bytecode is dependent. LLVM 1.3 provides these
1937 /// features, for use in future versions of LLVM.
1938 hasNoDependentLibraries = true;
1941 case 3: // LLVM 1.3 release version
1945 error("Unknown bytecode version number: " + itostr(RevisionNum));
1948 if (hasNoEndianness) Endianness = Module::AnyEndianness;
1949 if (hasNoPointerSize) PointerSize = Module::AnyPointerSize;
1951 TheModule->setEndianness(Endianness);
1952 TheModule->setPointerSize(PointerSize);
1954 if (Handler) Handler->handleVersionInfo(RevisionNum, Endianness, PointerSize);
1957 /// Parse a whole module.
1958 void BytecodeReader::ParseModule() {
1959 unsigned Type, Size;
1961 FunctionSignatureList.clear(); // Just in case...
1963 // Read into instance variables...
1967 bool SeenModuleGlobalInfo = false;
1968 bool SeenGlobalTypePlane = false;
1969 BufPtr MyEnd = BlockEnd;
1970 while (At < MyEnd) {
1972 read_block(Type, Size);
1976 case BytecodeFormat::GlobalTypePlaneBlockID:
1977 if (SeenGlobalTypePlane)
1978 error("Two GlobalTypePlane Blocks Encountered!");
1981 SeenGlobalTypePlane = true;
1984 case BytecodeFormat::ModuleGlobalInfoBlockID:
1985 if (SeenModuleGlobalInfo)
1986 error("Two ModuleGlobalInfo Blocks Encountered!");
1987 ParseModuleGlobalInfo();
1988 SeenModuleGlobalInfo = true;
1991 case BytecodeFormat::ConstantPoolBlockID:
1992 ParseConstantPool(ModuleValues, ModuleTypes,false);
1995 case BytecodeFormat::FunctionBlockID:
1996 ParseFunctionLazily();
1999 case BytecodeFormat::SymbolTableBlockID:
2000 ParseSymbolTable(0, &TheModule->getSymbolTable());
2006 error("Unexpected Block of Type #" + utostr(Type) + " encountered!");
2014 // After the module constant pool has been read, we can safely initialize
2015 // global variables...
2016 while (!GlobalInits.empty()) {
2017 GlobalVariable *GV = GlobalInits.back().first;
2018 unsigned Slot = GlobalInits.back().second;
2019 GlobalInits.pop_back();
2021 // Look up the initializer value...
2022 // FIXME: Preserve this type ID!
2024 const llvm::PointerType* GVType = GV->getType();
2025 unsigned TypeSlot = getTypeSlot(GVType->getElementType());
2026 if (Constant *CV = getConstantValue(TypeSlot, Slot)) {
2027 if (GV->hasInitializer())
2028 error("Global *already* has an initializer?!");
2029 if (Handler) Handler->handleGlobalInitializer(GV,CV);
2030 GV->setInitializer(CV);
2032 error("Cannot find initializer value.");
2035 /// Make sure we pulled them all out. If we didn't then there's a declaration
2036 /// but a missing body. That's not allowed.
2037 if (!FunctionSignatureList.empty())
2038 error("Function declared, but bytecode stream ended before definition");
2041 /// This function completely parses a bytecode buffer given by the \p Buf
2042 /// and \p Length parameters.
2043 void BytecodeReader::ParseBytecode(BufPtr Buf, unsigned Length,
2044 const std::string &ModuleID,
2045 bool processFunctions) {
2048 At = MemStart = BlockStart = Buf;
2049 MemEnd = BlockEnd = Buf + Length;
2051 // Create the module
2052 TheModule = new Module(ModuleID);
2054 if (Handler) Handler->handleStart(TheModule, Length);
2056 // Read and check signature...
2057 unsigned Sig = read_uint();
2058 if (Sig != ('l' | ('l' << 8) | ('v' << 16) | ('m' << 24))) {
2059 error("Invalid bytecode signature: " + utostr(Sig));
2062 // Tell the handler we're starting a module
2063 if (Handler) Handler->handleModuleBegin(ModuleID);
2065 // Get the module block and size and verify. This is handled specially
2066 // because the module block/size is always written in long format. Other
2067 // blocks are written in short format so the read_block method is used.
2068 unsigned Type, Size;
2071 if (Type != BytecodeFormat::ModuleBlockID) {
2072 error("Expected Module Block! Type:" + utostr(Type) + ", Size:"
2075 if (At + Size != MemEnd) {
2076 error("Invalid Top Level Block Length! Type:" + utostr(Type)
2077 + ", Size:" + utostr(Size));
2080 // Parse the module contents
2081 this->ParseModule();
2083 // Check for missing functions
2085 error("Function expected, but bytecode stream ended!");
2087 // Process all the function bodies now, if requested
2088 if (processFunctions)
2089 ParseAllFunctionBodies();
2091 // Tell the handler we're done with the module
2093 Handler->handleModuleEnd(ModuleID);
2095 // Tell the handler we're finished the parse
2096 if (Handler) Handler->handleFinish();
2098 } catch (std::string& errstr) {
2099 if (Handler) Handler->handleError(errstr);
2105 std::string msg("Unknown Exception Occurred");
2106 if (Handler) Handler->handleError(msg);
2114 //===----------------------------------------------------------------------===//
2115 //=== Default Implementations of Handler Methods
2116 //===----------------------------------------------------------------------===//
2118 BytecodeHandler::~BytecodeHandler() {}