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"
33 /// @brief A class for maintaining the slot number definition
34 /// as a placeholder for the actual definition for forward constants defs.
35 class ConstantPlaceHolder : public ConstantExpr {
37 ConstantPlaceHolder(); // DO NOT IMPLEMENT
38 void operator=(const ConstantPlaceHolder &); // DO NOT IMPLEMENT
40 ConstantPlaceHolder(const Type *Ty, unsigned id)
41 : ConstantExpr(Instruction::UserOp1, Constant::getNullValue(Ty), Ty),
43 unsigned getID() { return ID; }
48 // Provide some details on error
49 inline void BytecodeReader::error(std::string err) {
51 err += itostr(RevisionNum) ;
53 err += itostr(At-MemStart);
58 //===----------------------------------------------------------------------===//
59 // Bytecode Reading Methods
60 //===----------------------------------------------------------------------===//
62 /// Determine if the current block being read contains any more data.
63 inline bool BytecodeReader::moreInBlock() {
67 /// Throw an error if we've read past the end of the current block
68 inline void BytecodeReader::checkPastBlockEnd(const char * block_name) {
70 error(std::string("Attempt to read past the end of ") + block_name + " block.");
73 /// Align the buffer position to a 32 bit boundary
74 inline void BytecodeReader::align32() {
76 At = (const unsigned char *)((unsigned long)(At+3) & (~3UL));
78 if (Handler) Handler->handleAlignment(At - Save);
80 error("Ran out of data while aligning!");
83 /// Read a whole unsigned integer
84 inline unsigned BytecodeReader::read_uint() {
86 error("Ran out of data reading uint!");
88 return At[-4] | (At[-3] << 8) | (At[-2] << 16) | (At[-1] << 24);
91 /// Read a variable-bit-rate encoded unsigned integer
92 inline unsigned BytecodeReader::read_vbr_uint() {
99 error("Ran out of data reading vbr_uint!");
100 Result |= (unsigned)((*At++) & 0x7F) << Shift;
102 } while (At[-1] & 0x80);
103 if (Handler) Handler->handleVBR32(At-Save);
107 /// Read a variable-bit-rate encoded unsigned 64-bit integer.
108 inline uint64_t BytecodeReader::read_vbr_uint64() {
115 error("Ran out of data reading vbr_uint64!");
116 Result |= (uint64_t)((*At++) & 0x7F) << Shift;
118 } while (At[-1] & 0x80);
119 if (Handler) Handler->handleVBR64(At-Save);
123 /// Read a variable-bit-rate encoded signed 64-bit integer.
124 inline int64_t BytecodeReader::read_vbr_int64() {
125 uint64_t R = read_vbr_uint64();
128 return -(int64_t)(R >> 1);
129 else // There is no such thing as -0 with integers. "-0" really means
130 // 0x8000000000000000.
133 return (int64_t)(R >> 1);
136 /// Read a pascal-style string (length followed by text)
137 inline std::string BytecodeReader::read_str() {
138 unsigned Size = read_vbr_uint();
139 const unsigned char *OldAt = At;
141 if (At > BlockEnd) // Size invalid?
142 error("Ran out of data reading a string!");
143 return std::string((char*)OldAt, Size);
146 /// Read an arbitrary block of data
147 inline void BytecodeReader::read_data(void *Ptr, void *End) {
148 unsigned char *Start = (unsigned char *)Ptr;
149 unsigned Amount = (unsigned char *)End - Start;
150 if (At+Amount > BlockEnd)
151 error("Ran out of data!");
152 std::copy(At, At+Amount, Start);
156 /// Read a float value in little-endian order
157 inline void BytecodeReader::read_float(float& FloatVal) {
158 /// FIXME: This isn't optimal, it has size problems on some platforms
159 /// where FP is not IEEE.
164 FloatUnion.i = At[0] | (At[1] << 8) | (At[2] << 16) | (At[3] << 24);
165 At+=sizeof(uint32_t);
166 FloatVal = FloatUnion.f;
169 /// Read a double value in little-endian order
170 inline void BytecodeReader::read_double(double& DoubleVal) {
171 /// FIXME: This isn't optimal, it has size problems on some platforms
172 /// where FP is not IEEE.
177 DoubleUnion.i = (uint64_t(At[0]) << 0) | (uint64_t(At[1]) << 8) |
178 (uint64_t(At[2]) << 16) | (uint64_t(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].first;
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 for (unsigned i = 0, e = CompactionTypes.size(); i != e; ++i)
340 if (CompactionTypes[i].first == Ty)
341 return Type::FirstDerivedTyID + i;
343 error("Couldn't find type specified in compaction table!");
346 // Check the function level types first...
347 TypeListTy::iterator I = find(FunctionTypes.begin(), FunctionTypes.end(), Ty);
349 if (I != FunctionTypes.end())
350 return Type::FirstDerivedTyID + ModuleTypes.size() +
351 (&*I - &FunctionTypes[0]);
353 // Check the module level types now...
354 I = find(ModuleTypes.begin(), ModuleTypes.end(), Ty);
355 if (I == ModuleTypes.end())
356 error("Didn't find type in ModuleTypes.");
357 return Type::FirstDerivedTyID + (&*I - &ModuleTypes[0]);
360 /// This is just like getType, but when a compaction table is in use, it is
361 /// ignored. It also ignores function level types.
363 const Type *BytecodeReader::getGlobalTableType(unsigned Slot) {
364 if (Slot < Type::FirstDerivedTyID) {
365 const Type *Ty = Type::getPrimitiveType((Type::TypeID)Slot);
367 error("Not a primitive type ID?");
370 Slot -= Type::FirstDerivedTyID;
371 if (Slot >= ModuleTypes.size())
372 error("Illegal compaction table type reference!");
373 return ModuleTypes[Slot];
376 /// This is just like getTypeSlot, but when a compaction table is in use, it
377 /// is ignored. It also ignores function level types.
378 unsigned BytecodeReader::getGlobalTableTypeSlot(const Type *Ty) {
379 if (Ty->isPrimitiveType())
380 return Ty->getTypeID();
381 TypeListTy::iterator I = find(ModuleTypes.begin(),
382 ModuleTypes.end(), Ty);
383 if (I == ModuleTypes.end())
384 error("Didn't find type in ModuleTypes.");
385 return Type::FirstDerivedTyID + (&*I - &ModuleTypes[0]);
388 /// Retrieve a value of a given type and slot number, possibly creating
389 /// it if it doesn't already exist.
390 Value * BytecodeReader::getValue(unsigned type, unsigned oNum, bool Create) {
391 assert(type != Type::LabelTyID && "getValue() cannot get blocks!");
394 // If there is a compaction table active, it defines the low-level numbers.
395 // If not, the module values define the low-level numbers.
396 if (CompactionValues.size() > type && !CompactionValues[type].empty()) {
397 if (Num < CompactionValues[type].size())
398 return CompactionValues[type][Num];
399 Num -= CompactionValues[type].size();
401 // By default, the global type id is the type id passed in
402 unsigned GlobalTyID = type;
404 // If the type plane was compactified, figure out the global type ID by
405 // adding the derived type ids and the distance.
406 if (!CompactionTypes.empty() && type >= Type::FirstDerivedTyID)
407 GlobalTyID = CompactionTypes[type-Type::FirstDerivedTyID].second;
409 if (hasImplicitNull(GlobalTyID)) {
411 return Constant::getNullValue(getType(type));
415 if (GlobalTyID < ModuleValues.size() && ModuleValues[GlobalTyID]) {
416 if (Num < ModuleValues[GlobalTyID]->size())
417 return ModuleValues[GlobalTyID]->getOperand(Num);
418 Num -= ModuleValues[GlobalTyID]->size();
422 if (FunctionValues.size() > type &&
423 FunctionValues[type] &&
424 Num < FunctionValues[type]->size())
425 return FunctionValues[type]->getOperand(Num);
427 if (!Create) return 0; // Do not create a placeholder?
429 std::pair<unsigned,unsigned> KeyValue(type, oNum);
430 ForwardReferenceMap::iterator I = ForwardReferences.lower_bound(KeyValue);
431 if (I != ForwardReferences.end() && I->first == KeyValue)
432 return I->second; // We have already created this placeholder
434 Value *Val = new Argument(getType(type));
435 ForwardReferences.insert(I, std::make_pair(KeyValue, Val));
439 /// This is just like getValue, but when a compaction table is in use, it
440 /// is ignored. Also, no forward references or other fancy features are
442 Value* BytecodeReader::getGlobalTableValue(unsigned TyID, unsigned SlotNo) {
444 return Constant::getNullValue(getType(TyID));
446 if (!CompactionTypes.empty() && TyID >= Type::FirstDerivedTyID) {
447 TyID -= Type::FirstDerivedTyID;
448 if (TyID >= CompactionTypes.size())
449 error("Type ID out of range for compaction table!");
450 TyID = CompactionTypes[TyID].second;
455 if (TyID >= ModuleValues.size() || ModuleValues[TyID] == 0 ||
456 SlotNo >= ModuleValues[TyID]->size()) {
457 if (TyID >= ModuleValues.size() || ModuleValues[TyID] == 0)
458 error("Corrupt compaction table entry!"
459 + utostr(TyID) + ", " + utostr(SlotNo) + ": "
460 + utostr(ModuleValues.size()));
462 error("Corrupt compaction table entry!"
463 + utostr(TyID) + ", " + utostr(SlotNo) + ": "
464 + utostr(ModuleValues.size()) + ", "
465 + utohexstr(intptr_t((void*)ModuleValues[TyID])) + ", "
466 + utostr(ModuleValues[TyID]->size()));
468 return ModuleValues[TyID]->getOperand(SlotNo);
471 /// Just like getValue, except that it returns a null pointer
472 /// only on error. It always returns a constant (meaning that if the value is
473 /// defined, but is not a constant, that is an error). If the specified
474 /// constant hasn't been parsed yet, a placeholder is defined and used.
475 /// Later, after the real value is parsed, the placeholder is eliminated.
476 Constant* BytecodeReader::getConstantValue(unsigned TypeSlot, unsigned Slot) {
477 if (Value *V = getValue(TypeSlot, Slot, false))
478 if (Constant *C = dyn_cast<Constant>(V))
479 return C; // If we already have the value parsed, just return it
481 error("Value for slot " + utostr(Slot) +
482 " is expected to be a constant!");
484 const Type *Ty = getType(TypeSlot);
485 std::pair<const Type*, unsigned> Key(Ty, Slot);
486 ConstantRefsType::iterator I = ConstantFwdRefs.lower_bound(Key);
488 if (I != ConstantFwdRefs.end() && I->first == Key) {
491 // Create a placeholder for the constant reference and
492 // keep track of the fact that we have a forward ref to recycle it
493 Constant *C = new ConstantPlaceHolder(Ty, Slot);
495 // Keep track of the fact that we have a forward ref to recycle it
496 ConstantFwdRefs.insert(I, std::make_pair(Key, C));
501 //===----------------------------------------------------------------------===//
502 // IR Construction Methods
503 //===----------------------------------------------------------------------===//
505 /// As values are created, they are inserted into the appropriate place
506 /// with this method. The ValueTable argument must be one of ModuleValues
507 /// or FunctionValues data members of this class.
508 unsigned BytecodeReader::insertValue(Value *Val, unsigned type,
509 ValueTable &ValueTab) {
510 assert((!isa<Constant>(Val) || !cast<Constant>(Val)->isNullValue()) ||
511 !hasImplicitNull(type) &&
512 "Cannot read null values from bytecode!");
514 if (ValueTab.size() <= type)
515 ValueTab.resize(type+1);
517 if (!ValueTab[type]) ValueTab[type] = new ValueList();
519 ValueTab[type]->push_back(Val);
521 bool HasOffset = hasImplicitNull(type);
522 return ValueTab[type]->size()-1 + HasOffset;
525 /// Insert the arguments of a function as new values in the reader.
526 void BytecodeReader::insertArguments(Function* F) {
527 const FunctionType *FT = F->getFunctionType();
528 Function::aiterator AI = F->abegin();
529 for (FunctionType::param_iterator It = FT->param_begin();
530 It != FT->param_end(); ++It, ++AI)
531 insertValue(AI, getTypeSlot(AI->getType()), FunctionValues);
534 //===----------------------------------------------------------------------===//
535 // Bytecode Parsing Methods
536 //===----------------------------------------------------------------------===//
538 /// This method parses a single instruction. The instruction is
539 /// inserted at the end of the \p BB provided. The arguments of
540 /// the instruction are provided in the \p Args vector.
541 void BytecodeReader::ParseInstruction(std::vector<unsigned> &Oprnds,
545 // Clear instruction data
549 unsigned Op = read_uint();
551 // bits Instruction format: Common to all formats
552 // --------------------------
553 // 01-00: Opcode type, fixed to 1.
555 Opcode = (Op >> 2) & 63;
556 Oprnds.resize((Op >> 0) & 03);
558 // Extract the operands
559 switch (Oprnds.size()) {
561 // bits Instruction format:
562 // --------------------------
563 // 19-08: Resulting type plane
564 // 31-20: Operand #1 (if set to (2^12-1), then zero operands)
566 iType = (Op >> 8) & 4095;
567 Oprnds[0] = (Op >> 20) & 4095;
568 if (Oprnds[0] == 4095) // Handle special encoding for 0 operands...
572 // bits Instruction format:
573 // --------------------------
574 // 15-08: Resulting type plane
578 iType = (Op >> 8) & 255;
579 Oprnds[0] = (Op >> 16) & 255;
580 Oprnds[1] = (Op >> 24) & 255;
583 // bits Instruction format:
584 // --------------------------
585 // 13-08: Resulting type plane
590 iType = (Op >> 8) & 63;
591 Oprnds[0] = (Op >> 14) & 63;
592 Oprnds[1] = (Op >> 20) & 63;
593 Oprnds[2] = (Op >> 26) & 63;
596 At -= 4; // Hrm, try this again...
597 Opcode = read_vbr_uint();
599 iType = read_vbr_uint();
601 unsigned NumOprnds = read_vbr_uint();
602 Oprnds.resize(NumOprnds);
605 error("Zero-argument instruction found; this is invalid.");
607 for (unsigned i = 0; i != NumOprnds; ++i)
608 Oprnds[i] = read_vbr_uint();
613 const Type *InstTy = getSanitizedType(iType);
615 // We have enough info to inform the handler now.
616 if (Handler) Handler->handleInstruction(Opcode, InstTy, Oprnds, At-SaveAt);
618 // Declare the resulting instruction we'll build.
619 Instruction *Result = 0;
621 // Handle binary operators
622 if (Opcode >= Instruction::BinaryOpsBegin &&
623 Opcode < Instruction::BinaryOpsEnd && Oprnds.size() == 2)
624 Result = BinaryOperator::create((Instruction::BinaryOps)Opcode,
625 getValue(iType, Oprnds[0]),
626 getValue(iType, Oprnds[1]));
631 error("Illegal instruction read!");
633 case Instruction::VAArg:
634 Result = new VAArgInst(getValue(iType, Oprnds[0]),
635 getSanitizedType(Oprnds[1]));
637 case Instruction::VANext:
638 Result = new VANextInst(getValue(iType, Oprnds[0]),
639 getSanitizedType(Oprnds[1]));
641 case Instruction::Cast:
642 Result = new CastInst(getValue(iType, Oprnds[0]),
643 getSanitizedType(Oprnds[1]));
645 case Instruction::Select:
646 Result = new SelectInst(getValue(Type::BoolTyID, Oprnds[0]),
647 getValue(iType, Oprnds[1]),
648 getValue(iType, Oprnds[2]));
650 case Instruction::PHI: {
651 if (Oprnds.size() == 0 || (Oprnds.size() & 1))
652 error("Invalid phi node encountered!");
654 PHINode *PN = new PHINode(InstTy);
655 PN->op_reserve(Oprnds.size());
656 for (unsigned i = 0, e = Oprnds.size(); i != e; i += 2)
657 PN->addIncoming(getValue(iType, Oprnds[i]), getBasicBlock(Oprnds[i+1]));
662 case Instruction::Shl:
663 case Instruction::Shr:
664 Result = new ShiftInst((Instruction::OtherOps)Opcode,
665 getValue(iType, Oprnds[0]),
666 getValue(Type::UByteTyID, Oprnds[1]));
668 case Instruction::Ret:
669 if (Oprnds.size() == 0)
670 Result = new ReturnInst();
671 else if (Oprnds.size() == 1)
672 Result = new ReturnInst(getValue(iType, Oprnds[0]));
674 error("Unrecognized instruction!");
677 case Instruction::Br:
678 if (Oprnds.size() == 1)
679 Result = new BranchInst(getBasicBlock(Oprnds[0]));
680 else if (Oprnds.size() == 3)
681 Result = new BranchInst(getBasicBlock(Oprnds[0]),
682 getBasicBlock(Oprnds[1]), getValue(Type::BoolTyID , Oprnds[2]));
684 error("Invalid number of operands for a 'br' instruction!");
686 case Instruction::Switch: {
687 if (Oprnds.size() & 1)
688 error("Switch statement with odd number of arguments!");
690 SwitchInst *I = new SwitchInst(getValue(iType, Oprnds[0]),
691 getBasicBlock(Oprnds[1]));
692 for (unsigned i = 2, e = Oprnds.size(); i != e; i += 2)
693 I->addCase(cast<Constant>(getValue(iType, Oprnds[i])),
694 getBasicBlock(Oprnds[i+1]));
699 case Instruction::Call: {
700 if (Oprnds.size() == 0)
701 error("Invalid call instruction encountered!");
703 Value *F = getValue(iType, Oprnds[0]);
705 // Check to make sure we have a pointer to function type
706 const PointerType *PTy = dyn_cast<PointerType>(F->getType());
707 if (PTy == 0) error("Call to non function pointer value!");
708 const FunctionType *FTy = dyn_cast<FunctionType>(PTy->getElementType());
709 if (FTy == 0) error("Call to non function pointer value!");
711 std::vector<Value *> Params;
712 if (!FTy->isVarArg()) {
713 FunctionType::param_iterator It = FTy->param_begin();
715 for (unsigned i = 1, e = Oprnds.size(); i != e; ++i) {
716 if (It == FTy->param_end())
717 error("Invalid call instruction!");
718 Params.push_back(getValue(getTypeSlot(*It++), Oprnds[i]));
720 if (It != FTy->param_end())
721 error("Invalid call instruction!");
723 Oprnds.erase(Oprnds.begin(), Oprnds.begin()+1);
725 unsigned FirstVariableOperand;
726 if (Oprnds.size() < FTy->getNumParams())
727 error("Call instruction missing operands!");
729 // Read all of the fixed arguments
730 for (unsigned i = 0, e = FTy->getNumParams(); i != e; ++i)
731 Params.push_back(getValue(getTypeSlot(FTy->getParamType(i)),Oprnds[i]));
733 FirstVariableOperand = FTy->getNumParams();
735 if ((Oprnds.size()-FirstVariableOperand) & 1) // Must be pairs of type/value
736 error("Invalid call instruction!");
738 for (unsigned i = FirstVariableOperand, e = Oprnds.size();
740 Params.push_back(getValue(Oprnds[i], Oprnds[i+1]));
743 Result = new CallInst(F, Params);
746 case Instruction::Invoke: {
747 if (Oprnds.size() < 3)
748 error("Invalid invoke instruction!");
749 Value *F = getValue(iType, Oprnds[0]);
751 // Check to make sure we have a pointer to function type
752 const PointerType *PTy = dyn_cast<PointerType>(F->getType());
754 error("Invoke to non function pointer value!");
755 const FunctionType *FTy = dyn_cast<FunctionType>(PTy->getElementType());
757 error("Invoke to non function pointer value!");
759 std::vector<Value *> Params;
760 BasicBlock *Normal, *Except;
762 if (!FTy->isVarArg()) {
763 Normal = getBasicBlock(Oprnds[1]);
764 Except = getBasicBlock(Oprnds[2]);
766 FunctionType::param_iterator It = FTy->param_begin();
767 for (unsigned i = 3, e = Oprnds.size(); i != e; ++i) {
768 if (It == FTy->param_end())
769 error("Invalid invoke instruction!");
770 Params.push_back(getValue(getTypeSlot(*It++), Oprnds[i]));
772 if (It != FTy->param_end())
773 error("Invalid invoke instruction!");
775 Oprnds.erase(Oprnds.begin(), Oprnds.begin()+1);
777 Normal = getBasicBlock(Oprnds[0]);
778 Except = getBasicBlock(Oprnds[1]);
780 unsigned FirstVariableArgument = FTy->getNumParams()+2;
781 for (unsigned i = 2; i != FirstVariableArgument; ++i)
782 Params.push_back(getValue(getTypeSlot(FTy->getParamType(i-2)),
785 if (Oprnds.size()-FirstVariableArgument & 1) // Must be type/value pairs
786 error("Invalid invoke instruction!");
788 for (unsigned i = FirstVariableArgument; i < Oprnds.size(); i += 2)
789 Params.push_back(getValue(Oprnds[i], Oprnds[i+1]));
792 Result = new InvokeInst(F, Normal, Except, Params);
795 case Instruction::Malloc:
796 if (Oprnds.size() > 2)
797 error("Invalid malloc instruction!");
798 if (!isa<PointerType>(InstTy))
799 error("Invalid malloc instruction!");
801 Result = new MallocInst(cast<PointerType>(InstTy)->getElementType(),
802 Oprnds.size() ? getValue(Type::UIntTyID,
806 case Instruction::Alloca:
807 if (Oprnds.size() > 2)
808 error("Invalid alloca instruction!");
809 if (!isa<PointerType>(InstTy))
810 error("Invalid alloca instruction!");
812 Result = new AllocaInst(cast<PointerType>(InstTy)->getElementType(),
813 Oprnds.size() ? getValue(Type::UIntTyID,
816 case Instruction::Free:
817 if (!isa<PointerType>(InstTy))
818 error("Invalid free instruction!");
819 Result = new FreeInst(getValue(iType, Oprnds[0]));
821 case Instruction::GetElementPtr: {
822 if (Oprnds.size() == 0 || !isa<PointerType>(InstTy))
823 error("Invalid getelementptr instruction!");
825 std::vector<Value*> Idx;
827 const Type *NextTy = InstTy;
828 for (unsigned i = 1, e = Oprnds.size(); i != e; ++i) {
829 const CompositeType *TopTy = dyn_cast_or_null<CompositeType>(NextTy);
831 error("Invalid getelementptr instruction!");
833 unsigned ValIdx = Oprnds[i];
835 if (!hasRestrictedGEPTypes) {
836 // Struct indices are always uints, sequential type indices can be any
837 // of the 32 or 64-bit integer types. The actual choice of type is
838 // encoded in the low two bits of the slot number.
839 if (isa<StructType>(TopTy))
840 IdxTy = Type::UIntTyID;
842 switch (ValIdx & 3) {
844 case 0: IdxTy = Type::UIntTyID; break;
845 case 1: IdxTy = Type::IntTyID; break;
846 case 2: IdxTy = Type::ULongTyID; break;
847 case 3: IdxTy = Type::LongTyID; break;
852 IdxTy = isa<StructType>(TopTy) ? Type::UByteTyID : Type::LongTyID;
855 Idx.push_back(getValue(IdxTy, ValIdx));
857 // Convert ubyte struct indices into uint struct indices.
858 if (isa<StructType>(TopTy) && hasRestrictedGEPTypes)
859 if (ConstantUInt *C = dyn_cast<ConstantUInt>(Idx.back()))
860 Idx[Idx.size()-1] = ConstantExpr::getCast(C, Type::UIntTy);
862 NextTy = GetElementPtrInst::getIndexedType(InstTy, Idx, true);
865 Result = new GetElementPtrInst(getValue(iType, Oprnds[0]), Idx);
869 case 62: // volatile load
870 case Instruction::Load:
871 if (Oprnds.size() != 1 || !isa<PointerType>(InstTy))
872 error("Invalid load instruction!");
873 Result = new LoadInst(getValue(iType, Oprnds[0]), "", Opcode == 62);
876 case 63: // volatile store
877 case Instruction::Store: {
878 if (!isa<PointerType>(InstTy) || Oprnds.size() != 2)
879 error("Invalid store instruction!");
881 Value *Ptr = getValue(iType, Oprnds[1]);
882 const Type *ValTy = cast<PointerType>(Ptr->getType())->getElementType();
883 Result = new StoreInst(getValue(getTypeSlot(ValTy), Oprnds[0]), Ptr,
887 case Instruction::Unwind:
888 if (Oprnds.size() != 0)
889 error("Invalid unwind instruction!");
890 Result = new UnwindInst();
892 } // end switch(Opcode)
895 if (Result->getType() == InstTy)
898 TypeSlot = getTypeSlot(Result->getType());
900 insertValue(Result, TypeSlot, FunctionValues);
901 BB->getInstList().push_back(Result);
904 /// Get a particular numbered basic block, which might be a forward reference.
905 /// This works together with ParseBasicBlock to handle these forward references
906 /// in a clean manner. This function is used when constructing phi, br, switch,
907 /// and other instructions that reference basic blocks. Blocks are numbered
908 /// sequentially as they appear in the function.
909 BasicBlock *BytecodeReader::getBasicBlock(unsigned ID) {
910 // Make sure there is room in the table...
911 if (ParsedBasicBlocks.size() <= ID) ParsedBasicBlocks.resize(ID+1);
913 // First check to see if this is a backwards reference, i.e., ParseBasicBlock
914 // has already created this block, or if the forward reference has already
916 if (ParsedBasicBlocks[ID])
917 return ParsedBasicBlocks[ID];
919 // Otherwise, the basic block has not yet been created. Do so and add it to
920 // the ParsedBasicBlocks list.
921 return ParsedBasicBlocks[ID] = new BasicBlock();
924 /// In LLVM 1.0 bytecode files, we used to output one basicblock at a time.
925 /// This method reads in one of the basicblock packets. This method is not used
926 /// for bytecode files after LLVM 1.0
927 /// @returns The basic block constructed.
928 BasicBlock *BytecodeReader::ParseBasicBlock(unsigned BlockNo) {
929 if (Handler) Handler->handleBasicBlockBegin(BlockNo);
933 if (ParsedBasicBlocks.size() == BlockNo)
934 ParsedBasicBlocks.push_back(BB = new BasicBlock());
935 else if (ParsedBasicBlocks[BlockNo] == 0)
936 BB = ParsedBasicBlocks[BlockNo] = new BasicBlock();
938 BB = ParsedBasicBlocks[BlockNo];
940 std::vector<unsigned> Operands;
941 while (moreInBlock())
942 ParseInstruction(Operands, BB);
944 if (Handler) Handler->handleBasicBlockEnd(BlockNo);
948 /// Parse all of the BasicBlock's & Instruction's in the body of a function.
949 /// In post 1.0 bytecode files, we no longer emit basic block individually,
950 /// in order to avoid per-basic-block overhead.
951 /// @returns Rhe number of basic blocks encountered.
952 unsigned BytecodeReader::ParseInstructionList(Function* F) {
953 unsigned BlockNo = 0;
954 std::vector<unsigned> Args;
956 while (moreInBlock()) {
957 if (Handler) Handler->handleBasicBlockBegin(BlockNo);
959 if (ParsedBasicBlocks.size() == BlockNo)
960 ParsedBasicBlocks.push_back(BB = new BasicBlock());
961 else if (ParsedBasicBlocks[BlockNo] == 0)
962 BB = ParsedBasicBlocks[BlockNo] = new BasicBlock();
964 BB = ParsedBasicBlocks[BlockNo];
966 F->getBasicBlockList().push_back(BB);
968 // Read instructions into this basic block until we get to a terminator
969 while (moreInBlock() && !BB->getTerminator())
970 ParseInstruction(Args, BB);
972 if (!BB->getTerminator())
973 error("Non-terminated basic block found!");
975 if (Handler) Handler->handleBasicBlockEnd(BlockNo-1);
981 /// Parse a symbol table. This works for both module level and function
982 /// level symbol tables. For function level symbol tables, the CurrentFunction
983 /// parameter must be non-zero and the ST parameter must correspond to
984 /// CurrentFunction's symbol table. For Module level symbol tables, the
985 /// CurrentFunction argument must be zero.
986 void BytecodeReader::ParseSymbolTable(Function *CurrentFunction,
988 if (Handler) Handler->handleSymbolTableBegin(CurrentFunction,ST);
990 // Allow efficient basic block lookup by number.
991 std::vector<BasicBlock*> BBMap;
993 for (Function::iterator I = CurrentFunction->begin(),
994 E = CurrentFunction->end(); I != E; ++I)
997 /// In LLVM 1.3 we write types separately from values so
998 /// The types are always first in the symbol table. This is
999 /// because Type no longer derives from Value.
1000 if (!hasTypeDerivedFromValue) {
1001 // Symtab block header: [num entries]
1002 unsigned NumEntries = read_vbr_uint();
1003 for (unsigned i = 0; i < NumEntries; ++i) {
1004 // Symtab entry: [def slot #][name]
1005 unsigned slot = read_vbr_uint();
1006 std::string Name = read_str();
1007 const Type* T = getType(slot);
1008 ST->insert(Name, T);
1012 while (moreInBlock()) {
1013 // Symtab block header: [num entries][type id number]
1014 unsigned NumEntries = read_vbr_uint();
1016 bool isTypeType = read_typeid(Typ);
1017 const Type *Ty = getType(Typ);
1019 for (unsigned i = 0; i != NumEntries; ++i) {
1020 // Symtab entry: [def slot #][name]
1021 unsigned slot = read_vbr_uint();
1022 std::string Name = read_str();
1024 // if we're reading a pre 1.3 bytecode file and the type plane
1025 // is the "type type", handle it here
1027 const Type* T = getType(slot);
1029 error("Failed type look-up for name '" + Name + "'");
1030 ST->insert(Name, T);
1031 continue; // code below must be short circuited
1034 if (Typ == Type::LabelTyID) {
1035 if (slot < BBMap.size())
1038 V = getValue(Typ, slot, false); // Find mapping...
1041 error("Failed value look-up for name '" + Name + "'");
1042 V->setName(Name, ST);
1046 checkPastBlockEnd("Symbol Table");
1047 if (Handler) Handler->handleSymbolTableEnd();
1050 /// Read in the types portion of a compaction table.
1051 void BytecodeReader::ParseCompactionTypes(unsigned NumEntries) {
1052 for (unsigned i = 0; i != NumEntries; ++i) {
1053 unsigned TypeSlot = 0;
1054 if (read_typeid(TypeSlot))
1055 error("Invalid type in compaction table: type type");
1056 const Type *Typ = getGlobalTableType(TypeSlot);
1057 CompactionTypes.push_back(std::make_pair(Typ, TypeSlot));
1058 if (Handler) Handler->handleCompactionTableType(i, TypeSlot, Typ);
1062 /// Parse a compaction table.
1063 void BytecodeReader::ParseCompactionTable() {
1065 // Notify handler that we're beginning a compaction table.
1066 if (Handler) Handler->handleCompactionTableBegin();
1068 // In LLVM 1.3 Type no longer derives from Value. So,
1069 // we always write them first in the compaction table
1070 // because they can't occupy a "type plane" where the
1072 if (! hasTypeDerivedFromValue) {
1073 unsigned NumEntries = read_vbr_uint();
1074 ParseCompactionTypes(NumEntries);
1077 // Compaction tables live in separate blocks so we have to loop
1078 // until we've read the whole thing.
1079 while (moreInBlock()) {
1080 // Read the number of Value* entries in the compaction table
1081 unsigned NumEntries = read_vbr_uint();
1083 unsigned isTypeType = false;
1085 // Decode the type from value read in. Most compaction table
1086 // planes will have one or two entries in them. If that's the
1087 // case then the length is encoded in the bottom two bits and
1088 // the higher bits encode the type. This saves another VBR value.
1089 if ((NumEntries & 3) == 3) {
1090 // In this case, both low-order bits are set (value 3). This
1091 // is a signal that the typeid follows.
1093 isTypeType = read_typeid(Ty);
1095 // In this case, the low-order bits specify the number of entries
1096 // and the high order bits specify the type.
1097 Ty = NumEntries >> 2;
1098 isTypeType = sanitizeTypeId(Ty);
1102 // if we're reading a pre 1.3 bytecode file and the type plane
1103 // is the "type type", handle it here
1105 ParseCompactionTypes(NumEntries);
1107 // Make sure we have enough room for the plane.
1108 if (Ty >= CompactionValues.size())
1109 CompactionValues.resize(Ty+1);
1111 // Make sure the plane is empty or we have some kind of error.
1112 if (!CompactionValues[Ty].empty())
1113 error("Compaction table plane contains multiple entries!");
1115 // Notify handler about the plane.
1116 if (Handler) Handler->handleCompactionTablePlane(Ty, NumEntries);
1118 // Push the implicit zero.
1119 CompactionValues[Ty].push_back(Constant::getNullValue(getType(Ty)));
1121 // Read in each of the entries, put them in the compaction table
1122 // and notify the handler that we have a new compaction table value.
1123 for (unsigned i = 0; i != NumEntries; ++i) {
1124 unsigned ValSlot = read_vbr_uint();
1125 Value *V = getGlobalTableValue(Ty, ValSlot);
1126 CompactionValues[Ty].push_back(V);
1127 if (Handler) Handler->handleCompactionTableValue(i, Ty, ValSlot);
1131 // Notify handler that the compaction table is done.
1132 if (Handler) Handler->handleCompactionTableEnd();
1135 // Parse a single type. The typeid is read in first. If its a primitive type
1136 // then nothing else needs to be read, we know how to instantiate it. If its
1137 // a derived type, then additional data is read to fill out the type
1139 const Type *BytecodeReader::ParseType() {
1140 unsigned PrimType = 0;
1141 if (read_typeid(PrimType))
1142 error("Invalid type (type type) in type constants!");
1144 const Type *Result = 0;
1145 if ((Result = Type::getPrimitiveType((Type::TypeID)PrimType)))
1149 case Type::FunctionTyID: {
1150 const Type *RetType = readSanitizedType();
1152 unsigned NumParams = read_vbr_uint();
1154 std::vector<const Type*> Params;
1156 Params.push_back(readSanitizedType());
1158 bool isVarArg = Params.size() && Params.back() == Type::VoidTy;
1159 if (isVarArg) Params.pop_back();
1161 Result = FunctionType::get(RetType, Params, isVarArg);
1164 case Type::ArrayTyID: {
1165 const Type *ElementType = readSanitizedType();
1166 unsigned NumElements = read_vbr_uint();
1167 Result = ArrayType::get(ElementType, NumElements);
1170 case Type::StructTyID: {
1171 std::vector<const Type*> Elements;
1173 if (read_typeid(Typ))
1174 error("Invalid element type (type type) for structure!");
1176 while (Typ) { // List is terminated by void/0 typeid
1177 Elements.push_back(getType(Typ));
1178 if (read_typeid(Typ))
1179 error("Invalid element type (type type) for structure!");
1182 Result = StructType::get(Elements);
1185 case Type::PointerTyID: {
1186 Result = PointerType::get(readSanitizedType());
1190 case Type::OpaqueTyID: {
1191 Result = OpaqueType::get();
1196 error("Don't know how to deserialize primitive type " + utostr(PrimType));
1199 if (Handler) Handler->handleType(Result);
1203 // ParseType - We have to use this weird code to handle recursive
1204 // types. We know that recursive types will only reference the current slab of
1205 // values in the type plane, but they can forward reference types before they
1206 // have been read. For example, Type #0 might be '{ Ty#1 }' and Type #1 might
1207 // be 'Ty#0*'. When reading Type #0, type number one doesn't exist. To fix
1208 // this ugly problem, we pessimistically insert an opaque type for each type we
1209 // are about to read. This means that forward references will resolve to
1210 // something and when we reread the type later, we can replace the opaque type
1211 // with a new resolved concrete type.
1213 void BytecodeReader::ParseTypes(TypeListTy &Tab, unsigned NumEntries){
1214 assert(Tab.size() == 0 && "should not have read type constants in before!");
1216 // Insert a bunch of opaque types to be resolved later...
1217 Tab.reserve(NumEntries);
1218 for (unsigned i = 0; i != NumEntries; ++i)
1219 Tab.push_back(OpaqueType::get());
1221 // Loop through reading all of the types. Forward types will make use of the
1222 // opaque types just inserted.
1224 for (unsigned i = 0; i != NumEntries; ++i) {
1225 const Type* NewTy = ParseType();
1226 const Type* OldTy = Tab[i].get();
1228 error("Couldn't parse type!");
1230 // Don't directly push the new type on the Tab. Instead we want to replace
1231 // the opaque type we previously inserted with the new concrete value. This
1232 // approach helps with forward references to types. The refinement from the
1233 // abstract (opaque) type to the new type causes all uses of the abstract
1234 // type to use the concrete type (NewTy). This will also cause the opaque
1235 // type to be deleted.
1236 cast<DerivedType>(const_cast<Type*>(OldTy))->refineAbstractTypeTo(NewTy);
1238 // This should have replaced the old opaque type with the new type in the
1239 // value table... or with a preexisting type that was already in the system.
1240 // Let's just make sure it did.
1241 assert(Tab[i] != OldTy && "refineAbstractType didn't work!");
1245 /// Parse a single constant value
1246 Constant *BytecodeReader::ParseConstantValue(unsigned TypeID) {
1247 // We must check for a ConstantExpr before switching by type because
1248 // a ConstantExpr can be of any type, and has no explicit value.
1250 // 0 if not expr; numArgs if is expr
1251 unsigned isExprNumArgs = read_vbr_uint();
1253 if (isExprNumArgs) {
1254 // FIXME: Encoding of constant exprs could be much more compact!
1255 std::vector<Constant*> ArgVec;
1256 ArgVec.reserve(isExprNumArgs);
1257 unsigned Opcode = read_vbr_uint();
1259 // Read the slot number and types of each of the arguments
1260 for (unsigned i = 0; i != isExprNumArgs; ++i) {
1261 unsigned ArgValSlot = read_vbr_uint();
1262 unsigned ArgTypeSlot = 0;
1263 if (read_typeid(ArgTypeSlot))
1264 error("Invalid argument type (type type) for constant value");
1266 // Get the arg value from its slot if it exists, otherwise a placeholder
1267 ArgVec.push_back(getConstantValue(ArgTypeSlot, ArgValSlot));
1270 // Construct a ConstantExpr of the appropriate kind
1271 if (isExprNumArgs == 1) { // All one-operand expressions
1272 if (Opcode != Instruction::Cast)
1273 error("Only Cast instruction has one argument for ConstantExpr");
1275 Constant* Result = ConstantExpr::getCast(ArgVec[0], getType(TypeID));
1276 if (Handler) Handler->handleConstantExpression(Opcode, ArgVec, Result);
1278 } else if (Opcode == Instruction::GetElementPtr) { // GetElementPtr
1279 std::vector<Constant*> IdxList(ArgVec.begin()+1, ArgVec.end());
1281 if (hasRestrictedGEPTypes) {
1282 const Type *BaseTy = ArgVec[0]->getType();
1283 generic_gep_type_iterator<std::vector<Constant*>::iterator>
1284 GTI = gep_type_begin(BaseTy, IdxList.begin(), IdxList.end()),
1285 E = gep_type_end(BaseTy, IdxList.begin(), IdxList.end());
1286 for (unsigned i = 0; GTI != E; ++GTI, ++i)
1287 if (isa<StructType>(*GTI)) {
1288 if (IdxList[i]->getType() != Type::UByteTy)
1289 error("Invalid index for getelementptr!");
1290 IdxList[i] = ConstantExpr::getCast(IdxList[i], Type::UIntTy);
1294 Constant* Result = ConstantExpr::getGetElementPtr(ArgVec[0], IdxList);
1295 if (Handler) Handler->handleConstantExpression(Opcode, ArgVec, Result);
1297 } else if (Opcode == Instruction::Select) {
1298 if (ArgVec.size() != 3)
1299 error("Select instruction must have three arguments.");
1300 Constant* Result = ConstantExpr::getSelect(ArgVec[0], ArgVec[1],
1302 if (Handler) Handler->handleConstantExpression(Opcode, ArgVec, Result);
1304 } else { // All other 2-operand expressions
1305 Constant* Result = ConstantExpr::get(Opcode, ArgVec[0], ArgVec[1]);
1306 if (Handler) Handler->handleConstantExpression(Opcode, ArgVec, Result);
1311 // Ok, not an ConstantExpr. We now know how to read the given type...
1312 const Type *Ty = getType(TypeID);
1313 switch (Ty->getTypeID()) {
1314 case Type::BoolTyID: {
1315 unsigned Val = read_vbr_uint();
1316 if (Val != 0 && Val != 1)
1317 error("Invalid boolean value read.");
1318 Constant* Result = ConstantBool::get(Val == 1);
1319 if (Handler) Handler->handleConstantValue(Result);
1323 case Type::UByteTyID: // Unsigned integer types...
1324 case Type::UShortTyID:
1325 case Type::UIntTyID: {
1326 unsigned Val = read_vbr_uint();
1327 if (!ConstantUInt::isValueValidForType(Ty, Val))
1328 error("Invalid unsigned byte/short/int read.");
1329 Constant* Result = ConstantUInt::get(Ty, Val);
1330 if (Handler) Handler->handleConstantValue(Result);
1334 case Type::ULongTyID: {
1335 Constant* Result = ConstantUInt::get(Ty, read_vbr_uint64());
1336 if (Handler) Handler->handleConstantValue(Result);
1340 case Type::SByteTyID: // Signed integer types...
1341 case Type::ShortTyID:
1342 case Type::IntTyID: {
1343 case Type::LongTyID:
1344 int64_t Val = read_vbr_int64();
1345 if (!ConstantSInt::isValueValidForType(Ty, Val))
1346 error("Invalid signed byte/short/int/long read.");
1347 Constant* Result = ConstantSInt::get(Ty, Val);
1348 if (Handler) Handler->handleConstantValue(Result);
1352 case Type::FloatTyID: {
1355 Constant* Result = ConstantFP::get(Ty, Val);
1356 if (Handler) Handler->handleConstantValue(Result);
1360 case Type::DoubleTyID: {
1363 Constant* Result = ConstantFP::get(Ty, Val);
1364 if (Handler) Handler->handleConstantValue(Result);
1368 case Type::ArrayTyID: {
1369 const ArrayType *AT = cast<ArrayType>(Ty);
1370 unsigned NumElements = AT->getNumElements();
1371 unsigned TypeSlot = getTypeSlot(AT->getElementType());
1372 std::vector<Constant*> Elements;
1373 Elements.reserve(NumElements);
1374 while (NumElements--) // Read all of the elements of the constant.
1375 Elements.push_back(getConstantValue(TypeSlot,
1377 Constant* Result = ConstantArray::get(AT, Elements);
1378 if (Handler) Handler->handleConstantArray(AT, Elements, TypeSlot, Result);
1382 case Type::StructTyID: {
1383 const StructType *ST = cast<StructType>(Ty);
1385 std::vector<Constant *> Elements;
1386 Elements.reserve(ST->getNumElements());
1387 for (unsigned i = 0; i != ST->getNumElements(); ++i)
1388 Elements.push_back(getConstantValue(ST->getElementType(i),
1391 Constant* Result = ConstantStruct::get(ST, Elements);
1392 if (Handler) Handler->handleConstantStruct(ST, Elements, Result);
1396 case Type::PointerTyID: { // ConstantPointerRef value...
1397 const PointerType *PT = cast<PointerType>(Ty);
1398 unsigned Slot = read_vbr_uint();
1400 // Check to see if we have already read this global variable...
1401 Value *Val = getValue(TypeID, Slot, false);
1403 GlobalValue *GV = dyn_cast<GlobalValue>(Val);
1404 if (!GV) error("GlobalValue not in ValueTable!");
1405 if (Handler) Handler->handleConstantPointer(PT, Slot, GV);
1408 error("Forward references are not allowed here.");
1413 error("Don't know how to deserialize constant value of type '" +
1414 Ty->getDescription());
1420 /// Resolve references for constants. This function resolves the forward
1421 /// referenced constants in the ConstantFwdRefs map. It uses the
1422 /// replaceAllUsesWith method of Value class to substitute the placeholder
1423 /// instance with the actual instance.
1424 void BytecodeReader::ResolveReferencesToConstant(Constant *NewV, unsigned Slot){
1425 ConstantRefsType::iterator I =
1426 ConstantFwdRefs.find(std::make_pair(NewV->getType(), Slot));
1427 if (I == ConstantFwdRefs.end()) return; // Never forward referenced?
1429 Value *PH = I->second; // Get the placeholder...
1430 PH->replaceAllUsesWith(NewV);
1431 delete PH; // Delete the old placeholder
1432 ConstantFwdRefs.erase(I); // Remove the map entry for it
1435 /// Parse the constant strings section.
1436 void BytecodeReader::ParseStringConstants(unsigned NumEntries, ValueTable &Tab){
1437 for (; NumEntries; --NumEntries) {
1439 if (read_typeid(Typ))
1440 error("Invalid type (type type) for string constant");
1441 const Type *Ty = getType(Typ);
1442 if (!isa<ArrayType>(Ty))
1443 error("String constant data invalid!");
1445 const ArrayType *ATy = cast<ArrayType>(Ty);
1446 if (ATy->getElementType() != Type::SByteTy &&
1447 ATy->getElementType() != Type::UByteTy)
1448 error("String constant data invalid!");
1450 // Read character data. The type tells us how long the string is.
1451 char Data[ATy->getNumElements()];
1452 read_data(Data, Data+ATy->getNumElements());
1454 std::vector<Constant*> Elements(ATy->getNumElements());
1455 if (ATy->getElementType() == Type::SByteTy)
1456 for (unsigned i = 0, e = ATy->getNumElements(); i != e; ++i)
1457 Elements[i] = ConstantSInt::get(Type::SByteTy, (signed char)Data[i]);
1459 for (unsigned i = 0, e = ATy->getNumElements(); i != e; ++i)
1460 Elements[i] = ConstantUInt::get(Type::UByteTy, (unsigned char)Data[i]);
1462 // Create the constant, inserting it as needed.
1463 Constant *C = ConstantArray::get(ATy, Elements);
1464 unsigned Slot = insertValue(C, Typ, Tab);
1465 ResolveReferencesToConstant(C, Slot);
1466 if (Handler) Handler->handleConstantString(cast<ConstantArray>(C));
1470 /// Parse the constant pool.
1471 void BytecodeReader::ParseConstantPool(ValueTable &Tab,
1472 TypeListTy &TypeTab,
1474 if (Handler) Handler->handleGlobalConstantsBegin();
1476 /// In LLVM 1.3 Type does not derive from Value so the types
1477 /// do not occupy a plane. Consequently, we read the types
1478 /// first in the constant pool.
1479 if (isFunction && !hasTypeDerivedFromValue) {
1480 unsigned NumEntries = read_vbr_uint();
1481 ParseTypes(TypeTab, NumEntries);
1484 while (moreInBlock()) {
1485 unsigned NumEntries = read_vbr_uint();
1487 bool isTypeType = read_typeid(Typ);
1489 /// In LLVM 1.2 and before, Types were written to the
1490 /// bytecode file in the "Type Type" plane (#12).
1491 /// In 1.3 plane 12 is now the label plane. Handle this here.
1493 ParseTypes(TypeTab, NumEntries);
1494 } else if (Typ == Type::VoidTyID) {
1495 /// Use of Type::VoidTyID is a misnomer. It actually means
1496 /// that the following plane is constant strings
1497 assert(&Tab == &ModuleValues && "Cannot read strings in functions!");
1498 ParseStringConstants(NumEntries, Tab);
1500 for (unsigned i = 0; i < NumEntries; ++i) {
1501 Constant *C = ParseConstantValue(Typ);
1502 assert(C && "ParseConstantValue returned NULL!");
1503 unsigned Slot = insertValue(C, Typ, Tab);
1505 // If we are reading a function constant table, make sure that we adjust
1506 // the slot number to be the real global constant number.
1508 if (&Tab != &ModuleValues && Typ < ModuleValues.size() &&
1510 Slot += ModuleValues[Typ]->size();
1511 ResolveReferencesToConstant(C, Slot);
1515 checkPastBlockEnd("Constant Pool");
1516 if (Handler) Handler->handleGlobalConstantsEnd();
1519 /// Parse the contents of a function. Note that this function can be
1520 /// called lazily by materializeFunction
1521 /// @see materializeFunction
1522 void BytecodeReader::ParseFunctionBody(Function* F) {
1524 unsigned FuncSize = BlockEnd - At;
1525 GlobalValue::LinkageTypes Linkage = GlobalValue::ExternalLinkage;
1527 unsigned LinkageType = read_vbr_uint();
1528 switch (LinkageType) {
1529 case 0: Linkage = GlobalValue::ExternalLinkage; break;
1530 case 1: Linkage = GlobalValue::WeakLinkage; break;
1531 case 2: Linkage = GlobalValue::AppendingLinkage; break;
1532 case 3: Linkage = GlobalValue::InternalLinkage; break;
1533 case 4: Linkage = GlobalValue::LinkOnceLinkage; break;
1535 error("Invalid linkage type for Function.");
1536 Linkage = GlobalValue::InternalLinkage;
1540 F->setLinkage(Linkage);
1541 if (Handler) Handler->handleFunctionBegin(F,FuncSize);
1543 // Keep track of how many basic blocks we have read in...
1544 unsigned BlockNum = 0;
1545 bool InsertedArguments = false;
1547 BufPtr MyEnd = BlockEnd;
1548 while (At < MyEnd) {
1549 unsigned Type, Size;
1551 read_block(Type, Size);
1554 case BytecodeFormat::ConstantPoolBlockID:
1555 if (!InsertedArguments) {
1556 // Insert arguments into the value table before we parse the first basic
1557 // block in the function, but after we potentially read in the
1558 // compaction table.
1560 InsertedArguments = true;
1563 ParseConstantPool(FunctionValues, FunctionTypes, true);
1566 case BytecodeFormat::CompactionTableBlockID:
1567 ParseCompactionTable();
1570 case BytecodeFormat::BasicBlock: {
1571 if (!InsertedArguments) {
1572 // Insert arguments into the value table before we parse the first basic
1573 // block in the function, but after we potentially read in the
1574 // compaction table.
1576 InsertedArguments = true;
1579 BasicBlock *BB = ParseBasicBlock(BlockNum++);
1580 F->getBasicBlockList().push_back(BB);
1584 case BytecodeFormat::InstructionListBlockID: {
1585 // Insert arguments into the value table before we parse the instruction
1586 // list for the function, but after we potentially read in the compaction
1588 if (!InsertedArguments) {
1590 InsertedArguments = true;
1594 error("Already parsed basic blocks!");
1595 BlockNum = ParseInstructionList(F);
1599 case BytecodeFormat::SymbolTableBlockID:
1600 ParseSymbolTable(F, &F->getSymbolTable());
1606 error("Wrapped around reading bytecode.");
1611 // Malformed bc file if read past end of block.
1615 // Make sure there were no references to non-existant basic blocks.
1616 if (BlockNum != ParsedBasicBlocks.size())
1617 error("Illegal basic block operand reference");
1619 ParsedBasicBlocks.clear();
1621 // Resolve forward references. Replace any uses of a forward reference value
1622 // with the real value.
1624 // replaceAllUsesWith is very inefficient for instructions which have a LARGE
1625 // number of operands. PHI nodes often have forward references, and can also
1626 // often have a very large number of operands.
1628 // FIXME: REEVALUATE. replaceAllUsesWith is _much_ faster now, and this code
1629 // should be simplified back to using it!
1631 std::map<Value*, Value*> ForwardRefMapping;
1632 for (std::map<std::pair<unsigned,unsigned>, Value*>::iterator
1633 I = ForwardReferences.begin(), E = ForwardReferences.end();
1635 ForwardRefMapping[I->second] = getValue(I->first.first, I->first.second,
1638 for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
1639 for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
1640 for (unsigned i = 0, e = I->getNumOperands(); i != e; ++i)
1641 if (Argument *A = dyn_cast<Argument>(I->getOperand(i))) {
1642 std::map<Value*, Value*>::iterator It = ForwardRefMapping.find(A);
1643 if (It != ForwardRefMapping.end()) I->setOperand(i, It->second);
1646 while (!ForwardReferences.empty()) {
1647 std::map<std::pair<unsigned,unsigned>, Value*>::iterator I =
1648 ForwardReferences.begin();
1649 Value *PlaceHolder = I->second;
1650 ForwardReferences.erase(I);
1652 // Now that all the uses are gone, delete the placeholder...
1653 // If we couldn't find a def (error case), then leak a little
1654 // memory, because otherwise we can't remove all uses!
1658 // Clear out function-level types...
1659 FunctionTypes.clear();
1660 CompactionTypes.clear();
1661 CompactionValues.clear();
1662 freeTable(FunctionValues);
1664 if (Handler) Handler->handleFunctionEnd(F);
1667 /// This function parses LLVM functions lazily. It obtains the type of the
1668 /// function and records where the body of the function is in the bytecode
1669 /// buffer. The caller can then use the ParseNextFunction and
1670 /// ParseAllFunctionBodies to get handler events for the functions.
1671 void BytecodeReader::ParseFunctionLazily() {
1672 if (FunctionSignatureList.empty())
1673 error("FunctionSignatureList empty!");
1675 Function *Func = FunctionSignatureList.back();
1676 FunctionSignatureList.pop_back();
1678 // Save the information for future reading of the function
1679 LazyFunctionLoadMap[Func] = LazyFunctionInfo(BlockStart, BlockEnd);
1681 // Pretend we've `parsed' this function
1685 /// The ParserFunction method lazily parses one function. Use this method to
1686 /// casue the parser to parse a specific function in the module. Note that
1687 /// this will remove the function from what is to be included by
1688 /// ParseAllFunctionBodies.
1689 /// @see ParseAllFunctionBodies
1690 /// @see ParseBytecode
1691 void BytecodeReader::ParseFunction(Function* Func) {
1692 // Find {start, end} pointers and slot in the map. If not there, we're done.
1693 LazyFunctionMap::iterator Fi = LazyFunctionLoadMap.find(Func);
1695 // Make sure we found it
1696 if (Fi == LazyFunctionLoadMap.end()) {
1697 error("Unrecognized function of type " + Func->getType()->getDescription());
1701 BlockStart = At = Fi->second.Buf;
1702 BlockEnd = Fi->second.EndBuf;
1703 assert(Fi->first == Func && "Found wrong function?");
1705 LazyFunctionLoadMap.erase(Fi);
1707 this->ParseFunctionBody(Func);
1710 /// The ParseAllFunctionBodies method parses through all the previously
1711 /// unparsed functions in the bytecode file. If you want to completely parse
1712 /// a bytecode file, this method should be called after Parsebytecode because
1713 /// Parsebytecode only records the locations in the bytecode file of where
1714 /// the function definitions are located. This function uses that information
1715 /// to materialize the functions.
1716 /// @see ParseBytecode
1717 void BytecodeReader::ParseAllFunctionBodies() {
1718 LazyFunctionMap::iterator Fi = LazyFunctionLoadMap.begin();
1719 LazyFunctionMap::iterator Fe = LazyFunctionLoadMap.end();
1722 Function* Func = Fi->first;
1723 BlockStart = At = Fi->second.Buf;
1724 BlockEnd = Fi->second.EndBuf;
1725 this->ParseFunctionBody(Func);
1730 /// Parse the global type list
1731 void BytecodeReader::ParseGlobalTypes() {
1732 // Read the number of types
1733 unsigned NumEntries = read_vbr_uint();
1735 // Ignore the type plane identifier for types if the bc file is pre 1.3
1736 if (hasTypeDerivedFromValue)
1739 ParseTypes(ModuleTypes, NumEntries);
1742 /// Parse the Global info (types, global vars, constants)
1743 void BytecodeReader::ParseModuleGlobalInfo() {
1745 if (Handler) Handler->handleModuleGlobalsBegin();
1747 // Read global variables...
1748 unsigned VarType = read_vbr_uint();
1749 while (VarType != Type::VoidTyID) { // List is terminated by Void
1750 // VarType Fields: bit0 = isConstant, bit1 = hasInitializer, bit2,3,4 =
1751 // Linkage, bit4+ = slot#
1752 unsigned SlotNo = VarType >> 5;
1753 if (sanitizeTypeId(SlotNo))
1754 error("Invalid type (type type) for global var!");
1755 unsigned LinkageID = (VarType >> 2) & 7;
1756 bool isConstant = VarType & 1;
1757 bool hasInitializer = VarType & 2;
1758 GlobalValue::LinkageTypes Linkage;
1760 switch (LinkageID) {
1761 case 0: Linkage = GlobalValue::ExternalLinkage; break;
1762 case 1: Linkage = GlobalValue::WeakLinkage; break;
1763 case 2: Linkage = GlobalValue::AppendingLinkage; break;
1764 case 3: Linkage = GlobalValue::InternalLinkage; break;
1765 case 4: Linkage = GlobalValue::LinkOnceLinkage; break;
1767 error("Unknown linkage type: " + utostr(LinkageID));
1768 Linkage = GlobalValue::InternalLinkage;
1772 const Type *Ty = getType(SlotNo);
1774 error("Global has no type! SlotNo=" + utostr(SlotNo));
1777 if (!isa<PointerType>(Ty)) {
1778 error("Global not a pointer type! Ty= " + Ty->getDescription());
1781 const Type *ElTy = cast<PointerType>(Ty)->getElementType();
1783 // Create the global variable...
1784 GlobalVariable *GV = new GlobalVariable(ElTy, isConstant, Linkage,
1786 insertValue(GV, SlotNo, ModuleValues);
1788 unsigned initSlot = 0;
1789 if (hasInitializer) {
1790 initSlot = read_vbr_uint();
1791 GlobalInits.push_back(std::make_pair(GV, initSlot));
1794 // Notify handler about the global value.
1795 if (Handler) Handler->handleGlobalVariable(ElTy, isConstant, Linkage, SlotNo, initSlot);
1798 VarType = read_vbr_uint();
1801 // Read the function objects for all of the functions that are coming
1802 unsigned FnSignature = 0;
1803 if (read_typeid(FnSignature))
1804 error("Invalid function type (type type) found");
1806 while (FnSignature != Type::VoidTyID) { // List is terminated by Void
1807 const Type *Ty = getType(FnSignature);
1808 if (!isa<PointerType>(Ty) ||
1809 !isa<FunctionType>(cast<PointerType>(Ty)->getElementType())) {
1810 error("Function not a pointer to function type! Ty = " +
1811 Ty->getDescription());
1812 // FIXME: what should Ty be if handler continues?
1815 // We create functions by passing the underlying FunctionType to create...
1816 const FunctionType* FTy =
1817 cast<FunctionType>(cast<PointerType>(Ty)->getElementType());
1819 // Insert the place hodler
1820 Function* Func = new Function(FTy, GlobalValue::InternalLinkage,
1822 insertValue(Func, FnSignature, ModuleValues);
1824 // Save this for later so we know type of lazily instantiated functions
1825 FunctionSignatureList.push_back(Func);
1827 if (Handler) Handler->handleFunctionDeclaration(Func);
1829 // Get Next function signature
1830 if (read_typeid(FnSignature))
1831 error("Invalid function type (type type) found");
1834 // Now that the function signature list is set up, reverse it so that we can
1835 // remove elements efficiently from the back of the vector.
1836 std::reverse(FunctionSignatureList.begin(), FunctionSignatureList.end());
1838 // If this bytecode format has dependent library information in it ..
1839 if (!hasNoDependentLibraries) {
1840 // Read in the number of dependent library items that follow
1841 unsigned num_dep_libs = read_vbr_uint();
1842 std::string dep_lib;
1843 while( num_dep_libs-- ) {
1844 dep_lib = read_str();
1845 TheModule->addLibrary(dep_lib);
1848 // Read target triple and place into the module
1849 std::string triple = read_str();
1850 TheModule->setTargetTriple(triple);
1853 if (hasInconsistentModuleGlobalInfo)
1856 // This is for future proofing... in the future extra fields may be added that
1857 // we don't understand, so we transparently ignore them.
1861 if (Handler) Handler->handleModuleGlobalsEnd();
1864 /// Parse the version information and decode it by setting flags on the
1865 /// Reader that enable backward compatibility of the reader.
1866 void BytecodeReader::ParseVersionInfo() {
1867 unsigned Version = read_vbr_uint();
1869 // Unpack version number: low four bits are for flags, top bits = version
1870 Module::Endianness Endianness;
1871 Module::PointerSize PointerSize;
1872 Endianness = (Version & 1) ? Module::BigEndian : Module::LittleEndian;
1873 PointerSize = (Version & 2) ? Module::Pointer64 : Module::Pointer32;
1875 bool hasNoEndianness = Version & 4;
1876 bool hasNoPointerSize = Version & 8;
1878 RevisionNum = Version >> 4;
1880 // Default values for the current bytecode version
1881 hasInconsistentModuleGlobalInfo = false;
1882 hasExplicitPrimitiveZeros = false;
1883 hasRestrictedGEPTypes = false;
1884 hasTypeDerivedFromValue = false;
1885 hasLongBlockHeaders = false;
1886 has32BitTypes = false;
1887 hasNoDependentLibraries = false;
1889 switch (RevisionNum) {
1890 case 0: // LLVM 1.0, 1.1 release version
1891 // Base LLVM 1.0 bytecode format.
1892 hasInconsistentModuleGlobalInfo = true;
1893 hasExplicitPrimitiveZeros = true;
1897 case 1: // LLVM 1.2 release version
1898 // LLVM 1.2 added explicit support for emitting strings efficiently.
1900 // Also, it fixed the problem where the size of the ModuleGlobalInfo block
1901 // included the size for the alignment at the end, where the rest of the
1904 // LLVM 1.2 and before required that GEP indices be ubyte constants for
1905 // structures and longs for sequential types.
1906 hasRestrictedGEPTypes = true;
1908 // LLVM 1.2 and before had the Type class derive from Value class. This
1909 // changed in release 1.3 and consequently LLVM 1.3 bytecode files are
1910 // written differently because Types can no longer be part of the
1911 // type planes for Values.
1912 hasTypeDerivedFromValue = true;
1916 case 2: /// 1.2.5 (mid-release) version
1918 /// LLVM 1.2 and earlier had two-word block headers. This is a bit wasteful,
1919 /// especially for small files where the 8 bytes per block is a large fraction
1920 /// of the total block size. In LLVM 1.3, the block type and length are
1921 /// compressed into a single 32-bit unsigned integer. 27 bits for length, 5
1922 /// bits for block type.
1923 hasLongBlockHeaders = true;
1925 /// LLVM 1.2 and earlier wrote type slot numbers as vbr_uint32. In LLVM 1.3
1926 /// this has been reduced to vbr_uint24. It shouldn't make much difference
1927 /// since we haven't run into a module with > 24 million types, but for safety
1928 /// the 24-bit restriction has been enforced in 1.3 to free some bits in
1929 /// various places and to ensure consistency.
1930 has32BitTypes = true;
1932 /// LLVM 1.2 and earlier did not provide a target triple nor a list of
1933 /// libraries on which the bytecode is dependent. LLVM 1.3 provides these
1934 /// features, for use in future versions of LLVM.
1935 hasNoDependentLibraries = true;
1938 case 3: // LLVM 1.3 release version
1942 error("Unknown bytecode version number: " + itostr(RevisionNum));
1945 if (hasNoEndianness) Endianness = Module::AnyEndianness;
1946 if (hasNoPointerSize) PointerSize = Module::AnyPointerSize;
1948 TheModule->setEndianness(Endianness);
1949 TheModule->setPointerSize(PointerSize);
1951 if (Handler) Handler->handleVersionInfo(RevisionNum, Endianness, PointerSize);
1954 /// Parse a whole module.
1955 void BytecodeReader::ParseModule() {
1956 unsigned Type, Size;
1958 FunctionSignatureList.clear(); // Just in case...
1960 // Read into instance variables...
1964 bool SeenModuleGlobalInfo = false;
1965 bool SeenGlobalTypePlane = false;
1966 BufPtr MyEnd = BlockEnd;
1967 while (At < MyEnd) {
1969 read_block(Type, Size);
1973 case BytecodeFormat::GlobalTypePlaneBlockID:
1974 if (SeenGlobalTypePlane)
1975 error("Two GlobalTypePlane Blocks Encountered!");
1978 SeenGlobalTypePlane = true;
1981 case BytecodeFormat::ModuleGlobalInfoBlockID:
1982 if (SeenModuleGlobalInfo)
1983 error("Two ModuleGlobalInfo Blocks Encountered!");
1984 ParseModuleGlobalInfo();
1985 SeenModuleGlobalInfo = true;
1988 case BytecodeFormat::ConstantPoolBlockID:
1989 ParseConstantPool(ModuleValues, ModuleTypes,false);
1992 case BytecodeFormat::FunctionBlockID:
1993 ParseFunctionLazily();
1996 case BytecodeFormat::SymbolTableBlockID:
1997 ParseSymbolTable(0, &TheModule->getSymbolTable());
2003 error("Unexpected Block of Type #" + utostr(Type) + " encountered!");
2011 // After the module constant pool has been read, we can safely initialize
2012 // global variables...
2013 while (!GlobalInits.empty()) {
2014 GlobalVariable *GV = GlobalInits.back().first;
2015 unsigned Slot = GlobalInits.back().second;
2016 GlobalInits.pop_back();
2018 // Look up the initializer value...
2019 // FIXME: Preserve this type ID!
2021 const llvm::PointerType* GVType = GV->getType();
2022 unsigned TypeSlot = getTypeSlot(GVType->getElementType());
2023 if (Constant *CV = getConstantValue(TypeSlot, Slot)) {
2024 if (GV->hasInitializer())
2025 error("Global *already* has an initializer?!");
2026 if (Handler) Handler->handleGlobalInitializer(GV,CV);
2027 GV->setInitializer(CV);
2029 error("Cannot find initializer value.");
2032 /// Make sure we pulled them all out. If we didn't then there's a declaration
2033 /// but a missing body. That's not allowed.
2034 if (!FunctionSignatureList.empty())
2035 error("Function declared, but bytecode stream ended before definition");
2038 /// This function completely parses a bytecode buffer given by the \p Buf
2039 /// and \p Length parameters.
2040 void BytecodeReader::ParseBytecode(BufPtr Buf, unsigned Length,
2041 const std::string &ModuleID,
2042 bool processFunctions) {
2045 At = MemStart = BlockStart = Buf;
2046 MemEnd = BlockEnd = Buf + Length;
2048 // Create the module
2049 TheModule = new Module(ModuleID);
2051 if (Handler) Handler->handleStart(TheModule, Length);
2053 // Read and check signature...
2054 unsigned Sig = read_uint();
2055 if (Sig != ('l' | ('l' << 8) | ('v' << 16) | ('m' << 24))) {
2056 error("Invalid bytecode signature: " + utostr(Sig));
2059 // Tell the handler we're starting a module
2060 if (Handler) Handler->handleModuleBegin(ModuleID);
2062 // Get the module block and size and verify. This is handled specially
2063 // because the module block/size is always written in long format. Other
2064 // blocks are written in short format so the read_block method is used.
2065 unsigned Type, Size;
2068 if (Type != BytecodeFormat::ModuleBlockID) {
2069 error("Expected Module Block! Type:" + utostr(Type) + ", Size:"
2072 if (At + Size != MemEnd) {
2073 error("Invalid Top Level Block Length! Type:" + utostr(Type)
2074 + ", Size:" + utostr(Size));
2077 // Parse the module contents
2078 this->ParseModule();
2080 // Check for missing functions
2082 error("Function expected, but bytecode stream ended!");
2084 // Process all the function bodies now, if requested
2085 if (processFunctions)
2086 ParseAllFunctionBodies();
2088 // Tell the handler we're done with the module
2090 Handler->handleModuleEnd(ModuleID);
2092 // Tell the handler we're finished the parse
2093 if (Handler) Handler->handleFinish();
2095 } catch (std::string& errstr) {
2096 if (Handler) Handler->handleError(errstr);
2102 std::string msg("Unknown Exception Occurred");
2103 if (Handler) Handler->handleError(msg);
2111 //===----------------------------------------------------------------------===//
2112 //=== Default Implementations of Handler Methods
2113 //===----------------------------------------------------------------------===//
2115 BytecodeHandler::~BytecodeHandler() {}