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 "llvm/Support/Compressor.h"
28 #include "llvm/ADT/StringExtras.h"
35 /// @brief A class for maintaining the slot number definition
36 /// as a placeholder for the actual definition for forward constants defs.
37 class ConstantPlaceHolder : public ConstantExpr {
38 ConstantPlaceHolder(); // DO NOT IMPLEMENT
39 void operator=(const ConstantPlaceHolder &); // DO NOT IMPLEMENT
41 ConstantPlaceHolder(const Type *Ty)
42 : ConstantExpr(Instruction::UserOp1, Constant::getNullValue(Ty), Ty) {}
47 // Provide some details on error
48 inline void BytecodeReader::error(std::string err) {
50 err += itostr(RevisionNum) ;
52 err += itostr(At-MemStart);
57 //===----------------------------------------------------------------------===//
58 // Bytecode Reading Methods
59 //===----------------------------------------------------------------------===//
61 /// Determine if the current block being read contains any more data.
62 inline bool BytecodeReader::moreInBlock() {
66 /// Throw an error if we've read past the end of the current block
67 inline void BytecodeReader::checkPastBlockEnd(const char * block_name) {
69 error(std::string("Attempt to read past the end of ") + block_name +
73 /// Align the buffer position to a 32 bit boundary
74 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!");
85 /// Read a whole unsigned integer
86 inline unsigned BytecodeReader::read_uint() {
88 error("Ran out of data reading uint!");
90 return At[-4] | (At[-3] << 8) | (At[-2] << 16) | (At[-1] << 24);
93 /// Read a variable-bit-rate encoded unsigned integer
94 inline unsigned BytecodeReader::read_vbr_uint() {
101 error("Ran out of data reading vbr_uint!");
102 Result |= (unsigned)((*At++) & 0x7F) << Shift;
104 } while (At[-1] & 0x80);
105 if (Handler) Handler->handleVBR32(At-Save);
109 /// Read a variable-bit-rate encoded unsigned 64-bit integer.
110 inline uint64_t BytecodeReader::read_vbr_uint64() {
117 error("Ran out of data reading vbr_uint64!");
118 Result |= (uint64_t)((*At++) & 0x7F) << Shift;
120 } while (At[-1] & 0x80);
121 if (Handler) Handler->handleVBR64(At-Save);
125 /// Read a variable-bit-rate encoded signed 64-bit integer.
126 inline int64_t BytecodeReader::read_vbr_int64() {
127 uint64_t R = read_vbr_uint64();
130 return -(int64_t)(R >> 1);
131 else // There is no such thing as -0 with integers. "-0" really means
132 // 0x8000000000000000.
135 return (int64_t)(R >> 1);
138 /// Read a pascal-style string (length followed by text)
139 inline std::string BytecodeReader::read_str() {
140 unsigned Size = read_vbr_uint();
141 const unsigned char *OldAt = At;
143 if (At > BlockEnd) // Size invalid?
144 error("Ran out of data reading a string!");
145 return std::string((char*)OldAt, Size);
148 /// Read an arbitrary block of data
149 inline void BytecodeReader::read_data(void *Ptr, void *End) {
150 unsigned char *Start = (unsigned char *)Ptr;
151 unsigned Amount = (unsigned char *)End - Start;
152 if (At+Amount > BlockEnd)
153 error("Ran out of data!");
154 std::copy(At, At+Amount, Start);
158 /// Read a float value in little-endian order
159 inline void BytecodeReader::read_float(float& FloatVal) {
160 /// FIXME: This isn't optimal, it has size problems on some platforms
161 /// where FP is not IEEE.
166 FloatUnion.i = At[0] | (At[1] << 8) | (At[2] << 16) | (At[3] << 24);
167 At+=sizeof(uint32_t);
168 FloatVal = FloatUnion.f;
171 /// Read a double value in little-endian order
172 inline void BytecodeReader::read_double(double& DoubleVal) {
173 /// FIXME: This isn't optimal, it has size problems on some platforms
174 /// where FP is not IEEE.
179 DoubleUnion.i = (uint64_t(At[0]) << 0) | (uint64_t(At[1]) << 8) |
180 (uint64_t(At[2]) << 16) | (uint64_t(At[3]) << 24) |
181 (uint64_t(At[4]) << 32) | (uint64_t(At[5]) << 40) |
182 (uint64_t(At[6]) << 48) | (uint64_t(At[7]) << 56);
183 At+=sizeof(uint64_t);
184 DoubleVal = DoubleUnion.d;
187 /// Read a block header and obtain its type and size
188 inline void BytecodeReader::read_block(unsigned &Type, unsigned &Size) {
189 if ( hasLongBlockHeaders ) {
193 case BytecodeFormat::Reserved_DoNotUse :
194 error("Reserved_DoNotUse used as Module Type?");
195 Type = BytecodeFormat::ModuleBlockID; break;
196 case BytecodeFormat::Module:
197 Type = BytecodeFormat::ModuleBlockID; break;
198 case BytecodeFormat::Function:
199 Type = BytecodeFormat::FunctionBlockID; break;
200 case BytecodeFormat::ConstantPool:
201 Type = BytecodeFormat::ConstantPoolBlockID; break;
202 case BytecodeFormat::SymbolTable:
203 Type = BytecodeFormat::SymbolTableBlockID; break;
204 case BytecodeFormat::ModuleGlobalInfo:
205 Type = BytecodeFormat::ModuleGlobalInfoBlockID; break;
206 case BytecodeFormat::GlobalTypePlane:
207 Type = BytecodeFormat::GlobalTypePlaneBlockID; break;
208 case BytecodeFormat::InstructionList:
209 Type = BytecodeFormat::InstructionListBlockID; break;
210 case BytecodeFormat::CompactionTable:
211 Type = BytecodeFormat::CompactionTableBlockID; break;
212 case BytecodeFormat::BasicBlock:
213 /// This block type isn't used after version 1.1. However, we have to
214 /// still allow the value in case this is an old bc format file.
215 /// We just let its value creep thru.
218 error("Invalid block id found: " + utostr(Type));
223 Type = Size & 0x1F; // mask low order five bits
224 Size >>= 5; // get rid of five low order bits, leaving high 27
227 if (At + Size > BlockEnd)
228 error("Attempt to size a block past end of memory");
229 BlockEnd = At + Size;
230 if (Handler) Handler->handleBlock(Type, BlockStart, Size);
234 /// In LLVM 1.2 and before, Types were derived from Value and so they were
235 /// written as part of the type planes along with any other Value. In LLVM
236 /// 1.3 this changed so that Type does not derive from Value. Consequently,
237 /// the BytecodeReader's containers for Values can't contain Types because
238 /// there's no inheritance relationship. This means that the "Type Type"
239 /// plane is defunct along with the Type::TypeTyID TypeID. In LLVM 1.3
240 /// whenever a bytecode construct must have both types and values together,
241 /// the types are always read/written first and then the Values. Furthermore
242 /// since Type::TypeTyID no longer exists, its value (12) now corresponds to
243 /// Type::LabelTyID. In order to overcome this we must "sanitize" all the
244 /// type TypeIDs we encounter. For LLVM 1.3 bytecode files, there's no change.
245 /// For LLVM 1.2 and before, this function will decrement the type id by
246 /// one to account for the missing Type::TypeTyID enumerator if the value is
247 /// larger than 12 (Type::LabelTyID). If the value is exactly 12, then this
248 /// function returns true, otherwise false. This helps detect situations
249 /// where the pre 1.3 bytecode is indicating that what follows is a type.
250 /// @returns true iff type id corresponds to pre 1.3 "type type"
251 inline bool BytecodeReader::sanitizeTypeId(unsigned &TypeId) {
252 if (hasTypeDerivedFromValue) { /// do nothing if 1.3 or later
253 if (TypeId == Type::LabelTyID) {
254 TypeId = Type::VoidTyID; // sanitize it
255 return true; // indicate we got TypeTyID in pre 1.3 bytecode
256 } else if (TypeId > Type::LabelTyID)
257 --TypeId; // shift all planes down because type type plane is missing
262 /// Reads a vbr uint to read in a type id and does the necessary
263 /// conversion on it by calling sanitizeTypeId.
264 /// @returns true iff \p TypeId read corresponds to a pre 1.3 "type type"
265 /// @see sanitizeTypeId
266 inline bool BytecodeReader::read_typeid(unsigned &TypeId) {
267 TypeId = read_vbr_uint();
268 if ( !has32BitTypes )
269 if ( TypeId == 0x00FFFFFF )
270 TypeId = read_vbr_uint();
271 return sanitizeTypeId(TypeId);
274 //===----------------------------------------------------------------------===//
276 //===----------------------------------------------------------------------===//
278 /// Determine if a type id has an implicit null value
279 inline bool BytecodeReader::hasImplicitNull(unsigned TyID) {
280 if (!hasExplicitPrimitiveZeros)
281 return TyID != Type::LabelTyID && TyID != Type::VoidTyID;
282 return TyID >= Type::FirstDerivedTyID;
285 /// Obtain a type given a typeid and account for things like compaction tables,
286 /// function level vs module level, and the offsetting for the primitive types.
287 const Type *BytecodeReader::getType(unsigned ID) {
288 if (ID < Type::FirstDerivedTyID)
289 if (const Type *T = Type::getPrimitiveType((Type::TypeID)ID))
290 return T; // Asked for a primitive type...
292 // Otherwise, derived types need offset...
293 ID -= Type::FirstDerivedTyID;
295 if (!CompactionTypes.empty()) {
296 if (ID >= CompactionTypes.size())
297 error("Type ID out of range for compaction table!");
298 return CompactionTypes[ID].first;
301 // Is it a module-level type?
302 if (ID < ModuleTypes.size())
303 return ModuleTypes[ID].get();
305 // Nope, is it a function-level type?
306 ID -= ModuleTypes.size();
307 if (ID < FunctionTypes.size())
308 return FunctionTypes[ID].get();
310 error("Illegal type reference!");
314 /// Get a sanitized type id. This just makes sure that the \p ID
315 /// is both sanitized and not the "type type" of pre-1.3 bytecode.
316 /// @see sanitizeTypeId
317 inline const Type* BytecodeReader::getSanitizedType(unsigned& ID) {
318 if (sanitizeTypeId(ID))
319 error("Invalid type id encountered");
323 /// This method just saves some coding. It uses read_typeid to read
324 /// in a sanitized type id, errors that its not the type type, and
325 /// then calls getType to return the type value.
326 inline const Type* BytecodeReader::readSanitizedType() {
329 error("Invalid type id encountered");
333 /// Get the slot number associated with a type accounting for primitive
334 /// types, compaction tables, and function level vs module level.
335 unsigned BytecodeReader::getTypeSlot(const Type *Ty) {
336 if (Ty->isPrimitiveType())
337 return Ty->getTypeID();
339 // Scan the compaction table for the type if needed.
340 if (!CompactionTypes.empty()) {
341 for (unsigned i = 0, e = CompactionTypes.size(); i != e; ++i)
342 if (CompactionTypes[i].first == Ty)
343 return Type::FirstDerivedTyID + i;
345 error("Couldn't find type specified in compaction table!");
348 // Check the function level types first...
349 TypeListTy::iterator I = std::find(FunctionTypes.begin(),
350 FunctionTypes.end(), Ty);
352 if (I != FunctionTypes.end())
353 return Type::FirstDerivedTyID + ModuleTypes.size() +
354 (&*I - &FunctionTypes[0]);
356 // Check the module level types now...
357 I = std::find(ModuleTypes.begin(), ModuleTypes.end(), Ty);
358 if (I == ModuleTypes.end())
359 error("Didn't find type in ModuleTypes.");
360 return Type::FirstDerivedTyID + (&*I - &ModuleTypes[0]);
363 /// This is just like getType, but when a compaction table is in use, it is
364 /// ignored. It also ignores function level types.
366 const Type *BytecodeReader::getGlobalTableType(unsigned Slot) {
367 if (Slot < Type::FirstDerivedTyID) {
368 const Type *Ty = Type::getPrimitiveType((Type::TypeID)Slot);
370 error("Not a primitive type ID?");
373 Slot -= Type::FirstDerivedTyID;
374 if (Slot >= ModuleTypes.size())
375 error("Illegal compaction table type reference!");
376 return ModuleTypes[Slot];
379 /// This is just like getTypeSlot, but when a compaction table is in use, it
380 /// is ignored. It also ignores function level types.
381 unsigned BytecodeReader::getGlobalTableTypeSlot(const Type *Ty) {
382 if (Ty->isPrimitiveType())
383 return Ty->getTypeID();
384 TypeListTy::iterator I = std::find(ModuleTypes.begin(),
385 ModuleTypes.end(), Ty);
386 if (I == ModuleTypes.end())
387 error("Didn't find type in ModuleTypes.");
388 return Type::FirstDerivedTyID + (&*I - &ModuleTypes[0]);
391 /// Retrieve a value of a given type and slot number, possibly creating
392 /// it if it doesn't already exist.
393 Value * BytecodeReader::getValue(unsigned type, unsigned oNum, bool Create) {
394 assert(type != Type::LabelTyID && "getValue() cannot get blocks!");
397 // If there is a compaction table active, it defines the low-level numbers.
398 // If not, the module values define the low-level numbers.
399 if (CompactionValues.size() > type && !CompactionValues[type].empty()) {
400 if (Num < CompactionValues[type].size())
401 return CompactionValues[type][Num];
402 Num -= CompactionValues[type].size();
404 // By default, the global type id is the type id passed in
405 unsigned GlobalTyID = type;
407 // If the type plane was compactified, figure out the global type ID by
408 // adding the derived type ids and the distance.
409 if (!CompactionTypes.empty() && type >= Type::FirstDerivedTyID)
410 GlobalTyID = CompactionTypes[type-Type::FirstDerivedTyID].second;
412 if (hasImplicitNull(GlobalTyID)) {
414 return Constant::getNullValue(getType(type));
418 if (GlobalTyID < ModuleValues.size() && ModuleValues[GlobalTyID]) {
419 if (Num < ModuleValues[GlobalTyID]->size())
420 return ModuleValues[GlobalTyID]->getOperand(Num);
421 Num -= ModuleValues[GlobalTyID]->size();
425 if (FunctionValues.size() > type &&
426 FunctionValues[type] &&
427 Num < FunctionValues[type]->size())
428 return FunctionValues[type]->getOperand(Num);
430 if (!Create) return 0; // Do not create a placeholder?
432 // Did we already create a place holder?
433 std::pair<unsigned,unsigned> KeyValue(type, oNum);
434 ForwardReferenceMap::iterator I = ForwardReferences.lower_bound(KeyValue);
435 if (I != ForwardReferences.end() && I->first == KeyValue)
436 return I->second; // We have already created this placeholder
438 // If the type exists (it should)
439 if (const Type* Ty = getType(type)) {
440 // Create the place holder
441 Value *Val = new Argument(Ty);
442 ForwardReferences.insert(I, std::make_pair(KeyValue, Val));
445 throw "Can't create placeholder for value of type slot #" + utostr(type);
448 /// This is just like getValue, but when a compaction table is in use, it
449 /// is ignored. Also, no forward references or other fancy features are
451 Value* BytecodeReader::getGlobalTableValue(unsigned TyID, unsigned SlotNo) {
453 return Constant::getNullValue(getType(TyID));
455 if (!CompactionTypes.empty() && TyID >= Type::FirstDerivedTyID) {
456 TyID -= Type::FirstDerivedTyID;
457 if (TyID >= CompactionTypes.size())
458 error("Type ID out of range for compaction table!");
459 TyID = CompactionTypes[TyID].second;
464 if (TyID >= ModuleValues.size() || ModuleValues[TyID] == 0 ||
465 SlotNo >= ModuleValues[TyID]->size()) {
466 if (TyID >= ModuleValues.size() || ModuleValues[TyID] == 0)
467 error("Corrupt compaction table entry!"
468 + utostr(TyID) + ", " + utostr(SlotNo) + ": "
469 + utostr(ModuleValues.size()));
471 error("Corrupt compaction table entry!"
472 + utostr(TyID) + ", " + utostr(SlotNo) + ": "
473 + utostr(ModuleValues.size()) + ", "
474 + utohexstr(reinterpret_cast<uint64_t>(((void*)ModuleValues[TyID])))
476 + utostr(ModuleValues[TyID]->size()));
478 return ModuleValues[TyID]->getOperand(SlotNo);
481 /// Just like getValue, except that it returns a null pointer
482 /// only on error. It always returns a constant (meaning that if the value is
483 /// defined, but is not a constant, that is an error). If the specified
484 /// constant hasn't been parsed yet, a placeholder is defined and used.
485 /// Later, after the real value is parsed, the placeholder is eliminated.
486 Constant* BytecodeReader::getConstantValue(unsigned TypeSlot, unsigned Slot) {
487 if (Value *V = getValue(TypeSlot, Slot, false))
488 if (Constant *C = dyn_cast<Constant>(V))
489 return C; // If we already have the value parsed, just return it
491 error("Value for slot " + utostr(Slot) +
492 " is expected to be a constant!");
494 std::pair<unsigned, unsigned> Key(TypeSlot, Slot);
495 ConstantRefsType::iterator I = ConstantFwdRefs.lower_bound(Key);
497 if (I != ConstantFwdRefs.end() && I->first == Key) {
500 // Create a placeholder for the constant reference and
501 // keep track of the fact that we have a forward ref to recycle it
502 Constant *C = new ConstantPlaceHolder(getType(TypeSlot));
504 // Keep track of the fact that we have a forward ref to recycle it
505 ConstantFwdRefs.insert(I, std::make_pair(Key, C));
510 //===----------------------------------------------------------------------===//
511 // IR Construction Methods
512 //===----------------------------------------------------------------------===//
514 /// As values are created, they are inserted into the appropriate place
515 /// with this method. The ValueTable argument must be one of ModuleValues
516 /// or FunctionValues data members of this class.
517 unsigned BytecodeReader::insertValue(Value *Val, unsigned type,
518 ValueTable &ValueTab) {
519 assert((!isa<Constant>(Val) || !cast<Constant>(Val)->isNullValue()) ||
520 !hasImplicitNull(type) &&
521 "Cannot read null values from bytecode!");
523 if (ValueTab.size() <= type)
524 ValueTab.resize(type+1);
526 if (!ValueTab[type]) ValueTab[type] = new ValueList();
528 ValueTab[type]->push_back(Val);
530 bool HasOffset = hasImplicitNull(type);
531 return ValueTab[type]->size()-1 + HasOffset;
534 /// Insert the arguments of a function as new values in the reader.
535 void BytecodeReader::insertArguments(Function* F) {
536 const FunctionType *FT = F->getFunctionType();
537 Function::aiterator AI = F->abegin();
538 for (FunctionType::param_iterator It = FT->param_begin();
539 It != FT->param_end(); ++It, ++AI)
540 insertValue(AI, getTypeSlot(AI->getType()), FunctionValues);
543 //===----------------------------------------------------------------------===//
544 // Bytecode Parsing Methods
545 //===----------------------------------------------------------------------===//
547 /// This method parses a single instruction. The instruction is
548 /// inserted at the end of the \p BB provided. The arguments of
549 /// the instruction are provided in the \p Oprnds vector.
550 void BytecodeReader::ParseInstruction(std::vector<unsigned> &Oprnds,
554 // Clear instruction data
558 unsigned Op = read_uint();
560 // bits Instruction format: Common to all formats
561 // --------------------------
562 // 01-00: Opcode type, fixed to 1.
564 Opcode = (Op >> 2) & 63;
565 Oprnds.resize((Op >> 0) & 03);
567 // Extract the operands
568 switch (Oprnds.size()) {
570 // bits Instruction format:
571 // --------------------------
572 // 19-08: Resulting type plane
573 // 31-20: Operand #1 (if set to (2^12-1), then zero operands)
575 iType = (Op >> 8) & 4095;
576 Oprnds[0] = (Op >> 20) & 4095;
577 if (Oprnds[0] == 4095) // Handle special encoding for 0 operands...
581 // bits Instruction format:
582 // --------------------------
583 // 15-08: Resulting type plane
587 iType = (Op >> 8) & 255;
588 Oprnds[0] = (Op >> 16) & 255;
589 Oprnds[1] = (Op >> 24) & 255;
592 // bits Instruction format:
593 // --------------------------
594 // 13-08: Resulting type plane
599 iType = (Op >> 8) & 63;
600 Oprnds[0] = (Op >> 14) & 63;
601 Oprnds[1] = (Op >> 20) & 63;
602 Oprnds[2] = (Op >> 26) & 63;
605 At -= 4; // Hrm, try this again...
606 Opcode = read_vbr_uint();
608 iType = read_vbr_uint();
610 unsigned NumOprnds = read_vbr_uint();
611 Oprnds.resize(NumOprnds);
614 error("Zero-argument instruction found; this is invalid.");
616 for (unsigned i = 0; i != NumOprnds; ++i)
617 Oprnds[i] = read_vbr_uint();
622 const Type *InstTy = getSanitizedType(iType);
624 // We have enough info to inform the handler now.
625 if (Handler) Handler->handleInstruction(Opcode, InstTy, Oprnds, At-SaveAt);
627 // Declare the resulting instruction we'll build.
628 Instruction *Result = 0;
630 // If this is a bytecode format that did not include the unreachable
631 // instruction, bump up all opcodes numbers to make space.
632 if (hasNoUnreachableInst) {
633 if (Opcode >= Instruction::Unreachable &&
639 // Handle binary operators
640 if (Opcode >= Instruction::BinaryOpsBegin &&
641 Opcode < Instruction::BinaryOpsEnd && Oprnds.size() == 2)
642 Result = BinaryOperator::create((Instruction::BinaryOps)Opcode,
643 getValue(iType, Oprnds[0]),
644 getValue(iType, Oprnds[1]));
649 error("Illegal instruction read!");
651 case Instruction::VAArg:
652 Result = new VAArgInst(getValue(iType, Oprnds[0]),
653 getSanitizedType(Oprnds[1]));
655 case Instruction::VANext:
656 Result = new VANextInst(getValue(iType, Oprnds[0]),
657 getSanitizedType(Oprnds[1]));
659 case Instruction::Cast:
660 Result = new CastInst(getValue(iType, Oprnds[0]),
661 getSanitizedType(Oprnds[1]));
663 case Instruction::Select:
664 Result = new SelectInst(getValue(Type::BoolTyID, Oprnds[0]),
665 getValue(iType, Oprnds[1]),
666 getValue(iType, Oprnds[2]));
668 case Instruction::PHI: {
669 if (Oprnds.size() == 0 || (Oprnds.size() & 1))
670 error("Invalid phi node encountered!");
672 PHINode *PN = new PHINode(InstTy);
673 PN->op_reserve(Oprnds.size());
674 for (unsigned i = 0, e = Oprnds.size(); i != e; i += 2)
675 PN->addIncoming(getValue(iType, Oprnds[i]), getBasicBlock(Oprnds[i+1]));
680 case Instruction::Shl:
681 case Instruction::Shr:
682 Result = new ShiftInst((Instruction::OtherOps)Opcode,
683 getValue(iType, Oprnds[0]),
684 getValue(Type::UByteTyID, Oprnds[1]));
686 case Instruction::Ret:
687 if (Oprnds.size() == 0)
688 Result = new ReturnInst();
689 else if (Oprnds.size() == 1)
690 Result = new ReturnInst(getValue(iType, Oprnds[0]));
692 error("Unrecognized instruction!");
695 case Instruction::Br:
696 if (Oprnds.size() == 1)
697 Result = new BranchInst(getBasicBlock(Oprnds[0]));
698 else if (Oprnds.size() == 3)
699 Result = new BranchInst(getBasicBlock(Oprnds[0]),
700 getBasicBlock(Oprnds[1]), getValue(Type::BoolTyID , Oprnds[2]));
702 error("Invalid number of operands for a 'br' instruction!");
704 case Instruction::Switch: {
705 if (Oprnds.size() & 1)
706 error("Switch statement with odd number of arguments!");
708 SwitchInst *I = new SwitchInst(getValue(iType, Oprnds[0]),
709 getBasicBlock(Oprnds[1]));
710 for (unsigned i = 2, e = Oprnds.size(); i != e; i += 2)
711 I->addCase(cast<Constant>(getValue(iType, Oprnds[i])),
712 getBasicBlock(Oprnds[i+1]));
717 case Instruction::Call: {
718 if (Oprnds.size() == 0)
719 error("Invalid call instruction encountered!");
721 Value *F = getValue(iType, Oprnds[0]);
723 // Check to make sure we have a pointer to function type
724 const PointerType *PTy = dyn_cast<PointerType>(F->getType());
725 if (PTy == 0) error("Call to non function pointer value!");
726 const FunctionType *FTy = dyn_cast<FunctionType>(PTy->getElementType());
727 if (FTy == 0) error("Call to non function pointer value!");
729 std::vector<Value *> Params;
730 if (!FTy->isVarArg()) {
731 FunctionType::param_iterator It = FTy->param_begin();
733 for (unsigned i = 1, e = Oprnds.size(); i != e; ++i) {
734 if (It == FTy->param_end())
735 error("Invalid call instruction!");
736 Params.push_back(getValue(getTypeSlot(*It++), Oprnds[i]));
738 if (It != FTy->param_end())
739 error("Invalid call instruction!");
741 Oprnds.erase(Oprnds.begin(), Oprnds.begin()+1);
743 unsigned FirstVariableOperand;
744 if (Oprnds.size() < FTy->getNumParams())
745 error("Call instruction missing operands!");
747 // Read all of the fixed arguments
748 for (unsigned i = 0, e = FTy->getNumParams(); i != e; ++i)
749 Params.push_back(getValue(getTypeSlot(FTy->getParamType(i)),Oprnds[i]));
751 FirstVariableOperand = FTy->getNumParams();
753 if ((Oprnds.size()-FirstVariableOperand) & 1)
754 error("Invalid call instruction!"); // Must be pairs of type/value
756 for (unsigned i = FirstVariableOperand, e = Oprnds.size();
758 Params.push_back(getValue(Oprnds[i], Oprnds[i+1]));
761 Result = new CallInst(F, Params);
764 case Instruction::Invoke: {
765 if (Oprnds.size() < 3)
766 error("Invalid invoke instruction!");
767 Value *F = getValue(iType, Oprnds[0]);
769 // Check to make sure we have a pointer to function type
770 const PointerType *PTy = dyn_cast<PointerType>(F->getType());
772 error("Invoke to non function pointer value!");
773 const FunctionType *FTy = dyn_cast<FunctionType>(PTy->getElementType());
775 error("Invoke to non function pointer value!");
777 std::vector<Value *> Params;
778 BasicBlock *Normal, *Except;
780 if (!FTy->isVarArg()) {
781 Normal = getBasicBlock(Oprnds[1]);
782 Except = getBasicBlock(Oprnds[2]);
784 FunctionType::param_iterator It = FTy->param_begin();
785 for (unsigned i = 3, e = Oprnds.size(); i != e; ++i) {
786 if (It == FTy->param_end())
787 error("Invalid invoke instruction!");
788 Params.push_back(getValue(getTypeSlot(*It++), Oprnds[i]));
790 if (It != FTy->param_end())
791 error("Invalid invoke instruction!");
793 Oprnds.erase(Oprnds.begin(), Oprnds.begin()+1);
795 Normal = getBasicBlock(Oprnds[0]);
796 Except = getBasicBlock(Oprnds[1]);
798 unsigned FirstVariableArgument = FTy->getNumParams()+2;
799 for (unsigned i = 2; i != FirstVariableArgument; ++i)
800 Params.push_back(getValue(getTypeSlot(FTy->getParamType(i-2)),
803 if (Oprnds.size()-FirstVariableArgument & 1) // Must be type/value pairs
804 error("Invalid invoke instruction!");
806 for (unsigned i = FirstVariableArgument; i < Oprnds.size(); i += 2)
807 Params.push_back(getValue(Oprnds[i], Oprnds[i+1]));
810 Result = new InvokeInst(F, Normal, Except, Params);
813 case Instruction::Malloc:
814 if (Oprnds.size() > 2)
815 error("Invalid malloc instruction!");
816 if (!isa<PointerType>(InstTy))
817 error("Invalid malloc instruction!");
819 Result = new MallocInst(cast<PointerType>(InstTy)->getElementType(),
820 Oprnds.size() ? getValue(Type::UIntTyID,
824 case Instruction::Alloca:
825 if (Oprnds.size() > 2)
826 error("Invalid alloca instruction!");
827 if (!isa<PointerType>(InstTy))
828 error("Invalid alloca instruction!");
830 Result = new AllocaInst(cast<PointerType>(InstTy)->getElementType(),
831 Oprnds.size() ? getValue(Type::UIntTyID,
834 case Instruction::Free:
835 if (!isa<PointerType>(InstTy))
836 error("Invalid free instruction!");
837 Result = new FreeInst(getValue(iType, Oprnds[0]));
839 case Instruction::GetElementPtr: {
840 if (Oprnds.size() == 0 || !isa<PointerType>(InstTy))
841 error("Invalid getelementptr instruction!");
843 std::vector<Value*> Idx;
845 const Type *NextTy = InstTy;
846 for (unsigned i = 1, e = Oprnds.size(); i != e; ++i) {
847 const CompositeType *TopTy = dyn_cast_or_null<CompositeType>(NextTy);
849 error("Invalid getelementptr instruction!");
851 unsigned ValIdx = Oprnds[i];
853 if (!hasRestrictedGEPTypes) {
854 // Struct indices are always uints, sequential type indices can be any
855 // of the 32 or 64-bit integer types. The actual choice of type is
856 // encoded in the low two bits of the slot number.
857 if (isa<StructType>(TopTy))
858 IdxTy = Type::UIntTyID;
860 switch (ValIdx & 3) {
862 case 0: IdxTy = Type::UIntTyID; break;
863 case 1: IdxTy = Type::IntTyID; break;
864 case 2: IdxTy = Type::ULongTyID; break;
865 case 3: IdxTy = Type::LongTyID; break;
870 IdxTy = isa<StructType>(TopTy) ? Type::UByteTyID : Type::LongTyID;
873 Idx.push_back(getValue(IdxTy, ValIdx));
875 // Convert ubyte struct indices into uint struct indices.
876 if (isa<StructType>(TopTy) && hasRestrictedGEPTypes)
877 if (ConstantUInt *C = dyn_cast<ConstantUInt>(Idx.back()))
878 Idx[Idx.size()-1] = ConstantExpr::getCast(C, Type::UIntTy);
880 NextTy = GetElementPtrInst::getIndexedType(InstTy, Idx, true);
883 Result = new GetElementPtrInst(getValue(iType, Oprnds[0]), Idx);
887 case 62: // volatile load
888 case Instruction::Load:
889 if (Oprnds.size() != 1 || !isa<PointerType>(InstTy))
890 error("Invalid load instruction!");
891 Result = new LoadInst(getValue(iType, Oprnds[0]), "", Opcode == 62);
894 case 63: // volatile store
895 case Instruction::Store: {
896 if (!isa<PointerType>(InstTy) || Oprnds.size() != 2)
897 error("Invalid store instruction!");
899 Value *Ptr = getValue(iType, Oprnds[1]);
900 const Type *ValTy = cast<PointerType>(Ptr->getType())->getElementType();
901 Result = new StoreInst(getValue(getTypeSlot(ValTy), Oprnds[0]), Ptr,
905 case Instruction::Unwind:
906 if (Oprnds.size() != 0) error("Invalid unwind instruction!");
907 Result = new UnwindInst();
909 case Instruction::Unreachable:
910 if (Oprnds.size() != 0) error("Invalid unreachable instruction!");
911 Result = new UnreachableInst();
913 } // end switch(Opcode)
916 if (Result->getType() == InstTy)
919 TypeSlot = getTypeSlot(Result->getType());
921 insertValue(Result, TypeSlot, FunctionValues);
922 BB->getInstList().push_back(Result);
925 /// Get a particular numbered basic block, which might be a forward reference.
926 /// This works together with ParseBasicBlock to handle these forward references
927 /// in a clean manner. This function is used when constructing phi, br, switch,
928 /// and other instructions that reference basic blocks. Blocks are numbered
929 /// sequentially as they appear in the function.
930 BasicBlock *BytecodeReader::getBasicBlock(unsigned ID) {
931 // Make sure there is room in the table...
932 if (ParsedBasicBlocks.size() <= ID) ParsedBasicBlocks.resize(ID+1);
934 // First check to see if this is a backwards reference, i.e., ParseBasicBlock
935 // has already created this block, or if the forward reference has already
937 if (ParsedBasicBlocks[ID])
938 return ParsedBasicBlocks[ID];
940 // Otherwise, the basic block has not yet been created. Do so and add it to
941 // the ParsedBasicBlocks list.
942 return ParsedBasicBlocks[ID] = new BasicBlock();
945 /// In LLVM 1.0 bytecode files, we used to output one basicblock at a time.
946 /// This method reads in one of the basicblock packets. This method is not used
947 /// for bytecode files after LLVM 1.0
948 /// @returns The basic block constructed.
949 BasicBlock *BytecodeReader::ParseBasicBlock(unsigned BlockNo) {
950 if (Handler) Handler->handleBasicBlockBegin(BlockNo);
954 if (ParsedBasicBlocks.size() == BlockNo)
955 ParsedBasicBlocks.push_back(BB = new BasicBlock());
956 else if (ParsedBasicBlocks[BlockNo] == 0)
957 BB = ParsedBasicBlocks[BlockNo] = new BasicBlock();
959 BB = ParsedBasicBlocks[BlockNo];
961 std::vector<unsigned> Operands;
962 while (moreInBlock())
963 ParseInstruction(Operands, BB);
965 if (Handler) Handler->handleBasicBlockEnd(BlockNo);
969 /// Parse all of the BasicBlock's & Instruction's in the body of a function.
970 /// In post 1.0 bytecode files, we no longer emit basic block individually,
971 /// in order to avoid per-basic-block overhead.
972 /// @returns Rhe number of basic blocks encountered.
973 unsigned BytecodeReader::ParseInstructionList(Function* F) {
974 unsigned BlockNo = 0;
975 std::vector<unsigned> Args;
977 while (moreInBlock()) {
978 if (Handler) Handler->handleBasicBlockBegin(BlockNo);
980 if (ParsedBasicBlocks.size() == BlockNo)
981 ParsedBasicBlocks.push_back(BB = new BasicBlock());
982 else if (ParsedBasicBlocks[BlockNo] == 0)
983 BB = ParsedBasicBlocks[BlockNo] = new BasicBlock();
985 BB = ParsedBasicBlocks[BlockNo];
987 F->getBasicBlockList().push_back(BB);
989 // Read instructions into this basic block until we get to a terminator
990 while (moreInBlock() && !BB->getTerminator())
991 ParseInstruction(Args, BB);
993 if (!BB->getTerminator())
994 error("Non-terminated basic block found!");
996 if (Handler) Handler->handleBasicBlockEnd(BlockNo-1);
1002 /// Parse a symbol table. This works for both module level and function
1003 /// level symbol tables. For function level symbol tables, the CurrentFunction
1004 /// parameter must be non-zero and the ST parameter must correspond to
1005 /// CurrentFunction's symbol table. For Module level symbol tables, the
1006 /// CurrentFunction argument must be zero.
1007 void BytecodeReader::ParseSymbolTable(Function *CurrentFunction,
1009 if (Handler) Handler->handleSymbolTableBegin(CurrentFunction,ST);
1011 // Allow efficient basic block lookup by number.
1012 std::vector<BasicBlock*> BBMap;
1013 if (CurrentFunction)
1014 for (Function::iterator I = CurrentFunction->begin(),
1015 E = CurrentFunction->end(); I != E; ++I)
1018 /// In LLVM 1.3 we write types separately from values so
1019 /// The types are always first in the symbol table. This is
1020 /// because Type no longer derives from Value.
1021 if (!hasTypeDerivedFromValue) {
1022 // Symtab block header: [num entries]
1023 unsigned NumEntries = read_vbr_uint();
1024 for (unsigned i = 0; i < NumEntries; ++i) {
1025 // Symtab entry: [def slot #][name]
1026 unsigned slot = read_vbr_uint();
1027 std::string Name = read_str();
1028 const Type* T = getType(slot);
1029 ST->insert(Name, T);
1033 while (moreInBlock()) {
1034 // Symtab block header: [num entries][type id number]
1035 unsigned NumEntries = read_vbr_uint();
1037 bool isTypeType = read_typeid(Typ);
1038 const Type *Ty = getType(Typ);
1040 for (unsigned i = 0; i != NumEntries; ++i) {
1041 // Symtab entry: [def slot #][name]
1042 unsigned slot = read_vbr_uint();
1043 std::string Name = read_str();
1045 // if we're reading a pre 1.3 bytecode file and the type plane
1046 // is the "type type", handle it here
1048 const Type* T = getType(slot);
1050 error("Failed type look-up for name '" + Name + "'");
1051 ST->insert(Name, T);
1052 continue; // code below must be short circuited
1055 if (Typ == Type::LabelTyID) {
1056 if (slot < BBMap.size())
1059 V = getValue(Typ, slot, false); // Find mapping...
1062 error("Failed value look-up for name '" + Name + "'");
1063 V->setName(Name, ST);
1067 checkPastBlockEnd("Symbol Table");
1068 if (Handler) Handler->handleSymbolTableEnd();
1071 /// Read in the types portion of a compaction table.
1072 void BytecodeReader::ParseCompactionTypes(unsigned NumEntries) {
1073 for (unsigned i = 0; i != NumEntries; ++i) {
1074 unsigned TypeSlot = 0;
1075 if (read_typeid(TypeSlot))
1076 error("Invalid type in compaction table: type type");
1077 const Type *Typ = getGlobalTableType(TypeSlot);
1078 CompactionTypes.push_back(std::make_pair(Typ, TypeSlot));
1079 if (Handler) Handler->handleCompactionTableType(i, TypeSlot, Typ);
1083 /// Parse a compaction table.
1084 void BytecodeReader::ParseCompactionTable() {
1086 // Notify handler that we're beginning a compaction table.
1087 if (Handler) Handler->handleCompactionTableBegin();
1089 // In LLVM 1.3 Type no longer derives from Value. So,
1090 // we always write them first in the compaction table
1091 // because they can't occupy a "type plane" where the
1093 if (! hasTypeDerivedFromValue) {
1094 unsigned NumEntries = read_vbr_uint();
1095 ParseCompactionTypes(NumEntries);
1098 // Compaction tables live in separate blocks so we have to loop
1099 // until we've read the whole thing.
1100 while (moreInBlock()) {
1101 // Read the number of Value* entries in the compaction table
1102 unsigned NumEntries = read_vbr_uint();
1104 unsigned isTypeType = false;
1106 // Decode the type from value read in. Most compaction table
1107 // planes will have one or two entries in them. If that's the
1108 // case then the length is encoded in the bottom two bits and
1109 // the higher bits encode the type. This saves another VBR value.
1110 if ((NumEntries & 3) == 3) {
1111 // In this case, both low-order bits are set (value 3). This
1112 // is a signal that the typeid follows.
1114 isTypeType = read_typeid(Ty);
1116 // In this case, the low-order bits specify the number of entries
1117 // and the high order bits specify the type.
1118 Ty = NumEntries >> 2;
1119 isTypeType = sanitizeTypeId(Ty);
1123 // if we're reading a pre 1.3 bytecode file and the type plane
1124 // is the "type type", handle it here
1126 ParseCompactionTypes(NumEntries);
1128 // Make sure we have enough room for the plane.
1129 if (Ty >= CompactionValues.size())
1130 CompactionValues.resize(Ty+1);
1132 // Make sure the plane is empty or we have some kind of error.
1133 if (!CompactionValues[Ty].empty())
1134 error("Compaction table plane contains multiple entries!");
1136 // Notify handler about the plane.
1137 if (Handler) Handler->handleCompactionTablePlane(Ty, NumEntries);
1139 // Push the implicit zero.
1140 CompactionValues[Ty].push_back(Constant::getNullValue(getType(Ty)));
1142 // Read in each of the entries, put them in the compaction table
1143 // and notify the handler that we have a new compaction table value.
1144 for (unsigned i = 0; i != NumEntries; ++i) {
1145 unsigned ValSlot = read_vbr_uint();
1146 Value *V = getGlobalTableValue(Ty, ValSlot);
1147 CompactionValues[Ty].push_back(V);
1148 if (Handler) Handler->handleCompactionTableValue(i, Ty, ValSlot);
1152 // Notify handler that the compaction table is done.
1153 if (Handler) Handler->handleCompactionTableEnd();
1156 // Parse a single type. The typeid is read in first. If its a primitive type
1157 // then nothing else needs to be read, we know how to instantiate it. If its
1158 // a derived type, then additional data is read to fill out the type
1160 const Type *BytecodeReader::ParseType() {
1161 unsigned PrimType = 0;
1162 if (read_typeid(PrimType))
1163 error("Invalid type (type type) in type constants!");
1165 const Type *Result = 0;
1166 if ((Result = Type::getPrimitiveType((Type::TypeID)PrimType)))
1170 case Type::FunctionTyID: {
1171 const Type *RetType = readSanitizedType();
1173 unsigned NumParams = read_vbr_uint();
1175 std::vector<const Type*> Params;
1177 Params.push_back(readSanitizedType());
1179 bool isVarArg = Params.size() && Params.back() == Type::VoidTy;
1180 if (isVarArg) Params.pop_back();
1182 Result = FunctionType::get(RetType, Params, isVarArg);
1185 case Type::ArrayTyID: {
1186 const Type *ElementType = readSanitizedType();
1187 unsigned NumElements = read_vbr_uint();
1188 Result = ArrayType::get(ElementType, NumElements);
1191 case Type::PackedTyID: {
1192 const Type *ElementType = readSanitizedType();
1193 unsigned NumElements = read_vbr_uint();
1194 Result = PackedType::get(ElementType, NumElements);
1197 case Type::StructTyID: {
1198 std::vector<const Type*> Elements;
1200 if (read_typeid(Typ))
1201 error("Invalid element type (type type) for structure!");
1203 while (Typ) { // List is terminated by void/0 typeid
1204 Elements.push_back(getType(Typ));
1205 if (read_typeid(Typ))
1206 error("Invalid element type (type type) for structure!");
1209 Result = StructType::get(Elements);
1212 case Type::PointerTyID: {
1213 Result = PointerType::get(readSanitizedType());
1217 case Type::OpaqueTyID: {
1218 Result = OpaqueType::get();
1223 error("Don't know how to deserialize primitive type " + utostr(PrimType));
1226 if (Handler) Handler->handleType(Result);
1230 // ParseTypes - We have to use this weird code to handle recursive
1231 // types. We know that recursive types will only reference the current slab of
1232 // values in the type plane, but they can forward reference types before they
1233 // have been read. For example, Type #0 might be '{ Ty#1 }' and Type #1 might
1234 // be 'Ty#0*'. When reading Type #0, type number one doesn't exist. To fix
1235 // this ugly problem, we pessimistically insert an opaque type for each type we
1236 // are about to read. This means that forward references will resolve to
1237 // something and when we reread the type later, we can replace the opaque type
1238 // with a new resolved concrete type.
1240 void BytecodeReader::ParseTypes(TypeListTy &Tab, unsigned NumEntries){
1241 assert(Tab.size() == 0 && "should not have read type constants in before!");
1243 // Insert a bunch of opaque types to be resolved later...
1244 Tab.reserve(NumEntries);
1245 for (unsigned i = 0; i != NumEntries; ++i)
1246 Tab.push_back(OpaqueType::get());
1249 Handler->handleTypeList(NumEntries);
1251 // Loop through reading all of the types. Forward types will make use of the
1252 // opaque types just inserted.
1254 for (unsigned i = 0; i != NumEntries; ++i) {
1255 const Type* NewTy = ParseType();
1256 const Type* OldTy = Tab[i].get();
1258 error("Couldn't parse type!");
1260 // Don't directly push the new type on the Tab. Instead we want to replace
1261 // the opaque type we previously inserted with the new concrete value. This
1262 // approach helps with forward references to types. The refinement from the
1263 // abstract (opaque) type to the new type causes all uses of the abstract
1264 // type to use the concrete type (NewTy). This will also cause the opaque
1265 // type to be deleted.
1266 cast<DerivedType>(const_cast<Type*>(OldTy))->refineAbstractTypeTo(NewTy);
1268 // This should have replaced the old opaque type with the new type in the
1269 // value table... or with a preexisting type that was already in the system.
1270 // Let's just make sure it did.
1271 assert(Tab[i] != OldTy && "refineAbstractType didn't work!");
1275 /// Parse a single constant value
1276 Constant *BytecodeReader::ParseConstantValue(unsigned TypeID) {
1277 // We must check for a ConstantExpr before switching by type because
1278 // a ConstantExpr can be of any type, and has no explicit value.
1280 // 0 if not expr; numArgs if is expr
1281 unsigned isExprNumArgs = read_vbr_uint();
1283 if (isExprNumArgs) {
1284 // 'undef' is encoded with 'exprnumargs' == 1.
1285 if (!hasNoUndefValue)
1286 if (--isExprNumArgs == 0)
1287 return UndefValue::get(getType(TypeID));
1289 // FIXME: Encoding of constant exprs could be much more compact!
1290 std::vector<Constant*> ArgVec;
1291 ArgVec.reserve(isExprNumArgs);
1292 unsigned Opcode = read_vbr_uint();
1294 // Bytecode files before LLVM 1.4 need have a missing terminator inst.
1295 if (hasNoUnreachableInst) Opcode++;
1297 // Read the slot number and types of each of the arguments
1298 for (unsigned i = 0; i != isExprNumArgs; ++i) {
1299 unsigned ArgValSlot = read_vbr_uint();
1300 unsigned ArgTypeSlot = 0;
1301 if (read_typeid(ArgTypeSlot))
1302 error("Invalid argument type (type type) for constant value");
1304 // Get the arg value from its slot if it exists, otherwise a placeholder
1305 ArgVec.push_back(getConstantValue(ArgTypeSlot, ArgValSlot));
1308 // Construct a ConstantExpr of the appropriate kind
1309 if (isExprNumArgs == 1) { // All one-operand expressions
1310 if (Opcode != Instruction::Cast)
1311 error("Only cast instruction has one argument for ConstantExpr");
1313 Constant* Result = ConstantExpr::getCast(ArgVec[0], getType(TypeID));
1314 if (Handler) Handler->handleConstantExpression(Opcode, ArgVec, Result);
1316 } else if (Opcode == Instruction::GetElementPtr) { // GetElementPtr
1317 std::vector<Constant*> IdxList(ArgVec.begin()+1, ArgVec.end());
1319 if (hasRestrictedGEPTypes) {
1320 const Type *BaseTy = ArgVec[0]->getType();
1321 generic_gep_type_iterator<std::vector<Constant*>::iterator>
1322 GTI = gep_type_begin(BaseTy, IdxList.begin(), IdxList.end()),
1323 E = gep_type_end(BaseTy, IdxList.begin(), IdxList.end());
1324 for (unsigned i = 0; GTI != E; ++GTI, ++i)
1325 if (isa<StructType>(*GTI)) {
1326 if (IdxList[i]->getType() != Type::UByteTy)
1327 error("Invalid index for getelementptr!");
1328 IdxList[i] = ConstantExpr::getCast(IdxList[i], Type::UIntTy);
1332 Constant* Result = ConstantExpr::getGetElementPtr(ArgVec[0], IdxList);
1333 if (Handler) Handler->handleConstantExpression(Opcode, ArgVec, Result);
1335 } else if (Opcode == Instruction::Select) {
1336 if (ArgVec.size() != 3)
1337 error("Select instruction must have three arguments.");
1338 Constant* Result = ConstantExpr::getSelect(ArgVec[0], ArgVec[1],
1340 if (Handler) Handler->handleConstantExpression(Opcode, ArgVec, Result);
1342 } else { // All other 2-operand expressions
1343 Constant* Result = ConstantExpr::get(Opcode, ArgVec[0], ArgVec[1]);
1344 if (Handler) Handler->handleConstantExpression(Opcode, ArgVec, Result);
1349 // Ok, not an ConstantExpr. We now know how to read the given type...
1350 const Type *Ty = getType(TypeID);
1351 switch (Ty->getTypeID()) {
1352 case Type::BoolTyID: {
1353 unsigned Val = read_vbr_uint();
1354 if (Val != 0 && Val != 1)
1355 error("Invalid boolean value read.");
1356 Constant* Result = ConstantBool::get(Val == 1);
1357 if (Handler) Handler->handleConstantValue(Result);
1361 case Type::UByteTyID: // Unsigned integer types...
1362 case Type::UShortTyID:
1363 case Type::UIntTyID: {
1364 unsigned Val = read_vbr_uint();
1365 if (!ConstantUInt::isValueValidForType(Ty, Val))
1366 error("Invalid unsigned byte/short/int read.");
1367 Constant* Result = ConstantUInt::get(Ty, Val);
1368 if (Handler) Handler->handleConstantValue(Result);
1372 case Type::ULongTyID: {
1373 Constant* Result = ConstantUInt::get(Ty, read_vbr_uint64());
1374 if (Handler) Handler->handleConstantValue(Result);
1378 case Type::SByteTyID: // Signed integer types...
1379 case Type::ShortTyID:
1380 case Type::IntTyID: {
1381 case Type::LongTyID:
1382 int64_t Val = read_vbr_int64();
1383 if (!ConstantSInt::isValueValidForType(Ty, Val))
1384 error("Invalid signed byte/short/int/long read.");
1385 Constant* Result = ConstantSInt::get(Ty, Val);
1386 if (Handler) Handler->handleConstantValue(Result);
1390 case Type::FloatTyID: {
1393 Constant* Result = ConstantFP::get(Ty, Val);
1394 if (Handler) Handler->handleConstantValue(Result);
1398 case Type::DoubleTyID: {
1401 Constant* Result = ConstantFP::get(Ty, Val);
1402 if (Handler) Handler->handleConstantValue(Result);
1406 case Type::ArrayTyID: {
1407 const ArrayType *AT = cast<ArrayType>(Ty);
1408 unsigned NumElements = AT->getNumElements();
1409 unsigned TypeSlot = getTypeSlot(AT->getElementType());
1410 std::vector<Constant*> Elements;
1411 Elements.reserve(NumElements);
1412 while (NumElements--) // Read all of the elements of the constant.
1413 Elements.push_back(getConstantValue(TypeSlot,
1415 Constant* Result = ConstantArray::get(AT, Elements);
1416 if (Handler) Handler->handleConstantArray(AT, Elements, TypeSlot, Result);
1420 case Type::StructTyID: {
1421 const StructType *ST = cast<StructType>(Ty);
1423 std::vector<Constant *> Elements;
1424 Elements.reserve(ST->getNumElements());
1425 for (unsigned i = 0; i != ST->getNumElements(); ++i)
1426 Elements.push_back(getConstantValue(ST->getElementType(i),
1429 Constant* Result = ConstantStruct::get(ST, Elements);
1430 if (Handler) Handler->handleConstantStruct(ST, Elements, Result);
1434 case Type::PackedTyID: {
1435 const PackedType *PT = cast<PackedType>(Ty);
1436 unsigned NumElements = PT->getNumElements();
1437 unsigned TypeSlot = getTypeSlot(PT->getElementType());
1438 std::vector<Constant*> Elements;
1439 Elements.reserve(NumElements);
1440 while (NumElements--) // Read all of the elements of the constant.
1441 Elements.push_back(getConstantValue(TypeSlot,
1443 Constant* Result = ConstantPacked::get(PT, Elements);
1444 if (Handler) Handler->handleConstantPacked(PT, Elements, TypeSlot, Result);
1448 case Type::PointerTyID: { // ConstantPointerRef value (backwards compat).
1449 const PointerType *PT = cast<PointerType>(Ty);
1450 unsigned Slot = read_vbr_uint();
1452 // Check to see if we have already read this global variable...
1453 Value *Val = getValue(TypeID, Slot, false);
1455 GlobalValue *GV = dyn_cast<GlobalValue>(Val);
1456 if (!GV) error("GlobalValue not in ValueTable!");
1457 if (Handler) Handler->handleConstantPointer(PT, Slot, GV);
1460 error("Forward references are not allowed here.");
1465 error("Don't know how to deserialize constant value of type '" +
1466 Ty->getDescription());
1472 /// Resolve references for constants. This function resolves the forward
1473 /// referenced constants in the ConstantFwdRefs map. It uses the
1474 /// replaceAllUsesWith method of Value class to substitute the placeholder
1475 /// instance with the actual instance.
1476 void BytecodeReader::ResolveReferencesToConstant(Constant *NewV, unsigned Typ,
1478 ConstantRefsType::iterator I =
1479 ConstantFwdRefs.find(std::make_pair(Typ, Slot));
1480 if (I == ConstantFwdRefs.end()) return; // Never forward referenced?
1482 Value *PH = I->second; // Get the placeholder...
1483 PH->replaceAllUsesWith(NewV);
1484 delete PH; // Delete the old placeholder
1485 ConstantFwdRefs.erase(I); // Remove the map entry for it
1488 /// Parse the constant strings section.
1489 void BytecodeReader::ParseStringConstants(unsigned NumEntries, ValueTable &Tab){
1490 for (; NumEntries; --NumEntries) {
1492 if (read_typeid(Typ))
1493 error("Invalid type (type type) for string constant");
1494 const Type *Ty = getType(Typ);
1495 if (!isa<ArrayType>(Ty))
1496 error("String constant data invalid!");
1498 const ArrayType *ATy = cast<ArrayType>(Ty);
1499 if (ATy->getElementType() != Type::SByteTy &&
1500 ATy->getElementType() != Type::UByteTy)
1501 error("String constant data invalid!");
1503 // Read character data. The type tells us how long the string is.
1504 char Data[ATy->getNumElements()];
1505 read_data(Data, Data+ATy->getNumElements());
1507 std::vector<Constant*> Elements(ATy->getNumElements());
1508 if (ATy->getElementType() == Type::SByteTy)
1509 for (unsigned i = 0, e = ATy->getNumElements(); i != e; ++i)
1510 Elements[i] = ConstantSInt::get(Type::SByteTy, (signed char)Data[i]);
1512 for (unsigned i = 0, e = ATy->getNumElements(); i != e; ++i)
1513 Elements[i] = ConstantUInt::get(Type::UByteTy, (unsigned char)Data[i]);
1515 // Create the constant, inserting it as needed.
1516 Constant *C = ConstantArray::get(ATy, Elements);
1517 unsigned Slot = insertValue(C, Typ, Tab);
1518 ResolveReferencesToConstant(C, Typ, Slot);
1519 if (Handler) Handler->handleConstantString(cast<ConstantArray>(C));
1523 /// Parse the constant pool.
1524 void BytecodeReader::ParseConstantPool(ValueTable &Tab,
1525 TypeListTy &TypeTab,
1527 if (Handler) Handler->handleGlobalConstantsBegin();
1529 /// In LLVM 1.3 Type does not derive from Value so the types
1530 /// do not occupy a plane. Consequently, we read the types
1531 /// first in the constant pool.
1532 if (isFunction && !hasTypeDerivedFromValue) {
1533 unsigned NumEntries = read_vbr_uint();
1534 ParseTypes(TypeTab, NumEntries);
1537 while (moreInBlock()) {
1538 unsigned NumEntries = read_vbr_uint();
1540 bool isTypeType = read_typeid(Typ);
1542 /// In LLVM 1.2 and before, Types were written to the
1543 /// bytecode file in the "Type Type" plane (#12).
1544 /// In 1.3 plane 12 is now the label plane. Handle this here.
1546 ParseTypes(TypeTab, NumEntries);
1547 } else if (Typ == Type::VoidTyID) {
1548 /// Use of Type::VoidTyID is a misnomer. It actually means
1549 /// that the following plane is constant strings
1550 assert(&Tab == &ModuleValues && "Cannot read strings in functions!");
1551 ParseStringConstants(NumEntries, Tab);
1553 for (unsigned i = 0; i < NumEntries; ++i) {
1554 Constant *C = ParseConstantValue(Typ);
1555 assert(C && "ParseConstantValue returned NULL!");
1556 unsigned Slot = insertValue(C, Typ, Tab);
1558 // If we are reading a function constant table, make sure that we adjust
1559 // the slot number to be the real global constant number.
1561 if (&Tab != &ModuleValues && Typ < ModuleValues.size() &&
1563 Slot += ModuleValues[Typ]->size();
1564 ResolveReferencesToConstant(C, Typ, Slot);
1569 // After we have finished parsing the constant pool, we had better not have
1570 // any dangling references left.
1571 if (!ConstantFwdRefs.empty()) {
1572 ConstantRefsType::const_iterator I = ConstantFwdRefs.begin();
1573 Constant* missingConst = I->second;
1574 error(utostr(ConstantFwdRefs.size()) +
1575 " unresolved constant reference exist. First one is '" +
1576 missingConst->getName() + "' of type '" +
1577 missingConst->getType()->getDescription() + "'.");
1580 checkPastBlockEnd("Constant Pool");
1581 if (Handler) Handler->handleGlobalConstantsEnd();
1584 /// Parse the contents of a function. Note that this function can be
1585 /// called lazily by materializeFunction
1586 /// @see materializeFunction
1587 void BytecodeReader::ParseFunctionBody(Function* F) {
1589 unsigned FuncSize = BlockEnd - At;
1590 GlobalValue::LinkageTypes Linkage = GlobalValue::ExternalLinkage;
1592 unsigned LinkageType = read_vbr_uint();
1593 switch (LinkageType) {
1594 case 0: Linkage = GlobalValue::ExternalLinkage; break;
1595 case 1: Linkage = GlobalValue::WeakLinkage; break;
1596 case 2: Linkage = GlobalValue::AppendingLinkage; break;
1597 case 3: Linkage = GlobalValue::InternalLinkage; break;
1598 case 4: Linkage = GlobalValue::LinkOnceLinkage; break;
1600 error("Invalid linkage type for Function.");
1601 Linkage = GlobalValue::InternalLinkage;
1605 F->setLinkage(Linkage);
1606 if (Handler) Handler->handleFunctionBegin(F,FuncSize);
1608 // Keep track of how many basic blocks we have read in...
1609 unsigned BlockNum = 0;
1610 bool InsertedArguments = false;
1612 BufPtr MyEnd = BlockEnd;
1613 while (At < MyEnd) {
1614 unsigned Type, Size;
1616 read_block(Type, Size);
1619 case BytecodeFormat::ConstantPoolBlockID:
1620 if (!InsertedArguments) {
1621 // Insert arguments into the value table before we parse the first basic
1622 // block in the function, but after we potentially read in the
1623 // compaction table.
1625 InsertedArguments = true;
1628 ParseConstantPool(FunctionValues, FunctionTypes, true);
1631 case BytecodeFormat::CompactionTableBlockID:
1632 ParseCompactionTable();
1635 case BytecodeFormat::BasicBlock: {
1636 if (!InsertedArguments) {
1637 // Insert arguments into the value table before we parse the first basic
1638 // block in the function, but after we potentially read in the
1639 // compaction table.
1641 InsertedArguments = true;
1644 BasicBlock *BB = ParseBasicBlock(BlockNum++);
1645 F->getBasicBlockList().push_back(BB);
1649 case BytecodeFormat::InstructionListBlockID: {
1650 // Insert arguments into the value table before we parse the instruction
1651 // list for the function, but after we potentially read in the compaction
1653 if (!InsertedArguments) {
1655 InsertedArguments = true;
1659 error("Already parsed basic blocks!");
1660 BlockNum = ParseInstructionList(F);
1664 case BytecodeFormat::SymbolTableBlockID:
1665 ParseSymbolTable(F, &F->getSymbolTable());
1671 error("Wrapped around reading bytecode.");
1676 // Malformed bc file if read past end of block.
1680 // Make sure there were no references to non-existant basic blocks.
1681 if (BlockNum != ParsedBasicBlocks.size())
1682 error("Illegal basic block operand reference");
1684 ParsedBasicBlocks.clear();
1686 // Resolve forward references. Replace any uses of a forward reference value
1687 // with the real value.
1688 while (!ForwardReferences.empty()) {
1689 std::map<std::pair<unsigned,unsigned>, Value*>::iterator
1690 I = ForwardReferences.begin();
1691 Value *V = getValue(I->first.first, I->first.second, false);
1692 Value *PlaceHolder = I->second;
1693 PlaceHolder->replaceAllUsesWith(V);
1694 ForwardReferences.erase(I);
1698 // Clear out function-level types...
1699 FunctionTypes.clear();
1700 CompactionTypes.clear();
1701 CompactionValues.clear();
1702 freeTable(FunctionValues);
1704 if (Handler) Handler->handleFunctionEnd(F);
1707 /// This function parses LLVM functions lazily. It obtains the type of the
1708 /// function and records where the body of the function is in the bytecode
1709 /// buffer. The caller can then use the ParseNextFunction and
1710 /// ParseAllFunctionBodies to get handler events for the functions.
1711 void BytecodeReader::ParseFunctionLazily() {
1712 if (FunctionSignatureList.empty())
1713 error("FunctionSignatureList empty!");
1715 Function *Func = FunctionSignatureList.back();
1716 FunctionSignatureList.pop_back();
1718 // Save the information for future reading of the function
1719 LazyFunctionLoadMap[Func] = LazyFunctionInfo(BlockStart, BlockEnd);
1721 // This function has a body but it's not loaded so it appears `External'.
1722 // Mark it as a `Ghost' instead to notify the users that it has a body.
1723 Func->setLinkage(GlobalValue::GhostLinkage);
1725 // Pretend we've `parsed' this function
1729 /// The ParserFunction method lazily parses one function. Use this method to
1730 /// casue the parser to parse a specific function in the module. Note that
1731 /// this will remove the function from what is to be included by
1732 /// ParseAllFunctionBodies.
1733 /// @see ParseAllFunctionBodies
1734 /// @see ParseBytecode
1735 void BytecodeReader::ParseFunction(Function* Func) {
1736 // Find {start, end} pointers and slot in the map. If not there, we're done.
1737 LazyFunctionMap::iterator Fi = LazyFunctionLoadMap.find(Func);
1739 // Make sure we found it
1740 if (Fi == LazyFunctionLoadMap.end()) {
1741 error("Unrecognized function of type " + Func->getType()->getDescription());
1745 BlockStart = At = Fi->second.Buf;
1746 BlockEnd = Fi->second.EndBuf;
1747 assert(Fi->first == Func && "Found wrong function?");
1749 LazyFunctionLoadMap.erase(Fi);
1751 this->ParseFunctionBody(Func);
1754 /// The ParseAllFunctionBodies method parses through all the previously
1755 /// unparsed functions in the bytecode file. If you want to completely parse
1756 /// a bytecode file, this method should be called after Parsebytecode because
1757 /// Parsebytecode only records the locations in the bytecode file of where
1758 /// the function definitions are located. This function uses that information
1759 /// to materialize the functions.
1760 /// @see ParseBytecode
1761 void BytecodeReader::ParseAllFunctionBodies() {
1762 LazyFunctionMap::iterator Fi = LazyFunctionLoadMap.begin();
1763 LazyFunctionMap::iterator Fe = LazyFunctionLoadMap.end();
1766 Function* Func = Fi->first;
1767 BlockStart = At = Fi->second.Buf;
1768 BlockEnd = Fi->second.EndBuf;
1769 this->ParseFunctionBody(Func);
1774 /// Parse the global type list
1775 void BytecodeReader::ParseGlobalTypes() {
1776 // Read the number of types
1777 unsigned NumEntries = read_vbr_uint();
1779 // Ignore the type plane identifier for types if the bc file is pre 1.3
1780 if (hasTypeDerivedFromValue)
1783 ParseTypes(ModuleTypes, NumEntries);
1786 /// Parse the Global info (types, global vars, constants)
1787 void BytecodeReader::ParseModuleGlobalInfo() {
1789 if (Handler) Handler->handleModuleGlobalsBegin();
1791 // Read global variables...
1792 unsigned VarType = read_vbr_uint();
1793 while (VarType != Type::VoidTyID) { // List is terminated by Void
1794 // VarType Fields: bit0 = isConstant, bit1 = hasInitializer, bit2,3,4 =
1795 // Linkage, bit4+ = slot#
1796 unsigned SlotNo = VarType >> 5;
1797 if (sanitizeTypeId(SlotNo))
1798 error("Invalid type (type type) for global var!");
1799 unsigned LinkageID = (VarType >> 2) & 7;
1800 bool isConstant = VarType & 1;
1801 bool hasInitializer = VarType & 2;
1802 GlobalValue::LinkageTypes Linkage;
1804 switch (LinkageID) {
1805 case 0: Linkage = GlobalValue::ExternalLinkage; break;
1806 case 1: Linkage = GlobalValue::WeakLinkage; break;
1807 case 2: Linkage = GlobalValue::AppendingLinkage; break;
1808 case 3: Linkage = GlobalValue::InternalLinkage; break;
1809 case 4: Linkage = GlobalValue::LinkOnceLinkage; break;
1811 error("Unknown linkage type: " + utostr(LinkageID));
1812 Linkage = GlobalValue::InternalLinkage;
1816 const Type *Ty = getType(SlotNo);
1818 error("Global has no type! SlotNo=" + utostr(SlotNo));
1821 if (!isa<PointerType>(Ty)) {
1822 error("Global not a pointer type! Ty= " + Ty->getDescription());
1825 const Type *ElTy = cast<PointerType>(Ty)->getElementType();
1827 // Create the global variable...
1828 GlobalVariable *GV = new GlobalVariable(ElTy, isConstant, Linkage,
1830 insertValue(GV, SlotNo, ModuleValues);
1832 unsigned initSlot = 0;
1833 if (hasInitializer) {
1834 initSlot = read_vbr_uint();
1835 GlobalInits.push_back(std::make_pair(GV, initSlot));
1838 // Notify handler about the global value.
1840 Handler->handleGlobalVariable(ElTy, isConstant, Linkage, SlotNo,initSlot);
1843 VarType = read_vbr_uint();
1846 // Read the function objects for all of the functions that are coming
1847 unsigned FnSignature = read_vbr_uint();
1849 if (hasNoFlagsForFunctions)
1850 FnSignature = (FnSignature << 5) + 1;
1852 // List is terminated by VoidTy.
1853 while ((FnSignature >> 5) != Type::VoidTyID) {
1854 const Type *Ty = getType(FnSignature >> 5);
1855 if (!isa<PointerType>(Ty) ||
1856 !isa<FunctionType>(cast<PointerType>(Ty)->getElementType())) {
1857 error("Function not a pointer to function type! Ty = " +
1858 Ty->getDescription());
1861 // We create functions by passing the underlying FunctionType to create...
1862 const FunctionType* FTy =
1863 cast<FunctionType>(cast<PointerType>(Ty)->getElementType());
1866 // Insert the place holder.
1867 Function* Func = new Function(FTy, GlobalValue::ExternalLinkage,
1869 insertValue(Func, FnSignature >> 5, ModuleValues);
1871 // Flags are not used yet.
1872 unsigned Flags = FnSignature & 31;
1874 // Save this for later so we know type of lazily instantiated functions.
1875 // Note that known-external functions do not have FunctionInfo blocks, so we
1876 // do not add them to the FunctionSignatureList.
1877 if ((Flags & (1 << 4)) == 0)
1878 FunctionSignatureList.push_back(Func);
1880 if (Handler) Handler->handleFunctionDeclaration(Func);
1882 // Get the next function signature.
1883 FnSignature = read_vbr_uint();
1884 if (hasNoFlagsForFunctions)
1885 FnSignature = (FnSignature << 5) + 1;
1888 // Now that the function signature list is set up, reverse it so that we can
1889 // remove elements efficiently from the back of the vector.
1890 std::reverse(FunctionSignatureList.begin(), FunctionSignatureList.end());
1892 // If this bytecode format has dependent library information in it ..
1893 if (!hasNoDependentLibraries) {
1894 // Read in the number of dependent library items that follow
1895 unsigned num_dep_libs = read_vbr_uint();
1896 std::string dep_lib;
1897 while( num_dep_libs-- ) {
1898 dep_lib = read_str();
1899 TheModule->addLibrary(dep_lib);
1901 Handler->handleDependentLibrary(dep_lib);
1905 // Read target triple and place into the module
1906 std::string triple = read_str();
1907 TheModule->setTargetTriple(triple);
1909 Handler->handleTargetTriple(triple);
1912 if (hasInconsistentModuleGlobalInfo)
1915 // This is for future proofing... in the future extra fields may be added that
1916 // we don't understand, so we transparently ignore them.
1920 if (Handler) Handler->handleModuleGlobalsEnd();
1923 /// Parse the version information and decode it by setting flags on the
1924 /// Reader that enable backward compatibility of the reader.
1925 void BytecodeReader::ParseVersionInfo() {
1926 unsigned Version = read_vbr_uint();
1928 // Unpack version number: low four bits are for flags, top bits = version
1929 Module::Endianness Endianness;
1930 Module::PointerSize PointerSize;
1931 Endianness = (Version & 1) ? Module::BigEndian : Module::LittleEndian;
1932 PointerSize = (Version & 2) ? Module::Pointer64 : Module::Pointer32;
1934 bool hasNoEndianness = Version & 4;
1935 bool hasNoPointerSize = Version & 8;
1937 RevisionNum = Version >> 4;
1939 // Default values for the current bytecode version
1940 hasInconsistentModuleGlobalInfo = false;
1941 hasExplicitPrimitiveZeros = false;
1942 hasRestrictedGEPTypes = false;
1943 hasTypeDerivedFromValue = false;
1944 hasLongBlockHeaders = false;
1945 has32BitTypes = false;
1946 hasNoDependentLibraries = false;
1947 hasAlignment = false;
1948 hasInconsistentBBSlotNums = false;
1949 hasVBRByteTypes = false;
1950 hasUnnecessaryModuleBlockId = false;
1951 hasNoUndefValue = false;
1952 hasNoFlagsForFunctions = false;
1953 hasNoUnreachableInst = false;
1955 switch (RevisionNum) {
1956 case 0: // LLVM 1.0, 1.1 (Released)
1957 // Base LLVM 1.0 bytecode format.
1958 hasInconsistentModuleGlobalInfo = true;
1959 hasExplicitPrimitiveZeros = true;
1963 case 1: // LLVM 1.2 (Released)
1964 // LLVM 1.2 added explicit support for emitting strings efficiently.
1966 // Also, it fixed the problem where the size of the ModuleGlobalInfo block
1967 // included the size for the alignment at the end, where the rest of the
1970 // LLVM 1.2 and before required that GEP indices be ubyte constants for
1971 // structures and longs for sequential types.
1972 hasRestrictedGEPTypes = true;
1974 // LLVM 1.2 and before had the Type class derive from Value class. This
1975 // changed in release 1.3 and consequently LLVM 1.3 bytecode files are
1976 // written differently because Types can no longer be part of the
1977 // type planes for Values.
1978 hasTypeDerivedFromValue = true;
1982 case 2: // 1.2.5 (Not Released)
1984 // LLVM 1.2 and earlier had two-word block headers. This is a bit wasteful,
1985 // especially for small files where the 8 bytes per block is a large
1986 // fraction of the total block size. In LLVM 1.3, the block type and length
1987 // are compressed into a single 32-bit unsigned integer. 27 bits for length,
1988 // 5 bits for block type.
1989 hasLongBlockHeaders = true;
1991 // LLVM 1.2 and earlier wrote type slot numbers as vbr_uint32. In LLVM 1.3
1992 // this has been reduced to vbr_uint24. It shouldn't make much difference
1993 // since we haven't run into a module with > 24 million types, but for
1994 // safety the 24-bit restriction has been enforced in 1.3 to free some bits
1995 // in various places and to ensure consistency.
1996 has32BitTypes = true;
1998 // LLVM 1.2 and earlier did not provide a target triple nor a list of
1999 // libraries on which the bytecode is dependent. LLVM 1.3 provides these
2000 // features, for use in future versions of LLVM.
2001 hasNoDependentLibraries = true;
2005 case 3: // LLVM 1.3 (Released)
2006 // LLVM 1.3 and earlier caused alignment bytes to be written on some block
2007 // boundaries and at the end of some strings. In extreme cases (e.g. lots
2008 // of GEP references to a constant array), this can increase the file size
2009 // by 30% or more. In version 1.4 alignment is done away with completely.
2010 hasAlignment = true;
2014 case 4: // 1.3.1 (Not Released)
2015 // In version 4, we did not support the 'undef' constant.
2016 hasNoUndefValue = true;
2018 // In version 4 and above, we did not include space for flags for functions
2019 // in the module info block.
2020 hasNoFlagsForFunctions = true;
2022 // In version 4 and above, we did not include the 'unreachable' instruction
2023 // in the opcode numbering in the bytecode file.
2024 hasNoUnreachableInst = true;
2029 case 5: // 1.x.x (Not Released)
2031 // FIXME: NONE of this is implemented yet!
2033 // In version 5, basic blocks have a minimum index of 0 whereas all the
2034 // other primitives have a minimum index of 1 (because 0 is the "null"
2035 // value. In version 5, we made this consistent.
2036 hasInconsistentBBSlotNums = true;
2038 // In version 5, the types SByte and UByte were encoded as vbr_uint so that
2039 // signed values > 63 and unsigned values >127 would be encoded as two
2040 // bytes. In version 5, they are encoded directly in a single byte.
2041 hasVBRByteTypes = true;
2043 // In version 5, modules begin with a "Module Block" which encodes a 4-byte
2044 // integer value 0x01 to identify the module block. This is unnecessary and
2045 // removed in version 5.
2046 hasUnnecessaryModuleBlockId = true;
2049 error("Unknown bytecode version number: " + itostr(RevisionNum));
2052 if (hasNoEndianness) Endianness = Module::AnyEndianness;
2053 if (hasNoPointerSize) PointerSize = Module::AnyPointerSize;
2055 TheModule->setEndianness(Endianness);
2056 TheModule->setPointerSize(PointerSize);
2058 if (Handler) Handler->handleVersionInfo(RevisionNum, Endianness, PointerSize);
2061 /// Parse a whole module.
2062 void BytecodeReader::ParseModule() {
2063 unsigned Type, Size;
2065 FunctionSignatureList.clear(); // Just in case...
2067 // Read into instance variables...
2071 bool SeenModuleGlobalInfo = false;
2072 bool SeenGlobalTypePlane = false;
2073 BufPtr MyEnd = BlockEnd;
2074 while (At < MyEnd) {
2076 read_block(Type, Size);
2080 case BytecodeFormat::GlobalTypePlaneBlockID:
2081 if (SeenGlobalTypePlane)
2082 error("Two GlobalTypePlane Blocks Encountered!");
2086 SeenGlobalTypePlane = true;
2089 case BytecodeFormat::ModuleGlobalInfoBlockID:
2090 if (SeenModuleGlobalInfo)
2091 error("Two ModuleGlobalInfo Blocks Encountered!");
2092 ParseModuleGlobalInfo();
2093 SeenModuleGlobalInfo = true;
2096 case BytecodeFormat::ConstantPoolBlockID:
2097 ParseConstantPool(ModuleValues, ModuleTypes,false);
2100 case BytecodeFormat::FunctionBlockID:
2101 ParseFunctionLazily();
2104 case BytecodeFormat::SymbolTableBlockID:
2105 ParseSymbolTable(0, &TheModule->getSymbolTable());
2111 error("Unexpected Block of Type #" + utostr(Type) + " encountered!");
2119 // After the module constant pool has been read, we can safely initialize
2120 // global variables...
2121 while (!GlobalInits.empty()) {
2122 GlobalVariable *GV = GlobalInits.back().first;
2123 unsigned Slot = GlobalInits.back().second;
2124 GlobalInits.pop_back();
2126 // Look up the initializer value...
2127 // FIXME: Preserve this type ID!
2129 const llvm::PointerType* GVType = GV->getType();
2130 unsigned TypeSlot = getTypeSlot(GVType->getElementType());
2131 if (Constant *CV = getConstantValue(TypeSlot, Slot)) {
2132 if (GV->hasInitializer())
2133 error("Global *already* has an initializer?!");
2134 if (Handler) Handler->handleGlobalInitializer(GV,CV);
2135 GV->setInitializer(CV);
2137 error("Cannot find initializer value.");
2140 /// Make sure we pulled them all out. If we didn't then there's a declaration
2141 /// but a missing body. That's not allowed.
2142 if (!FunctionSignatureList.empty())
2143 error("Function declared, but bytecode stream ended before definition");
2146 /// This function completely parses a bytecode buffer given by the \p Buf
2147 /// and \p Length parameters.
2148 void BytecodeReader::ParseBytecode(BufPtr Buf, unsigned Length,
2149 const std::string &ModuleID) {
2153 At = MemStart = BlockStart = Buf;
2154 MemEnd = BlockEnd = Buf + Length;
2156 // Create the module
2157 TheModule = new Module(ModuleID);
2159 if (Handler) Handler->handleStart(TheModule, Length);
2161 // Read the four bytes of the signature.
2162 unsigned Sig = read_uint();
2164 // If this is a compressed file
2165 if (Sig == ('l' | ('l' << 8) | ('v' << 16) | ('c' << 24))) {
2167 // Invoke the decompression of the bytecode. Note that we have to skip the
2168 // file's magic number which is not part of the compressed block. Hence,
2169 // the Buf+4 and Length-4. The result goes into decompressedBlock, a data
2170 // member for retention until BytecodeReader is destructed.
2171 unsigned decompressedLength = Compressor::decompressToNewBuffer(
2172 (char*)Buf+4,Length-4,decompressedBlock);
2174 // We must adjust the buffer pointers used by the bytecode reader to point
2175 // into the new decompressed block. After decompression, the
2176 // decompressedBlock will point to a contiguous memory area that has
2177 // the decompressed data.
2178 At = MemStart = BlockStart = Buf = (BufPtr) decompressedBlock;
2179 MemEnd = BlockEnd = Buf + decompressedLength;
2181 // else if this isn't a regular (uncompressed) bytecode file, then its
2182 // and error, generate that now.
2183 } else if (Sig != ('l' | ('l' << 8) | ('v' << 16) | ('m' << 24))) {
2184 error("Invalid bytecode signature: " + utohexstr(Sig));
2187 // Tell the handler we're starting a module
2188 if (Handler) Handler->handleModuleBegin(ModuleID);
2190 // Get the module block and size and verify. This is handled specially
2191 // because the module block/size is always written in long format. Other
2192 // blocks are written in short format so the read_block method is used.
2193 unsigned Type, Size;
2196 if (Type != BytecodeFormat::ModuleBlockID) {
2197 error("Expected Module Block! Type:" + utostr(Type) + ", Size:"
2201 // It looks like the darwin ranlib program is broken, and adds trailing
2202 // garbage to the end of some bytecode files. This hack allows the bc
2203 // reader to ignore trailing garbage on bytecode files.
2204 if (At + Size < MemEnd)
2205 MemEnd = BlockEnd = At+Size;
2207 if (At + Size != MemEnd)
2208 error("Invalid Top Level Block Length! Type:" + utostr(Type)
2209 + ", Size:" + utostr(Size));
2211 // Parse the module contents
2212 this->ParseModule();
2214 // Check for missing functions
2216 error("Function expected, but bytecode stream ended!");
2218 // Tell the handler we're done with the module
2220 Handler->handleModuleEnd(ModuleID);
2222 // Tell the handler we're finished the parse
2223 if (Handler) Handler->handleFinish();
2225 } catch (std::string& errstr) {
2226 if (Handler) Handler->handleError(errstr);
2230 if (decompressedBlock != 0 ) {
2231 ::free(decompressedBlock);
2232 decompressedBlock = 0;
2236 std::string msg("Unknown Exception Occurred");
2237 if (Handler) Handler->handleError(msg);
2241 if (decompressedBlock != 0) {
2242 ::free(decompressedBlock);
2243 decompressedBlock = 0;
2249 //===----------------------------------------------------------------------===//
2250 //=== Default Implementations of Handler Methods
2251 //===----------------------------------------------------------------------===//
2253 BytecodeHandler::~BytecodeHandler() {}