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/Config/alloca.h"
23 #include "llvm/Constants.h"
24 #include "llvm/Instructions.h"
25 #include "llvm/SymbolTable.h"
26 #include "llvm/Bytecode/Format.h"
27 #include "llvm/Support/GetElementPtrTypeIterator.h"
28 #include "llvm/Support/Compressor.h"
29 #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
42 ConstantPlaceHolder(const Type *Ty)
43 : ConstantExpr(Ty, Instruction::UserOp1, &Op, 1),
44 Op(UndefValue::get(Type::IntTy), this) {
49 // Provide some details on error
50 inline void BytecodeReader::error(std::string err) {
52 err += itostr(RevisionNum) ;
54 err += itostr(At-MemStart);
59 //===----------------------------------------------------------------------===//
60 // Bytecode Reading Methods
61 //===----------------------------------------------------------------------===//
63 /// Determine if the current block being read contains any more data.
64 inline bool BytecodeReader::moreInBlock() {
68 /// Throw an error if we've read past the end of the current block
69 inline void BytecodeReader::checkPastBlockEnd(const char * block_name) {
71 error(std::string("Attempt to read past the end of ") + block_name +
75 /// Align the buffer position to a 32 bit boundary
76 inline void BytecodeReader::align32() {
79 At = (const unsigned char *)((unsigned long)(At+3) & (~3UL));
81 if (Handler) Handler->handleAlignment(At - Save);
83 error("Ran out of data while aligning!");
87 /// Read a whole unsigned integer
88 inline unsigned BytecodeReader::read_uint() {
90 error("Ran out of data reading uint!");
92 return At[-4] | (At[-3] << 8) | (At[-2] << 16) | (At[-1] << 24);
95 /// Read a variable-bit-rate encoded unsigned integer
96 inline unsigned BytecodeReader::read_vbr_uint() {
103 error("Ran out of data reading vbr_uint!");
104 Result |= (unsigned)((*At++) & 0x7F) << Shift;
106 } while (At[-1] & 0x80);
107 if (Handler) Handler->handleVBR32(At-Save);
111 /// Read a variable-bit-rate encoded unsigned 64-bit integer.
112 inline uint64_t BytecodeReader::read_vbr_uint64() {
119 error("Ran out of data reading vbr_uint64!");
120 Result |= (uint64_t)((*At++) & 0x7F) << Shift;
122 } while (At[-1] & 0x80);
123 if (Handler) Handler->handleVBR64(At-Save);
127 /// Read a variable-bit-rate encoded signed 64-bit integer.
128 inline int64_t BytecodeReader::read_vbr_int64() {
129 uint64_t R = read_vbr_uint64();
132 return -(int64_t)(R >> 1);
133 else // There is no such thing as -0 with integers. "-0" really means
134 // 0x8000000000000000.
137 return (int64_t)(R >> 1);
140 /// Read a pascal-style string (length followed by text)
141 inline std::string BytecodeReader::read_str() {
142 unsigned Size = read_vbr_uint();
143 const unsigned char *OldAt = At;
145 if (At > BlockEnd) // Size invalid?
146 error("Ran out of data reading a string!");
147 return std::string((char*)OldAt, Size);
150 /// Read an arbitrary block of data
151 inline void BytecodeReader::read_data(void *Ptr, void *End) {
152 unsigned char *Start = (unsigned char *)Ptr;
153 unsigned Amount = (unsigned char *)End - Start;
154 if (At+Amount > BlockEnd)
155 error("Ran out of data!");
156 std::copy(At, At+Amount, Start);
160 /// Read a float value in little-endian order
161 inline void BytecodeReader::read_float(float& FloatVal) {
162 /// FIXME: This isn't optimal, it has size problems on some platforms
163 /// where FP is not IEEE.
168 FloatUnion.i = At[0] | (At[1] << 8) | (At[2] << 16) | (At[3] << 24);
169 At+=sizeof(uint32_t);
170 FloatVal = FloatUnion.f;
173 /// Read a double value in little-endian order
174 inline void BytecodeReader::read_double(double& DoubleVal) {
175 /// FIXME: This isn't optimal, it has size problems on some platforms
176 /// where FP is not IEEE.
181 DoubleUnion.i = (uint64_t(At[0]) << 0) | (uint64_t(At[1]) << 8) |
182 (uint64_t(At[2]) << 16) | (uint64_t(At[3]) << 24) |
183 (uint64_t(At[4]) << 32) | (uint64_t(At[5]) << 40) |
184 (uint64_t(At[6]) << 48) | (uint64_t(At[7]) << 56);
185 At+=sizeof(uint64_t);
186 DoubleVal = DoubleUnion.d;
189 /// Read a block header and obtain its type and size
190 inline void BytecodeReader::read_block(unsigned &Type, unsigned &Size) {
191 if ( hasLongBlockHeaders ) {
195 case BytecodeFormat::Reserved_DoNotUse :
196 error("Reserved_DoNotUse used as Module Type?");
197 Type = BytecodeFormat::ModuleBlockID; break;
198 case BytecodeFormat::Module:
199 Type = BytecodeFormat::ModuleBlockID; break;
200 case BytecodeFormat::Function:
201 Type = BytecodeFormat::FunctionBlockID; break;
202 case BytecodeFormat::ConstantPool:
203 Type = BytecodeFormat::ConstantPoolBlockID; break;
204 case BytecodeFormat::SymbolTable:
205 Type = BytecodeFormat::SymbolTableBlockID; break;
206 case BytecodeFormat::ModuleGlobalInfo:
207 Type = BytecodeFormat::ModuleGlobalInfoBlockID; break;
208 case BytecodeFormat::GlobalTypePlane:
209 Type = BytecodeFormat::GlobalTypePlaneBlockID; break;
210 case BytecodeFormat::InstructionList:
211 Type = BytecodeFormat::InstructionListBlockID; break;
212 case BytecodeFormat::CompactionTable:
213 Type = BytecodeFormat::CompactionTableBlockID; break;
214 case BytecodeFormat::BasicBlock:
215 /// This block type isn't used after version 1.1. However, we have to
216 /// still allow the value in case this is an old bc format file.
217 /// We just let its value creep thru.
220 error("Invalid block id found: " + utostr(Type));
225 Type = Size & 0x1F; // mask low order five bits
226 Size >>= 5; // get rid of five low order bits, leaving high 27
229 if (At + Size > BlockEnd)
230 error("Attempt to size a block past end of memory");
231 BlockEnd = At + Size;
232 if (Handler) Handler->handleBlock(Type, BlockStart, Size);
236 /// In LLVM 1.2 and before, Types were derived from Value and so they were
237 /// written as part of the type planes along with any other Value. In LLVM
238 /// 1.3 this changed so that Type does not derive from Value. Consequently,
239 /// the BytecodeReader's containers for Values can't contain Types because
240 /// there's no inheritance relationship. This means that the "Type Type"
241 /// plane is defunct along with the Type::TypeTyID TypeID. In LLVM 1.3
242 /// whenever a bytecode construct must have both types and values together,
243 /// the types are always read/written first and then the Values. Furthermore
244 /// since Type::TypeTyID no longer exists, its value (12) now corresponds to
245 /// Type::LabelTyID. In order to overcome this we must "sanitize" all the
246 /// type TypeIDs we encounter. For LLVM 1.3 bytecode files, there's no change.
247 /// For LLVM 1.2 and before, this function will decrement the type id by
248 /// one to account for the missing Type::TypeTyID enumerator if the value is
249 /// larger than 12 (Type::LabelTyID). If the value is exactly 12, then this
250 /// function returns true, otherwise false. This helps detect situations
251 /// where the pre 1.3 bytecode is indicating that what follows is a type.
252 /// @returns true iff type id corresponds to pre 1.3 "type type"
253 inline bool BytecodeReader::sanitizeTypeId(unsigned &TypeId) {
254 if (hasTypeDerivedFromValue) { /// do nothing if 1.3 or later
255 if (TypeId == Type::LabelTyID) {
256 TypeId = Type::VoidTyID; // sanitize it
257 return true; // indicate we got TypeTyID in pre 1.3 bytecode
258 } else if (TypeId > Type::LabelTyID)
259 --TypeId; // shift all planes down because type type plane is missing
264 /// Reads a vbr uint to read in a type id and does the necessary
265 /// conversion on it by calling sanitizeTypeId.
266 /// @returns true iff \p TypeId read corresponds to a pre 1.3 "type type"
267 /// @see sanitizeTypeId
268 inline bool BytecodeReader::read_typeid(unsigned &TypeId) {
269 TypeId = read_vbr_uint();
270 if ( !has32BitTypes )
271 if ( TypeId == 0x00FFFFFF )
272 TypeId = read_vbr_uint();
273 return sanitizeTypeId(TypeId);
276 //===----------------------------------------------------------------------===//
278 //===----------------------------------------------------------------------===//
280 /// Determine if a type id has an implicit null value
281 inline bool BytecodeReader::hasImplicitNull(unsigned TyID) {
282 if (!hasExplicitPrimitiveZeros)
283 return TyID != Type::LabelTyID && TyID != Type::VoidTyID;
284 return TyID >= Type::FirstDerivedTyID;
287 /// Obtain a type given a typeid and account for things like compaction tables,
288 /// function level vs module level, and the offsetting for the primitive types.
289 const Type *BytecodeReader::getType(unsigned ID) {
290 if (ID < Type::FirstDerivedTyID)
291 if (const Type *T = Type::getPrimitiveType((Type::TypeID)ID))
292 return T; // Asked for a primitive type...
294 // Otherwise, derived types need offset...
295 ID -= Type::FirstDerivedTyID;
297 if (!CompactionTypes.empty()) {
298 if (ID >= CompactionTypes.size())
299 error("Type ID out of range for compaction table!");
300 return CompactionTypes[ID].first;
303 // Is it a module-level type?
304 if (ID < ModuleTypes.size())
305 return ModuleTypes[ID].get();
307 // Nope, is it a function-level type?
308 ID -= ModuleTypes.size();
309 if (ID < FunctionTypes.size())
310 return FunctionTypes[ID].get();
312 error("Illegal type reference!");
316 /// Get a sanitized type id. This just makes sure that the \p ID
317 /// is both sanitized and not the "type type" of pre-1.3 bytecode.
318 /// @see sanitizeTypeId
319 inline const Type* BytecodeReader::getSanitizedType(unsigned& ID) {
320 if (sanitizeTypeId(ID))
321 error("Invalid type id encountered");
325 /// This method just saves some coding. It uses read_typeid to read
326 /// in a sanitized type id, errors that its not the type type, and
327 /// then calls getType to return the type value.
328 inline const Type* BytecodeReader::readSanitizedType() {
331 error("Invalid type id encountered");
335 /// Get the slot number associated with a type accounting for primitive
336 /// types, compaction tables, and function level vs module level.
337 unsigned BytecodeReader::getTypeSlot(const Type *Ty) {
338 if (Ty->isPrimitiveType())
339 return Ty->getTypeID();
341 // Scan the compaction table for the type if needed.
342 if (!CompactionTypes.empty()) {
343 for (unsigned i = 0, e = CompactionTypes.size(); i != e; ++i)
344 if (CompactionTypes[i].first == Ty)
345 return Type::FirstDerivedTyID + i;
347 error("Couldn't find type specified in compaction table!");
350 // Check the function level types first...
351 TypeListTy::iterator I = std::find(FunctionTypes.begin(),
352 FunctionTypes.end(), Ty);
354 if (I != FunctionTypes.end())
355 return Type::FirstDerivedTyID + ModuleTypes.size() +
356 (&*I - &FunctionTypes[0]);
358 // Check the module level types now...
359 I = std::find(ModuleTypes.begin(), ModuleTypes.end(), Ty);
360 if (I == ModuleTypes.end())
361 error("Didn't find type in ModuleTypes.");
362 return Type::FirstDerivedTyID + (&*I - &ModuleTypes[0]);
365 /// This is just like getType, but when a compaction table is in use, it is
366 /// ignored. It also ignores function level types.
368 const Type *BytecodeReader::getGlobalTableType(unsigned Slot) {
369 if (Slot < Type::FirstDerivedTyID) {
370 const Type *Ty = Type::getPrimitiveType((Type::TypeID)Slot);
372 error("Not a primitive type ID?");
375 Slot -= Type::FirstDerivedTyID;
376 if (Slot >= ModuleTypes.size())
377 error("Illegal compaction table type reference!");
378 return ModuleTypes[Slot];
381 /// This is just like getTypeSlot, but when a compaction table is in use, it
382 /// is ignored. It also ignores function level types.
383 unsigned BytecodeReader::getGlobalTableTypeSlot(const Type *Ty) {
384 if (Ty->isPrimitiveType())
385 return Ty->getTypeID();
386 TypeListTy::iterator I = std::find(ModuleTypes.begin(),
387 ModuleTypes.end(), Ty);
388 if (I == ModuleTypes.end())
389 error("Didn't find type in ModuleTypes.");
390 return Type::FirstDerivedTyID + (&*I - &ModuleTypes[0]);
393 /// Retrieve a value of a given type and slot number, possibly creating
394 /// it if it doesn't already exist.
395 Value * BytecodeReader::getValue(unsigned type, unsigned oNum, bool Create) {
396 assert(type != Type::LabelTyID && "getValue() cannot get blocks!");
399 // If there is a compaction table active, it defines the low-level numbers.
400 // If not, the module values define the low-level numbers.
401 if (CompactionValues.size() > type && !CompactionValues[type].empty()) {
402 if (Num < CompactionValues[type].size())
403 return CompactionValues[type][Num];
404 Num -= CompactionValues[type].size();
406 // By default, the global type id is the type id passed in
407 unsigned GlobalTyID = type;
409 // If the type plane was compactified, figure out the global type ID by
410 // adding the derived type ids and the distance.
411 if (!CompactionTypes.empty() && type >= Type::FirstDerivedTyID)
412 GlobalTyID = CompactionTypes[type-Type::FirstDerivedTyID].second;
414 if (hasImplicitNull(GlobalTyID)) {
415 const Type *Ty = getType(type);
416 if (!isa<OpaqueType>(Ty)) {
418 return Constant::getNullValue(Ty);
423 if (GlobalTyID < ModuleValues.size() && ModuleValues[GlobalTyID]) {
424 if (Num < ModuleValues[GlobalTyID]->size())
425 return ModuleValues[GlobalTyID]->getOperand(Num);
426 Num -= ModuleValues[GlobalTyID]->size();
430 if (FunctionValues.size() > type &&
431 FunctionValues[type] &&
432 Num < FunctionValues[type]->size())
433 return FunctionValues[type]->getOperand(Num);
435 if (!Create) return 0; // Do not create a placeholder?
437 // Did we already create a place holder?
438 std::pair<unsigned,unsigned> KeyValue(type, oNum);
439 ForwardReferenceMap::iterator I = ForwardReferences.lower_bound(KeyValue);
440 if (I != ForwardReferences.end() && I->first == KeyValue)
441 return I->second; // We have already created this placeholder
443 // If the type exists (it should)
444 if (const Type* Ty = getType(type)) {
445 // Create the place holder
446 Value *Val = new Argument(Ty);
447 ForwardReferences.insert(I, std::make_pair(KeyValue, Val));
450 throw "Can't create placeholder for value of type slot #" + utostr(type);
453 /// This is just like getValue, but when a compaction table is in use, it
454 /// is ignored. Also, no forward references or other fancy features are
456 Value* BytecodeReader::getGlobalTableValue(unsigned TyID, unsigned SlotNo) {
458 return Constant::getNullValue(getType(TyID));
460 if (!CompactionTypes.empty() && TyID >= Type::FirstDerivedTyID) {
461 TyID -= Type::FirstDerivedTyID;
462 if (TyID >= CompactionTypes.size())
463 error("Type ID out of range for compaction table!");
464 TyID = CompactionTypes[TyID].second;
469 if (TyID >= ModuleValues.size() || ModuleValues[TyID] == 0 ||
470 SlotNo >= ModuleValues[TyID]->size()) {
471 if (TyID >= ModuleValues.size() || ModuleValues[TyID] == 0)
472 error("Corrupt compaction table entry!"
473 + utostr(TyID) + ", " + utostr(SlotNo) + ": "
474 + utostr(ModuleValues.size()));
476 error("Corrupt compaction table entry!"
477 + utostr(TyID) + ", " + utostr(SlotNo) + ": "
478 + utostr(ModuleValues.size()) + ", "
479 + utohexstr(reinterpret_cast<uint64_t>(((void*)ModuleValues[TyID])))
481 + utostr(ModuleValues[TyID]->size()));
483 return ModuleValues[TyID]->getOperand(SlotNo);
486 /// Just like getValue, except that it returns a null pointer
487 /// only on error. It always returns a constant (meaning that if the value is
488 /// defined, but is not a constant, that is an error). If the specified
489 /// constant hasn't been parsed yet, a placeholder is defined and used.
490 /// Later, after the real value is parsed, the placeholder is eliminated.
491 Constant* BytecodeReader::getConstantValue(unsigned TypeSlot, unsigned Slot) {
492 if (Value *V = getValue(TypeSlot, Slot, false))
493 if (Constant *C = dyn_cast<Constant>(V))
494 return C; // If we already have the value parsed, just return it
496 error("Value for slot " + utostr(Slot) +
497 " is expected to be a constant!");
499 std::pair<unsigned, unsigned> Key(TypeSlot, Slot);
500 ConstantRefsType::iterator I = ConstantFwdRefs.lower_bound(Key);
502 if (I != ConstantFwdRefs.end() && I->first == Key) {
505 // Create a placeholder for the constant reference and
506 // keep track of the fact that we have a forward ref to recycle it
507 Constant *C = new ConstantPlaceHolder(getType(TypeSlot));
509 // Keep track of the fact that we have a forward ref to recycle it
510 ConstantFwdRefs.insert(I, std::make_pair(Key, C));
515 //===----------------------------------------------------------------------===//
516 // IR Construction Methods
517 //===----------------------------------------------------------------------===//
519 /// As values are created, they are inserted into the appropriate place
520 /// with this method. The ValueTable argument must be one of ModuleValues
521 /// or FunctionValues data members of this class.
522 unsigned BytecodeReader::insertValue(Value *Val, unsigned type,
523 ValueTable &ValueTab) {
524 assert((!isa<Constant>(Val) || !cast<Constant>(Val)->isNullValue()) ||
525 !hasImplicitNull(type) &&
526 "Cannot read null values from bytecode!");
528 if (ValueTab.size() <= type)
529 ValueTab.resize(type+1);
531 if (!ValueTab[type]) ValueTab[type] = new ValueList();
533 ValueTab[type]->push_back(Val);
535 bool HasOffset = hasImplicitNull(type) && !isa<OpaqueType>(Val->getType());
536 return ValueTab[type]->size()-1 + HasOffset;
539 /// Insert the arguments of a function as new values in the reader.
540 void BytecodeReader::insertArguments(Function* F) {
541 const FunctionType *FT = F->getFunctionType();
542 Function::arg_iterator AI = F->arg_begin();
543 for (FunctionType::param_iterator It = FT->param_begin();
544 It != FT->param_end(); ++It, ++AI)
545 insertValue(AI, getTypeSlot(AI->getType()), FunctionValues);
548 //===----------------------------------------------------------------------===//
549 // Bytecode Parsing Methods
550 //===----------------------------------------------------------------------===//
552 /// This method parses a single instruction. The instruction is
553 /// inserted at the end of the \p BB provided. The arguments of
554 /// the instruction are provided in the \p Oprnds vector.
555 void BytecodeReader::ParseInstruction(std::vector<unsigned> &Oprnds,
559 // Clear instruction data
563 unsigned Op = read_uint();
565 // bits Instruction format: Common to all formats
566 // --------------------------
567 // 01-00: Opcode type, fixed to 1.
569 Opcode = (Op >> 2) & 63;
570 Oprnds.resize((Op >> 0) & 03);
572 // Extract the operands
573 switch (Oprnds.size()) {
575 // bits Instruction format:
576 // --------------------------
577 // 19-08: Resulting type plane
578 // 31-20: Operand #1 (if set to (2^12-1), then zero operands)
580 iType = (Op >> 8) & 4095;
581 Oprnds[0] = (Op >> 20) & 4095;
582 if (Oprnds[0] == 4095) // Handle special encoding for 0 operands...
586 // bits Instruction format:
587 // --------------------------
588 // 15-08: Resulting type plane
592 iType = (Op >> 8) & 255;
593 Oprnds[0] = (Op >> 16) & 255;
594 Oprnds[1] = (Op >> 24) & 255;
597 // bits Instruction format:
598 // --------------------------
599 // 13-08: Resulting type plane
604 iType = (Op >> 8) & 63;
605 Oprnds[0] = (Op >> 14) & 63;
606 Oprnds[1] = (Op >> 20) & 63;
607 Oprnds[2] = (Op >> 26) & 63;
610 At -= 4; // Hrm, try this again...
611 Opcode = read_vbr_uint();
613 iType = read_vbr_uint();
615 unsigned NumOprnds = read_vbr_uint();
616 Oprnds.resize(NumOprnds);
619 error("Zero-argument instruction found; this is invalid.");
621 for (unsigned i = 0; i != NumOprnds; ++i)
622 Oprnds[i] = read_vbr_uint();
627 const Type *InstTy = getSanitizedType(iType);
629 // We have enough info to inform the handler now.
630 if (Handler) Handler->handleInstruction(Opcode, InstTy, Oprnds, At-SaveAt);
632 // Declare the resulting instruction we'll build.
633 Instruction *Result = 0;
635 // If this is a bytecode format that did not include the unreachable
636 // instruction, bump up all opcodes numbers to make space.
637 if (hasNoUnreachableInst) {
638 if (Opcode >= Instruction::Unreachable &&
644 // Handle binary operators
645 if (Opcode >= Instruction::BinaryOpsBegin &&
646 Opcode < Instruction::BinaryOpsEnd && Oprnds.size() == 2)
647 Result = BinaryOperator::create((Instruction::BinaryOps)Opcode,
648 getValue(iType, Oprnds[0]),
649 getValue(iType, Oprnds[1]));
654 error("Illegal instruction read!");
656 case Instruction::VAArg:
657 Result = new VAArgInst(getValue(iType, Oprnds[0]),
658 getSanitizedType(Oprnds[1]));
660 case Instruction::VANext:
661 Result = new VANextInst(getValue(iType, Oprnds[0]),
662 getSanitizedType(Oprnds[1]));
664 case Instruction::Cast:
665 Result = new CastInst(getValue(iType, Oprnds[0]),
666 getSanitizedType(Oprnds[1]));
668 case Instruction::Select:
669 Result = new SelectInst(getValue(Type::BoolTyID, Oprnds[0]),
670 getValue(iType, Oprnds[1]),
671 getValue(iType, Oprnds[2]));
673 case Instruction::PHI: {
674 if (Oprnds.size() == 0 || (Oprnds.size() & 1))
675 error("Invalid phi node encountered!");
677 PHINode *PN = new PHINode(InstTy);
678 PN->reserveOperandSpace(Oprnds.size());
679 for (unsigned i = 0, e = Oprnds.size(); i != e; i += 2)
680 PN->addIncoming(getValue(iType, Oprnds[i]), getBasicBlock(Oprnds[i+1]));
685 case Instruction::Shl:
686 case Instruction::Shr:
687 Result = new ShiftInst((Instruction::OtherOps)Opcode,
688 getValue(iType, Oprnds[0]),
689 getValue(Type::UByteTyID, Oprnds[1]));
691 case Instruction::Ret:
692 if (Oprnds.size() == 0)
693 Result = new ReturnInst();
694 else if (Oprnds.size() == 1)
695 Result = new ReturnInst(getValue(iType, Oprnds[0]));
697 error("Unrecognized instruction!");
700 case Instruction::Br:
701 if (Oprnds.size() == 1)
702 Result = new BranchInst(getBasicBlock(Oprnds[0]));
703 else if (Oprnds.size() == 3)
704 Result = new BranchInst(getBasicBlock(Oprnds[0]),
705 getBasicBlock(Oprnds[1]), getValue(Type::BoolTyID , Oprnds[2]));
707 error("Invalid number of operands for a 'br' instruction!");
709 case Instruction::Switch: {
710 if (Oprnds.size() & 1)
711 error("Switch statement with odd number of arguments!");
713 SwitchInst *I = new SwitchInst(getValue(iType, Oprnds[0]),
714 getBasicBlock(Oprnds[1]),
716 for (unsigned i = 2, e = Oprnds.size(); i != e; i += 2)
717 I->addCase(cast<ConstantInt>(getValue(iType, Oprnds[i])),
718 getBasicBlock(Oprnds[i+1]));
723 case 61: // tail call
724 case Instruction::Call: {
725 if (Oprnds.size() == 0)
726 error("Invalid call instruction encountered!");
728 Value *F = getValue(iType, Oprnds[0]);
730 // Check to make sure we have a pointer to function type
731 const PointerType *PTy = dyn_cast<PointerType>(F->getType());
732 if (PTy == 0) error("Call to non function pointer value!");
733 const FunctionType *FTy = dyn_cast<FunctionType>(PTy->getElementType());
734 if (FTy == 0) error("Call to non function pointer value!");
736 std::vector<Value *> Params;
737 if (!FTy->isVarArg()) {
738 FunctionType::param_iterator It = FTy->param_begin();
740 for (unsigned i = 1, e = Oprnds.size(); i != e; ++i) {
741 if (It == FTy->param_end())
742 error("Invalid call instruction!");
743 Params.push_back(getValue(getTypeSlot(*It++), Oprnds[i]));
745 if (It != FTy->param_end())
746 error("Invalid call instruction!");
748 Oprnds.erase(Oprnds.begin(), Oprnds.begin()+1);
750 unsigned FirstVariableOperand;
751 if (Oprnds.size() < FTy->getNumParams())
752 error("Call instruction missing operands!");
754 // Read all of the fixed arguments
755 for (unsigned i = 0, e = FTy->getNumParams(); i != e; ++i)
756 Params.push_back(getValue(getTypeSlot(FTy->getParamType(i)),Oprnds[i]));
758 FirstVariableOperand = FTy->getNumParams();
760 if ((Oprnds.size()-FirstVariableOperand) & 1)
761 error("Invalid call instruction!"); // Must be pairs of type/value
763 for (unsigned i = FirstVariableOperand, e = Oprnds.size();
765 Params.push_back(getValue(Oprnds[i], Oprnds[i+1]));
768 Result = new CallInst(F, Params);
769 if (Opcode == 61) cast<CallInst>(Result)->setTailCall(true);
772 case Instruction::Invoke: {
773 if (Oprnds.size() < 3)
774 error("Invalid invoke instruction!");
775 Value *F = getValue(iType, Oprnds[0]);
777 // Check to make sure we have a pointer to function type
778 const PointerType *PTy = dyn_cast<PointerType>(F->getType());
780 error("Invoke to non function pointer value!");
781 const FunctionType *FTy = dyn_cast<FunctionType>(PTy->getElementType());
783 error("Invoke to non function pointer value!");
785 std::vector<Value *> Params;
786 BasicBlock *Normal, *Except;
788 if (!FTy->isVarArg()) {
789 Normal = getBasicBlock(Oprnds[1]);
790 Except = getBasicBlock(Oprnds[2]);
792 FunctionType::param_iterator It = FTy->param_begin();
793 for (unsigned i = 3, e = Oprnds.size(); i != e; ++i) {
794 if (It == FTy->param_end())
795 error("Invalid invoke instruction!");
796 Params.push_back(getValue(getTypeSlot(*It++), Oprnds[i]));
798 if (It != FTy->param_end())
799 error("Invalid invoke instruction!");
801 Oprnds.erase(Oprnds.begin(), Oprnds.begin()+1);
803 Normal = getBasicBlock(Oprnds[0]);
804 Except = getBasicBlock(Oprnds[1]);
806 unsigned FirstVariableArgument = FTy->getNumParams()+2;
807 for (unsigned i = 2; i != FirstVariableArgument; ++i)
808 Params.push_back(getValue(getTypeSlot(FTy->getParamType(i-2)),
811 if (Oprnds.size()-FirstVariableArgument & 1) // Must be type/value pairs
812 error("Invalid invoke instruction!");
814 for (unsigned i = FirstVariableArgument; i < Oprnds.size(); i += 2)
815 Params.push_back(getValue(Oprnds[i], Oprnds[i+1]));
818 Result = new InvokeInst(F, Normal, Except, Params);
821 case Instruction::Malloc:
822 if (Oprnds.size() > 2)
823 error("Invalid malloc instruction!");
824 if (!isa<PointerType>(InstTy))
825 error("Invalid malloc instruction!");
827 Result = new MallocInst(cast<PointerType>(InstTy)->getElementType(),
828 Oprnds.size() ? getValue(Type::UIntTyID,
832 case Instruction::Alloca:
833 if (Oprnds.size() > 2)
834 error("Invalid alloca instruction!");
835 if (!isa<PointerType>(InstTy))
836 error("Invalid alloca instruction!");
838 Result = new AllocaInst(cast<PointerType>(InstTy)->getElementType(),
839 Oprnds.size() ? getValue(Type::UIntTyID,
842 case Instruction::Free:
843 if (!isa<PointerType>(InstTy))
844 error("Invalid free instruction!");
845 Result = new FreeInst(getValue(iType, Oprnds[0]));
847 case Instruction::GetElementPtr: {
848 if (Oprnds.size() == 0 || !isa<PointerType>(InstTy))
849 error("Invalid getelementptr instruction!");
851 std::vector<Value*> Idx;
853 const Type *NextTy = InstTy;
854 for (unsigned i = 1, e = Oprnds.size(); i != e; ++i) {
855 const CompositeType *TopTy = dyn_cast_or_null<CompositeType>(NextTy);
857 error("Invalid getelementptr instruction!");
859 unsigned ValIdx = Oprnds[i];
861 if (!hasRestrictedGEPTypes) {
862 // Struct indices are always uints, sequential type indices can be any
863 // of the 32 or 64-bit integer types. The actual choice of type is
864 // encoded in the low two bits of the slot number.
865 if (isa<StructType>(TopTy))
866 IdxTy = Type::UIntTyID;
868 switch (ValIdx & 3) {
870 case 0: IdxTy = Type::UIntTyID; break;
871 case 1: IdxTy = Type::IntTyID; break;
872 case 2: IdxTy = Type::ULongTyID; break;
873 case 3: IdxTy = Type::LongTyID; break;
878 IdxTy = isa<StructType>(TopTy) ? Type::UByteTyID : Type::LongTyID;
881 Idx.push_back(getValue(IdxTy, ValIdx));
883 // Convert ubyte struct indices into uint struct indices.
884 if (isa<StructType>(TopTy) && hasRestrictedGEPTypes)
885 if (ConstantUInt *C = dyn_cast<ConstantUInt>(Idx.back()))
886 Idx[Idx.size()-1] = ConstantExpr::getCast(C, Type::UIntTy);
888 NextTy = GetElementPtrInst::getIndexedType(InstTy, Idx, true);
891 Result = new GetElementPtrInst(getValue(iType, Oprnds[0]), Idx);
895 case 62: // volatile load
896 case Instruction::Load:
897 if (Oprnds.size() != 1 || !isa<PointerType>(InstTy))
898 error("Invalid load instruction!");
899 Result = new LoadInst(getValue(iType, Oprnds[0]), "", Opcode == 62);
902 case 63: // volatile store
903 case Instruction::Store: {
904 if (!isa<PointerType>(InstTy) || Oprnds.size() != 2)
905 error("Invalid store instruction!");
907 Value *Ptr = getValue(iType, Oprnds[1]);
908 const Type *ValTy = cast<PointerType>(Ptr->getType())->getElementType();
909 Result = new StoreInst(getValue(getTypeSlot(ValTy), Oprnds[0]), Ptr,
913 case Instruction::Unwind:
914 if (Oprnds.size() != 0) error("Invalid unwind instruction!");
915 Result = new UnwindInst();
917 case Instruction::Unreachable:
918 if (Oprnds.size() != 0) error("Invalid unreachable instruction!");
919 Result = new UnreachableInst();
921 } // end switch(Opcode)
924 if (Result->getType() == InstTy)
927 TypeSlot = getTypeSlot(Result->getType());
929 insertValue(Result, TypeSlot, FunctionValues);
930 BB->getInstList().push_back(Result);
933 /// Get a particular numbered basic block, which might be a forward reference.
934 /// This works together with ParseBasicBlock to handle these forward references
935 /// in a clean manner. This function is used when constructing phi, br, switch,
936 /// and other instructions that reference basic blocks. Blocks are numbered
937 /// sequentially as they appear in the function.
938 BasicBlock *BytecodeReader::getBasicBlock(unsigned ID) {
939 // Make sure there is room in the table...
940 if (ParsedBasicBlocks.size() <= ID) ParsedBasicBlocks.resize(ID+1);
942 // First check to see if this is a backwards reference, i.e., ParseBasicBlock
943 // has already created this block, or if the forward reference has already
945 if (ParsedBasicBlocks[ID])
946 return ParsedBasicBlocks[ID];
948 // Otherwise, the basic block has not yet been created. Do so and add it to
949 // the ParsedBasicBlocks list.
950 return ParsedBasicBlocks[ID] = new BasicBlock();
953 /// In LLVM 1.0 bytecode files, we used to output one basicblock at a time.
954 /// This method reads in one of the basicblock packets. This method is not used
955 /// for bytecode files after LLVM 1.0
956 /// @returns The basic block constructed.
957 BasicBlock *BytecodeReader::ParseBasicBlock(unsigned BlockNo) {
958 if (Handler) Handler->handleBasicBlockBegin(BlockNo);
962 if (ParsedBasicBlocks.size() == BlockNo)
963 ParsedBasicBlocks.push_back(BB = new BasicBlock());
964 else if (ParsedBasicBlocks[BlockNo] == 0)
965 BB = ParsedBasicBlocks[BlockNo] = new BasicBlock();
967 BB = ParsedBasicBlocks[BlockNo];
969 std::vector<unsigned> Operands;
970 while (moreInBlock())
971 ParseInstruction(Operands, BB);
973 if (Handler) Handler->handleBasicBlockEnd(BlockNo);
977 /// Parse all of the BasicBlock's & Instruction's in the body of a function.
978 /// In post 1.0 bytecode files, we no longer emit basic block individually,
979 /// in order to avoid per-basic-block overhead.
980 /// @returns Rhe number of basic blocks encountered.
981 unsigned BytecodeReader::ParseInstructionList(Function* F) {
982 unsigned BlockNo = 0;
983 std::vector<unsigned> Args;
985 while (moreInBlock()) {
986 if (Handler) Handler->handleBasicBlockBegin(BlockNo);
988 if (ParsedBasicBlocks.size() == BlockNo)
989 ParsedBasicBlocks.push_back(BB = new BasicBlock());
990 else if (ParsedBasicBlocks[BlockNo] == 0)
991 BB = ParsedBasicBlocks[BlockNo] = new BasicBlock();
993 BB = ParsedBasicBlocks[BlockNo];
995 F->getBasicBlockList().push_back(BB);
997 // Read instructions into this basic block until we get to a terminator
998 while (moreInBlock() && !BB->getTerminator())
999 ParseInstruction(Args, BB);
1001 if (!BB->getTerminator())
1002 error("Non-terminated basic block found!");
1004 if (Handler) Handler->handleBasicBlockEnd(BlockNo-1);
1010 /// Parse a symbol table. This works for both module level and function
1011 /// level symbol tables. For function level symbol tables, the CurrentFunction
1012 /// parameter must be non-zero and the ST parameter must correspond to
1013 /// CurrentFunction's symbol table. For Module level symbol tables, the
1014 /// CurrentFunction argument must be zero.
1015 void BytecodeReader::ParseSymbolTable(Function *CurrentFunction,
1017 if (Handler) Handler->handleSymbolTableBegin(CurrentFunction,ST);
1019 // Allow efficient basic block lookup by number.
1020 std::vector<BasicBlock*> BBMap;
1021 if (CurrentFunction)
1022 for (Function::iterator I = CurrentFunction->begin(),
1023 E = CurrentFunction->end(); I != E; ++I)
1026 /// In LLVM 1.3 we write types separately from values so
1027 /// The types are always first in the symbol table. This is
1028 /// because Type no longer derives from Value.
1029 if (!hasTypeDerivedFromValue) {
1030 // Symtab block header: [num entries]
1031 unsigned NumEntries = read_vbr_uint();
1032 for (unsigned i = 0; i < NumEntries; ++i) {
1033 // Symtab entry: [def slot #][name]
1034 unsigned slot = read_vbr_uint();
1035 std::string Name = read_str();
1036 const Type* T = getType(slot);
1037 ST->insert(Name, T);
1041 while (moreInBlock()) {
1042 // Symtab block header: [num entries][type id number]
1043 unsigned NumEntries = read_vbr_uint();
1045 bool isTypeType = read_typeid(Typ);
1046 const Type *Ty = getType(Typ);
1048 for (unsigned i = 0; i != NumEntries; ++i) {
1049 // Symtab entry: [def slot #][name]
1050 unsigned slot = read_vbr_uint();
1051 std::string Name = read_str();
1053 // if we're reading a pre 1.3 bytecode file and the type plane
1054 // is the "type type", handle it here
1056 const Type* T = getType(slot);
1058 error("Failed type look-up for name '" + Name + "'");
1059 ST->insert(Name, T);
1060 continue; // code below must be short circuited
1063 if (Typ == Type::LabelTyID) {
1064 if (slot < BBMap.size())
1067 V = getValue(Typ, slot, false); // Find mapping...
1070 error("Failed value look-up for name '" + Name + "'");
1075 checkPastBlockEnd("Symbol Table");
1076 if (Handler) Handler->handleSymbolTableEnd();
1079 /// Read in the types portion of a compaction table.
1080 void BytecodeReader::ParseCompactionTypes(unsigned NumEntries) {
1081 for (unsigned i = 0; i != NumEntries; ++i) {
1082 unsigned TypeSlot = 0;
1083 if (read_typeid(TypeSlot))
1084 error("Invalid type in compaction table: type type");
1085 const Type *Typ = getGlobalTableType(TypeSlot);
1086 CompactionTypes.push_back(std::make_pair(Typ, TypeSlot));
1087 if (Handler) Handler->handleCompactionTableType(i, TypeSlot, Typ);
1091 /// Parse a compaction table.
1092 void BytecodeReader::ParseCompactionTable() {
1094 // Notify handler that we're beginning a compaction table.
1095 if (Handler) Handler->handleCompactionTableBegin();
1097 // In LLVM 1.3 Type no longer derives from Value. So,
1098 // we always write them first in the compaction table
1099 // because they can't occupy a "type plane" where the
1101 if (! hasTypeDerivedFromValue) {
1102 unsigned NumEntries = read_vbr_uint();
1103 ParseCompactionTypes(NumEntries);
1106 // Compaction tables live in separate blocks so we have to loop
1107 // until we've read the whole thing.
1108 while (moreInBlock()) {
1109 // Read the number of Value* entries in the compaction table
1110 unsigned NumEntries = read_vbr_uint();
1112 unsigned isTypeType = false;
1114 // Decode the type from value read in. Most compaction table
1115 // planes will have one or two entries in them. If that's the
1116 // case then the length is encoded in the bottom two bits and
1117 // the higher bits encode the type. This saves another VBR value.
1118 if ((NumEntries & 3) == 3) {
1119 // In this case, both low-order bits are set (value 3). This
1120 // is a signal that the typeid follows.
1122 isTypeType = read_typeid(Ty);
1124 // In this case, the low-order bits specify the number of entries
1125 // and the high order bits specify the type.
1126 Ty = NumEntries >> 2;
1127 isTypeType = sanitizeTypeId(Ty);
1131 // if we're reading a pre 1.3 bytecode file and the type plane
1132 // is the "type type", handle it here
1134 ParseCompactionTypes(NumEntries);
1136 // Make sure we have enough room for the plane.
1137 if (Ty >= CompactionValues.size())
1138 CompactionValues.resize(Ty+1);
1140 // Make sure the plane is empty or we have some kind of error.
1141 if (!CompactionValues[Ty].empty())
1142 error("Compaction table plane contains multiple entries!");
1144 // Notify handler about the plane.
1145 if (Handler) Handler->handleCompactionTablePlane(Ty, NumEntries);
1147 // Push the implicit zero.
1148 CompactionValues[Ty].push_back(Constant::getNullValue(getType(Ty)));
1150 // Read in each of the entries, put them in the compaction table
1151 // and notify the handler that we have a new compaction table value.
1152 for (unsigned i = 0; i != NumEntries; ++i) {
1153 unsigned ValSlot = read_vbr_uint();
1154 Value *V = getGlobalTableValue(Ty, ValSlot);
1155 CompactionValues[Ty].push_back(V);
1156 if (Handler) Handler->handleCompactionTableValue(i, Ty, ValSlot);
1160 // Notify handler that the compaction table is done.
1161 if (Handler) Handler->handleCompactionTableEnd();
1164 // Parse a single type. The typeid is read in first. If its a primitive type
1165 // then nothing else needs to be read, we know how to instantiate it. If its
1166 // a derived type, then additional data is read to fill out the type
1168 const Type *BytecodeReader::ParseType() {
1169 unsigned PrimType = 0;
1170 if (read_typeid(PrimType))
1171 error("Invalid type (type type) in type constants!");
1173 const Type *Result = 0;
1174 if ((Result = Type::getPrimitiveType((Type::TypeID)PrimType)))
1178 case Type::FunctionTyID: {
1179 const Type *RetType = readSanitizedType();
1181 unsigned NumParams = read_vbr_uint();
1183 std::vector<const Type*> Params;
1185 Params.push_back(readSanitizedType());
1187 bool isVarArg = Params.size() && Params.back() == Type::VoidTy;
1188 if (isVarArg) Params.pop_back();
1190 Result = FunctionType::get(RetType, Params, isVarArg);
1193 case Type::ArrayTyID: {
1194 const Type *ElementType = readSanitizedType();
1195 unsigned NumElements = read_vbr_uint();
1196 Result = ArrayType::get(ElementType, NumElements);
1199 case Type::PackedTyID: {
1200 const Type *ElementType = readSanitizedType();
1201 unsigned NumElements = read_vbr_uint();
1202 Result = PackedType::get(ElementType, NumElements);
1205 case Type::StructTyID: {
1206 std::vector<const Type*> Elements;
1208 if (read_typeid(Typ))
1209 error("Invalid element type (type type) for structure!");
1211 while (Typ) { // List is terminated by void/0 typeid
1212 Elements.push_back(getType(Typ));
1213 if (read_typeid(Typ))
1214 error("Invalid element type (type type) for structure!");
1217 Result = StructType::get(Elements);
1220 case Type::PointerTyID: {
1221 Result = PointerType::get(readSanitizedType());
1225 case Type::OpaqueTyID: {
1226 Result = OpaqueType::get();
1231 error("Don't know how to deserialize primitive type " + utostr(PrimType));
1234 if (Handler) Handler->handleType(Result);
1238 // ParseTypes - We have to use this weird code to handle recursive
1239 // types. We know that recursive types will only reference the current slab of
1240 // values in the type plane, but they can forward reference types before they
1241 // have been read. For example, Type #0 might be '{ Ty#1 }' and Type #1 might
1242 // be 'Ty#0*'. When reading Type #0, type number one doesn't exist. To fix
1243 // this ugly problem, we pessimistically insert an opaque type for each type we
1244 // are about to read. This means that forward references will resolve to
1245 // something and when we reread the type later, we can replace the opaque type
1246 // with a new resolved concrete type.
1248 void BytecodeReader::ParseTypes(TypeListTy &Tab, unsigned NumEntries){
1249 assert(Tab.size() == 0 && "should not have read type constants in before!");
1251 // Insert a bunch of opaque types to be resolved later...
1252 Tab.reserve(NumEntries);
1253 for (unsigned i = 0; i != NumEntries; ++i)
1254 Tab.push_back(OpaqueType::get());
1257 Handler->handleTypeList(NumEntries);
1259 // Loop through reading all of the types. Forward types will make use of the
1260 // opaque types just inserted.
1262 for (unsigned i = 0; i != NumEntries; ++i) {
1263 const Type* NewTy = ParseType();
1264 const Type* OldTy = Tab[i].get();
1266 error("Couldn't parse type!");
1268 // Don't directly push the new type on the Tab. Instead we want to replace
1269 // the opaque type we previously inserted with the new concrete value. This
1270 // approach helps with forward references to types. The refinement from the
1271 // abstract (opaque) type to the new type causes all uses of the abstract
1272 // type to use the concrete type (NewTy). This will also cause the opaque
1273 // type to be deleted.
1274 cast<DerivedType>(const_cast<Type*>(OldTy))->refineAbstractTypeTo(NewTy);
1276 // This should have replaced the old opaque type with the new type in the
1277 // value table... or with a preexisting type that was already in the system.
1278 // Let's just make sure it did.
1279 assert(Tab[i] != OldTy && "refineAbstractType didn't work!");
1283 /// Parse a single constant value
1284 Constant *BytecodeReader::ParseConstantValue(unsigned TypeID) {
1285 // We must check for a ConstantExpr before switching by type because
1286 // a ConstantExpr can be of any type, and has no explicit value.
1288 // 0 if not expr; numArgs if is expr
1289 unsigned isExprNumArgs = read_vbr_uint();
1291 if (isExprNumArgs) {
1292 // 'undef' is encoded with 'exprnumargs' == 1.
1293 if (!hasNoUndefValue)
1294 if (--isExprNumArgs == 0)
1295 return UndefValue::get(getType(TypeID));
1297 // FIXME: Encoding of constant exprs could be much more compact!
1298 std::vector<Constant*> ArgVec;
1299 ArgVec.reserve(isExprNumArgs);
1300 unsigned Opcode = read_vbr_uint();
1302 // Bytecode files before LLVM 1.4 need have a missing terminator inst.
1303 if (hasNoUnreachableInst) Opcode++;
1305 // Read the slot number and types of each of the arguments
1306 for (unsigned i = 0; i != isExprNumArgs; ++i) {
1307 unsigned ArgValSlot = read_vbr_uint();
1308 unsigned ArgTypeSlot = 0;
1309 if (read_typeid(ArgTypeSlot))
1310 error("Invalid argument type (type type) for constant value");
1312 // Get the arg value from its slot if it exists, otherwise a placeholder
1313 ArgVec.push_back(getConstantValue(ArgTypeSlot, ArgValSlot));
1316 // Construct a ConstantExpr of the appropriate kind
1317 if (isExprNumArgs == 1) { // All one-operand expressions
1318 if (Opcode != Instruction::Cast)
1319 error("Only cast instruction has one argument for ConstantExpr");
1321 Constant* Result = ConstantExpr::getCast(ArgVec[0], getType(TypeID));
1322 if (Handler) Handler->handleConstantExpression(Opcode, ArgVec, Result);
1324 } else if (Opcode == Instruction::GetElementPtr) { // GetElementPtr
1325 std::vector<Constant*> IdxList(ArgVec.begin()+1, ArgVec.end());
1327 if (hasRestrictedGEPTypes) {
1328 const Type *BaseTy = ArgVec[0]->getType();
1329 generic_gep_type_iterator<std::vector<Constant*>::iterator>
1330 GTI = gep_type_begin(BaseTy, IdxList.begin(), IdxList.end()),
1331 E = gep_type_end(BaseTy, IdxList.begin(), IdxList.end());
1332 for (unsigned i = 0; GTI != E; ++GTI, ++i)
1333 if (isa<StructType>(*GTI)) {
1334 if (IdxList[i]->getType() != Type::UByteTy)
1335 error("Invalid index for getelementptr!");
1336 IdxList[i] = ConstantExpr::getCast(IdxList[i], Type::UIntTy);
1340 Constant* Result = ConstantExpr::getGetElementPtr(ArgVec[0], IdxList);
1341 if (Handler) Handler->handleConstantExpression(Opcode, ArgVec, Result);
1343 } else if (Opcode == Instruction::Select) {
1344 if (ArgVec.size() != 3)
1345 error("Select instruction must have three arguments.");
1346 Constant* Result = ConstantExpr::getSelect(ArgVec[0], ArgVec[1],
1348 if (Handler) Handler->handleConstantExpression(Opcode, ArgVec, Result);
1350 } else { // All other 2-operand expressions
1351 Constant* Result = ConstantExpr::get(Opcode, ArgVec[0], ArgVec[1]);
1352 if (Handler) Handler->handleConstantExpression(Opcode, ArgVec, Result);
1357 // Ok, not an ConstantExpr. We now know how to read the given type...
1358 const Type *Ty = getType(TypeID);
1359 switch (Ty->getTypeID()) {
1360 case Type::BoolTyID: {
1361 unsigned Val = read_vbr_uint();
1362 if (Val != 0 && Val != 1)
1363 error("Invalid boolean value read.");
1364 Constant* Result = ConstantBool::get(Val == 1);
1365 if (Handler) Handler->handleConstantValue(Result);
1369 case Type::UByteTyID: // Unsigned integer types...
1370 case Type::UShortTyID:
1371 case Type::UIntTyID: {
1372 unsigned Val = read_vbr_uint();
1373 if (!ConstantUInt::isValueValidForType(Ty, Val))
1374 error("Invalid unsigned byte/short/int read.");
1375 Constant* Result = ConstantUInt::get(Ty, Val);
1376 if (Handler) Handler->handleConstantValue(Result);
1380 case Type::ULongTyID: {
1381 Constant* Result = ConstantUInt::get(Ty, read_vbr_uint64());
1382 if (Handler) Handler->handleConstantValue(Result);
1386 case Type::SByteTyID: // Signed integer types...
1387 case Type::ShortTyID:
1388 case Type::IntTyID: {
1389 case Type::LongTyID:
1390 int64_t Val = read_vbr_int64();
1391 if (!ConstantSInt::isValueValidForType(Ty, Val))
1392 error("Invalid signed byte/short/int/long read.");
1393 Constant* Result = ConstantSInt::get(Ty, Val);
1394 if (Handler) Handler->handleConstantValue(Result);
1398 case Type::FloatTyID: {
1401 Constant* Result = ConstantFP::get(Ty, Val);
1402 if (Handler) Handler->handleConstantValue(Result);
1406 case Type::DoubleTyID: {
1409 Constant* Result = ConstantFP::get(Ty, Val);
1410 if (Handler) Handler->handleConstantValue(Result);
1414 case Type::ArrayTyID: {
1415 const ArrayType *AT = cast<ArrayType>(Ty);
1416 unsigned NumElements = AT->getNumElements();
1417 unsigned TypeSlot = getTypeSlot(AT->getElementType());
1418 std::vector<Constant*> Elements;
1419 Elements.reserve(NumElements);
1420 while (NumElements--) // Read all of the elements of the constant.
1421 Elements.push_back(getConstantValue(TypeSlot,
1423 Constant* Result = ConstantArray::get(AT, Elements);
1424 if (Handler) Handler->handleConstantArray(AT, Elements, TypeSlot, Result);
1428 case Type::StructTyID: {
1429 const StructType *ST = cast<StructType>(Ty);
1431 std::vector<Constant *> Elements;
1432 Elements.reserve(ST->getNumElements());
1433 for (unsigned i = 0; i != ST->getNumElements(); ++i)
1434 Elements.push_back(getConstantValue(ST->getElementType(i),
1437 Constant* Result = ConstantStruct::get(ST, Elements);
1438 if (Handler) Handler->handleConstantStruct(ST, Elements, Result);
1442 case Type::PackedTyID: {
1443 const PackedType *PT = cast<PackedType>(Ty);
1444 unsigned NumElements = PT->getNumElements();
1445 unsigned TypeSlot = getTypeSlot(PT->getElementType());
1446 std::vector<Constant*> Elements;
1447 Elements.reserve(NumElements);
1448 while (NumElements--) // Read all of the elements of the constant.
1449 Elements.push_back(getConstantValue(TypeSlot,
1451 Constant* Result = ConstantPacked::get(PT, Elements);
1452 if (Handler) Handler->handleConstantPacked(PT, Elements, TypeSlot, Result);
1456 case Type::PointerTyID: { // ConstantPointerRef value (backwards compat).
1457 const PointerType *PT = cast<PointerType>(Ty);
1458 unsigned Slot = read_vbr_uint();
1460 // Check to see if we have already read this global variable...
1461 Value *Val = getValue(TypeID, Slot, false);
1463 GlobalValue *GV = dyn_cast<GlobalValue>(Val);
1464 if (!GV) error("GlobalValue not in ValueTable!");
1465 if (Handler) Handler->handleConstantPointer(PT, Slot, GV);
1468 error("Forward references are not allowed here.");
1473 error("Don't know how to deserialize constant value of type '" +
1474 Ty->getDescription());
1480 /// Resolve references for constants. This function resolves the forward
1481 /// referenced constants in the ConstantFwdRefs map. It uses the
1482 /// replaceAllUsesWith method of Value class to substitute the placeholder
1483 /// instance with the actual instance.
1484 void BytecodeReader::ResolveReferencesToConstant(Constant *NewV, unsigned Typ,
1486 ConstantRefsType::iterator I =
1487 ConstantFwdRefs.find(std::make_pair(Typ, Slot));
1488 if (I == ConstantFwdRefs.end()) return; // Never forward referenced?
1490 Value *PH = I->second; // Get the placeholder...
1491 PH->replaceAllUsesWith(NewV);
1492 delete PH; // Delete the old placeholder
1493 ConstantFwdRefs.erase(I); // Remove the map entry for it
1496 /// Parse the constant strings section.
1497 void BytecodeReader::ParseStringConstants(unsigned NumEntries, ValueTable &Tab){
1498 for (; NumEntries; --NumEntries) {
1500 if (read_typeid(Typ))
1501 error("Invalid type (type type) for string constant");
1502 const Type *Ty = getType(Typ);
1503 if (!isa<ArrayType>(Ty))
1504 error("String constant data invalid!");
1506 const ArrayType *ATy = cast<ArrayType>(Ty);
1507 if (ATy->getElementType() != Type::SByteTy &&
1508 ATy->getElementType() != Type::UByteTy)
1509 error("String constant data invalid!");
1511 // Read character data. The type tells us how long the string is.
1512 char *Data = reinterpret_cast<char *>(alloca(ATy->getNumElements()));
1513 read_data(Data, Data+ATy->getNumElements());
1515 std::vector<Constant*> Elements(ATy->getNumElements());
1516 if (ATy->getElementType() == Type::SByteTy)
1517 for (unsigned i = 0, e = ATy->getNumElements(); i != e; ++i)
1518 Elements[i] = ConstantSInt::get(Type::SByteTy, (signed char)Data[i]);
1520 for (unsigned i = 0, e = ATy->getNumElements(); i != e; ++i)
1521 Elements[i] = ConstantUInt::get(Type::UByteTy, (unsigned char)Data[i]);
1523 // Create the constant, inserting it as needed.
1524 Constant *C = ConstantArray::get(ATy, Elements);
1525 unsigned Slot = insertValue(C, Typ, Tab);
1526 ResolveReferencesToConstant(C, Typ, Slot);
1527 if (Handler) Handler->handleConstantString(cast<ConstantArray>(C));
1531 /// Parse the constant pool.
1532 void BytecodeReader::ParseConstantPool(ValueTable &Tab,
1533 TypeListTy &TypeTab,
1535 if (Handler) Handler->handleGlobalConstantsBegin();
1537 /// In LLVM 1.3 Type does not derive from Value so the types
1538 /// do not occupy a plane. Consequently, we read the types
1539 /// first in the constant pool.
1540 if (isFunction && !hasTypeDerivedFromValue) {
1541 unsigned NumEntries = read_vbr_uint();
1542 ParseTypes(TypeTab, NumEntries);
1545 while (moreInBlock()) {
1546 unsigned NumEntries = read_vbr_uint();
1548 bool isTypeType = read_typeid(Typ);
1550 /// In LLVM 1.2 and before, Types were written to the
1551 /// bytecode file in the "Type Type" plane (#12).
1552 /// In 1.3 plane 12 is now the label plane. Handle this here.
1554 ParseTypes(TypeTab, NumEntries);
1555 } else if (Typ == Type::VoidTyID) {
1556 /// Use of Type::VoidTyID is a misnomer. It actually means
1557 /// that the following plane is constant strings
1558 assert(&Tab == &ModuleValues && "Cannot read strings in functions!");
1559 ParseStringConstants(NumEntries, Tab);
1561 for (unsigned i = 0; i < NumEntries; ++i) {
1562 Constant *C = ParseConstantValue(Typ);
1563 assert(C && "ParseConstantValue returned NULL!");
1564 unsigned Slot = insertValue(C, Typ, Tab);
1566 // If we are reading a function constant table, make sure that we adjust
1567 // the slot number to be the real global constant number.
1569 if (&Tab != &ModuleValues && Typ < ModuleValues.size() &&
1571 Slot += ModuleValues[Typ]->size();
1572 ResolveReferencesToConstant(C, Typ, Slot);
1577 // After we have finished parsing the constant pool, we had better not have
1578 // any dangling references left.
1579 if (!ConstantFwdRefs.empty()) {
1580 ConstantRefsType::const_iterator I = ConstantFwdRefs.begin();
1581 Constant* missingConst = I->second;
1582 error(utostr(ConstantFwdRefs.size()) +
1583 " unresolved constant reference exist. First one is '" +
1584 missingConst->getName() + "' of type '" +
1585 missingConst->getType()->getDescription() + "'.");
1588 checkPastBlockEnd("Constant Pool");
1589 if (Handler) Handler->handleGlobalConstantsEnd();
1592 /// Parse the contents of a function. Note that this function can be
1593 /// called lazily by materializeFunction
1594 /// @see materializeFunction
1595 void BytecodeReader::ParseFunctionBody(Function* F) {
1597 unsigned FuncSize = BlockEnd - At;
1598 GlobalValue::LinkageTypes Linkage = GlobalValue::ExternalLinkage;
1600 unsigned LinkageType = read_vbr_uint();
1601 switch (LinkageType) {
1602 case 0: Linkage = GlobalValue::ExternalLinkage; break;
1603 case 1: Linkage = GlobalValue::WeakLinkage; break;
1604 case 2: Linkage = GlobalValue::AppendingLinkage; break;
1605 case 3: Linkage = GlobalValue::InternalLinkage; break;
1606 case 4: Linkage = GlobalValue::LinkOnceLinkage; break;
1608 error("Invalid linkage type for Function.");
1609 Linkage = GlobalValue::InternalLinkage;
1613 F->setLinkage(Linkage);
1614 if (Handler) Handler->handleFunctionBegin(F,FuncSize);
1616 // Keep track of how many basic blocks we have read in...
1617 unsigned BlockNum = 0;
1618 bool InsertedArguments = false;
1620 BufPtr MyEnd = BlockEnd;
1621 while (At < MyEnd) {
1622 unsigned Type, Size;
1624 read_block(Type, Size);
1627 case BytecodeFormat::ConstantPoolBlockID:
1628 if (!InsertedArguments) {
1629 // Insert arguments into the value table before we parse the first basic
1630 // block in the function, but after we potentially read in the
1631 // compaction table.
1633 InsertedArguments = true;
1636 ParseConstantPool(FunctionValues, FunctionTypes, true);
1639 case BytecodeFormat::CompactionTableBlockID:
1640 ParseCompactionTable();
1643 case BytecodeFormat::BasicBlock: {
1644 if (!InsertedArguments) {
1645 // Insert arguments into the value table before we parse the first basic
1646 // block in the function, but after we potentially read in the
1647 // compaction table.
1649 InsertedArguments = true;
1652 BasicBlock *BB = ParseBasicBlock(BlockNum++);
1653 F->getBasicBlockList().push_back(BB);
1657 case BytecodeFormat::InstructionListBlockID: {
1658 // Insert arguments into the value table before we parse the instruction
1659 // list for the function, but after we potentially read in the compaction
1661 if (!InsertedArguments) {
1663 InsertedArguments = true;
1667 error("Already parsed basic blocks!");
1668 BlockNum = ParseInstructionList(F);
1672 case BytecodeFormat::SymbolTableBlockID:
1673 ParseSymbolTable(F, &F->getSymbolTable());
1679 error("Wrapped around reading bytecode.");
1684 // Malformed bc file if read past end of block.
1688 // Make sure there were no references to non-existant basic blocks.
1689 if (BlockNum != ParsedBasicBlocks.size())
1690 error("Illegal basic block operand reference");
1692 ParsedBasicBlocks.clear();
1694 // Resolve forward references. Replace any uses of a forward reference value
1695 // with the real value.
1696 while (!ForwardReferences.empty()) {
1697 std::map<std::pair<unsigned,unsigned>, Value*>::iterator
1698 I = ForwardReferences.begin();
1699 Value *V = getValue(I->first.first, I->first.second, false);
1700 Value *PlaceHolder = I->second;
1701 PlaceHolder->replaceAllUsesWith(V);
1702 ForwardReferences.erase(I);
1706 // Clear out function-level types...
1707 FunctionTypes.clear();
1708 CompactionTypes.clear();
1709 CompactionValues.clear();
1710 freeTable(FunctionValues);
1712 if (Handler) Handler->handleFunctionEnd(F);
1715 /// This function parses LLVM functions lazily. It obtains the type of the
1716 /// function and records where the body of the function is in the bytecode
1717 /// buffer. The caller can then use the ParseNextFunction and
1718 /// ParseAllFunctionBodies to get handler events for the functions.
1719 void BytecodeReader::ParseFunctionLazily() {
1720 if (FunctionSignatureList.empty())
1721 error("FunctionSignatureList empty!");
1723 Function *Func = FunctionSignatureList.back();
1724 FunctionSignatureList.pop_back();
1726 // Save the information for future reading of the function
1727 LazyFunctionLoadMap[Func] = LazyFunctionInfo(BlockStart, BlockEnd);
1729 // This function has a body but it's not loaded so it appears `External'.
1730 // Mark it as a `Ghost' instead to notify the users that it has a body.
1731 Func->setLinkage(GlobalValue::GhostLinkage);
1733 // Pretend we've `parsed' this function
1737 /// The ParserFunction method lazily parses one function. Use this method to
1738 /// casue the parser to parse a specific function in the module. Note that
1739 /// this will remove the function from what is to be included by
1740 /// ParseAllFunctionBodies.
1741 /// @see ParseAllFunctionBodies
1742 /// @see ParseBytecode
1743 void BytecodeReader::ParseFunction(Function* Func) {
1744 // Find {start, end} pointers and slot in the map. If not there, we're done.
1745 LazyFunctionMap::iterator Fi = LazyFunctionLoadMap.find(Func);
1747 // Make sure we found it
1748 if (Fi == LazyFunctionLoadMap.end()) {
1749 error("Unrecognized function of type " + Func->getType()->getDescription());
1753 BlockStart = At = Fi->second.Buf;
1754 BlockEnd = Fi->second.EndBuf;
1755 assert(Fi->first == Func && "Found wrong function?");
1757 LazyFunctionLoadMap.erase(Fi);
1759 this->ParseFunctionBody(Func);
1762 /// The ParseAllFunctionBodies method parses through all the previously
1763 /// unparsed functions in the bytecode file. If you want to completely parse
1764 /// a bytecode file, this method should be called after Parsebytecode because
1765 /// Parsebytecode only records the locations in the bytecode file of where
1766 /// the function definitions are located. This function uses that information
1767 /// to materialize the functions.
1768 /// @see ParseBytecode
1769 void BytecodeReader::ParseAllFunctionBodies() {
1770 LazyFunctionMap::iterator Fi = LazyFunctionLoadMap.begin();
1771 LazyFunctionMap::iterator Fe = LazyFunctionLoadMap.end();
1774 Function* Func = Fi->first;
1775 BlockStart = At = Fi->second.Buf;
1776 BlockEnd = Fi->second.EndBuf;
1777 ParseFunctionBody(Func);
1780 LazyFunctionLoadMap.clear();
1783 /// Parse the global type list
1784 void BytecodeReader::ParseGlobalTypes() {
1785 // Read the number of types
1786 unsigned NumEntries = read_vbr_uint();
1788 // Ignore the type plane identifier for types if the bc file is pre 1.3
1789 if (hasTypeDerivedFromValue)
1792 ParseTypes(ModuleTypes, NumEntries);
1795 /// Parse the Global info (types, global vars, constants)
1796 void BytecodeReader::ParseModuleGlobalInfo() {
1798 if (Handler) Handler->handleModuleGlobalsBegin();
1800 // Read global variables...
1801 unsigned VarType = read_vbr_uint();
1802 while (VarType != Type::VoidTyID) { // List is terminated by Void
1803 // VarType Fields: bit0 = isConstant, bit1 = hasInitializer, bit2,3,4 =
1804 // Linkage, bit4+ = slot#
1805 unsigned SlotNo = VarType >> 5;
1806 if (sanitizeTypeId(SlotNo))
1807 error("Invalid type (type type) for global var!");
1808 unsigned LinkageID = (VarType >> 2) & 7;
1809 bool isConstant = VarType & 1;
1810 bool hasInitializer = VarType & 2;
1811 GlobalValue::LinkageTypes Linkage;
1813 switch (LinkageID) {
1814 case 0: Linkage = GlobalValue::ExternalLinkage; break;
1815 case 1: Linkage = GlobalValue::WeakLinkage; break;
1816 case 2: Linkage = GlobalValue::AppendingLinkage; break;
1817 case 3: Linkage = GlobalValue::InternalLinkage; break;
1818 case 4: Linkage = GlobalValue::LinkOnceLinkage; break;
1820 error("Unknown linkage type: " + utostr(LinkageID));
1821 Linkage = GlobalValue::InternalLinkage;
1825 const Type *Ty = getType(SlotNo);
1827 error("Global has no type! SlotNo=" + utostr(SlotNo));
1830 if (!isa<PointerType>(Ty)) {
1831 error("Global not a pointer type! Ty= " + Ty->getDescription());
1834 const Type *ElTy = cast<PointerType>(Ty)->getElementType();
1836 // Create the global variable...
1837 GlobalVariable *GV = new GlobalVariable(ElTy, isConstant, Linkage,
1839 insertValue(GV, SlotNo, ModuleValues);
1841 unsigned initSlot = 0;
1842 if (hasInitializer) {
1843 initSlot = read_vbr_uint();
1844 GlobalInits.push_back(std::make_pair(GV, initSlot));
1847 // Notify handler about the global value.
1849 Handler->handleGlobalVariable(ElTy, isConstant, Linkage, SlotNo,initSlot);
1852 VarType = read_vbr_uint();
1855 // Read the function objects for all of the functions that are coming
1856 unsigned FnSignature = read_vbr_uint();
1858 if (hasNoFlagsForFunctions)
1859 FnSignature = (FnSignature << 5) + 1;
1861 // List is terminated by VoidTy.
1862 while ((FnSignature >> 5) != Type::VoidTyID) {
1863 const Type *Ty = getType(FnSignature >> 5);
1864 if (!isa<PointerType>(Ty) ||
1865 !isa<FunctionType>(cast<PointerType>(Ty)->getElementType())) {
1866 error("Function not a pointer to function type! Ty = " +
1867 Ty->getDescription());
1870 // We create functions by passing the underlying FunctionType to create...
1871 const FunctionType* FTy =
1872 cast<FunctionType>(cast<PointerType>(Ty)->getElementType());
1875 // Insert the place holder.
1876 Function* Func = new Function(FTy, GlobalValue::ExternalLinkage,
1878 insertValue(Func, FnSignature >> 5, ModuleValues);
1880 // Flags are not used yet.
1881 unsigned Flags = FnSignature & 31;
1883 // Save this for later so we know type of lazily instantiated functions.
1884 // Note that known-external functions do not have FunctionInfo blocks, so we
1885 // do not add them to the FunctionSignatureList.
1886 if ((Flags & (1 << 4)) == 0)
1887 FunctionSignatureList.push_back(Func);
1889 // Look at the low bits. If there is a calling conv here, apply it,
1890 // read it as a vbr.
1893 Func->setCallingConv(Flags-1);
1895 Func->setCallingConv(read_vbr_uint());
1897 if (Handler) Handler->handleFunctionDeclaration(Func);
1899 // Get the next function signature.
1900 FnSignature = read_vbr_uint();
1901 if (hasNoFlagsForFunctions)
1902 FnSignature = (FnSignature << 5) + 1;
1905 // Now that the function signature list is set up, reverse it so that we can
1906 // remove elements efficiently from the back of the vector.
1907 std::reverse(FunctionSignatureList.begin(), FunctionSignatureList.end());
1909 // If this bytecode format has dependent library information in it ..
1910 if (!hasNoDependentLibraries) {
1911 // Read in the number of dependent library items that follow
1912 unsigned num_dep_libs = read_vbr_uint();
1913 std::string dep_lib;
1914 while( num_dep_libs-- ) {
1915 dep_lib = read_str();
1916 TheModule->addLibrary(dep_lib);
1918 Handler->handleDependentLibrary(dep_lib);
1922 // Read target triple and place into the module
1923 std::string triple = read_str();
1924 TheModule->setTargetTriple(triple);
1926 Handler->handleTargetTriple(triple);
1929 if (hasInconsistentModuleGlobalInfo)
1932 // This is for future proofing... in the future extra fields may be added that
1933 // we don't understand, so we transparently ignore them.
1937 if (Handler) Handler->handleModuleGlobalsEnd();
1940 /// Parse the version information and decode it by setting flags on the
1941 /// Reader that enable backward compatibility of the reader.
1942 void BytecodeReader::ParseVersionInfo() {
1943 unsigned Version = read_vbr_uint();
1945 // Unpack version number: low four bits are for flags, top bits = version
1946 Module::Endianness Endianness;
1947 Module::PointerSize PointerSize;
1948 Endianness = (Version & 1) ? Module::BigEndian : Module::LittleEndian;
1949 PointerSize = (Version & 2) ? Module::Pointer64 : Module::Pointer32;
1951 bool hasNoEndianness = Version & 4;
1952 bool hasNoPointerSize = Version & 8;
1954 RevisionNum = Version >> 4;
1956 // Default values for the current bytecode version
1957 hasInconsistentModuleGlobalInfo = false;
1958 hasExplicitPrimitiveZeros = false;
1959 hasRestrictedGEPTypes = false;
1960 hasTypeDerivedFromValue = false;
1961 hasLongBlockHeaders = false;
1962 has32BitTypes = false;
1963 hasNoDependentLibraries = false;
1964 hasAlignment = false;
1965 hasInconsistentBBSlotNums = false;
1966 hasVBRByteTypes = false;
1967 hasUnnecessaryModuleBlockId = false;
1968 hasNoUndefValue = false;
1969 hasNoFlagsForFunctions = false;
1970 hasNoUnreachableInst = false;
1972 switch (RevisionNum) {
1973 case 0: // LLVM 1.0, 1.1 (Released)
1974 // Base LLVM 1.0 bytecode format.
1975 hasInconsistentModuleGlobalInfo = true;
1976 hasExplicitPrimitiveZeros = true;
1980 case 1: // LLVM 1.2 (Released)
1981 // LLVM 1.2 added explicit support for emitting strings efficiently.
1983 // Also, it fixed the problem where the size of the ModuleGlobalInfo block
1984 // included the size for the alignment at the end, where the rest of the
1987 // LLVM 1.2 and before required that GEP indices be ubyte constants for
1988 // structures and longs for sequential types.
1989 hasRestrictedGEPTypes = true;
1991 // LLVM 1.2 and before had the Type class derive from Value class. This
1992 // changed in release 1.3 and consequently LLVM 1.3 bytecode files are
1993 // written differently because Types can no longer be part of the
1994 // type planes for Values.
1995 hasTypeDerivedFromValue = true;
1999 case 2: // 1.2.5 (Not Released)
2001 // LLVM 1.2 and earlier had two-word block headers. This is a bit wasteful,
2002 // especially for small files where the 8 bytes per block is a large
2003 // fraction of the total block size. In LLVM 1.3, the block type and length
2004 // are compressed into a single 32-bit unsigned integer. 27 bits for length,
2005 // 5 bits for block type.
2006 hasLongBlockHeaders = true;
2008 // LLVM 1.2 and earlier wrote type slot numbers as vbr_uint32. In LLVM 1.3
2009 // this has been reduced to vbr_uint24. It shouldn't make much difference
2010 // since we haven't run into a module with > 24 million types, but for
2011 // safety the 24-bit restriction has been enforced in 1.3 to free some bits
2012 // in various places and to ensure consistency.
2013 has32BitTypes = true;
2015 // LLVM 1.2 and earlier did not provide a target triple nor a list of
2016 // libraries on which the bytecode is dependent. LLVM 1.3 provides these
2017 // features, for use in future versions of LLVM.
2018 hasNoDependentLibraries = true;
2022 case 3: // LLVM 1.3 (Released)
2023 // LLVM 1.3 and earlier caused alignment bytes to be written on some block
2024 // boundaries and at the end of some strings. In extreme cases (e.g. lots
2025 // of GEP references to a constant array), this can increase the file size
2026 // by 30% or more. In version 1.4 alignment is done away with completely.
2027 hasAlignment = true;
2031 case 4: // 1.3.1 (Not Released)
2032 // In version 4, we did not support the 'undef' constant.
2033 hasNoUndefValue = true;
2035 // In version 4 and above, we did not include space for flags for functions
2036 // in the module info block.
2037 hasNoFlagsForFunctions = true;
2039 // In version 4 and above, we did not include the 'unreachable' instruction
2040 // in the opcode numbering in the bytecode file.
2041 hasNoUnreachableInst = true;
2046 case 5: // 1.x.x (Not Released)
2048 // FIXME: NONE of this is implemented yet!
2050 // In version 5, basic blocks have a minimum index of 0 whereas all the
2051 // other primitives have a minimum index of 1 (because 0 is the "null"
2052 // value. In version 5, we made this consistent.
2053 hasInconsistentBBSlotNums = true;
2055 // In version 5, the types SByte and UByte were encoded as vbr_uint so that
2056 // signed values > 63 and unsigned values >127 would be encoded as two
2057 // bytes. In version 5, they are encoded directly in a single byte.
2058 hasVBRByteTypes = true;
2060 // In version 5, modules begin with a "Module Block" which encodes a 4-byte
2061 // integer value 0x01 to identify the module block. This is unnecessary and
2062 // removed in version 5.
2063 hasUnnecessaryModuleBlockId = true;
2066 error("Unknown bytecode version number: " + itostr(RevisionNum));
2069 if (hasNoEndianness) Endianness = Module::AnyEndianness;
2070 if (hasNoPointerSize) PointerSize = Module::AnyPointerSize;
2072 TheModule->setEndianness(Endianness);
2073 TheModule->setPointerSize(PointerSize);
2075 if (Handler) Handler->handleVersionInfo(RevisionNum, Endianness, PointerSize);
2078 /// Parse a whole module.
2079 void BytecodeReader::ParseModule() {
2080 unsigned Type, Size;
2082 FunctionSignatureList.clear(); // Just in case...
2084 // Read into instance variables...
2088 bool SeenModuleGlobalInfo = false;
2089 bool SeenGlobalTypePlane = false;
2090 BufPtr MyEnd = BlockEnd;
2091 while (At < MyEnd) {
2093 read_block(Type, Size);
2097 case BytecodeFormat::GlobalTypePlaneBlockID:
2098 if (SeenGlobalTypePlane)
2099 error("Two GlobalTypePlane Blocks Encountered!");
2103 SeenGlobalTypePlane = true;
2106 case BytecodeFormat::ModuleGlobalInfoBlockID:
2107 if (SeenModuleGlobalInfo)
2108 error("Two ModuleGlobalInfo Blocks Encountered!");
2109 ParseModuleGlobalInfo();
2110 SeenModuleGlobalInfo = true;
2113 case BytecodeFormat::ConstantPoolBlockID:
2114 ParseConstantPool(ModuleValues, ModuleTypes,false);
2117 case BytecodeFormat::FunctionBlockID:
2118 ParseFunctionLazily();
2121 case BytecodeFormat::SymbolTableBlockID:
2122 ParseSymbolTable(0, &TheModule->getSymbolTable());
2128 error("Unexpected Block of Type #" + utostr(Type) + " encountered!");
2136 // After the module constant pool has been read, we can safely initialize
2137 // global variables...
2138 while (!GlobalInits.empty()) {
2139 GlobalVariable *GV = GlobalInits.back().first;
2140 unsigned Slot = GlobalInits.back().second;
2141 GlobalInits.pop_back();
2143 // Look up the initializer value...
2144 // FIXME: Preserve this type ID!
2146 const llvm::PointerType* GVType = GV->getType();
2147 unsigned TypeSlot = getTypeSlot(GVType->getElementType());
2148 if (Constant *CV = getConstantValue(TypeSlot, Slot)) {
2149 if (GV->hasInitializer())
2150 error("Global *already* has an initializer?!");
2151 if (Handler) Handler->handleGlobalInitializer(GV,CV);
2152 GV->setInitializer(CV);
2154 error("Cannot find initializer value.");
2157 if (!ConstantFwdRefs.empty())
2158 error("Use of undefined constants in a module");
2160 /// Make sure we pulled them all out. If we didn't then there's a declaration
2161 /// but a missing body. That's not allowed.
2162 if (!FunctionSignatureList.empty())
2163 error("Function declared, but bytecode stream ended before definition");
2166 /// This function completely parses a bytecode buffer given by the \p Buf
2167 /// and \p Length parameters.
2168 void BytecodeReader::ParseBytecode(BufPtr Buf, unsigned Length,
2169 const std::string &ModuleID) {
2173 At = MemStart = BlockStart = Buf;
2174 MemEnd = BlockEnd = Buf + Length;
2176 // Create the module
2177 TheModule = new Module(ModuleID);
2179 if (Handler) Handler->handleStart(TheModule, Length);
2181 // Read the four bytes of the signature.
2182 unsigned Sig = read_uint();
2184 // If this is a compressed file
2185 if (Sig == ('l' | ('l' << 8) | ('v' << 16) | ('c' << 24))) {
2187 // Invoke the decompression of the bytecode. Note that we have to skip the
2188 // file's magic number which is not part of the compressed block. Hence,
2189 // the Buf+4 and Length-4. The result goes into decompressedBlock, a data
2190 // member for retention until BytecodeReader is destructed.
2191 unsigned decompressedLength = Compressor::decompressToNewBuffer(
2192 (char*)Buf+4,Length-4,decompressedBlock);
2194 // We must adjust the buffer pointers used by the bytecode reader to point
2195 // into the new decompressed block. After decompression, the
2196 // decompressedBlock will point to a contiguous memory area that has
2197 // the decompressed data.
2198 At = MemStart = BlockStart = Buf = (BufPtr) decompressedBlock;
2199 MemEnd = BlockEnd = Buf + decompressedLength;
2201 // else if this isn't a regular (uncompressed) bytecode file, then its
2202 // and error, generate that now.
2203 } else if (Sig != ('l' | ('l' << 8) | ('v' << 16) | ('m' << 24))) {
2204 error("Invalid bytecode signature: " + utohexstr(Sig));
2207 // Tell the handler we're starting a module
2208 if (Handler) Handler->handleModuleBegin(ModuleID);
2210 // Get the module block and size and verify. This is handled specially
2211 // because the module block/size is always written in long format. Other
2212 // blocks are written in short format so the read_block method is used.
2213 unsigned Type, Size;
2216 if (Type != BytecodeFormat::ModuleBlockID) {
2217 error("Expected Module Block! Type:" + utostr(Type) + ", Size:"
2221 // It looks like the darwin ranlib program is broken, and adds trailing
2222 // garbage to the end of some bytecode files. This hack allows the bc
2223 // reader to ignore trailing garbage on bytecode files.
2224 if (At + Size < MemEnd)
2225 MemEnd = BlockEnd = At+Size;
2227 if (At + Size != MemEnd)
2228 error("Invalid Top Level Block Length! Type:" + utostr(Type)
2229 + ", Size:" + utostr(Size));
2231 // Parse the module contents
2232 this->ParseModule();
2234 // Check for missing functions
2236 error("Function expected, but bytecode stream ended!");
2238 // Tell the handler we're done with the module
2240 Handler->handleModuleEnd(ModuleID);
2242 // Tell the handler we're finished the parse
2243 if (Handler) Handler->handleFinish();
2245 } catch (std::string& errstr) {
2246 if (Handler) Handler->handleError(errstr);
2250 if (decompressedBlock != 0 ) {
2251 ::free(decompressedBlock);
2252 decompressedBlock = 0;
2256 std::string msg("Unknown Exception Occurred");
2257 if (Handler) Handler->handleError(msg);
2261 if (decompressedBlock != 0) {
2262 ::free(decompressedBlock);
2263 decompressedBlock = 0;
2269 //===----------------------------------------------------------------------===//
2270 //=== Default Implementations of Handler Methods
2271 //===----------------------------------------------------------------------===//
2273 BytecodeHandler::~BytecodeHandler() {}