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 //===----------------------------------------------------------------------===//
19 #include "ReaderInternals.h"
20 #include "llvm/Bytecode/Reader.h"
21 #include "llvm/Bytecode/Format.h"
22 #include "llvm/Module.h"
23 #include "Support/StringExtras.h"
26 unsigned BytecodeParser::getTypeSlot(const Type *Ty) {
27 if (Ty->isPrimitiveType())
28 return Ty->getPrimitiveID();
30 // Scan the compaction table for the type if needed.
31 if (CompactionTable.size() > Type::TypeTyID) {
32 std::vector<Value*> &Plane = CompactionTable[Type::TypeTyID];
34 std::vector<Value*>::iterator I = find(Plane.begin(), Plane.end(),
35 const_cast<Type*>(Ty));
37 throw std::string("Couldn't find type specified in compaction table!");
38 return Type::FirstDerivedTyID + (&*I - &Plane[0]);
42 // Check the function level types first...
43 TypeValuesListTy::iterator I = find(FunctionTypeValues.begin(),
44 FunctionTypeValues.end(), Ty);
45 if (I != FunctionTypeValues.end())
46 return Type::FirstDerivedTyID + ModuleTypeValues.size() +
47 (&*I - &FunctionTypeValues[0]);
49 I = find(ModuleTypeValues.begin(), ModuleTypeValues.end(), Ty);
50 if (I == ModuleTypeValues.end())
51 throw std::string("Didn't find type in ModuleTypeValues.");
52 return Type::FirstDerivedTyID + (&*I - &ModuleTypeValues[0]);
55 const Type *BytecodeParser::getType(unsigned ID) {
56 //cerr << "Looking up Type ID: " << ID << "\n";
58 if (ID < Type::FirstDerivedTyID)
59 if (const Type *T = Type::getPrimitiveType((Type::PrimitiveID)ID))
60 return T; // Asked for a primitive type...
62 // Otherwise, derived types need offset...
63 ID -= Type::FirstDerivedTyID;
65 if (CompactionTable.size() > Type::TypeTyID &&
66 !CompactionTable[Type::TypeTyID].empty()) {
67 if (ID >= CompactionTable[Type::TypeTyID].size())
68 throw std::string("Type ID out of range for compaction table!");
69 return cast<Type>(CompactionTable[Type::TypeTyID][ID]);
72 // Is it a module-level type?
73 if (ID < ModuleTypeValues.size())
74 return ModuleTypeValues[ID].get();
76 // Nope, is it a function-level type?
77 ID -= ModuleTypeValues.size();
78 if (ID < FunctionTypeValues.size())
79 return FunctionTypeValues[ID].get();
81 throw std::string("Illegal type reference!");
84 static inline bool hasImplicitNull(unsigned TyID, bool EncodesPrimitiveZeros) {
85 if (!EncodesPrimitiveZeros)
86 return TyID != Type::LabelTyID && TyID != Type::TypeTyID &&
87 TyID != Type::VoidTyID;
88 return TyID >= Type::FirstDerivedTyID;
91 unsigned BytecodeParser::insertValue(Value *Val, unsigned type,
92 ValueTable &ValueTab) {
93 assert((!isa<Constant>(Val) || !cast<Constant>(Val)->isNullValue()) ||
94 !hasImplicitNull(type, hasExplicitPrimitiveZeros) &&
95 "Cannot read null values from bytecode!");
96 assert(type != Type::TypeTyID && "Types should never be insertValue'd!");
98 if (ValueTab.size() <= type)
99 ValueTab.resize(type+1);
101 if (!ValueTab[type]) ValueTab[type] = new ValueList();
103 //cerr << "insertValue Values[" << type << "][" << ValueTab[type].size()
104 // << "] = " << Val << "\n";
105 ValueTab[type]->push_back(Val);
107 bool HasOffset = hasImplicitNull(type, hasExplicitPrimitiveZeros);
108 return ValueTab[type]->size()-1 + HasOffset;
111 Value *BytecodeParser::getValue(unsigned type, unsigned oNum, bool Create) {
112 assert(type != Type::TypeTyID && "getValue() cannot get types!");
113 assert(type != Type::LabelTyID && "getValue() cannot get blocks!");
116 // If there is a compaction table active, it defines the low-level numbers.
117 // If not, the module values define the low-level numbers.
118 if (CompactionTable.size() > type && !CompactionTable[type].empty()) {
119 if (Num < CompactionTable[type].size())
120 return CompactionTable[type][Num];
121 Num -= CompactionTable[type].size();
123 // If the type plane was compactified, figure out the global type ID.
124 unsigned GlobalTyID = type;
125 if (CompactionTable.size() > Type::TypeTyID &&
126 !CompactionTable[Type::TypeTyID].empty() &&
127 type >= Type::FirstDerivedTyID) {
128 std::vector<Value*> &TypePlane = CompactionTable[Type::TypeTyID];
129 const Type *Ty = cast<Type>(TypePlane[type-Type::FirstDerivedTyID]);
130 TypeValuesListTy::iterator I =
131 find(ModuleTypeValues.begin(), ModuleTypeValues.end(), Ty);
132 assert(I != ModuleTypeValues.end());
133 GlobalTyID = Type::FirstDerivedTyID + (&*I - &ModuleTypeValues[0]);
136 if (hasImplicitNull(GlobalTyID, hasExplicitPrimitiveZeros)) {
138 return Constant::getNullValue(getType(type));
142 if (GlobalTyID < ModuleValues.size() && ModuleValues[GlobalTyID]) {
143 if (Num < ModuleValues[GlobalTyID]->size())
144 return ModuleValues[GlobalTyID]->getOperand(Num);
145 Num -= ModuleValues[GlobalTyID]->size();
149 if (Values.size() > type && Values[type] && Num < Values[type]->size())
150 return Values[type]->getOperand(Num);
152 if (!Create) return 0; // Do not create a placeholder?
154 std::pair<unsigned,unsigned> KeyValue(type, oNum);
155 std::map<std::pair<unsigned,unsigned>, Value*>::iterator I =
156 ForwardReferences.lower_bound(KeyValue);
157 if (I != ForwardReferences.end() && I->first == KeyValue)
158 return I->second; // We have already created this placeholder
160 Value *Val = new Argument(getType(type));
161 ForwardReferences.insert(I, std::make_pair(KeyValue, Val));
165 /// getBasicBlock - Get a particular numbered basic block, which might be a
166 /// forward reference. This works together with ParseBasicBlock to handle these
167 /// forward references in a clean manner.
169 BasicBlock *BytecodeParser::getBasicBlock(unsigned ID) {
170 // Make sure there is room in the table...
171 if (ParsedBasicBlocks.size() <= ID) ParsedBasicBlocks.resize(ID+1);
173 // First check to see if this is a backwards reference, i.e., ParseBasicBlock
174 // has already created this block, or if the forward reference has already
176 if (ParsedBasicBlocks[ID])
177 return ParsedBasicBlocks[ID];
179 // Otherwise, the basic block has not yet been created. Do so and add it to
180 // the ParsedBasicBlocks list.
181 return ParsedBasicBlocks[ID] = new BasicBlock();
184 /// getConstantValue - Just like getValue, except that it returns a null pointer
185 /// only on error. It always returns a constant (meaning that if the value is
186 /// defined, but is not a constant, that is an error). If the specified
187 /// constant hasn't been parsed yet, a placeholder is defined and used. Later,
188 /// after the real value is parsed, the placeholder is eliminated.
190 Constant *BytecodeParser::getConstantValue(unsigned TypeSlot, unsigned Slot) {
191 if (Value *V = getValue(TypeSlot, Slot, false))
192 if (Constant *C = dyn_cast<Constant>(V))
193 return C; // If we already have the value parsed, just return it
194 else if (GlobalValue *GV = dyn_cast<GlobalValue>(V))
195 // ConstantPointerRef's are an abomination, but at least they don't have
196 // to infest bytecode files.
197 return ConstantPointerRef::get(GV);
199 throw std::string("Reference of a value is expected to be a constant!");
201 const Type *Ty = getType(TypeSlot);
202 std::pair<const Type*, unsigned> Key(Ty, Slot);
203 ConstantRefsType::iterator I = ConstantFwdRefs.lower_bound(Key);
205 if (I != ConstantFwdRefs.end() && I->first == Key) {
206 BCR_TRACE(5, "Previous forward ref found!\n");
209 // Create a placeholder for the constant reference and
210 // keep track of the fact that we have a forward ref to recycle it
211 BCR_TRACE(5, "Creating new forward ref to a constant!\n");
212 Constant *C = new ConstPHolder(Ty, Slot);
214 // Keep track of the fact that we have a forward ref to recycle it
215 ConstantFwdRefs.insert(I, std::make_pair(Key, C));
220 /// ParseBasicBlock - In LLVM 1.0 bytecode files, we used to output one
221 /// basicblock at a time. This method reads in one of the basicblock packets.
222 BasicBlock *BytecodeParser::ParseBasicBlock(const unsigned char *&Buf,
223 const unsigned char *EndBuf,
226 if (ParsedBasicBlocks.size() == BlockNo)
227 ParsedBasicBlocks.push_back(BB = new BasicBlock());
228 else if (ParsedBasicBlocks[BlockNo] == 0)
229 BB = ParsedBasicBlocks[BlockNo] = new BasicBlock();
231 BB = ParsedBasicBlocks[BlockNo];
233 std::vector<unsigned> Args;
235 ParseInstruction(Buf, EndBuf, Args, BB);
241 /// ParseInstructionList - Parse all of the BasicBlock's & Instruction's in the
242 /// body of a function. In post 1.0 bytecode files, we no longer emit basic
243 /// block individually, in order to avoid per-basic-block overhead.
244 unsigned BytecodeParser::ParseInstructionList(Function *F,
245 const unsigned char *&Buf,
246 const unsigned char *EndBuf) {
247 unsigned BlockNo = 0;
248 std::vector<unsigned> Args;
250 while (Buf < EndBuf) {
252 if (ParsedBasicBlocks.size() == BlockNo)
253 ParsedBasicBlocks.push_back(BB = new BasicBlock());
254 else if (ParsedBasicBlocks[BlockNo] == 0)
255 BB = ParsedBasicBlocks[BlockNo] = new BasicBlock();
257 BB = ParsedBasicBlocks[BlockNo];
259 F->getBasicBlockList().push_back(BB);
261 // Read instructions into this basic block until we get to a terminator
262 while (Buf < EndBuf && !BB->getTerminator())
263 ParseInstruction(Buf, EndBuf, Args, BB);
265 if (!BB->getTerminator())
266 throw std::string("Non-terminated basic block found!");
272 void BytecodeParser::ParseSymbolTable(const unsigned char *&Buf,
273 const unsigned char *EndBuf,
275 Function *CurrentFunction) {
276 // Allow efficient basic block lookup by number.
277 std::vector<BasicBlock*> BBMap;
279 for (Function::iterator I = CurrentFunction->begin(),
280 E = CurrentFunction->end(); I != E; ++I)
283 while (Buf < EndBuf) {
284 // Symtab block header: [num entries][type id number]
285 unsigned NumEntries = read_vbr_uint(Buf, EndBuf);
286 unsigned Typ = read_vbr_uint(Buf, EndBuf);
287 const Type *Ty = getType(Typ);
288 BCR_TRACE(3, "Plane Type: '" << *Ty << "' with " << NumEntries <<
291 for (unsigned i = 0; i != NumEntries; ++i) {
292 // Symtab entry: [def slot #][name]
293 unsigned slot = read_vbr_uint(Buf, EndBuf);
294 std::string Name = read_str(Buf, EndBuf);
297 if (Typ == Type::TypeTyID)
298 V = (Value*)getType(slot);
299 else if (Typ == Type::LabelTyID) {
300 if (slot < BBMap.size())
303 V = getValue(Typ, slot, false); // Find mapping...
306 throw "Failed value look-up for name '" + Name + "'";
307 BCR_TRACE(4, "Map: '" << Name << "' to #" << slot << ":" << *V;
308 if (!isa<Instruction>(V)) std::cerr << "\n");
310 V->setName(Name, ST);
314 if (Buf > EndBuf) throw std::string("Tried to read past end of buffer.");
317 void BytecodeParser::ResolveReferencesToConstant(Constant *NewV, unsigned Slot){
318 ConstantRefsType::iterator I =
319 ConstantFwdRefs.find(std::make_pair(NewV->getType(), Slot));
320 if (I == ConstantFwdRefs.end()) return; // Never forward referenced?
322 BCR_TRACE(3, "Mutating forward refs!\n");
323 Value *PH = I->second; // Get the placeholder...
324 PH->replaceAllUsesWith(NewV);
325 delete PH; // Delete the old placeholder
326 ConstantFwdRefs.erase(I); // Remove the map entry for it
329 void BytecodeParser::ParseFunction(const unsigned char *&Buf,
330 const unsigned char *EndBuf) {
331 if (FunctionSignatureList.empty())
332 throw std::string("FunctionSignatureList empty!");
334 Function *F = FunctionSignatureList.back();
335 FunctionSignatureList.pop_back();
337 // Save the information for future reading of the function
338 LazyFunctionLoadMap[F] = LazyFunctionInfo(Buf, EndBuf);
339 // Pretend we've `parsed' this function
343 void BytecodeParser::materializeFunction(Function* F) {
344 // Find {start, end} pointers and slot in the map. If not there, we're done.
345 std::map<Function*, LazyFunctionInfo>::iterator Fi =
346 LazyFunctionLoadMap.find(F);
347 if (Fi == LazyFunctionLoadMap.end()) return;
349 const unsigned char *Buf = Fi->second.Buf;
350 const unsigned char *EndBuf = Fi->second.EndBuf;
351 LazyFunctionLoadMap.erase(Fi);
353 GlobalValue::LinkageTypes Linkage = GlobalValue::ExternalLinkage;
355 unsigned LinkageType = read_vbr_uint(Buf, EndBuf);
357 throw std::string("Invalid linkage type for Function.");
358 switch (LinkageType) {
359 case 0: Linkage = GlobalValue::ExternalLinkage; break;
360 case 1: Linkage = GlobalValue::WeakLinkage; break;
361 case 2: Linkage = GlobalValue::AppendingLinkage; break;
362 case 3: Linkage = GlobalValue::InternalLinkage; break;
363 case 4: Linkage = GlobalValue::LinkOnceLinkage; break;
366 F->setLinkage(Linkage);
368 // Keep track of how many basic blocks we have read in...
369 unsigned BlockNum = 0;
370 bool InsertedArguments = false;
372 while (Buf < EndBuf) {
374 const unsigned char *OldBuf = Buf;
375 readBlock(Buf, EndBuf, Type, Size);
378 case BytecodeFormat::ConstantPool:
379 if (!InsertedArguments) {
380 // Insert arguments into the value table before we parse the first basic
381 // block in the function, but after we potentially read in the
383 const FunctionType *FT = F->getFunctionType();
384 Function::aiterator AI = F->abegin();
385 for (FunctionType::param_iterator It = FT->param_begin();
386 It != FT->param_end(); ++It, ++AI)
387 insertValue(AI, getTypeSlot(AI->getType()), Values);
388 InsertedArguments = true;
391 BCR_TRACE(2, "BLOCK BytecodeFormat::ConstantPool: {\n");
392 ParseConstantPool(Buf, Buf+Size, Values, FunctionTypeValues);
395 case BytecodeFormat::CompactionTable:
396 BCR_TRACE(2, "BLOCK BytecodeFormat::CompactionTable: {\n");
397 ParseCompactionTable(Buf, Buf+Size);
400 case BytecodeFormat::BasicBlock: {
401 if (!InsertedArguments) {
402 // Insert arguments into the value table before we parse the first basic
403 // block in the function, but after we potentially read in the
405 const FunctionType *FT = F->getFunctionType();
406 Function::aiterator AI = F->abegin();
407 for (FunctionType::param_iterator It = FT->param_begin();
408 It != FT->param_end(); ++It, ++AI)
409 insertValue(AI, getTypeSlot(AI->getType()), Values);
410 InsertedArguments = true;
413 BCR_TRACE(2, "BLOCK BytecodeFormat::BasicBlock: {\n");
414 BasicBlock *BB = ParseBasicBlock(Buf, Buf+Size, BlockNum++);
415 F->getBasicBlockList().push_back(BB);
419 case BytecodeFormat::InstructionList: {
420 // Insert arguments into the value table before we parse the instruction
421 // list for the function, but after we potentially read in the compaction
423 if (!InsertedArguments) {
424 const FunctionType *FT = F->getFunctionType();
425 Function::aiterator AI = F->abegin();
426 for (FunctionType::param_iterator It = FT->param_begin();
427 It != FT->param_end(); ++It, ++AI)
428 insertValue(AI, getTypeSlot(AI->getType()), Values);
429 InsertedArguments = true;
432 BCR_TRACE(2, "BLOCK BytecodeFormat::InstructionList: {\n");
433 if (BlockNum) throw std::string("Already parsed basic blocks!");
434 BlockNum = ParseInstructionList(F, Buf, Buf+Size);
438 case BytecodeFormat::SymbolTable:
439 BCR_TRACE(2, "BLOCK BytecodeFormat::SymbolTable: {\n");
440 ParseSymbolTable(Buf, Buf+Size, &F->getSymbolTable(), F);
444 BCR_TRACE(2, "BLOCK <unknown>:ignored! {\n");
447 throw std::string("Wrapped around reading bytecode.");
450 BCR_TRACE(2, "} end block\n");
452 // Malformed bc file if read past end of block.
453 align32(Buf, EndBuf);
456 // Make sure there were no references to non-existant basic blocks.
457 if (BlockNum != ParsedBasicBlocks.size())
458 throw std::string("Illegal basic block operand reference");
459 ParsedBasicBlocks.clear();
461 // Resolve forward references. Replace any uses of a forward reference value
462 // with the real value.
464 // replaceAllUsesWith is very inefficient for instructions which have a LARGE
465 // number of operands. PHI nodes often have forward references, and can also
466 // often have a very large number of operands.
468 // FIXME: REEVALUATE. replaceAllUsesWith is _much_ faster now, and this code
469 // should be simplified back to using it!
471 std::map<Value*, Value*> ForwardRefMapping;
472 for (std::map<std::pair<unsigned,unsigned>, Value*>::iterator
473 I = ForwardReferences.begin(), E = ForwardReferences.end();
475 ForwardRefMapping[I->second] = getValue(I->first.first, I->first.second,
478 for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
479 for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
480 for (unsigned i = 0, e = I->getNumOperands(); i != e; ++i)
481 if (Argument *A = dyn_cast<Argument>(I->getOperand(i))) {
482 std::map<Value*, Value*>::iterator It = ForwardRefMapping.find(A);
483 if (It != ForwardRefMapping.end()) I->setOperand(i, It->second);
486 while (!ForwardReferences.empty()) {
487 std::map<std::pair<unsigned,unsigned>, Value*>::iterator I =
488 ForwardReferences.begin();
489 Value *PlaceHolder = I->second;
490 ForwardReferences.erase(I);
492 // Now that all the uses are gone, delete the placeholder...
493 // If we couldn't find a def (error case), then leak a little
494 // memory, because otherwise we can't remove all uses!
498 // Clear out function-level types...
499 FunctionTypeValues.clear();
500 CompactionTable.clear();
504 void BytecodeParser::ParseCompactionTable(const unsigned char *&Buf,
505 const unsigned char *End) {
508 unsigned NumEntries = read_vbr_uint(Buf, End);
511 if ((NumEntries & 3) == 3) {
513 Ty = read_vbr_uint(Buf, End);
515 Ty = NumEntries >> 2;
519 if (Ty >= CompactionTable.size())
520 CompactionTable.resize(Ty+1);
522 if (!CompactionTable[Ty].empty())
523 throw std::string("Compaction table plane contains multiple entries!");
525 if (Ty == Type::TypeTyID) {
526 for (unsigned i = 0; i != NumEntries; ++i) {
527 const Type *Typ = getGlobalTableType(read_vbr_uint(Buf, End));
528 CompactionTable[Type::TypeTyID].push_back(const_cast<Type*>(Typ));
531 CompactionTable.resize(NumEntries+Type::FirstDerivedTyID);
533 const Type *Typ = getType(Ty);
534 // Push the implicit zero
535 CompactionTable[Ty].push_back(Constant::getNullValue(Typ));
536 for (unsigned i = 0; i != NumEntries; ++i) {
537 Value *V = getGlobalTableValue(Typ, read_vbr_uint(Buf, End));
538 CompactionTable[Ty].push_back(V);
547 void BytecodeParser::ParseModuleGlobalInfo(const unsigned char *&Buf,
548 const unsigned char *End) {
549 if (!FunctionSignatureList.empty())
550 throw std::string("Two ModuleGlobalInfo packets found!");
552 // Read global variables...
553 unsigned VarType = read_vbr_uint(Buf, End);
554 while (VarType != Type::VoidTyID) { // List is terminated by Void
555 // VarType Fields: bit0 = isConstant, bit1 = hasInitializer, bit2,3,4 =
556 // Linkage, bit4+ = slot#
557 unsigned SlotNo = VarType >> 5;
558 unsigned LinkageID = (VarType >> 2) & 7;
559 GlobalValue::LinkageTypes Linkage;
562 default: assert(0 && "Unknown linkage type!");
563 case 0: Linkage = GlobalValue::ExternalLinkage; break;
564 case 1: Linkage = GlobalValue::WeakLinkage; break;
565 case 2: Linkage = GlobalValue::AppendingLinkage; break;
566 case 3: Linkage = GlobalValue::InternalLinkage; break;
567 case 4: Linkage = GlobalValue::LinkOnceLinkage; break;
570 const Type *Ty = getType(SlotNo);
571 if (!isa<PointerType>(Ty))
572 throw std::string("Global not pointer type! Ty = " +
573 Ty->getDescription());
575 const Type *ElTy = cast<PointerType>(Ty)->getElementType();
577 // Create the global variable...
578 GlobalVariable *GV = new GlobalVariable(ElTy, VarType & 1, Linkage,
580 BCR_TRACE(2, "Global Variable of type: " << *Ty << "\n");
581 insertValue(GV, SlotNo, ModuleValues);
583 if (VarType & 2) // Does it have an initializer?
584 GlobalInits.push_back(std::make_pair(GV, read_vbr_uint(Buf, End)));
585 VarType = read_vbr_uint(Buf, End);
588 // Read the function objects for all of the functions that are coming
589 unsigned FnSignature = read_vbr_uint(Buf, End);
590 while (FnSignature != Type::VoidTyID) { // List is terminated by Void
591 const Type *Ty = getType(FnSignature);
592 if (!isa<PointerType>(Ty) ||
593 !isa<FunctionType>(cast<PointerType>(Ty)->getElementType()))
594 throw std::string("Function not ptr to func type! Ty = " +
595 Ty->getDescription());
597 // We create functions by passing the underlying FunctionType to create...
598 Ty = cast<PointerType>(Ty)->getElementType();
600 // When the ModuleGlobalInfo section is read, we load the type of each
601 // function and the 'ModuleValues' slot that it lands in. We then load a
602 // placeholder into its slot to reserve it. When the function is loaded,
603 // this placeholder is replaced.
605 // Insert the placeholder...
606 Function *Func = new Function(cast<FunctionType>(Ty),
607 GlobalValue::InternalLinkage, "", TheModule);
608 insertValue(Func, FnSignature, ModuleValues);
610 // Keep track of this information in a list that is emptied as functions are
613 FunctionSignatureList.push_back(Func);
615 FnSignature = read_vbr_uint(Buf, End);
616 BCR_TRACE(2, "Function of type: " << Ty << "\n");
619 if (hasInconsistentModuleGlobalInfo)
622 // Now that the function signature list is set up, reverse it so that we can
623 // remove elements efficiently from the back of the vector.
624 std::reverse(FunctionSignatureList.begin(), FunctionSignatureList.end());
626 // This is for future proofing... in the future extra fields may be added that
627 // we don't understand, so we transparently ignore them.
632 void BytecodeParser::ParseVersionInfo(const unsigned char *&Buf,
633 const unsigned char *EndBuf) {
634 unsigned Version = read_vbr_uint(Buf, EndBuf);
636 // Unpack version number: low four bits are for flags, top bits = version
637 Module::Endianness Endianness;
638 Module::PointerSize PointerSize;
639 Endianness = (Version & 1) ? Module::BigEndian : Module::LittleEndian;
640 PointerSize = (Version & 2) ? Module::Pointer64 : Module::Pointer32;
642 bool hasNoEndianness = Version & 4;
643 bool hasNoPointerSize = Version & 8;
645 RevisionNum = Version >> 4;
647 // Default values for the current bytecode version
648 hasInconsistentModuleGlobalInfo = false;
649 hasExplicitPrimitiveZeros = false;
650 hasRestrictedGEPTypes = false;
652 switch (RevisionNum) {
653 case 0: // LLVM 1.0, 1.1 release version
654 // Base LLVM 1.0 bytecode format.
655 hasInconsistentModuleGlobalInfo = true;
656 hasExplicitPrimitiveZeros = true;
658 case 1: // LLVM 1.2 release version
659 // LLVM 1.2 added explicit support for emitting strings efficiently.
661 // Also, it fixed the problem where the size of the ModuleGlobalInfo block
662 // included the size for the alignment at the end, where the rest of the
665 // LLVM 1.2 and before required that GEP indices be ubyte constants for
666 // structures and longs for sequential types.
667 hasRestrictedGEPTypes = true;
670 case 2: // LLVM 1.3 release version
674 throw std::string("Unknown bytecode version number!");
677 if (hasNoEndianness) Endianness = Module::AnyEndianness;
678 if (hasNoPointerSize) PointerSize = Module::AnyPointerSize;
680 TheModule->setEndianness(Endianness);
681 TheModule->setPointerSize(PointerSize);
682 BCR_TRACE(1, "Bytecode Rev = " << (unsigned)RevisionNum << "\n");
683 BCR_TRACE(1, "Endianness/PointerSize = " << Endianness << ","
684 << PointerSize << "\n");
687 void BytecodeParser::ParseModule(const unsigned char *Buf,
688 const unsigned char *EndBuf) {
690 readBlock(Buf, EndBuf, Type, Size);
691 if (Type != BytecodeFormat::Module || Buf+Size != EndBuf)
692 throw std::string("Expected Module packet! B: "+
693 utostr((unsigned)(intptr_t)Buf) + ", S: "+utostr(Size)+
694 " E: "+utostr((unsigned)(intptr_t)EndBuf)); // Hrm, not a class?
696 BCR_TRACE(0, "BLOCK BytecodeFormat::Module: {\n");
697 FunctionSignatureList.clear(); // Just in case...
699 // Read into instance variables...
700 ParseVersionInfo(Buf, EndBuf);
701 align32(Buf, EndBuf);
703 while (Buf < EndBuf) {
704 const unsigned char *OldBuf = Buf;
705 readBlock(Buf, EndBuf, Type, Size);
707 case BytecodeFormat::GlobalTypePlane:
708 BCR_TRACE(1, "BLOCK BytecodeFormat::GlobalTypePlane: {\n");
709 ParseGlobalTypes(Buf, Buf+Size);
712 case BytecodeFormat::ModuleGlobalInfo:
713 BCR_TRACE(1, "BLOCK BytecodeFormat::ModuleGlobalInfo: {\n");
714 ParseModuleGlobalInfo(Buf, Buf+Size);
717 case BytecodeFormat::ConstantPool:
718 BCR_TRACE(1, "BLOCK BytecodeFormat::ConstantPool: {\n");
719 ParseConstantPool(Buf, Buf+Size, ModuleValues, ModuleTypeValues);
722 case BytecodeFormat::Function: {
723 BCR_TRACE(1, "BLOCK BytecodeFormat::Function: {\n");
724 ParseFunction(Buf, Buf+Size);
728 case BytecodeFormat::SymbolTable:
729 BCR_TRACE(1, "BLOCK BytecodeFormat::SymbolTable: {\n");
730 ParseSymbolTable(Buf, Buf+Size, &TheModule->getSymbolTable(), 0);
734 if (OldBuf > Buf) throw std::string("Expected Module Block!");
737 BCR_TRACE(1, "} end block\n");
738 align32(Buf, EndBuf);
741 // After the module constant pool has been read, we can safely initialize
742 // global variables...
743 while (!GlobalInits.empty()) {
744 GlobalVariable *GV = GlobalInits.back().first;
745 unsigned Slot = GlobalInits.back().second;
746 GlobalInits.pop_back();
748 // Look up the initializer value...
749 // FIXME: Preserve this type ID!
750 unsigned TypeSlot = getTypeSlot(GV->getType()->getElementType());
751 if (Constant *CV = getConstantValue(TypeSlot, Slot)) {
752 if (GV->hasInitializer())
753 throw std::string("Global *already* has an initializer?!");
754 GV->setInitializer(CV);
756 throw std::string("Cannot find initializer value.");
759 if (!FunctionSignatureList.empty())
760 throw std::string("Function expected, but bytecode stream ended!");
762 BCR_TRACE(0, "} end block\n\n");
765 void BytecodeParser::ParseBytecode(const unsigned char *Buf, unsigned Length,
766 const std::string &ModuleID) {
768 unsigned char *EndBuf = (unsigned char*)(Buf + Length);
770 // Read and check signature...
771 unsigned Sig = read(Buf, EndBuf);
772 if (Sig != ('l' | ('l' << 8) | ('v' << 16) | ('m' << 24)))
773 throw std::string("Invalid bytecode signature!");
775 TheModule = new Module(ModuleID);
777 ParseModule(Buf, EndBuf);
778 } catch (std::string &Error) {
779 freeState(); // Must destroy handles before deleting module!