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/CallingConv.h"
23 #include "llvm/Constants.h"
24 #include "llvm/InlineAsm.h"
25 #include "llvm/Instructions.h"
26 #include "llvm/SymbolTable.h"
27 #include "llvm/Bytecode/Format.h"
28 #include "llvm/Config/alloca.h"
29 #include "llvm/Support/GetElementPtrTypeIterator.h"
30 #include "llvm/Support/Compressor.h"
31 #include "llvm/Support/MathExtras.h"
32 #include "llvm/ADT/StringExtras.h"
38 /// @brief A class for maintaining the slot number definition
39 /// as a placeholder for the actual definition for forward constants defs.
40 class ConstantPlaceHolder : public ConstantExpr {
41 ConstantPlaceHolder(); // DO NOT IMPLEMENT
42 void operator=(const ConstantPlaceHolder &); // DO NOT IMPLEMENT
45 ConstantPlaceHolder(const Type *Ty)
46 : ConstantExpr(Ty, Instruction::UserOp1, &Op, 1),
47 Op(UndefValue::get(Type::IntTy), this) {
52 // Provide some details on error
53 inline void BytecodeReader::error(const std::string& err) {
54 ErrorMsg = err + " (Vers=" + itostr(RevisionNum) + ", Pos="
55 + 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 /// Read a whole unsigned integer
76 inline unsigned BytecodeReader::read_uint() {
78 error("Ran out of data reading uint!");
80 return At[-4] | (At[-3] << 8) | (At[-2] << 16) | (At[-1] << 24);
83 /// Read a variable-bit-rate encoded unsigned integer
84 inline unsigned BytecodeReader::read_vbr_uint() {
91 error("Ran out of data reading vbr_uint!");
92 Result |= (unsigned)((*At++) & 0x7F) << Shift;
94 } while (At[-1] & 0x80);
95 if (Handler) Handler->handleVBR32(At-Save);
99 /// Read a variable-bit-rate encoded unsigned 64-bit integer.
100 inline uint64_t BytecodeReader::read_vbr_uint64() {
107 error("Ran out of data reading vbr_uint64!");
108 Result |= (uint64_t)((*At++) & 0x7F) << Shift;
110 } while (At[-1] & 0x80);
111 if (Handler) Handler->handleVBR64(At-Save);
115 /// Read a variable-bit-rate encoded signed 64-bit integer.
116 inline int64_t BytecodeReader::read_vbr_int64() {
117 uint64_t R = read_vbr_uint64();
120 return -(int64_t)(R >> 1);
121 else // There is no such thing as -0 with integers. "-0" really means
122 // 0x8000000000000000.
125 return (int64_t)(R >> 1);
128 /// Read a pascal-style string (length followed by text)
129 inline std::string BytecodeReader::read_str() {
130 unsigned Size = read_vbr_uint();
131 const unsigned char *OldAt = At;
133 if (At > BlockEnd) // Size invalid?
134 error("Ran out of data reading a string!");
135 return std::string((char*)OldAt, Size);
138 /// Read an arbitrary block of data
139 inline void BytecodeReader::read_data(void *Ptr, void *End) {
140 unsigned char *Start = (unsigned char *)Ptr;
141 unsigned Amount = (unsigned char *)End - Start;
142 if (At+Amount > BlockEnd)
143 error("Ran out of data!");
144 std::copy(At, At+Amount, Start);
148 /// Read a float value in little-endian order
149 inline void BytecodeReader::read_float(float& FloatVal) {
150 /// FIXME: This isn't optimal, it has size problems on some platforms
151 /// where FP is not IEEE.
152 FloatVal = BitsToFloat(At[0] | (At[1] << 8) | (At[2] << 16) | (At[3] << 24));
153 At+=sizeof(uint32_t);
156 /// Read a double value in little-endian order
157 inline void BytecodeReader::read_double(double& DoubleVal) {
158 /// FIXME: This isn't optimal, it has size problems on some platforms
159 /// where FP is not IEEE.
160 DoubleVal = BitsToDouble((uint64_t(At[0]) << 0) | (uint64_t(At[1]) << 8) |
161 (uint64_t(At[2]) << 16) | (uint64_t(At[3]) << 24) |
162 (uint64_t(At[4]) << 32) | (uint64_t(At[5]) << 40) |
163 (uint64_t(At[6]) << 48) | (uint64_t(At[7]) << 56));
164 At+=sizeof(uint64_t);
167 /// Read a block header and obtain its type and size
168 inline void BytecodeReader::read_block(unsigned &Type, unsigned &Size) {
169 Size = read_uint(); // Read the header
170 Type = Size & 0x1F; // mask low order five bits to get type
171 Size >>= 5; // high order 27 bits is the size
173 if (At + Size > BlockEnd)
174 error("Attempt to size a block past end of memory");
175 BlockEnd = At + Size;
176 if (Handler) Handler->handleBlock(Type, BlockStart, Size);
179 //===----------------------------------------------------------------------===//
181 //===----------------------------------------------------------------------===//
183 /// Determine if a type id has an implicit null value
184 inline bool BytecodeReader::hasImplicitNull(unsigned TyID) {
185 return TyID != Type::LabelTyID && TyID != Type::VoidTyID;
188 /// Obtain a type given a typeid and account for things like compaction tables,
189 /// function level vs module level, and the offsetting for the primitive types.
190 const Type *BytecodeReader::getType(unsigned ID) {
191 if (ID < Type::FirstDerivedTyID)
192 if (const Type *T = Type::getPrimitiveType((Type::TypeID)ID))
193 return T; // Asked for a primitive type...
195 // Otherwise, derived types need offset...
196 ID -= Type::FirstDerivedTyID;
198 if (!CompactionTypes.empty()) {
199 if (ID >= CompactionTypes.size())
200 error("Type ID out of range for compaction table!");
201 return CompactionTypes[ID].first;
204 // Is it a module-level type?
205 if (ID < ModuleTypes.size())
206 return ModuleTypes[ID].get();
208 // Nope, is it a function-level type?
209 ID -= ModuleTypes.size();
210 if (ID < FunctionTypes.size())
211 return FunctionTypes[ID].get();
213 error("Illegal type reference!");
217 /// This method just saves some coding. It uses read_vbr_uint to read in a
218 /// type id, errors that its not the type type, and then calls getType to
219 /// return the type value.
220 inline const Type* BytecodeReader::readType() {
221 return getType(read_vbr_uint());
224 /// Get the slot number associated with a type accounting for primitive
225 /// types, compaction tables, and function level vs module level.
226 unsigned BytecodeReader::getTypeSlot(const Type *Ty) {
227 if (Ty->isPrimitiveType())
228 return Ty->getTypeID();
230 // Scan the compaction table for the type if needed.
231 if (!CompactionTypes.empty()) {
232 for (unsigned i = 0, e = CompactionTypes.size(); i != e; ++i)
233 if (CompactionTypes[i].first == Ty)
234 return Type::FirstDerivedTyID + i;
236 error("Couldn't find type specified in compaction table!");
239 // Check the function level types first...
240 TypeListTy::iterator I = std::find(FunctionTypes.begin(),
241 FunctionTypes.end(), Ty);
243 if (I != FunctionTypes.end())
244 return Type::FirstDerivedTyID + ModuleTypes.size() +
245 (&*I - &FunctionTypes[0]);
247 // If we don't have our cache yet, build it now.
248 if (ModuleTypeIDCache.empty()) {
250 ModuleTypeIDCache.reserve(ModuleTypes.size());
251 for (TypeListTy::iterator I = ModuleTypes.begin(), E = ModuleTypes.end();
253 ModuleTypeIDCache.push_back(std::make_pair(*I, N));
255 std::sort(ModuleTypeIDCache.begin(), ModuleTypeIDCache.end());
258 // Binary search the cache for the entry.
259 std::vector<std::pair<const Type*, unsigned> >::iterator IT =
260 std::lower_bound(ModuleTypeIDCache.begin(), ModuleTypeIDCache.end(),
261 std::make_pair(Ty, 0U));
262 if (IT == ModuleTypeIDCache.end() || IT->first != Ty)
263 error("Didn't find type in ModuleTypes.");
265 return Type::FirstDerivedTyID + IT->second;
268 /// This is just like getType, but when a compaction table is in use, it is
269 /// ignored. It also ignores function level types.
271 const Type *BytecodeReader::getGlobalTableType(unsigned Slot) {
272 if (Slot < Type::FirstDerivedTyID) {
273 const Type *Ty = Type::getPrimitiveType((Type::TypeID)Slot);
275 error("Not a primitive type ID?");
278 Slot -= Type::FirstDerivedTyID;
279 if (Slot >= ModuleTypes.size())
280 error("Illegal compaction table type reference!");
281 return ModuleTypes[Slot];
284 /// This is just like getTypeSlot, but when a compaction table is in use, it
285 /// is ignored. It also ignores function level types.
286 unsigned BytecodeReader::getGlobalTableTypeSlot(const Type *Ty) {
287 if (Ty->isPrimitiveType())
288 return Ty->getTypeID();
290 // If we don't have our cache yet, build it now.
291 if (ModuleTypeIDCache.empty()) {
293 ModuleTypeIDCache.reserve(ModuleTypes.size());
294 for (TypeListTy::iterator I = ModuleTypes.begin(), E = ModuleTypes.end();
296 ModuleTypeIDCache.push_back(std::make_pair(*I, N));
298 std::sort(ModuleTypeIDCache.begin(), ModuleTypeIDCache.end());
301 // Binary search the cache for the entry.
302 std::vector<std::pair<const Type*, unsigned> >::iterator IT =
303 std::lower_bound(ModuleTypeIDCache.begin(), ModuleTypeIDCache.end(),
304 std::make_pair(Ty, 0U));
305 if (IT == ModuleTypeIDCache.end() || IT->first != Ty)
306 error("Didn't find type in ModuleTypes.");
308 return Type::FirstDerivedTyID + IT->second;
311 /// Retrieve a value of a given type and slot number, possibly creating
312 /// it if it doesn't already exist.
313 Value * BytecodeReader::getValue(unsigned type, unsigned oNum, bool Create) {
314 assert(type != Type::LabelTyID && "getValue() cannot get blocks!");
317 // If there is a compaction table active, it defines the low-level numbers.
318 // If not, the module values define the low-level numbers.
319 if (CompactionValues.size() > type && !CompactionValues[type].empty()) {
320 if (Num < CompactionValues[type].size())
321 return CompactionValues[type][Num];
322 Num -= CompactionValues[type].size();
324 // By default, the global type id is the type id passed in
325 unsigned GlobalTyID = type;
327 // If the type plane was compactified, figure out the global type ID by
328 // adding the derived type ids and the distance.
329 if (!CompactionTypes.empty() && type >= Type::FirstDerivedTyID)
330 GlobalTyID = CompactionTypes[type-Type::FirstDerivedTyID].second;
332 if (hasImplicitNull(GlobalTyID)) {
333 const Type *Ty = getType(type);
334 if (!isa<OpaqueType>(Ty)) {
336 return Constant::getNullValue(Ty);
341 if (GlobalTyID < ModuleValues.size() && ModuleValues[GlobalTyID]) {
342 if (Num < ModuleValues[GlobalTyID]->size())
343 return ModuleValues[GlobalTyID]->getOperand(Num);
344 Num -= ModuleValues[GlobalTyID]->size();
348 if (FunctionValues.size() > type &&
349 FunctionValues[type] &&
350 Num < FunctionValues[type]->size())
351 return FunctionValues[type]->getOperand(Num);
353 if (!Create) return 0; // Do not create a placeholder?
355 // Did we already create a place holder?
356 std::pair<unsigned,unsigned> KeyValue(type, oNum);
357 ForwardReferenceMap::iterator I = ForwardReferences.lower_bound(KeyValue);
358 if (I != ForwardReferences.end() && I->first == KeyValue)
359 return I->second; // We have already created this placeholder
361 // If the type exists (it should)
362 if (const Type* Ty = getType(type)) {
363 // Create the place holder
364 Value *Val = new Argument(Ty);
365 ForwardReferences.insert(I, std::make_pair(KeyValue, Val));
368 error("Can't create placeholder for value of type slot #" + utostr(type));
369 return 0; // just silence warning, error calls longjmp
372 /// This is just like getValue, but when a compaction table is in use, it
373 /// is ignored. Also, no forward references or other fancy features are
375 Value* BytecodeReader::getGlobalTableValue(unsigned TyID, unsigned SlotNo) {
377 return Constant::getNullValue(getType(TyID));
379 if (!CompactionTypes.empty() && TyID >= Type::FirstDerivedTyID) {
380 TyID -= Type::FirstDerivedTyID;
381 if (TyID >= CompactionTypes.size())
382 error("Type ID out of range for compaction table!");
383 TyID = CompactionTypes[TyID].second;
388 if (TyID >= ModuleValues.size() || ModuleValues[TyID] == 0 ||
389 SlotNo >= ModuleValues[TyID]->size()) {
390 if (TyID >= ModuleValues.size() || ModuleValues[TyID] == 0)
391 error("Corrupt compaction table entry!"
392 + utostr(TyID) + ", " + utostr(SlotNo) + ": "
393 + utostr(ModuleValues.size()));
395 error("Corrupt compaction table entry!"
396 + utostr(TyID) + ", " + utostr(SlotNo) + ": "
397 + utostr(ModuleValues.size()) + ", "
398 + utohexstr(reinterpret_cast<uint64_t>(((void*)ModuleValues[TyID])))
400 + utostr(ModuleValues[TyID]->size()));
402 return ModuleValues[TyID]->getOperand(SlotNo);
405 /// Just like getValue, except that it returns a null pointer
406 /// only on error. It always returns a constant (meaning that if the value is
407 /// defined, but is not a constant, that is an error). If the specified
408 /// constant hasn't been parsed yet, a placeholder is defined and used.
409 /// Later, after the real value is parsed, the placeholder is eliminated.
410 Constant* BytecodeReader::getConstantValue(unsigned TypeSlot, unsigned Slot) {
411 if (Value *V = getValue(TypeSlot, Slot, false))
412 if (Constant *C = dyn_cast<Constant>(V))
413 return C; // If we already have the value parsed, just return it
415 error("Value for slot " + utostr(Slot) +
416 " is expected to be a constant!");
418 std::pair<unsigned, unsigned> Key(TypeSlot, Slot);
419 ConstantRefsType::iterator I = ConstantFwdRefs.lower_bound(Key);
421 if (I != ConstantFwdRefs.end() && I->first == Key) {
424 // Create a placeholder for the constant reference and
425 // keep track of the fact that we have a forward ref to recycle it
426 Constant *C = new ConstantPlaceHolder(getType(TypeSlot));
428 // Keep track of the fact that we have a forward ref to recycle it
429 ConstantFwdRefs.insert(I, std::make_pair(Key, C));
434 //===----------------------------------------------------------------------===//
435 // IR Construction Methods
436 //===----------------------------------------------------------------------===//
438 /// As values are created, they are inserted into the appropriate place
439 /// with this method. The ValueTable argument must be one of ModuleValues
440 /// or FunctionValues data members of this class.
441 unsigned BytecodeReader::insertValue(Value *Val, unsigned type,
442 ValueTable &ValueTab) {
443 if (ValueTab.size() <= type)
444 ValueTab.resize(type+1);
446 if (!ValueTab[type]) ValueTab[type] = new ValueList();
448 ValueTab[type]->push_back(Val);
450 bool HasOffset = hasImplicitNull(type) && !isa<OpaqueType>(Val->getType());
451 return ValueTab[type]->size()-1 + HasOffset;
454 /// Insert the arguments of a function as new values in the reader.
455 void BytecodeReader::insertArguments(Function* F) {
456 const FunctionType *FT = F->getFunctionType();
457 Function::arg_iterator AI = F->arg_begin();
458 for (FunctionType::param_iterator It = FT->param_begin();
459 It != FT->param_end(); ++It, ++AI)
460 insertValue(AI, getTypeSlot(AI->getType()), FunctionValues);
463 //===----------------------------------------------------------------------===//
464 // Bytecode Parsing Methods
465 //===----------------------------------------------------------------------===//
467 /// This method parses a single instruction. The instruction is
468 /// inserted at the end of the \p BB provided. The arguments of
469 /// the instruction are provided in the \p Oprnds vector.
470 void BytecodeReader::ParseInstruction(std::vector<unsigned> &Oprnds,
474 // Clear instruction data
478 unsigned Op = read_uint();
480 // bits Instruction format: Common to all formats
481 // --------------------------
482 // 01-00: Opcode type, fixed to 1.
484 Opcode = (Op >> 2) & 63;
485 Oprnds.resize((Op >> 0) & 03);
487 // Extract the operands
488 switch (Oprnds.size()) {
490 // bits Instruction format:
491 // --------------------------
492 // 19-08: Resulting type plane
493 // 31-20: Operand #1 (if set to (2^12-1), then zero operands)
495 iType = (Op >> 8) & 4095;
496 Oprnds[0] = (Op >> 20) & 4095;
497 if (Oprnds[0] == 4095) // Handle special encoding for 0 operands...
501 // bits Instruction format:
502 // --------------------------
503 // 15-08: Resulting type plane
507 iType = (Op >> 8) & 255;
508 Oprnds[0] = (Op >> 16) & 255;
509 Oprnds[1] = (Op >> 24) & 255;
512 // bits Instruction format:
513 // --------------------------
514 // 13-08: Resulting type plane
519 iType = (Op >> 8) & 63;
520 Oprnds[0] = (Op >> 14) & 63;
521 Oprnds[1] = (Op >> 20) & 63;
522 Oprnds[2] = (Op >> 26) & 63;
525 At -= 4; // Hrm, try this again...
526 Opcode = read_vbr_uint();
528 iType = read_vbr_uint();
530 unsigned NumOprnds = read_vbr_uint();
531 Oprnds.resize(NumOprnds);
534 error("Zero-argument instruction found; this is invalid.");
536 for (unsigned i = 0; i != NumOprnds; ++i)
537 Oprnds[i] = read_vbr_uint();
541 const Type *InstTy = getType(iType);
543 // Make the necessary adjustments for dealing with backwards compatibility
545 Instruction* Result = 0;
547 // We have enough info to inform the handler now.
549 Handler->handleInstruction(Opcode, InstTy, Oprnds, At-SaveAt);
551 // First, handle the easy binary operators case
552 if (Opcode >= Instruction::BinaryOpsBegin &&
553 Opcode < Instruction::BinaryOpsEnd && Oprnds.size() == 2) {
554 Result = BinaryOperator::create(Instruction::BinaryOps(Opcode),
555 getValue(iType, Oprnds[0]),
556 getValue(iType, Oprnds[1]));
558 // Indicate that we don't think this is a call instruction (yet).
559 // Process based on the Opcode read
561 default: // There was an error, this shouldn't happen.
563 error("Illegal instruction read!");
565 case Instruction::VAArg:
566 if (Oprnds.size() != 2)
567 error("Invalid VAArg instruction!");
568 Result = new VAArgInst(getValue(iType, Oprnds[0]),
571 case Instruction::ExtractElement: {
572 if (Oprnds.size() != 2)
573 error("Invalid extractelement instruction!");
574 Value *V1 = getValue(iType, Oprnds[0]);
575 Value *V2 = getValue(Type::UIntTyID, Oprnds[1]);
577 if (!ExtractElementInst::isValidOperands(V1, V2))
578 error("Invalid extractelement instruction!");
580 Result = new ExtractElementInst(V1, V2);
583 case Instruction::InsertElement: {
584 const PackedType *PackedTy = dyn_cast<PackedType>(InstTy);
585 if (!PackedTy || Oprnds.size() != 3)
586 error("Invalid insertelement instruction!");
588 Value *V1 = getValue(iType, Oprnds[0]);
589 Value *V2 = getValue(getTypeSlot(PackedTy->getElementType()),Oprnds[1]);
590 Value *V3 = getValue(Type::UIntTyID, Oprnds[2]);
592 if (!InsertElementInst::isValidOperands(V1, V2, V3))
593 error("Invalid insertelement instruction!");
594 Result = new InsertElementInst(V1, V2, V3);
597 case Instruction::ShuffleVector: {
598 const PackedType *PackedTy = dyn_cast<PackedType>(InstTy);
599 if (!PackedTy || Oprnds.size() != 3)
600 error("Invalid shufflevector instruction!");
601 Value *V1 = getValue(iType, Oprnds[0]);
602 Value *V2 = getValue(iType, Oprnds[1]);
603 const PackedType *EltTy =
604 PackedType::get(Type::UIntTy, PackedTy->getNumElements());
605 Value *V3 = getValue(getTypeSlot(EltTy), Oprnds[2]);
606 if (!ShuffleVectorInst::isValidOperands(V1, V2, V3))
607 error("Invalid shufflevector instruction!");
608 Result = new ShuffleVectorInst(V1, V2, V3);
611 case Instruction::Trunc:
612 if (Oprnds.size() != 2)
613 error("Invalid cast instruction!");
614 Result = new TruncInst(getValue(iType, Oprnds[0]),
617 case Instruction::ZExt:
618 if (Oprnds.size() != 2)
619 error("Invalid cast instruction!");
620 Result = new ZExtInst(getValue(iType, Oprnds[0]),
623 case Instruction::SExt:
624 if (Oprnds.size() != 2)
625 error("Invalid Cast instruction!");
626 Result = new SExtInst(getValue(iType, Oprnds[0]),
629 case Instruction::FPTrunc:
630 if (Oprnds.size() != 2)
631 error("Invalid cast instruction!");
632 Result = new FPTruncInst(getValue(iType, Oprnds[0]),
635 case Instruction::FPExt:
636 if (Oprnds.size() != 2)
637 error("Invalid cast instruction!");
638 Result = new FPExtInst(getValue(iType, Oprnds[0]),
641 case Instruction::UIToFP:
642 if (Oprnds.size() != 2)
643 error("Invalid cast instruction!");
644 Result = new UIToFPInst(getValue(iType, Oprnds[0]),
647 case Instruction::SIToFP:
648 if (Oprnds.size() != 2)
649 error("Invalid cast instruction!");
650 Result = new SIToFPInst(getValue(iType, Oprnds[0]),
653 case Instruction::FPToUI:
654 if (Oprnds.size() != 2)
655 error("Invalid cast instruction!");
656 Result = new FPToUIInst(getValue(iType, Oprnds[0]),
659 case Instruction::FPToSI:
660 if (Oprnds.size() != 2)
661 error("Invalid cast instruction!");
662 Result = new FPToSIInst(getValue(iType, Oprnds[0]),
665 case Instruction::IntToPtr:
666 if (Oprnds.size() != 2)
667 error("Invalid cast instruction!");
668 Result = new IntToPtrInst(getValue(iType, Oprnds[0]),
671 case Instruction::PtrToInt:
672 if (Oprnds.size() != 2)
673 error("Invalid cast instruction!");
674 Result = new PtrToIntInst(getValue(iType, Oprnds[0]),
677 case Instruction::BitCast:
678 if (Oprnds.size() != 2)
679 error("Invalid cast instruction!");
680 Result = new BitCastInst(getValue(iType, Oprnds[0]),
683 case Instruction::Select:
684 if (Oprnds.size() != 3)
685 error("Invalid Select instruction!");
686 Result = new SelectInst(getValue(Type::BoolTyID, Oprnds[0]),
687 getValue(iType, Oprnds[1]),
688 getValue(iType, Oprnds[2]));
690 case Instruction::PHI: {
691 if (Oprnds.size() == 0 || (Oprnds.size() & 1))
692 error("Invalid phi node encountered!");
694 PHINode *PN = new PHINode(InstTy);
695 PN->reserveOperandSpace(Oprnds.size());
696 for (unsigned i = 0, e = Oprnds.size(); i != e; i += 2)
698 getValue(iType, Oprnds[i]), getBasicBlock(Oprnds[i+1]));
702 case Instruction::ICmp:
703 case Instruction::FCmp:
704 if (Oprnds.size() != 3)
705 error("Cmp instructions requires 3 operands");
706 // These instructions encode the comparison predicate as the 3rd operand.
707 Result = CmpInst::create(Instruction::OtherOps(Opcode),
708 static_cast<unsigned short>(Oprnds[2]),
709 getValue(iType, Oprnds[0]), getValue(iType, Oprnds[1]));
711 case Instruction::Shl:
712 case Instruction::LShr:
713 case Instruction::AShr:
714 Result = new ShiftInst(Instruction::OtherOps(Opcode),
715 getValue(iType, Oprnds[0]),
716 getValue(Type::UByteTyID, Oprnds[1]));
718 case Instruction::Ret:
719 if (Oprnds.size() == 0)
720 Result = new ReturnInst();
721 else if (Oprnds.size() == 1)
722 Result = new ReturnInst(getValue(iType, Oprnds[0]));
724 error("Unrecognized instruction!");
727 case Instruction::Br:
728 if (Oprnds.size() == 1)
729 Result = new BranchInst(getBasicBlock(Oprnds[0]));
730 else if (Oprnds.size() == 3)
731 Result = new BranchInst(getBasicBlock(Oprnds[0]),
732 getBasicBlock(Oprnds[1]), getValue(Type::BoolTyID , Oprnds[2]));
734 error("Invalid number of operands for a 'br' instruction!");
736 case Instruction::Switch: {
737 if (Oprnds.size() & 1)
738 error("Switch statement with odd number of arguments!");
740 SwitchInst *I = new SwitchInst(getValue(iType, Oprnds[0]),
741 getBasicBlock(Oprnds[1]),
743 for (unsigned i = 2, e = Oprnds.size(); i != e; i += 2)
744 I->addCase(cast<ConstantInt>(getValue(iType, Oprnds[i])),
745 getBasicBlock(Oprnds[i+1]));
749 case 58: // Call with extra operand for calling conv
750 case 59: // tail call, Fast CC
751 case 60: // normal call, Fast CC
752 case 61: // tail call, C Calling Conv
753 case Instruction::Call: { // Normal Call, C Calling Convention
754 if (Oprnds.size() == 0)
755 error("Invalid call instruction encountered!");
756 Value *F = getValue(iType, Oprnds[0]);
758 unsigned CallingConv = CallingConv::C;
759 bool isTailCall = false;
761 if (Opcode == 61 || Opcode == 59)
765 isTailCall = Oprnds.back() & 1;
766 CallingConv = Oprnds.back() >> 1;
768 } else if (Opcode == 59 || Opcode == 60) {
769 CallingConv = CallingConv::Fast;
772 // Check to make sure we have a pointer to function type
773 const PointerType *PTy = dyn_cast<PointerType>(F->getType());
774 if (PTy == 0) error("Call to non function pointer value!");
775 const FunctionType *FTy = dyn_cast<FunctionType>(PTy->getElementType());
776 if (FTy == 0) error("Call to non function pointer value!");
778 std::vector<Value *> Params;
779 if (!FTy->isVarArg()) {
780 FunctionType::param_iterator It = FTy->param_begin();
782 for (unsigned i = 1, e = Oprnds.size(); i != e; ++i) {
783 if (It == FTy->param_end())
784 error("Invalid call instruction!");
785 Params.push_back(getValue(getTypeSlot(*It++), Oprnds[i]));
787 if (It != FTy->param_end())
788 error("Invalid call instruction!");
790 Oprnds.erase(Oprnds.begin(), Oprnds.begin()+1);
792 unsigned FirstVariableOperand;
793 if (Oprnds.size() < FTy->getNumParams())
794 error("Call instruction missing operands!");
796 // Read all of the fixed arguments
797 for (unsigned i = 0, e = FTy->getNumParams(); i != e; ++i)
799 getValue(getTypeSlot(FTy->getParamType(i)),Oprnds[i]));
801 FirstVariableOperand = FTy->getNumParams();
803 if ((Oprnds.size()-FirstVariableOperand) & 1)
804 error("Invalid call instruction!"); // Must be pairs of type/value
806 for (unsigned i = FirstVariableOperand, e = Oprnds.size();
808 Params.push_back(getValue(Oprnds[i], Oprnds[i+1]));
811 Result = new CallInst(F, Params);
812 if (isTailCall) cast<CallInst>(Result)->setTailCall();
813 if (CallingConv) cast<CallInst>(Result)->setCallingConv(CallingConv);
816 case Instruction::Invoke: { // Invoke C CC
817 if (Oprnds.size() < 3)
818 error("Invalid invoke instruction!");
819 Value *F = getValue(iType, Oprnds[0]);
821 // Check to make sure we have a pointer to function type
822 const PointerType *PTy = dyn_cast<PointerType>(F->getType());
824 error("Invoke to non function pointer value!");
825 const FunctionType *FTy = dyn_cast<FunctionType>(PTy->getElementType());
827 error("Invoke to non function pointer value!");
829 std::vector<Value *> Params;
830 BasicBlock *Normal, *Except;
831 unsigned CallingConv = Oprnds.back();
834 if (!FTy->isVarArg()) {
835 Normal = getBasicBlock(Oprnds[1]);
836 Except = getBasicBlock(Oprnds[2]);
838 FunctionType::param_iterator It = FTy->param_begin();
839 for (unsigned i = 3, e = Oprnds.size(); i != e; ++i) {
840 if (It == FTy->param_end())
841 error("Invalid invoke instruction!");
842 Params.push_back(getValue(getTypeSlot(*It++), Oprnds[i]));
844 if (It != FTy->param_end())
845 error("Invalid invoke instruction!");
847 Oprnds.erase(Oprnds.begin(), Oprnds.begin()+1);
849 Normal = getBasicBlock(Oprnds[0]);
850 Except = getBasicBlock(Oprnds[1]);
852 unsigned FirstVariableArgument = FTy->getNumParams()+2;
853 for (unsigned i = 2; i != FirstVariableArgument; ++i)
854 Params.push_back(getValue(getTypeSlot(FTy->getParamType(i-2)),
857 // Must be type/value pairs. If not, error out.
858 if (Oprnds.size()-FirstVariableArgument & 1)
859 error("Invalid invoke instruction!");
861 for (unsigned i = FirstVariableArgument; i < Oprnds.size(); i += 2)
862 Params.push_back(getValue(Oprnds[i], Oprnds[i+1]));
865 Result = new InvokeInst(F, Normal, Except, Params);
866 if (CallingConv) cast<InvokeInst>(Result)->setCallingConv(CallingConv);
869 case Instruction::Malloc: {
871 if (Oprnds.size() == 2)
872 Align = (1 << Oprnds[1]) >> 1;
873 else if (Oprnds.size() > 2)
874 error("Invalid malloc instruction!");
875 if (!isa<PointerType>(InstTy))
876 error("Invalid malloc instruction!");
878 Result = new MallocInst(cast<PointerType>(InstTy)->getElementType(),
879 getValue(Type::UIntTyID, Oprnds[0]), Align);
882 case Instruction::Alloca: {
884 if (Oprnds.size() == 2)
885 Align = (1 << Oprnds[1]) >> 1;
886 else if (Oprnds.size() > 2)
887 error("Invalid alloca instruction!");
888 if (!isa<PointerType>(InstTy))
889 error("Invalid alloca instruction!");
891 Result = new AllocaInst(cast<PointerType>(InstTy)->getElementType(),
892 getValue(Type::UIntTyID, Oprnds[0]), Align);
895 case Instruction::Free:
896 if (!isa<PointerType>(InstTy))
897 error("Invalid free instruction!");
898 Result = new FreeInst(getValue(iType, Oprnds[0]));
900 case Instruction::GetElementPtr: {
901 if (Oprnds.size() == 0 || !isa<PointerType>(InstTy))
902 error("Invalid getelementptr instruction!");
904 std::vector<Value*> Idx;
906 const Type *NextTy = InstTy;
907 for (unsigned i = 1, e = Oprnds.size(); i != e; ++i) {
908 const CompositeType *TopTy = dyn_cast_or_null<CompositeType>(NextTy);
910 error("Invalid getelementptr instruction!");
912 unsigned ValIdx = Oprnds[i];
914 // Struct indices are always uints, sequential type indices can be
915 // any of the 32 or 64-bit integer types. The actual choice of
916 // type is encoded in the low two bits of the slot number.
917 if (isa<StructType>(TopTy))
918 IdxTy = Type::UIntTyID;
920 switch (ValIdx & 3) {
922 case 0: IdxTy = Type::UIntTyID; break;
923 case 1: IdxTy = Type::IntTyID; break;
924 case 2: IdxTy = Type::ULongTyID; break;
925 case 3: IdxTy = Type::LongTyID; break;
929 Idx.push_back(getValue(IdxTy, ValIdx));
930 NextTy = GetElementPtrInst::getIndexedType(InstTy, Idx, true);
933 Result = new GetElementPtrInst(getValue(iType, Oprnds[0]), Idx);
936 case 62: // volatile load
937 case Instruction::Load:
938 if (Oprnds.size() != 1 || !isa<PointerType>(InstTy))
939 error("Invalid load instruction!");
940 Result = new LoadInst(getValue(iType, Oprnds[0]), "", Opcode == 62);
942 case 63: // volatile store
943 case Instruction::Store: {
944 if (!isa<PointerType>(InstTy) || Oprnds.size() != 2)
945 error("Invalid store instruction!");
947 Value *Ptr = getValue(iType, Oprnds[1]);
948 const Type *ValTy = cast<PointerType>(Ptr->getType())->getElementType();
949 Result = new StoreInst(getValue(getTypeSlot(ValTy), Oprnds[0]), Ptr,
953 case Instruction::Unwind:
954 if (Oprnds.size() != 0) error("Invalid unwind instruction!");
955 Result = new UnwindInst();
957 case Instruction::Unreachable:
958 if (Oprnds.size() != 0) error("Invalid unreachable instruction!");
959 Result = new UnreachableInst();
961 } // end switch(Opcode)
964 BB->getInstList().push_back(Result);
967 if (Result->getType() == InstTy)
970 TypeSlot = getTypeSlot(Result->getType());
972 insertValue(Result, TypeSlot, FunctionValues);
975 /// Get a particular numbered basic block, which might be a forward reference.
976 /// This works together with ParseInstructionList to handle these forward
977 /// references in a clean manner. This function is used when constructing
978 /// phi, br, switch, and other instructions that reference basic blocks.
979 /// Blocks are numbered sequentially as they appear in the function.
980 BasicBlock *BytecodeReader::getBasicBlock(unsigned ID) {
981 // Make sure there is room in the table...
982 if (ParsedBasicBlocks.size() <= ID) ParsedBasicBlocks.resize(ID+1);
984 // First check to see if this is a backwards reference, i.e. this block
985 // has already been created, or if the forward reference has already
987 if (ParsedBasicBlocks[ID])
988 return ParsedBasicBlocks[ID];
990 // Otherwise, the basic block has not yet been created. Do so and add it to
991 // the ParsedBasicBlocks list.
992 return ParsedBasicBlocks[ID] = new BasicBlock();
995 /// Parse all of the BasicBlock's & Instruction's in the body of a function.
996 /// In post 1.0 bytecode files, we no longer emit basic block individually,
997 /// in order to avoid per-basic-block overhead.
998 /// @returns the number of basic blocks encountered.
999 unsigned BytecodeReader::ParseInstructionList(Function* F) {
1000 unsigned BlockNo = 0;
1001 std::vector<unsigned> Args;
1003 while (moreInBlock()) {
1004 if (Handler) Handler->handleBasicBlockBegin(BlockNo);
1006 if (ParsedBasicBlocks.size() == BlockNo)
1007 ParsedBasicBlocks.push_back(BB = new BasicBlock());
1008 else if (ParsedBasicBlocks[BlockNo] == 0)
1009 BB = ParsedBasicBlocks[BlockNo] = new BasicBlock();
1011 BB = ParsedBasicBlocks[BlockNo];
1013 F->getBasicBlockList().push_back(BB);
1015 // Read instructions into this basic block until we get to a terminator
1016 while (moreInBlock() && !BB->getTerminator())
1017 ParseInstruction(Args, BB);
1019 if (!BB->getTerminator())
1020 error("Non-terminated basic block found!");
1022 if (Handler) Handler->handleBasicBlockEnd(BlockNo-1);
1028 /// Parse a symbol table. This works for both module level and function
1029 /// level symbol tables. For function level symbol tables, the CurrentFunction
1030 /// parameter must be non-zero and the ST parameter must correspond to
1031 /// CurrentFunction's symbol table. For Module level symbol tables, the
1032 /// CurrentFunction argument must be zero.
1033 void BytecodeReader::ParseSymbolTable(Function *CurrentFunction,
1035 if (Handler) Handler->handleSymbolTableBegin(CurrentFunction,ST);
1037 // Allow efficient basic block lookup by number.
1038 std::vector<BasicBlock*> BBMap;
1039 if (CurrentFunction)
1040 for (Function::iterator I = CurrentFunction->begin(),
1041 E = CurrentFunction->end(); I != E; ++I)
1044 // Symtab block header: [num entries]
1045 unsigned NumEntries = read_vbr_uint();
1046 for (unsigned i = 0; i < NumEntries; ++i) {
1047 // Symtab entry: [def slot #][name]
1048 unsigned slot = read_vbr_uint();
1049 std::string Name = read_str();
1050 const Type* T = getType(slot);
1051 ST->insert(Name, T);
1054 while (moreInBlock()) {
1055 // Symtab block header: [num entries][type id number]
1056 unsigned NumEntries = read_vbr_uint();
1057 unsigned Typ = read_vbr_uint();
1059 for (unsigned i = 0; i != NumEntries; ++i) {
1060 // Symtab entry: [def slot #][name]
1061 unsigned slot = read_vbr_uint();
1062 std::string Name = read_str();
1064 if (Typ == Type::LabelTyID) {
1065 if (slot < BBMap.size())
1068 V = getValue(Typ, slot, false); // Find mapping...
1071 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 = read_vbr_uint();
1083 const Type *Typ = getGlobalTableType(TypeSlot);
1084 CompactionTypes.push_back(std::make_pair(Typ, TypeSlot));
1085 if (Handler) Handler->handleCompactionTableType(i, TypeSlot, Typ);
1089 /// Parse a compaction table.
1090 void BytecodeReader::ParseCompactionTable() {
1092 // Notify handler that we're beginning a compaction table.
1093 if (Handler) Handler->handleCompactionTableBegin();
1095 // Get the types for the compaction table.
1096 unsigned NumEntries = read_vbr_uint();
1097 ParseCompactionTypes(NumEntries);
1099 // Compaction tables live in separate blocks so we have to loop
1100 // until we've read the whole thing.
1101 while (moreInBlock()) {
1102 // Read the number of Value* entries in the compaction table
1103 unsigned NumEntries = read_vbr_uint();
1106 // Decode the type from value read in. Most compaction table
1107 // planes will have one or two entries in them. If that's the
1108 // case then the length is encoded in the bottom two bits and
1109 // the higher bits encode the type. This saves another VBR value.
1110 if ((NumEntries & 3) == 3) {
1111 // In this case, both low-order bits are set (value 3). This
1112 // is a signal that the typeid follows.
1114 Ty = read_vbr_uint();
1116 // In this case, the low-order bits specify the number of entries
1117 // and the high order bits specify the type.
1118 Ty = NumEntries >> 2;
1122 // Make sure we have enough room for the plane.
1123 if (Ty >= CompactionValues.size())
1124 CompactionValues.resize(Ty+1);
1126 // Make sure the plane is empty or we have some kind of error.
1127 if (!CompactionValues[Ty].empty())
1128 error("Compaction table plane contains multiple entries!");
1130 // Notify handler about the plane.
1131 if (Handler) Handler->handleCompactionTablePlane(Ty, NumEntries);
1133 // Push the implicit zero.
1134 CompactionValues[Ty].push_back(Constant::getNullValue(getType(Ty)));
1136 // Read in each of the entries, put them in the compaction table
1137 // and notify the handler that we have a new compaction table value.
1138 for (unsigned i = 0; i != NumEntries; ++i) {
1139 unsigned ValSlot = read_vbr_uint();
1140 Value *V = getGlobalTableValue(Ty, ValSlot);
1141 CompactionValues[Ty].push_back(V);
1142 if (Handler) Handler->handleCompactionTableValue(i, Ty, ValSlot);
1145 // Notify handler that the compaction table is done.
1146 if (Handler) Handler->handleCompactionTableEnd();
1149 // Parse a single type. The typeid is read in first. If its a primitive type
1150 // then nothing else needs to be read, we know how to instantiate it. If its
1151 // a derived type, then additional data is read to fill out the type
1153 const Type *BytecodeReader::ParseType() {
1154 unsigned PrimType = read_vbr_uint();
1155 const Type *Result = 0;
1156 if ((Result = Type::getPrimitiveType((Type::TypeID)PrimType)))
1160 case Type::FunctionTyID: {
1161 const Type *RetType = readType();
1163 unsigned NumParams = read_vbr_uint();
1165 std::vector<const Type*> Params;
1167 Params.push_back(readType());
1169 bool isVarArg = Params.size() && Params.back() == Type::VoidTy;
1170 if (isVarArg) Params.pop_back();
1172 Result = FunctionType::get(RetType, Params, isVarArg);
1175 case Type::ArrayTyID: {
1176 const Type *ElementType = readType();
1177 unsigned NumElements = read_vbr_uint();
1178 Result = ArrayType::get(ElementType, NumElements);
1181 case Type::PackedTyID: {
1182 const Type *ElementType = readType();
1183 unsigned NumElements = read_vbr_uint();
1184 Result = PackedType::get(ElementType, NumElements);
1187 case Type::StructTyID: {
1188 std::vector<const Type*> Elements;
1189 unsigned Typ = read_vbr_uint();
1190 while (Typ) { // List is terminated by void/0 typeid
1191 Elements.push_back(getType(Typ));
1192 Typ = read_vbr_uint();
1195 Result = StructType::get(Elements);
1198 case Type::PointerTyID: {
1199 Result = PointerType::get(readType());
1203 case Type::OpaqueTyID: {
1204 Result = OpaqueType::get();
1209 error("Don't know how to deserialize primitive type " + utostr(PrimType));
1212 if (Handler) Handler->handleType(Result);
1216 // ParseTypes - We have to use this weird code to handle recursive
1217 // types. We know that recursive types will only reference the current slab of
1218 // values in the type plane, but they can forward reference types before they
1219 // have been read. For example, Type #0 might be '{ Ty#1 }' and Type #1 might
1220 // be 'Ty#0*'. When reading Type #0, type number one doesn't exist. To fix
1221 // this ugly problem, we pessimistically insert an opaque type for each type we
1222 // are about to read. This means that forward references will resolve to
1223 // something and when we reread the type later, we can replace the opaque type
1224 // with a new resolved concrete type.
1226 void BytecodeReader::ParseTypes(TypeListTy &Tab, unsigned NumEntries){
1227 assert(Tab.size() == 0 && "should not have read type constants in before!");
1229 // Insert a bunch of opaque types to be resolved later...
1230 Tab.reserve(NumEntries);
1231 for (unsigned i = 0; i != NumEntries; ++i)
1232 Tab.push_back(OpaqueType::get());
1235 Handler->handleTypeList(NumEntries);
1237 // If we are about to resolve types, make sure the type cache is clear.
1239 ModuleTypeIDCache.clear();
1241 // Loop through reading all of the types. Forward types will make use of the
1242 // opaque types just inserted.
1244 for (unsigned i = 0; i != NumEntries; ++i) {
1245 const Type* NewTy = ParseType();
1246 const Type* OldTy = Tab[i].get();
1248 error("Couldn't parse type!");
1250 // Don't directly push the new type on the Tab. Instead we want to replace
1251 // the opaque type we previously inserted with the new concrete value. This
1252 // approach helps with forward references to types. The refinement from the
1253 // abstract (opaque) type to the new type causes all uses of the abstract
1254 // type to use the concrete type (NewTy). This will also cause the opaque
1255 // type to be deleted.
1256 cast<DerivedType>(const_cast<Type*>(OldTy))->refineAbstractTypeTo(NewTy);
1258 // This should have replaced the old opaque type with the new type in the
1259 // value table... or with a preexisting type that was already in the system.
1260 // Let's just make sure it did.
1261 assert(Tab[i] != OldTy && "refineAbstractType didn't work!");
1265 /// Parse a single constant value
1266 Value *BytecodeReader::ParseConstantPoolValue(unsigned TypeID) {
1267 // We must check for a ConstantExpr before switching by type because
1268 // a ConstantExpr can be of any type, and has no explicit value.
1270 // 0 if not expr; numArgs if is expr
1271 unsigned isExprNumArgs = read_vbr_uint();
1273 if (isExprNumArgs) {
1274 // 'undef' is encoded with 'exprnumargs' == 1.
1275 if (isExprNumArgs == 1)
1276 return UndefValue::get(getType(TypeID));
1278 // Inline asm is encoded with exprnumargs == ~0U.
1279 if (isExprNumArgs == ~0U) {
1280 std::string AsmStr = read_str();
1281 std::string ConstraintStr = read_str();
1282 unsigned Flags = read_vbr_uint();
1284 const PointerType *PTy = dyn_cast<PointerType>(getType(TypeID));
1285 const FunctionType *FTy =
1286 PTy ? dyn_cast<FunctionType>(PTy->getElementType()) : 0;
1288 if (!FTy || !InlineAsm::Verify(FTy, ConstraintStr))
1289 error("Invalid constraints for inline asm");
1291 error("Invalid flags for inline asm");
1292 bool HasSideEffects = Flags & 1;
1293 return InlineAsm::get(FTy, AsmStr, ConstraintStr, HasSideEffects);
1298 // FIXME: Encoding of constant exprs could be much more compact!
1299 std::vector<Constant*> ArgVec;
1300 ArgVec.reserve(isExprNumArgs);
1301 unsigned Opcode = read_vbr_uint();
1303 // Read the slot number and types of each of the arguments
1304 for (unsigned i = 0; i != isExprNumArgs; ++i) {
1305 unsigned ArgValSlot = read_vbr_uint();
1306 unsigned ArgTypeSlot = read_vbr_uint();
1308 // Get the arg value from its slot if it exists, otherwise a placeholder
1309 ArgVec.push_back(getConstantValue(ArgTypeSlot, ArgValSlot));
1312 // Construct a ConstantExpr of the appropriate kind
1313 if (isExprNumArgs == 1) { // All one-operand expressions
1314 if (!Instruction::isCast(Opcode))
1315 error("Only cast instruction has one argument for ConstantExpr");
1317 Constant *Result = ConstantExpr::getCast(ArgVec[0], getType(TypeID));
1318 if (Handler) Handler->handleConstantExpression(Opcode, ArgVec, Result);
1320 } else if (Opcode == Instruction::GetElementPtr) { // GetElementPtr
1321 std::vector<Constant*> IdxList(ArgVec.begin()+1, ArgVec.end());
1322 Constant *Result = ConstantExpr::getGetElementPtr(ArgVec[0], IdxList);
1323 if (Handler) Handler->handleConstantExpression(Opcode, ArgVec, Result);
1325 } else if (Opcode == Instruction::Select) {
1326 if (ArgVec.size() != 3)
1327 error("Select instruction must have three arguments.");
1328 Constant* Result = ConstantExpr::getSelect(ArgVec[0], ArgVec[1],
1330 if (Handler) Handler->handleConstantExpression(Opcode, ArgVec, Result);
1332 } else if (Opcode == Instruction::ExtractElement) {
1333 if (ArgVec.size() != 2 ||
1334 !ExtractElementInst::isValidOperands(ArgVec[0], ArgVec[1]))
1335 error("Invalid extractelement constand expr arguments");
1336 Constant* Result = ConstantExpr::getExtractElement(ArgVec[0], ArgVec[1]);
1337 if (Handler) Handler->handleConstantExpression(Opcode, ArgVec, Result);
1339 } else if (Opcode == Instruction::InsertElement) {
1340 if (ArgVec.size() != 3 ||
1341 !InsertElementInst::isValidOperands(ArgVec[0], ArgVec[1], ArgVec[2]))
1342 error("Invalid insertelement constand expr arguments");
1345 ConstantExpr::getInsertElement(ArgVec[0], ArgVec[1], ArgVec[2]);
1346 if (Handler) Handler->handleConstantExpression(Opcode, ArgVec, Result);
1348 } else if (Opcode == Instruction::ShuffleVector) {
1349 if (ArgVec.size() != 3 ||
1350 !ShuffleVectorInst::isValidOperands(ArgVec[0], ArgVec[1], ArgVec[2]))
1351 error("Invalid shufflevector constant expr arguments.");
1353 ConstantExpr::getShuffleVector(ArgVec[0], ArgVec[1], ArgVec[2]);
1354 if (Handler) Handler->handleConstantExpression(Opcode, ArgVec, Result);
1356 } else if (Opcode == Instruction::ICmp) {
1357 if (ArgVec.size() != 2)
1358 error("Invalid ICmp constant expr arguments");
1359 unsigned short pred = read_vbr_uint();
1360 return ConstantExpr::getICmp(pred, ArgVec[0], ArgVec[1]);
1361 } else if (Opcode == Instruction::FCmp) {
1362 if (ArgVec.size() != 2)
1363 error("Invalid FCmp constant expr arguments");
1364 unsigned short pred = read_vbr_uint();
1365 return ConstantExpr::getFCmp(pred, ArgVec[0], ArgVec[1]);
1366 } else { // All other 2-operand expressions
1367 Constant* Result = ConstantExpr::get(Opcode, ArgVec[0], ArgVec[1]);
1368 if (Handler) Handler->handleConstantExpression(Opcode, ArgVec, Result);
1373 // Ok, not an ConstantExpr. We now know how to read the given type...
1374 const Type *Ty = getType(TypeID);
1375 Constant *Result = 0;
1376 switch (Ty->getTypeID()) {
1377 case Type::BoolTyID: {
1378 unsigned Val = read_vbr_uint();
1379 if (Val != 0 && Val != 1)
1380 error("Invalid boolean value read.");
1381 Result = ConstantBool::get(Val == 1);
1382 if (Handler) Handler->handleConstantValue(Result);
1386 case Type::UByteTyID: // Unsigned integer types...
1387 case Type::UShortTyID:
1388 case Type::UIntTyID: {
1389 unsigned Val = read_vbr_uint();
1390 if (!ConstantInt::isValueValidForType(Ty, uint64_t(Val)))
1391 error("Invalid unsigned byte/short/int read.");
1392 Result = ConstantInt::get(Ty, Val);
1393 if (Handler) Handler->handleConstantValue(Result);
1397 case Type::ULongTyID:
1398 Result = ConstantInt::get(Ty, read_vbr_uint64());
1399 if (Handler) Handler->handleConstantValue(Result);
1402 case Type::SByteTyID: // Signed integer types...
1403 case Type::ShortTyID:
1405 case Type::LongTyID: {
1406 int64_t Val = read_vbr_int64();
1407 if (!ConstantInt::isValueValidForType(Ty, Val))
1408 error("Invalid signed byte/short/int/long read.");
1409 Result = ConstantInt::get(Ty, Val);
1410 if (Handler) Handler->handleConstantValue(Result);
1414 case Type::FloatTyID: {
1417 Result = ConstantFP::get(Ty, Val);
1418 if (Handler) Handler->handleConstantValue(Result);
1422 case Type::DoubleTyID: {
1425 Result = ConstantFP::get(Ty, Val);
1426 if (Handler) Handler->handleConstantValue(Result);
1430 case Type::ArrayTyID: {
1431 const ArrayType *AT = cast<ArrayType>(Ty);
1432 unsigned NumElements = AT->getNumElements();
1433 unsigned TypeSlot = getTypeSlot(AT->getElementType());
1434 std::vector<Constant*> Elements;
1435 Elements.reserve(NumElements);
1436 while (NumElements--) // Read all of the elements of the constant.
1437 Elements.push_back(getConstantValue(TypeSlot,
1439 Result = ConstantArray::get(AT, Elements);
1440 if (Handler) Handler->handleConstantArray(AT, Elements, TypeSlot, Result);
1444 case Type::StructTyID: {
1445 const StructType *ST = cast<StructType>(Ty);
1447 std::vector<Constant *> Elements;
1448 Elements.reserve(ST->getNumElements());
1449 for (unsigned i = 0; i != ST->getNumElements(); ++i)
1450 Elements.push_back(getConstantValue(ST->getElementType(i),
1453 Result = ConstantStruct::get(ST, Elements);
1454 if (Handler) Handler->handleConstantStruct(ST, Elements, Result);
1458 case Type::PackedTyID: {
1459 const PackedType *PT = cast<PackedType>(Ty);
1460 unsigned NumElements = PT->getNumElements();
1461 unsigned TypeSlot = getTypeSlot(PT->getElementType());
1462 std::vector<Constant*> Elements;
1463 Elements.reserve(NumElements);
1464 while (NumElements--) // Read all of the elements of the constant.
1465 Elements.push_back(getConstantValue(TypeSlot,
1467 Result = ConstantPacked::get(PT, Elements);
1468 if (Handler) Handler->handleConstantPacked(PT, Elements, TypeSlot, Result);
1472 case Type::PointerTyID: { // ConstantPointerRef value (backwards compat).
1473 const PointerType *PT = cast<PointerType>(Ty);
1474 unsigned Slot = read_vbr_uint();
1476 // Check to see if we have already read this global variable...
1477 Value *Val = getValue(TypeID, Slot, false);
1479 GlobalValue *GV = dyn_cast<GlobalValue>(Val);
1480 if (!GV) error("GlobalValue not in ValueTable!");
1481 if (Handler) Handler->handleConstantPointer(PT, Slot, GV);
1484 error("Forward references are not allowed here.");
1489 error("Don't know how to deserialize constant value of type '" +
1490 Ty->getDescription());
1494 // Check that we didn't read a null constant if they are implicit for this
1495 // type plane. Do not do this check for constantexprs, as they may be folded
1496 // to a null value in a way that isn't predicted when a .bc file is initially
1498 assert((!isa<Constant>(Result) || !cast<Constant>(Result)->isNullValue()) ||
1499 !hasImplicitNull(TypeID) &&
1500 "Cannot read null values from bytecode!");
1504 /// Resolve references for constants. This function resolves the forward
1505 /// referenced constants in the ConstantFwdRefs map. It uses the
1506 /// replaceAllUsesWith method of Value class to substitute the placeholder
1507 /// instance with the actual instance.
1508 void BytecodeReader::ResolveReferencesToConstant(Constant *NewV, unsigned Typ,
1510 ConstantRefsType::iterator I =
1511 ConstantFwdRefs.find(std::make_pair(Typ, Slot));
1512 if (I == ConstantFwdRefs.end()) return; // Never forward referenced?
1514 Value *PH = I->second; // Get the placeholder...
1515 PH->replaceAllUsesWith(NewV);
1516 delete PH; // Delete the old placeholder
1517 ConstantFwdRefs.erase(I); // Remove the map entry for it
1520 /// Parse the constant strings section.
1521 void BytecodeReader::ParseStringConstants(unsigned NumEntries, ValueTable &Tab){
1522 for (; NumEntries; --NumEntries) {
1523 unsigned Typ = read_vbr_uint();
1524 const Type *Ty = getType(Typ);
1525 if (!isa<ArrayType>(Ty))
1526 error("String constant data invalid!");
1528 const ArrayType *ATy = cast<ArrayType>(Ty);
1529 if (ATy->getElementType() != Type::SByteTy &&
1530 ATy->getElementType() != Type::UByteTy)
1531 error("String constant data invalid!");
1533 // Read character data. The type tells us how long the string is.
1534 char *Data = reinterpret_cast<char *>(alloca(ATy->getNumElements()));
1535 read_data(Data, Data+ATy->getNumElements());
1537 std::vector<Constant*> Elements(ATy->getNumElements());
1538 const Type* ElemType = ATy->getElementType();
1539 for (unsigned i = 0, e = ATy->getNumElements(); i != e; ++i)
1540 Elements[i] = ConstantInt::get(ElemType, (unsigned char)Data[i]);
1542 // Create the constant, inserting it as needed.
1543 Constant *C = ConstantArray::get(ATy, Elements);
1544 unsigned Slot = insertValue(C, Typ, Tab);
1545 ResolveReferencesToConstant(C, Typ, Slot);
1546 if (Handler) Handler->handleConstantString(cast<ConstantArray>(C));
1550 /// Parse the constant pool.
1551 void BytecodeReader::ParseConstantPool(ValueTable &Tab,
1552 TypeListTy &TypeTab,
1554 if (Handler) Handler->handleGlobalConstantsBegin();
1556 /// In LLVM 1.3 Type does not derive from Value so the types
1557 /// do not occupy a plane. Consequently, we read the types
1558 /// first in the constant pool.
1560 unsigned NumEntries = read_vbr_uint();
1561 ParseTypes(TypeTab, NumEntries);
1564 while (moreInBlock()) {
1565 unsigned NumEntries = read_vbr_uint();
1566 unsigned Typ = read_vbr_uint();
1568 if (Typ == Type::VoidTyID) {
1569 /// Use of Type::VoidTyID is a misnomer. It actually means
1570 /// that the following plane is constant strings
1571 assert(&Tab == &ModuleValues && "Cannot read strings in functions!");
1572 ParseStringConstants(NumEntries, Tab);
1574 for (unsigned i = 0; i < NumEntries; ++i) {
1575 Value *V = ParseConstantPoolValue(Typ);
1576 assert(V && "ParseConstantPoolValue returned NULL!");
1577 unsigned Slot = insertValue(V, Typ, Tab);
1579 // If we are reading a function constant table, make sure that we adjust
1580 // the slot number to be the real global constant number.
1582 if (&Tab != &ModuleValues && Typ < ModuleValues.size() &&
1584 Slot += ModuleValues[Typ]->size();
1585 if (Constant *C = dyn_cast<Constant>(V))
1586 ResolveReferencesToConstant(C, Typ, Slot);
1591 // After we have finished parsing the constant pool, we had better not have
1592 // any dangling references left.
1593 if (!ConstantFwdRefs.empty()) {
1594 ConstantRefsType::const_iterator I = ConstantFwdRefs.begin();
1595 Constant* missingConst = I->second;
1596 error(utostr(ConstantFwdRefs.size()) +
1597 " unresolved constant reference exist. First one is '" +
1598 missingConst->getName() + "' of type '" +
1599 missingConst->getType()->getDescription() + "'.");
1602 checkPastBlockEnd("Constant Pool");
1603 if (Handler) Handler->handleGlobalConstantsEnd();
1606 /// Parse the contents of a function. Note that this function can be
1607 /// called lazily by materializeFunction
1608 /// @see materializeFunction
1609 void BytecodeReader::ParseFunctionBody(Function* F) {
1611 unsigned FuncSize = BlockEnd - At;
1612 GlobalValue::LinkageTypes Linkage = GlobalValue::ExternalLinkage;
1614 unsigned LinkageType = read_vbr_uint();
1615 switch (LinkageType) {
1616 case 0: Linkage = GlobalValue::ExternalLinkage; break;
1617 case 1: Linkage = GlobalValue::WeakLinkage; break;
1618 case 2: Linkage = GlobalValue::AppendingLinkage; break;
1619 case 3: Linkage = GlobalValue::InternalLinkage; break;
1620 case 4: Linkage = GlobalValue::LinkOnceLinkage; break;
1621 case 5: Linkage = GlobalValue::DLLImportLinkage; break;
1622 case 6: Linkage = GlobalValue::DLLExportLinkage; break;
1623 case 7: Linkage = GlobalValue::ExternalWeakLinkage; break;
1625 error("Invalid linkage type for Function.");
1626 Linkage = GlobalValue::InternalLinkage;
1630 F->setLinkage(Linkage);
1631 if (Handler) Handler->handleFunctionBegin(F,FuncSize);
1633 // Keep track of how many basic blocks we have read in...
1634 unsigned BlockNum = 0;
1635 bool InsertedArguments = false;
1637 BufPtr MyEnd = BlockEnd;
1638 while (At < MyEnd) {
1639 unsigned Type, Size;
1641 read_block(Type, Size);
1644 case BytecodeFormat::ConstantPoolBlockID:
1645 if (!InsertedArguments) {
1646 // Insert arguments into the value table before we parse the first basic
1647 // block in the function, but after we potentially read in the
1648 // compaction table.
1650 InsertedArguments = true;
1653 ParseConstantPool(FunctionValues, FunctionTypes, true);
1656 case BytecodeFormat::CompactionTableBlockID:
1657 ParseCompactionTable();
1660 case BytecodeFormat::InstructionListBlockID: {
1661 // Insert arguments into the value table before we parse the instruction
1662 // list for the function, but after we potentially read in the compaction
1664 if (!InsertedArguments) {
1666 InsertedArguments = true;
1670 error("Already parsed basic blocks!");
1671 BlockNum = ParseInstructionList(F);
1675 case BytecodeFormat::SymbolTableBlockID:
1676 ParseSymbolTable(F, &F->getSymbolTable());
1682 error("Wrapped around reading bytecode.");
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 bool BytecodeReader::ParseFunction(Function* Func, std::string* ErrMsg) {
1745 if (setjmp(context))
1748 // Find {start, end} pointers and slot in the map. If not there, we're done.
1749 LazyFunctionMap::iterator Fi = LazyFunctionLoadMap.find(Func);
1751 // Make sure we found it
1752 if (Fi == LazyFunctionLoadMap.end()) {
1753 error("Unrecognized function of type " + Func->getType()->getDescription());
1757 BlockStart = At = Fi->second.Buf;
1758 BlockEnd = Fi->second.EndBuf;
1759 assert(Fi->first == Func && "Found wrong function?");
1761 LazyFunctionLoadMap.erase(Fi);
1763 this->ParseFunctionBody(Func);
1767 /// The ParseAllFunctionBodies method parses through all the previously
1768 /// unparsed functions in the bytecode file. If you want to completely parse
1769 /// a bytecode file, this method should be called after Parsebytecode because
1770 /// Parsebytecode only records the locations in the bytecode file of where
1771 /// the function definitions are located. This function uses that information
1772 /// to materialize the functions.
1773 /// @see ParseBytecode
1774 bool BytecodeReader::ParseAllFunctionBodies(std::string* ErrMsg) {
1775 if (setjmp(context))
1778 LazyFunctionMap::iterator Fi = LazyFunctionLoadMap.begin();
1779 LazyFunctionMap::iterator Fe = LazyFunctionLoadMap.end();
1782 Function* Func = Fi->first;
1783 BlockStart = At = Fi->second.Buf;
1784 BlockEnd = Fi->second.EndBuf;
1785 ParseFunctionBody(Func);
1788 LazyFunctionLoadMap.clear();
1792 /// Parse the global type list
1793 void BytecodeReader::ParseGlobalTypes() {
1794 // Read the number of types
1795 unsigned NumEntries = read_vbr_uint();
1796 ParseTypes(ModuleTypes, NumEntries);
1799 /// Parse the Global info (types, global vars, constants)
1800 void BytecodeReader::ParseModuleGlobalInfo() {
1802 if (Handler) Handler->handleModuleGlobalsBegin();
1804 // SectionID - If a global has an explicit section specified, this map
1805 // remembers the ID until we can translate it into a string.
1806 std::map<GlobalValue*, unsigned> SectionID;
1808 // Read global variables...
1809 unsigned VarType = read_vbr_uint();
1810 while (VarType != Type::VoidTyID) { // List is terminated by Void
1811 // VarType Fields: bit0 = isConstant, bit1 = hasInitializer, bit2,3,4 =
1812 // Linkage, bit4+ = slot#
1813 unsigned SlotNo = VarType >> 5;
1814 unsigned LinkageID = (VarType >> 2) & 7;
1815 bool isConstant = VarType & 1;
1816 bool hasInitializer = (VarType & 2) != 0;
1817 unsigned Alignment = 0;
1818 unsigned GlobalSectionID = 0;
1820 // An extension word is present when linkage = 3 (internal) and hasinit = 0.
1821 if (LinkageID == 3 && !hasInitializer) {
1822 unsigned ExtWord = read_vbr_uint();
1823 // The extension word has this format: bit 0 = has initializer, bit 1-3 =
1824 // linkage, bit 4-8 = alignment (log2), bits 10+ = future use.
1825 hasInitializer = ExtWord & 1;
1826 LinkageID = (ExtWord >> 1) & 7;
1827 Alignment = (1 << ((ExtWord >> 4) & 31)) >> 1;
1829 if (ExtWord & (1 << 9)) // Has a section ID.
1830 GlobalSectionID = read_vbr_uint();
1833 GlobalValue::LinkageTypes Linkage;
1834 switch (LinkageID) {
1835 case 0: Linkage = GlobalValue::ExternalLinkage; break;
1836 case 1: Linkage = GlobalValue::WeakLinkage; break;
1837 case 2: Linkage = GlobalValue::AppendingLinkage; break;
1838 case 3: Linkage = GlobalValue::InternalLinkage; break;
1839 case 4: Linkage = GlobalValue::LinkOnceLinkage; break;
1840 case 5: Linkage = GlobalValue::DLLImportLinkage; break;
1841 case 6: Linkage = GlobalValue::DLLExportLinkage; break;
1842 case 7: Linkage = GlobalValue::ExternalWeakLinkage; break;
1844 error("Unknown linkage type: " + utostr(LinkageID));
1845 Linkage = GlobalValue::InternalLinkage;
1849 const Type *Ty = getType(SlotNo);
1851 error("Global has no type! SlotNo=" + utostr(SlotNo));
1853 if (!isa<PointerType>(Ty))
1854 error("Global not a pointer type! Ty= " + Ty->getDescription());
1856 const Type *ElTy = cast<PointerType>(Ty)->getElementType();
1858 // Create the global variable...
1859 GlobalVariable *GV = new GlobalVariable(ElTy, isConstant, Linkage,
1861 GV->setAlignment(Alignment);
1862 insertValue(GV, SlotNo, ModuleValues);
1864 if (GlobalSectionID != 0)
1865 SectionID[GV] = GlobalSectionID;
1867 unsigned initSlot = 0;
1868 if (hasInitializer) {
1869 initSlot = read_vbr_uint();
1870 GlobalInits.push_back(std::make_pair(GV, initSlot));
1873 // Notify handler about the global value.
1875 Handler->handleGlobalVariable(ElTy, isConstant, Linkage, SlotNo,initSlot);
1878 VarType = read_vbr_uint();
1881 // Read the function objects for all of the functions that are coming
1882 unsigned FnSignature = read_vbr_uint();
1884 // List is terminated by VoidTy.
1885 while (((FnSignature & (~0U >> 1)) >> 5) != Type::VoidTyID) {
1886 const Type *Ty = getType((FnSignature & (~0U >> 1)) >> 5);
1887 if (!isa<PointerType>(Ty) ||
1888 !isa<FunctionType>(cast<PointerType>(Ty)->getElementType())) {
1889 error("Function not a pointer to function type! Ty = " +
1890 Ty->getDescription());
1893 // We create functions by passing the underlying FunctionType to create...
1894 const FunctionType* FTy =
1895 cast<FunctionType>(cast<PointerType>(Ty)->getElementType());
1897 // Insert the place holder.
1898 Function *Func = new Function(FTy, GlobalValue::ExternalLinkage,
1901 insertValue(Func, (FnSignature & (~0U >> 1)) >> 5, ModuleValues);
1903 // Flags are not used yet.
1904 unsigned Flags = FnSignature & 31;
1906 // Save this for later so we know type of lazily instantiated functions.
1907 // Note that known-external functions do not have FunctionInfo blocks, so we
1908 // do not add them to the FunctionSignatureList.
1909 if ((Flags & (1 << 4)) == 0)
1910 FunctionSignatureList.push_back(Func);
1912 // Get the calling convention from the low bits.
1913 unsigned CC = Flags & 15;
1914 unsigned Alignment = 0;
1915 if (FnSignature & (1 << 31)) { // Has extension word?
1916 unsigned ExtWord = read_vbr_uint();
1917 Alignment = (1 << (ExtWord & 31)) >> 1;
1918 CC |= ((ExtWord >> 5) & 15) << 4;
1920 if (ExtWord & (1 << 10)) // Has a section ID.
1921 SectionID[Func] = read_vbr_uint();
1923 // Parse external declaration linkage
1924 switch ((ExtWord >> 11) & 3) {
1926 case 1: Func->setLinkage(Function::DLLImportLinkage); break;
1927 case 2: Func->setLinkage(Function::ExternalWeakLinkage); break;
1928 default: assert(0 && "Unsupported external linkage");
1932 Func->setCallingConv(CC-1);
1933 Func->setAlignment(Alignment);
1935 if (Handler) Handler->handleFunctionDeclaration(Func);
1937 // Get the next function signature.
1938 FnSignature = read_vbr_uint();
1941 // Now that the function signature list is set up, reverse it so that we can
1942 // remove elements efficiently from the back of the vector.
1943 std::reverse(FunctionSignatureList.begin(), FunctionSignatureList.end());
1945 /// SectionNames - This contains the list of section names encoded in the
1946 /// moduleinfoblock. Functions and globals with an explicit section index
1947 /// into this to get their section name.
1948 std::vector<std::string> SectionNames;
1950 // Read in the dependent library information.
1951 unsigned num_dep_libs = read_vbr_uint();
1952 std::string dep_lib;
1953 while (num_dep_libs--) {
1954 dep_lib = read_str();
1955 TheModule->addLibrary(dep_lib);
1957 Handler->handleDependentLibrary(dep_lib);
1960 // Read target triple and place into the module.
1961 std::string triple = read_str();
1962 TheModule->setTargetTriple(triple);
1964 Handler->handleTargetTriple(triple);
1966 if (At != BlockEnd) {
1967 // If the file has section info in it, read the section names now.
1968 unsigned NumSections = read_vbr_uint();
1969 while (NumSections--)
1970 SectionNames.push_back(read_str());
1973 // If the file has module-level inline asm, read it now.
1975 TheModule->setModuleInlineAsm(read_str());
1977 // If any globals are in specified sections, assign them now.
1978 for (std::map<GlobalValue*, unsigned>::iterator I = SectionID.begin(), E =
1979 SectionID.end(); I != E; ++I)
1981 if (I->second > SectionID.size())
1982 error("SectionID out of range for global!");
1983 I->first->setSection(SectionNames[I->second-1]);
1986 // This is for future proofing... in the future extra fields may be added that
1987 // we don't understand, so we transparently ignore them.
1991 if (Handler) Handler->handleModuleGlobalsEnd();
1994 /// Parse the version information and decode it by setting flags on the
1995 /// Reader that enable backward compatibility of the reader.
1996 void BytecodeReader::ParseVersionInfo() {
1997 unsigned Version = read_vbr_uint();
1999 // Unpack version number: low four bits are for flags, top bits = version
2000 Module::Endianness Endianness;
2001 Module::PointerSize PointerSize;
2002 Endianness = (Version & 1) ? Module::BigEndian : Module::LittleEndian;
2003 PointerSize = (Version & 2) ? Module::Pointer64 : Module::Pointer32;
2005 bool hasNoEndianness = Version & 4;
2006 bool hasNoPointerSize = Version & 8;
2008 RevisionNum = Version >> 4;
2010 // We don't provide backwards compatibility in the Reader any more. To
2011 // upgrade, the user should use llvm-upgrade.
2012 if (RevisionNum < 7)
2013 error("Bytecode formats < 7 are no longer supported. Use llvm-upgrade.");
2015 if (hasNoEndianness) Endianness = Module::AnyEndianness;
2016 if (hasNoPointerSize) PointerSize = Module::AnyPointerSize;
2018 TheModule->setEndianness(Endianness);
2019 TheModule->setPointerSize(PointerSize);
2021 if (Handler) Handler->handleVersionInfo(RevisionNum, Endianness, PointerSize);
2024 /// Parse a whole module.
2025 void BytecodeReader::ParseModule() {
2026 unsigned Type, Size;
2028 FunctionSignatureList.clear(); // Just in case...
2030 // Read into instance variables...
2033 bool SeenModuleGlobalInfo = false;
2034 bool SeenGlobalTypePlane = false;
2035 BufPtr MyEnd = BlockEnd;
2036 while (At < MyEnd) {
2038 read_block(Type, Size);
2042 case BytecodeFormat::GlobalTypePlaneBlockID:
2043 if (SeenGlobalTypePlane)
2044 error("Two GlobalTypePlane Blocks Encountered!");
2048 SeenGlobalTypePlane = true;
2051 case BytecodeFormat::ModuleGlobalInfoBlockID:
2052 if (SeenModuleGlobalInfo)
2053 error("Two ModuleGlobalInfo Blocks Encountered!");
2054 ParseModuleGlobalInfo();
2055 SeenModuleGlobalInfo = true;
2058 case BytecodeFormat::ConstantPoolBlockID:
2059 ParseConstantPool(ModuleValues, ModuleTypes,false);
2062 case BytecodeFormat::FunctionBlockID:
2063 ParseFunctionLazily();
2066 case BytecodeFormat::SymbolTableBlockID:
2067 ParseSymbolTable(0, &TheModule->getSymbolTable());
2073 error("Unexpected Block of Type #" + utostr(Type) + " encountered!");
2080 // After the module constant pool has been read, we can safely initialize
2081 // global variables...
2082 while (!GlobalInits.empty()) {
2083 GlobalVariable *GV = GlobalInits.back().first;
2084 unsigned Slot = GlobalInits.back().second;
2085 GlobalInits.pop_back();
2087 // Look up the initializer value...
2088 // FIXME: Preserve this type ID!
2090 const llvm::PointerType* GVType = GV->getType();
2091 unsigned TypeSlot = getTypeSlot(GVType->getElementType());
2092 if (Constant *CV = getConstantValue(TypeSlot, Slot)) {
2093 if (GV->hasInitializer())
2094 error("Global *already* has an initializer?!");
2095 if (Handler) Handler->handleGlobalInitializer(GV,CV);
2096 GV->setInitializer(CV);
2098 error("Cannot find initializer value.");
2101 if (!ConstantFwdRefs.empty())
2102 error("Use of undefined constants in a module");
2104 /// Make sure we pulled them all out. If we didn't then there's a declaration
2105 /// but a missing body. That's not allowed.
2106 if (!FunctionSignatureList.empty())
2107 error("Function declared, but bytecode stream ended before definition");
2110 /// This function completely parses a bytecode buffer given by the \p Buf
2111 /// and \p Length parameters.
2112 bool BytecodeReader::ParseBytecode(volatile BufPtr Buf, unsigned Length,
2113 const std::string &ModuleID,
2114 std::string* ErrMsg) {
2116 /// We handle errors by
2117 if (setjmp(context)) {
2118 // Cleanup after error
2119 if (Handler) Handler->handleError(ErrorMsg);
2123 if (decompressedBlock != 0 ) {
2124 ::free(decompressedBlock);
2125 decompressedBlock = 0;
2127 // Set caller's error message, if requested
2130 // Indicate an error occurred
2135 At = MemStart = BlockStart = Buf;
2136 MemEnd = BlockEnd = Buf + Length;
2138 // Create the module
2139 TheModule = new Module(ModuleID);
2141 if (Handler) Handler->handleStart(TheModule, Length);
2143 // Read the four bytes of the signature.
2144 unsigned Sig = read_uint();
2146 // If this is a compressed file
2147 if (Sig == ('l' | ('l' << 8) | ('v' << 16) | ('c' << 24))) {
2149 // Invoke the decompression of the bytecode. Note that we have to skip the
2150 // file's magic number which is not part of the compressed block. Hence,
2151 // the Buf+4 and Length-4. The result goes into decompressedBlock, a data
2152 // member for retention until BytecodeReader is destructed.
2153 unsigned decompressedLength = Compressor::decompressToNewBuffer(
2154 (char*)Buf+4,Length-4,decompressedBlock);
2156 // We must adjust the buffer pointers used by the bytecode reader to point
2157 // into the new decompressed block. After decompression, the
2158 // decompressedBlock will point to a contiguous memory area that has
2159 // the decompressed data.
2160 At = MemStart = BlockStart = Buf = (BufPtr) decompressedBlock;
2161 MemEnd = BlockEnd = Buf + decompressedLength;
2163 // else if this isn't a regular (uncompressed) bytecode file, then its
2164 // and error, generate that now.
2165 } else if (Sig != ('l' | ('l' << 8) | ('v' << 16) | ('m' << 24))) {
2166 error("Invalid bytecode signature: " + utohexstr(Sig));
2169 // Tell the handler we're starting a module
2170 if (Handler) Handler->handleModuleBegin(ModuleID);
2172 // Get the module block and size and verify. This is handled specially
2173 // because the module block/size is always written in long format. Other
2174 // blocks are written in short format so the read_block method is used.
2175 unsigned Type, Size;
2178 if (Type != BytecodeFormat::ModuleBlockID) {
2179 error("Expected Module Block! Type:" + utostr(Type) + ", Size:"
2183 // It looks like the darwin ranlib program is broken, and adds trailing
2184 // garbage to the end of some bytecode files. This hack allows the bc
2185 // reader to ignore trailing garbage on bytecode files.
2186 if (At + Size < MemEnd)
2187 MemEnd = BlockEnd = At+Size;
2189 if (At + Size != MemEnd)
2190 error("Invalid Top Level Block Length! Type:" + utostr(Type)
2191 + ", Size:" + utostr(Size));
2193 // Parse the module contents
2194 this->ParseModule();
2196 // Check for missing functions
2198 error("Function expected, but bytecode stream ended!");
2200 // Tell the handler we're done with the module
2202 Handler->handleModuleEnd(ModuleID);
2204 // Tell the handler we're finished the parse
2205 if (Handler) Handler->handleFinish();
2211 //===----------------------------------------------------------------------===//
2212 //=== Default Implementations of Handler Methods
2213 //===----------------------------------------------------------------------===//
2215 BytecodeHandler::~BytecodeHandler() {}