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::Int32Ty), 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::Int32TyID, 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::Int32TyID, 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::Int32Ty, 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::Int8TyID, 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::Int32TyID, 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::Int32TyID, 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 bit of the slot number.
917 if (isa<StructType>(TopTy))
918 IdxTy = Type::Int32TyID;
920 switch (ValIdx & 1) {
922 case 0: IdxTy = Type::Int32TyID; break;
923 case 1: IdxTy = Type::Int64TyID; break;
927 Idx.push_back(getValue(IdxTy, ValIdx));
928 NextTy = GetElementPtrInst::getIndexedType(InstTy, Idx, true);
931 Result = new GetElementPtrInst(getValue(iType, Oprnds[0]), Idx);
934 case 62: // volatile load
935 case Instruction::Load:
936 if (Oprnds.size() != 1 || !isa<PointerType>(InstTy))
937 error("Invalid load instruction!");
938 Result = new LoadInst(getValue(iType, Oprnds[0]), "", Opcode == 62);
940 case 63: // volatile store
941 case Instruction::Store: {
942 if (!isa<PointerType>(InstTy) || Oprnds.size() != 2)
943 error("Invalid store instruction!");
945 Value *Ptr = getValue(iType, Oprnds[1]);
946 const Type *ValTy = cast<PointerType>(Ptr->getType())->getElementType();
947 Result = new StoreInst(getValue(getTypeSlot(ValTy), Oprnds[0]), Ptr,
951 case Instruction::Unwind:
952 if (Oprnds.size() != 0) error("Invalid unwind instruction!");
953 Result = new UnwindInst();
955 case Instruction::Unreachable:
956 if (Oprnds.size() != 0) error("Invalid unreachable instruction!");
957 Result = new UnreachableInst();
959 } // end switch(Opcode)
962 BB->getInstList().push_back(Result);
965 if (Result->getType() == InstTy)
968 TypeSlot = getTypeSlot(Result->getType());
970 insertValue(Result, TypeSlot, FunctionValues);
973 /// Get a particular numbered basic block, which might be a forward reference.
974 /// This works together with ParseInstructionList to handle these forward
975 /// references in a clean manner. This function is used when constructing
976 /// phi, br, switch, and other instructions that reference basic blocks.
977 /// Blocks are numbered sequentially as they appear in the function.
978 BasicBlock *BytecodeReader::getBasicBlock(unsigned ID) {
979 // Make sure there is room in the table...
980 if (ParsedBasicBlocks.size() <= ID) ParsedBasicBlocks.resize(ID+1);
982 // First check to see if this is a backwards reference, i.e. this block
983 // has already been created, or if the forward reference has already
985 if (ParsedBasicBlocks[ID])
986 return ParsedBasicBlocks[ID];
988 // Otherwise, the basic block has not yet been created. Do so and add it to
989 // the ParsedBasicBlocks list.
990 return ParsedBasicBlocks[ID] = new BasicBlock();
993 /// Parse all of the BasicBlock's & Instruction's in the body of a function.
994 /// In post 1.0 bytecode files, we no longer emit basic block individually,
995 /// in order to avoid per-basic-block overhead.
996 /// @returns the number of basic blocks encountered.
997 unsigned BytecodeReader::ParseInstructionList(Function* F) {
998 unsigned BlockNo = 0;
999 std::vector<unsigned> Args;
1001 while (moreInBlock()) {
1002 if (Handler) Handler->handleBasicBlockBegin(BlockNo);
1004 if (ParsedBasicBlocks.size() == BlockNo)
1005 ParsedBasicBlocks.push_back(BB = new BasicBlock());
1006 else if (ParsedBasicBlocks[BlockNo] == 0)
1007 BB = ParsedBasicBlocks[BlockNo] = new BasicBlock();
1009 BB = ParsedBasicBlocks[BlockNo];
1011 F->getBasicBlockList().push_back(BB);
1013 // Read instructions into this basic block until we get to a terminator
1014 while (moreInBlock() && !BB->getTerminator())
1015 ParseInstruction(Args, BB);
1017 if (!BB->getTerminator())
1018 error("Non-terminated basic block found!");
1020 if (Handler) Handler->handleBasicBlockEnd(BlockNo-1);
1026 /// Parse a symbol table. This works for both module level and function
1027 /// level symbol tables. For function level symbol tables, the CurrentFunction
1028 /// parameter must be non-zero and the ST parameter must correspond to
1029 /// CurrentFunction's symbol table. For Module level symbol tables, the
1030 /// CurrentFunction argument must be zero.
1031 void BytecodeReader::ParseSymbolTable(Function *CurrentFunction,
1033 if (Handler) Handler->handleSymbolTableBegin(CurrentFunction,ST);
1035 // Allow efficient basic block lookup by number.
1036 std::vector<BasicBlock*> BBMap;
1037 if (CurrentFunction)
1038 for (Function::iterator I = CurrentFunction->begin(),
1039 E = CurrentFunction->end(); I != E; ++I)
1042 // Symtab block header: [num entries]
1043 unsigned NumEntries = read_vbr_uint();
1044 for (unsigned i = 0; i < NumEntries; ++i) {
1045 // Symtab entry: [def slot #][name]
1046 unsigned slot = read_vbr_uint();
1047 std::string Name = read_str();
1048 const Type* T = getType(slot);
1049 ST->insert(Name, T);
1052 while (moreInBlock()) {
1053 // Symtab block header: [num entries][type id number]
1054 unsigned NumEntries = read_vbr_uint();
1055 unsigned Typ = read_vbr_uint();
1057 for (unsigned i = 0; i != NumEntries; ++i) {
1058 // Symtab entry: [def slot #][name]
1059 unsigned slot = read_vbr_uint();
1060 std::string Name = read_str();
1062 if (Typ == Type::LabelTyID) {
1063 if (slot < BBMap.size())
1066 V = getValue(Typ, slot, false); // Find mapping...
1069 error("Failed value look-up for name '" + Name + "'");
1073 checkPastBlockEnd("Symbol Table");
1074 if (Handler) Handler->handleSymbolTableEnd();
1077 /// Read in the types portion of a compaction table.
1078 void BytecodeReader::ParseCompactionTypes(unsigned NumEntries) {
1079 for (unsigned i = 0; i != NumEntries; ++i) {
1080 unsigned TypeSlot = read_vbr_uint();
1081 const Type *Typ = getGlobalTableType(TypeSlot);
1082 CompactionTypes.push_back(std::make_pair(Typ, TypeSlot));
1083 if (Handler) Handler->handleCompactionTableType(i, TypeSlot, Typ);
1087 /// Parse a compaction table.
1088 void BytecodeReader::ParseCompactionTable() {
1090 // Notify handler that we're beginning a compaction table.
1091 if (Handler) Handler->handleCompactionTableBegin();
1093 // Get the types for the compaction table.
1094 unsigned NumEntries = read_vbr_uint();
1095 ParseCompactionTypes(NumEntries);
1097 // Compaction tables live in separate blocks so we have to loop
1098 // until we've read the whole thing.
1099 while (moreInBlock()) {
1100 // Read the number of Value* entries in the compaction table
1101 unsigned NumEntries = read_vbr_uint();
1104 // Decode the type from value read in. Most compaction table
1105 // planes will have one or two entries in them. If that's the
1106 // case then the length is encoded in the bottom two bits and
1107 // the higher bits encode the type. This saves another VBR value.
1108 if ((NumEntries & 3) == 3) {
1109 // In this case, both low-order bits are set (value 3). This
1110 // is a signal that the typeid follows.
1112 Ty = read_vbr_uint();
1114 // In this case, the low-order bits specify the number of entries
1115 // and the high order bits specify the type.
1116 Ty = NumEntries >> 2;
1120 // Make sure we have enough room for the plane.
1121 if (Ty >= CompactionValues.size())
1122 CompactionValues.resize(Ty+1);
1124 // Make sure the plane is empty or we have some kind of error.
1125 if (!CompactionValues[Ty].empty())
1126 error("Compaction table plane contains multiple entries!");
1128 // Notify handler about the plane.
1129 if (Handler) Handler->handleCompactionTablePlane(Ty, NumEntries);
1131 // Push the implicit zero.
1132 CompactionValues[Ty].push_back(Constant::getNullValue(getType(Ty)));
1134 // Read in each of the entries, put them in the compaction table
1135 // and notify the handler that we have a new compaction table value.
1136 for (unsigned i = 0; i != NumEntries; ++i) {
1137 unsigned ValSlot = read_vbr_uint();
1138 Value *V = getGlobalTableValue(Ty, ValSlot);
1139 CompactionValues[Ty].push_back(V);
1140 if (Handler) Handler->handleCompactionTableValue(i, Ty, ValSlot);
1143 // Notify handler that the compaction table is done.
1144 if (Handler) Handler->handleCompactionTableEnd();
1147 // Parse a single type. The typeid is read in first. If its a primitive type
1148 // then nothing else needs to be read, we know how to instantiate it. If its
1149 // a derived type, then additional data is read to fill out the type
1151 const Type *BytecodeReader::ParseType() {
1152 unsigned PrimType = read_vbr_uint();
1153 const Type *Result = 0;
1154 if ((Result = Type::getPrimitiveType((Type::TypeID)PrimType)))
1158 case Type::FunctionTyID: {
1159 const Type *RetType = readType();
1160 unsigned RetAttr = read_vbr_uint();
1162 unsigned NumParams = read_vbr_uint();
1164 std::vector<const Type*> Params;
1165 std::vector<FunctionType::ParameterAttributes> Attrs;
1166 Attrs.push_back(FunctionType::ParameterAttributes(RetAttr));
1167 while (NumParams--) {
1168 Params.push_back(readType());
1169 if (Params.back() != Type::VoidTy)
1170 Attrs.push_back(FunctionType::ParameterAttributes(read_vbr_uint()));
1173 bool isVarArg = Params.size() && Params.back() == Type::VoidTy;
1174 if (isVarArg) Params.pop_back();
1176 Result = FunctionType::get(RetType, Params, isVarArg, Attrs);
1179 case Type::ArrayTyID: {
1180 const Type *ElementType = readType();
1181 unsigned NumElements = read_vbr_uint();
1182 Result = ArrayType::get(ElementType, NumElements);
1185 case Type::PackedTyID: {
1186 const Type *ElementType = readType();
1187 unsigned NumElements = read_vbr_uint();
1188 Result = PackedType::get(ElementType, NumElements);
1191 case Type::StructTyID: {
1192 std::vector<const Type*> Elements;
1193 unsigned Typ = read_vbr_uint();
1194 while (Typ) { // List is terminated by void/0 typeid
1195 Elements.push_back(getType(Typ));
1196 Typ = read_vbr_uint();
1199 Result = StructType::get(Elements, false);
1202 case Type::BC_ONLY_PackedStructTyID: {
1203 std::vector<const Type*> Elements;
1204 unsigned Typ = read_vbr_uint();
1205 while (Typ) { // List is terminated by void/0 typeid
1206 Elements.push_back(getType(Typ));
1207 Typ = read_vbr_uint();
1210 Result = StructType::get(Elements, true);
1213 case Type::PointerTyID: {
1214 Result = PointerType::get(readType());
1218 case Type::OpaqueTyID: {
1219 Result = OpaqueType::get();
1224 error("Don't know how to deserialize primitive type " + utostr(PrimType));
1227 if (Handler) Handler->handleType(Result);
1231 // ParseTypes - We have to use this weird code to handle recursive
1232 // types. We know that recursive types will only reference the current slab of
1233 // values in the type plane, but they can forward reference types before they
1234 // have been read. For example, Type #0 might be '{ Ty#1 }' and Type #1 might
1235 // be 'Ty#0*'. When reading Type #0, type number one doesn't exist. To fix
1236 // this ugly problem, we pessimistically insert an opaque type for each type we
1237 // are about to read. This means that forward references will resolve to
1238 // something and when we reread the type later, we can replace the opaque type
1239 // with a new resolved concrete type.
1241 void BytecodeReader::ParseTypes(TypeListTy &Tab, unsigned NumEntries){
1242 assert(Tab.size() == 0 && "should not have read type constants in before!");
1244 // Insert a bunch of opaque types to be resolved later...
1245 Tab.reserve(NumEntries);
1246 for (unsigned i = 0; i != NumEntries; ++i)
1247 Tab.push_back(OpaqueType::get());
1250 Handler->handleTypeList(NumEntries);
1252 // If we are about to resolve types, make sure the type cache is clear.
1254 ModuleTypeIDCache.clear();
1256 // Loop through reading all of the types. Forward types will make use of the
1257 // opaque types just inserted.
1259 for (unsigned i = 0; i != NumEntries; ++i) {
1260 const Type* NewTy = ParseType();
1261 const Type* OldTy = Tab[i].get();
1263 error("Couldn't parse type!");
1265 // Don't directly push the new type on the Tab. Instead we want to replace
1266 // the opaque type we previously inserted with the new concrete value. This
1267 // approach helps with forward references to types. The refinement from the
1268 // abstract (opaque) type to the new type causes all uses of the abstract
1269 // type to use the concrete type (NewTy). This will also cause the opaque
1270 // type to be deleted.
1271 cast<DerivedType>(const_cast<Type*>(OldTy))->refineAbstractTypeTo(NewTy);
1273 // This should have replaced the old opaque type with the new type in the
1274 // value table... or with a preexisting type that was already in the system.
1275 // Let's just make sure it did.
1276 assert(Tab[i] != OldTy && "refineAbstractType didn't work!");
1280 /// Parse a single constant value
1281 Value *BytecodeReader::ParseConstantPoolValue(unsigned TypeID) {
1282 // We must check for a ConstantExpr before switching by type because
1283 // a ConstantExpr can be of any type, and has no explicit value.
1285 // 0 if not expr; numArgs if is expr
1286 unsigned isExprNumArgs = read_vbr_uint();
1288 if (isExprNumArgs) {
1289 // 'undef' is encoded with 'exprnumargs' == 1.
1290 if (isExprNumArgs == 1)
1291 return UndefValue::get(getType(TypeID));
1293 // Inline asm is encoded with exprnumargs == ~0U.
1294 if (isExprNumArgs == ~0U) {
1295 std::string AsmStr = read_str();
1296 std::string ConstraintStr = read_str();
1297 unsigned Flags = read_vbr_uint();
1299 const PointerType *PTy = dyn_cast<PointerType>(getType(TypeID));
1300 const FunctionType *FTy =
1301 PTy ? dyn_cast<FunctionType>(PTy->getElementType()) : 0;
1303 if (!FTy || !InlineAsm::Verify(FTy, ConstraintStr))
1304 error("Invalid constraints for inline asm");
1306 error("Invalid flags for inline asm");
1307 bool HasSideEffects = Flags & 1;
1308 return InlineAsm::get(FTy, AsmStr, ConstraintStr, HasSideEffects);
1313 // FIXME: Encoding of constant exprs could be much more compact!
1314 std::vector<Constant*> ArgVec;
1315 ArgVec.reserve(isExprNumArgs);
1316 unsigned Opcode = read_vbr_uint();
1318 // Read the slot number and types of each of the arguments
1319 for (unsigned i = 0; i != isExprNumArgs; ++i) {
1320 unsigned ArgValSlot = read_vbr_uint();
1321 unsigned ArgTypeSlot = read_vbr_uint();
1323 // Get the arg value from its slot if it exists, otherwise a placeholder
1324 ArgVec.push_back(getConstantValue(ArgTypeSlot, ArgValSlot));
1327 // Construct a ConstantExpr of the appropriate kind
1328 if (isExprNumArgs == 1) { // All one-operand expressions
1329 if (!Instruction::isCast(Opcode))
1330 error("Only cast instruction has one argument for ConstantExpr");
1332 Constant *Result = ConstantExpr::getCast(Opcode, ArgVec[0],
1334 if (Handler) Handler->handleConstantExpression(Opcode, ArgVec, Result);
1336 } else if (Opcode == Instruction::GetElementPtr) { // GetElementPtr
1337 std::vector<Constant*> IdxList(ArgVec.begin()+1, ArgVec.end());
1338 Constant *Result = ConstantExpr::getGetElementPtr(ArgVec[0], IdxList);
1339 if (Handler) Handler->handleConstantExpression(Opcode, ArgVec, Result);
1341 } else if (Opcode == Instruction::Select) {
1342 if (ArgVec.size() != 3)
1343 error("Select instruction must have three arguments.");
1344 Constant* Result = ConstantExpr::getSelect(ArgVec[0], ArgVec[1],
1346 if (Handler) Handler->handleConstantExpression(Opcode, ArgVec, Result);
1348 } else if (Opcode == Instruction::ExtractElement) {
1349 if (ArgVec.size() != 2 ||
1350 !ExtractElementInst::isValidOperands(ArgVec[0], ArgVec[1]))
1351 error("Invalid extractelement constand expr arguments");
1352 Constant* Result = ConstantExpr::getExtractElement(ArgVec[0], ArgVec[1]);
1353 if (Handler) Handler->handleConstantExpression(Opcode, ArgVec, Result);
1355 } else if (Opcode == Instruction::InsertElement) {
1356 if (ArgVec.size() != 3 ||
1357 !InsertElementInst::isValidOperands(ArgVec[0], ArgVec[1], ArgVec[2]))
1358 error("Invalid insertelement constand expr arguments");
1361 ConstantExpr::getInsertElement(ArgVec[0], ArgVec[1], ArgVec[2]);
1362 if (Handler) Handler->handleConstantExpression(Opcode, ArgVec, Result);
1364 } else if (Opcode == Instruction::ShuffleVector) {
1365 if (ArgVec.size() != 3 ||
1366 !ShuffleVectorInst::isValidOperands(ArgVec[0], ArgVec[1], ArgVec[2]))
1367 error("Invalid shufflevector constant expr arguments.");
1369 ConstantExpr::getShuffleVector(ArgVec[0], ArgVec[1], ArgVec[2]);
1370 if (Handler) Handler->handleConstantExpression(Opcode, ArgVec, Result);
1372 } else if (Opcode == Instruction::ICmp) {
1373 if (ArgVec.size() != 2)
1374 error("Invalid ICmp constant expr arguments.");
1375 unsigned predicate = read_vbr_uint();
1376 Constant *Result = ConstantExpr::getICmp(predicate, ArgVec[0], ArgVec[1]);
1377 if (Handler) Handler->handleConstantExpression(Opcode, ArgVec, Result);
1379 } else if (Opcode == Instruction::FCmp) {
1380 if (ArgVec.size() != 2)
1381 error("Invalid FCmp constant expr arguments.");
1382 unsigned predicate = read_vbr_uint();
1383 Constant *Result = ConstantExpr::getFCmp(predicate, ArgVec[0], ArgVec[1]);
1384 if (Handler) Handler->handleConstantExpression(Opcode, ArgVec, Result);
1386 } else { // All other 2-operand expressions
1387 Constant* Result = ConstantExpr::get(Opcode, ArgVec[0], ArgVec[1]);
1388 if (Handler) Handler->handleConstantExpression(Opcode, ArgVec, Result);
1393 // Ok, not an ConstantExpr. We now know how to read the given type...
1394 const Type *Ty = getType(TypeID);
1395 Constant *Result = 0;
1396 switch (Ty->getTypeID()) {
1397 case Type::BoolTyID: {
1398 unsigned Val = read_vbr_uint();
1399 if (Val != 0 && Val != 1)
1400 error("Invalid boolean value read.");
1401 Result = ConstantBool::get(Val == 1);
1402 if (Handler) Handler->handleConstantValue(Result);
1406 case Type::Int8TyID: // Unsigned integer types...
1407 case Type::Int16TyID:
1408 case Type::Int32TyID: {
1409 unsigned Val = read_vbr_uint();
1410 if (!ConstantInt::isValueValidForType(Ty, uint64_t(Val)))
1411 error("Invalid unsigned byte/short/int read.");
1412 Result = ConstantInt::get(Ty, Val);
1413 if (Handler) Handler->handleConstantValue(Result);
1417 case Type::Int64TyID: {
1418 uint64_t Val = read_vbr_uint64();
1419 if (!ConstantInt::isValueValidForType(Ty, Val))
1420 error("Invalid constant integer read.");
1421 Result = ConstantInt::get(Ty, Val);
1422 if (Handler) Handler->handleConstantValue(Result);
1425 case Type::FloatTyID: {
1428 Result = ConstantFP::get(Ty, Val);
1429 if (Handler) Handler->handleConstantValue(Result);
1433 case Type::DoubleTyID: {
1436 Result = ConstantFP::get(Ty, Val);
1437 if (Handler) Handler->handleConstantValue(Result);
1441 case Type::ArrayTyID: {
1442 const ArrayType *AT = cast<ArrayType>(Ty);
1443 unsigned NumElements = AT->getNumElements();
1444 unsigned TypeSlot = getTypeSlot(AT->getElementType());
1445 std::vector<Constant*> Elements;
1446 Elements.reserve(NumElements);
1447 while (NumElements--) // Read all of the elements of the constant.
1448 Elements.push_back(getConstantValue(TypeSlot,
1450 Result = ConstantArray::get(AT, Elements);
1451 if (Handler) Handler->handleConstantArray(AT, Elements, TypeSlot, Result);
1455 case Type::StructTyID: {
1456 const StructType *ST = cast<StructType>(Ty);
1458 std::vector<Constant *> Elements;
1459 Elements.reserve(ST->getNumElements());
1460 for (unsigned i = 0; i != ST->getNumElements(); ++i)
1461 Elements.push_back(getConstantValue(ST->getElementType(i),
1464 Result = ConstantStruct::get(ST, Elements);
1465 if (Handler) Handler->handleConstantStruct(ST, Elements, Result);
1469 case Type::PackedTyID: {
1470 const PackedType *PT = cast<PackedType>(Ty);
1471 unsigned NumElements = PT->getNumElements();
1472 unsigned TypeSlot = getTypeSlot(PT->getElementType());
1473 std::vector<Constant*> Elements;
1474 Elements.reserve(NumElements);
1475 while (NumElements--) // Read all of the elements of the constant.
1476 Elements.push_back(getConstantValue(TypeSlot,
1478 Result = ConstantPacked::get(PT, Elements);
1479 if (Handler) Handler->handleConstantPacked(PT, Elements, TypeSlot, Result);
1483 case Type::PointerTyID: { // ConstantPointerRef value (backwards compat).
1484 const PointerType *PT = cast<PointerType>(Ty);
1485 unsigned Slot = read_vbr_uint();
1487 // Check to see if we have already read this global variable...
1488 Value *Val = getValue(TypeID, Slot, false);
1490 GlobalValue *GV = dyn_cast<GlobalValue>(Val);
1491 if (!GV) error("GlobalValue not in ValueTable!");
1492 if (Handler) Handler->handleConstantPointer(PT, Slot, GV);
1495 error("Forward references are not allowed here.");
1500 error("Don't know how to deserialize constant value of type '" +
1501 Ty->getDescription());
1505 // Check that we didn't read a null constant if they are implicit for this
1506 // type plane. Do not do this check for constantexprs, as they may be folded
1507 // to a null value in a way that isn't predicted when a .bc file is initially
1509 assert((!isa<Constant>(Result) || !cast<Constant>(Result)->isNullValue()) ||
1510 !hasImplicitNull(TypeID) &&
1511 "Cannot read null values from bytecode!");
1515 /// Resolve references for constants. This function resolves the forward
1516 /// referenced constants in the ConstantFwdRefs map. It uses the
1517 /// replaceAllUsesWith method of Value class to substitute the placeholder
1518 /// instance with the actual instance.
1519 void BytecodeReader::ResolveReferencesToConstant(Constant *NewV, unsigned Typ,
1521 ConstantRefsType::iterator I =
1522 ConstantFwdRefs.find(std::make_pair(Typ, Slot));
1523 if (I == ConstantFwdRefs.end()) return; // Never forward referenced?
1525 Value *PH = I->second; // Get the placeholder...
1526 PH->replaceAllUsesWith(NewV);
1527 delete PH; // Delete the old placeholder
1528 ConstantFwdRefs.erase(I); // Remove the map entry for it
1531 /// Parse the constant strings section.
1532 void BytecodeReader::ParseStringConstants(unsigned NumEntries, ValueTable &Tab){
1533 for (; NumEntries; --NumEntries) {
1534 unsigned Typ = read_vbr_uint();
1535 const Type *Ty = getType(Typ);
1536 if (!isa<ArrayType>(Ty))
1537 error("String constant data invalid!");
1539 const ArrayType *ATy = cast<ArrayType>(Ty);
1540 if (ATy->getElementType() != Type::Int8Ty &&
1541 ATy->getElementType() != Type::Int8Ty)
1542 error("String constant data invalid!");
1544 // Read character data. The type tells us how long the string is.
1545 char *Data = reinterpret_cast<char *>(alloca(ATy->getNumElements()));
1546 read_data(Data, Data+ATy->getNumElements());
1548 std::vector<Constant*> Elements(ATy->getNumElements());
1549 const Type* ElemType = ATy->getElementType();
1550 for (unsigned i = 0, e = ATy->getNumElements(); i != e; ++i)
1551 Elements[i] = ConstantInt::get(ElemType, (unsigned char)Data[i]);
1553 // Create the constant, inserting it as needed.
1554 Constant *C = ConstantArray::get(ATy, Elements);
1555 unsigned Slot = insertValue(C, Typ, Tab);
1556 ResolveReferencesToConstant(C, Typ, Slot);
1557 if (Handler) Handler->handleConstantString(cast<ConstantArray>(C));
1561 /// Parse the constant pool.
1562 void BytecodeReader::ParseConstantPool(ValueTable &Tab,
1563 TypeListTy &TypeTab,
1565 if (Handler) Handler->handleGlobalConstantsBegin();
1567 /// In LLVM 1.3 Type does not derive from Value so the types
1568 /// do not occupy a plane. Consequently, we read the types
1569 /// first in the constant pool.
1571 unsigned NumEntries = read_vbr_uint();
1572 ParseTypes(TypeTab, NumEntries);
1575 while (moreInBlock()) {
1576 unsigned NumEntries = read_vbr_uint();
1577 unsigned Typ = read_vbr_uint();
1579 if (Typ == Type::VoidTyID) {
1580 /// Use of Type::VoidTyID is a misnomer. It actually means
1581 /// that the following plane is constant strings
1582 assert(&Tab == &ModuleValues && "Cannot read strings in functions!");
1583 ParseStringConstants(NumEntries, Tab);
1585 for (unsigned i = 0; i < NumEntries; ++i) {
1586 Value *V = ParseConstantPoolValue(Typ);
1587 assert(V && "ParseConstantPoolValue returned NULL!");
1588 unsigned Slot = insertValue(V, Typ, Tab);
1590 // If we are reading a function constant table, make sure that we adjust
1591 // the slot number to be the real global constant number.
1593 if (&Tab != &ModuleValues && Typ < ModuleValues.size() &&
1595 Slot += ModuleValues[Typ]->size();
1596 if (Constant *C = dyn_cast<Constant>(V))
1597 ResolveReferencesToConstant(C, Typ, Slot);
1602 // After we have finished parsing the constant pool, we had better not have
1603 // any dangling references left.
1604 if (!ConstantFwdRefs.empty()) {
1605 ConstantRefsType::const_iterator I = ConstantFwdRefs.begin();
1606 Constant* missingConst = I->second;
1607 error(utostr(ConstantFwdRefs.size()) +
1608 " unresolved constant reference exist. First one is '" +
1609 missingConst->getName() + "' of type '" +
1610 missingConst->getType()->getDescription() + "'.");
1613 checkPastBlockEnd("Constant Pool");
1614 if (Handler) Handler->handleGlobalConstantsEnd();
1617 /// Parse the contents of a function. Note that this function can be
1618 /// called lazily by materializeFunction
1619 /// @see materializeFunction
1620 void BytecodeReader::ParseFunctionBody(Function* F) {
1622 unsigned FuncSize = BlockEnd - At;
1623 GlobalValue::LinkageTypes Linkage = GlobalValue::ExternalLinkage;
1625 unsigned LinkageType = read_vbr_uint();
1626 switch (LinkageType) {
1627 case 0: Linkage = GlobalValue::ExternalLinkage; break;
1628 case 1: Linkage = GlobalValue::WeakLinkage; break;
1629 case 2: Linkage = GlobalValue::AppendingLinkage; break;
1630 case 3: Linkage = GlobalValue::InternalLinkage; break;
1631 case 4: Linkage = GlobalValue::LinkOnceLinkage; break;
1632 case 5: Linkage = GlobalValue::DLLImportLinkage; break;
1633 case 6: Linkage = GlobalValue::DLLExportLinkage; break;
1634 case 7: Linkage = GlobalValue::ExternalWeakLinkage; break;
1636 error("Invalid linkage type for Function.");
1637 Linkage = GlobalValue::InternalLinkage;
1641 F->setLinkage(Linkage);
1642 if (Handler) Handler->handleFunctionBegin(F,FuncSize);
1644 // Keep track of how many basic blocks we have read in...
1645 unsigned BlockNum = 0;
1646 bool InsertedArguments = false;
1648 BufPtr MyEnd = BlockEnd;
1649 while (At < MyEnd) {
1650 unsigned Type, Size;
1652 read_block(Type, Size);
1655 case BytecodeFormat::ConstantPoolBlockID:
1656 if (!InsertedArguments) {
1657 // Insert arguments into the value table before we parse the first basic
1658 // block in the function, but after we potentially read in the
1659 // compaction table.
1661 InsertedArguments = true;
1664 ParseConstantPool(FunctionValues, FunctionTypes, true);
1667 case BytecodeFormat::CompactionTableBlockID:
1668 ParseCompactionTable();
1671 case BytecodeFormat::InstructionListBlockID: {
1672 // Insert arguments into the value table before we parse the instruction
1673 // list for the function, but after we potentially read in the compaction
1675 if (!InsertedArguments) {
1677 InsertedArguments = true;
1681 error("Already parsed basic blocks!");
1682 BlockNum = ParseInstructionList(F);
1686 case BytecodeFormat::SymbolTableBlockID:
1687 ParseSymbolTable(F, &F->getSymbolTable());
1693 error("Wrapped around reading bytecode.");
1699 // Make sure there were no references to non-existant basic blocks.
1700 if (BlockNum != ParsedBasicBlocks.size())
1701 error("Illegal basic block operand reference");
1703 ParsedBasicBlocks.clear();
1705 // Resolve forward references. Replace any uses of a forward reference value
1706 // with the real value.
1707 while (!ForwardReferences.empty()) {
1708 std::map<std::pair<unsigned,unsigned>, Value*>::iterator
1709 I = ForwardReferences.begin();
1710 Value *V = getValue(I->first.first, I->first.second, false);
1711 Value *PlaceHolder = I->second;
1712 PlaceHolder->replaceAllUsesWith(V);
1713 ForwardReferences.erase(I);
1717 // Clear out function-level types...
1718 FunctionTypes.clear();
1719 CompactionTypes.clear();
1720 CompactionValues.clear();
1721 freeTable(FunctionValues);
1723 if (Handler) Handler->handleFunctionEnd(F);
1726 /// This function parses LLVM functions lazily. It obtains the type of the
1727 /// function and records where the body of the function is in the bytecode
1728 /// buffer. The caller can then use the ParseNextFunction and
1729 /// ParseAllFunctionBodies to get handler events for the functions.
1730 void BytecodeReader::ParseFunctionLazily() {
1731 if (FunctionSignatureList.empty())
1732 error("FunctionSignatureList empty!");
1734 Function *Func = FunctionSignatureList.back();
1735 FunctionSignatureList.pop_back();
1737 // Save the information for future reading of the function
1738 LazyFunctionLoadMap[Func] = LazyFunctionInfo(BlockStart, BlockEnd);
1740 // This function has a body but it's not loaded so it appears `External'.
1741 // Mark it as a `Ghost' instead to notify the users that it has a body.
1742 Func->setLinkage(GlobalValue::GhostLinkage);
1744 // Pretend we've `parsed' this function
1748 /// The ParserFunction method lazily parses one function. Use this method to
1749 /// casue the parser to parse a specific function in the module. Note that
1750 /// this will remove the function from what is to be included by
1751 /// ParseAllFunctionBodies.
1752 /// @see ParseAllFunctionBodies
1753 /// @see ParseBytecode
1754 bool BytecodeReader::ParseFunction(Function* Func, std::string* ErrMsg) {
1756 if (setjmp(context)) {
1757 // Set caller's error message, if requested
1760 // Indicate an error occurred
1764 // Find {start, end} pointers and slot in the map. If not there, we're done.
1765 LazyFunctionMap::iterator Fi = LazyFunctionLoadMap.find(Func);
1767 // Make sure we found it
1768 if (Fi == LazyFunctionLoadMap.end()) {
1769 error("Unrecognized function of type " + Func->getType()->getDescription());
1773 BlockStart = At = Fi->second.Buf;
1774 BlockEnd = Fi->second.EndBuf;
1775 assert(Fi->first == Func && "Found wrong function?");
1777 LazyFunctionLoadMap.erase(Fi);
1779 this->ParseFunctionBody(Func);
1783 /// The ParseAllFunctionBodies method parses through all the previously
1784 /// unparsed functions in the bytecode file. If you want to completely parse
1785 /// a bytecode file, this method should be called after Parsebytecode because
1786 /// Parsebytecode only records the locations in the bytecode file of where
1787 /// the function definitions are located. This function uses that information
1788 /// to materialize the functions.
1789 /// @see ParseBytecode
1790 bool BytecodeReader::ParseAllFunctionBodies(std::string* ErrMsg) {
1791 if (setjmp(context)) {
1792 // Set caller's error message, if requested
1795 // Indicate an error occurred
1799 LazyFunctionMap::iterator Fi = LazyFunctionLoadMap.begin();
1800 LazyFunctionMap::iterator Fe = LazyFunctionLoadMap.end();
1803 Function* Func = Fi->first;
1804 BlockStart = At = Fi->second.Buf;
1805 BlockEnd = Fi->second.EndBuf;
1806 ParseFunctionBody(Func);
1809 LazyFunctionLoadMap.clear();
1813 /// Parse the global type list
1814 void BytecodeReader::ParseGlobalTypes() {
1815 // Read the number of types
1816 unsigned NumEntries = read_vbr_uint();
1817 ParseTypes(ModuleTypes, NumEntries);
1820 /// Parse the Global info (types, global vars, constants)
1821 void BytecodeReader::ParseModuleGlobalInfo() {
1823 if (Handler) Handler->handleModuleGlobalsBegin();
1825 // SectionID - If a global has an explicit section specified, this map
1826 // remembers the ID until we can translate it into a string.
1827 std::map<GlobalValue*, unsigned> SectionID;
1829 // Read global variables...
1830 unsigned VarType = read_vbr_uint();
1831 while (VarType != Type::VoidTyID) { // List is terminated by Void
1832 // VarType Fields: bit0 = isConstant, bit1 = hasInitializer, bit2,3,4 =
1833 // Linkage, bit4+ = slot#
1834 unsigned SlotNo = VarType >> 5;
1835 unsigned LinkageID = (VarType >> 2) & 7;
1836 bool isConstant = VarType & 1;
1837 bool hasInitializer = (VarType & 2) != 0;
1838 unsigned Alignment = 0;
1839 unsigned GlobalSectionID = 0;
1841 // An extension word is present when linkage = 3 (internal) and hasinit = 0.
1842 if (LinkageID == 3 && !hasInitializer) {
1843 unsigned ExtWord = read_vbr_uint();
1844 // The extension word has this format: bit 0 = has initializer, bit 1-3 =
1845 // linkage, bit 4-8 = alignment (log2), bits 10+ = future use.
1846 hasInitializer = ExtWord & 1;
1847 LinkageID = (ExtWord >> 1) & 7;
1848 Alignment = (1 << ((ExtWord >> 4) & 31)) >> 1;
1850 if (ExtWord & (1 << 9)) // Has a section ID.
1851 GlobalSectionID = read_vbr_uint();
1854 GlobalValue::LinkageTypes Linkage;
1855 switch (LinkageID) {
1856 case 0: Linkage = GlobalValue::ExternalLinkage; break;
1857 case 1: Linkage = GlobalValue::WeakLinkage; break;
1858 case 2: Linkage = GlobalValue::AppendingLinkage; break;
1859 case 3: Linkage = GlobalValue::InternalLinkage; break;
1860 case 4: Linkage = GlobalValue::LinkOnceLinkage; break;
1861 case 5: Linkage = GlobalValue::DLLImportLinkage; break;
1862 case 6: Linkage = GlobalValue::DLLExportLinkage; break;
1863 case 7: Linkage = GlobalValue::ExternalWeakLinkage; break;
1865 error("Unknown linkage type: " + utostr(LinkageID));
1866 Linkage = GlobalValue::InternalLinkage;
1870 const Type *Ty = getType(SlotNo);
1872 error("Global has no type! SlotNo=" + utostr(SlotNo));
1874 if (!isa<PointerType>(Ty))
1875 error("Global not a pointer type! Ty= " + Ty->getDescription());
1877 const Type *ElTy = cast<PointerType>(Ty)->getElementType();
1879 // Create the global variable...
1880 GlobalVariable *GV = new GlobalVariable(ElTy, isConstant, Linkage,
1882 GV->setAlignment(Alignment);
1883 insertValue(GV, SlotNo, ModuleValues);
1885 if (GlobalSectionID != 0)
1886 SectionID[GV] = GlobalSectionID;
1888 unsigned initSlot = 0;
1889 if (hasInitializer) {
1890 initSlot = read_vbr_uint();
1891 GlobalInits.push_back(std::make_pair(GV, initSlot));
1894 // Notify handler about the global value.
1896 Handler->handleGlobalVariable(ElTy, isConstant, Linkage, SlotNo,initSlot);
1899 VarType = read_vbr_uint();
1902 // Read the function objects for all of the functions that are coming
1903 unsigned FnSignature = read_vbr_uint();
1905 // List is terminated by VoidTy.
1906 while (((FnSignature & (~0U >> 1)) >> 5) != Type::VoidTyID) {
1907 const Type *Ty = getType((FnSignature & (~0U >> 1)) >> 5);
1908 if (!isa<PointerType>(Ty) ||
1909 !isa<FunctionType>(cast<PointerType>(Ty)->getElementType())) {
1910 error("Function not a pointer to function type! Ty = " +
1911 Ty->getDescription());
1914 // We create functions by passing the underlying FunctionType to create...
1915 const FunctionType* FTy =
1916 cast<FunctionType>(cast<PointerType>(Ty)->getElementType());
1918 // Insert the place holder.
1919 Function *Func = new Function(FTy, GlobalValue::ExternalLinkage,
1922 insertValue(Func, (FnSignature & (~0U >> 1)) >> 5, ModuleValues);
1924 // Flags are not used yet.
1925 unsigned Flags = FnSignature & 31;
1927 // Save this for later so we know type of lazily instantiated functions.
1928 // Note that known-external functions do not have FunctionInfo blocks, so we
1929 // do not add them to the FunctionSignatureList.
1930 if ((Flags & (1 << 4)) == 0)
1931 FunctionSignatureList.push_back(Func);
1933 // Get the calling convention from the low bits.
1934 unsigned CC = Flags & 15;
1935 unsigned Alignment = 0;
1936 if (FnSignature & (1 << 31)) { // Has extension word?
1937 unsigned ExtWord = read_vbr_uint();
1938 Alignment = (1 << (ExtWord & 31)) >> 1;
1939 CC |= ((ExtWord >> 5) & 15) << 4;
1941 if (ExtWord & (1 << 10)) // Has a section ID.
1942 SectionID[Func] = read_vbr_uint();
1944 // Parse external declaration linkage
1945 switch ((ExtWord >> 11) & 3) {
1947 case 1: Func->setLinkage(Function::DLLImportLinkage); break;
1948 case 2: Func->setLinkage(Function::ExternalWeakLinkage); break;
1949 default: assert(0 && "Unsupported external linkage");
1953 Func->setCallingConv(CC-1);
1954 Func->setAlignment(Alignment);
1956 if (Handler) Handler->handleFunctionDeclaration(Func);
1958 // Get the next function signature.
1959 FnSignature = read_vbr_uint();
1962 // Now that the function signature list is set up, reverse it so that we can
1963 // remove elements efficiently from the back of the vector.
1964 std::reverse(FunctionSignatureList.begin(), FunctionSignatureList.end());
1966 /// SectionNames - This contains the list of section names encoded in the
1967 /// moduleinfoblock. Functions and globals with an explicit section index
1968 /// into this to get their section name.
1969 std::vector<std::string> SectionNames;
1971 // Read in the dependent library information.
1972 unsigned num_dep_libs = read_vbr_uint();
1973 std::string dep_lib;
1974 while (num_dep_libs--) {
1975 dep_lib = read_str();
1976 TheModule->addLibrary(dep_lib);
1978 Handler->handleDependentLibrary(dep_lib);
1981 // Read target triple and place into the module.
1982 std::string triple = read_str();
1983 TheModule->setTargetTriple(triple);
1985 Handler->handleTargetTriple(triple);
1987 if (At != BlockEnd) {
1988 // If the file has section info in it, read the section names now.
1989 unsigned NumSections = read_vbr_uint();
1990 while (NumSections--)
1991 SectionNames.push_back(read_str());
1994 // If the file has module-level inline asm, read it now.
1996 TheModule->setModuleInlineAsm(read_str());
1998 // If any globals are in specified sections, assign them now.
1999 for (std::map<GlobalValue*, unsigned>::iterator I = SectionID.begin(), E =
2000 SectionID.end(); I != E; ++I)
2002 if (I->second > SectionID.size())
2003 error("SectionID out of range for global!");
2004 I->first->setSection(SectionNames[I->second-1]);
2007 // This is for future proofing... in the future extra fields may be added that
2008 // we don't understand, so we transparently ignore them.
2012 if (Handler) Handler->handleModuleGlobalsEnd();
2015 /// Parse the version information and decode it by setting flags on the
2016 /// Reader that enable backward compatibility of the reader.
2017 void BytecodeReader::ParseVersionInfo() {
2018 unsigned Version = read_vbr_uint();
2020 // Unpack version number: low four bits are for flags, top bits = version
2021 Module::Endianness Endianness;
2022 Module::PointerSize PointerSize;
2023 Endianness = (Version & 1) ? Module::BigEndian : Module::LittleEndian;
2024 PointerSize = (Version & 2) ? Module::Pointer64 : Module::Pointer32;
2026 bool hasNoEndianness = Version & 4;
2027 bool hasNoPointerSize = Version & 8;
2029 RevisionNum = Version >> 4;
2031 // We don't provide backwards compatibility in the Reader any more. To
2032 // upgrade, the user should use llvm-upgrade.
2033 if (RevisionNum < 7)
2034 error("Bytecode formats < 7 are no longer supported. Use llvm-upgrade.");
2036 if (hasNoEndianness) Endianness = Module::AnyEndianness;
2037 if (hasNoPointerSize) PointerSize = Module::AnyPointerSize;
2039 TheModule->setEndianness(Endianness);
2040 TheModule->setPointerSize(PointerSize);
2042 if (Handler) Handler->handleVersionInfo(RevisionNum, Endianness, PointerSize);
2045 /// Parse a whole module.
2046 void BytecodeReader::ParseModule() {
2047 unsigned Type, Size;
2049 FunctionSignatureList.clear(); // Just in case...
2051 // Read into instance variables...
2054 bool SeenModuleGlobalInfo = false;
2055 bool SeenGlobalTypePlane = false;
2056 BufPtr MyEnd = BlockEnd;
2057 while (At < MyEnd) {
2059 read_block(Type, Size);
2063 case BytecodeFormat::GlobalTypePlaneBlockID:
2064 if (SeenGlobalTypePlane)
2065 error("Two GlobalTypePlane Blocks Encountered!");
2069 SeenGlobalTypePlane = true;
2072 case BytecodeFormat::ModuleGlobalInfoBlockID:
2073 if (SeenModuleGlobalInfo)
2074 error("Two ModuleGlobalInfo Blocks Encountered!");
2075 ParseModuleGlobalInfo();
2076 SeenModuleGlobalInfo = true;
2079 case BytecodeFormat::ConstantPoolBlockID:
2080 ParseConstantPool(ModuleValues, ModuleTypes,false);
2083 case BytecodeFormat::FunctionBlockID:
2084 ParseFunctionLazily();
2087 case BytecodeFormat::SymbolTableBlockID:
2088 ParseSymbolTable(0, &TheModule->getSymbolTable());
2094 error("Unexpected Block of Type #" + utostr(Type) + " encountered!");
2101 // After the module constant pool has been read, we can safely initialize
2102 // global variables...
2103 while (!GlobalInits.empty()) {
2104 GlobalVariable *GV = GlobalInits.back().first;
2105 unsigned Slot = GlobalInits.back().second;
2106 GlobalInits.pop_back();
2108 // Look up the initializer value...
2109 // FIXME: Preserve this type ID!
2111 const llvm::PointerType* GVType = GV->getType();
2112 unsigned TypeSlot = getTypeSlot(GVType->getElementType());
2113 if (Constant *CV = getConstantValue(TypeSlot, Slot)) {
2114 if (GV->hasInitializer())
2115 error("Global *already* has an initializer?!");
2116 if (Handler) Handler->handleGlobalInitializer(GV,CV);
2117 GV->setInitializer(CV);
2119 error("Cannot find initializer value.");
2122 if (!ConstantFwdRefs.empty())
2123 error("Use of undefined constants in a module");
2125 /// Make sure we pulled them all out. If we didn't then there's a declaration
2126 /// but a missing body. That's not allowed.
2127 if (!FunctionSignatureList.empty())
2128 error("Function declared, but bytecode stream ended before definition");
2131 /// This function completely parses a bytecode buffer given by the \p Buf
2132 /// and \p Length parameters.
2133 bool BytecodeReader::ParseBytecode(volatile BufPtr Buf, unsigned Length,
2134 const std::string &ModuleID,
2135 std::string* ErrMsg) {
2137 /// We handle errors by
2138 if (setjmp(context)) {
2139 // Cleanup after error
2140 if (Handler) Handler->handleError(ErrorMsg);
2144 if (decompressedBlock != 0 ) {
2145 ::free(decompressedBlock);
2146 decompressedBlock = 0;
2148 // Set caller's error message, if requested
2151 // Indicate an error occurred
2156 At = MemStart = BlockStart = Buf;
2157 MemEnd = BlockEnd = Buf + Length;
2159 // Create the module
2160 TheModule = new Module(ModuleID);
2162 if (Handler) Handler->handleStart(TheModule, Length);
2164 // Read the four bytes of the signature.
2165 unsigned Sig = read_uint();
2167 // If this is a compressed file
2168 if (Sig == ('l' | ('l' << 8) | ('v' << 16) | ('c' << 24))) {
2170 // Invoke the decompression of the bytecode. Note that we have to skip the
2171 // file's magic number which is not part of the compressed block. Hence,
2172 // the Buf+4 and Length-4. The result goes into decompressedBlock, a data
2173 // member for retention until BytecodeReader is destructed.
2174 unsigned decompressedLength = Compressor::decompressToNewBuffer(
2175 (char*)Buf+4,Length-4,decompressedBlock);
2177 // We must adjust the buffer pointers used by the bytecode reader to point
2178 // into the new decompressed block. After decompression, the
2179 // decompressedBlock will point to a contiguous memory area that has
2180 // the decompressed data.
2181 At = MemStart = BlockStart = Buf = (BufPtr) decompressedBlock;
2182 MemEnd = BlockEnd = Buf + decompressedLength;
2184 // else if this isn't a regular (uncompressed) bytecode file, then its
2185 // and error, generate that now.
2186 } else if (Sig != ('l' | ('l' << 8) | ('v' << 16) | ('m' << 24))) {
2187 error("Invalid bytecode signature: " + utohexstr(Sig));
2190 // Tell the handler we're starting a module
2191 if (Handler) Handler->handleModuleBegin(ModuleID);
2193 // Get the module block and size and verify. This is handled specially
2194 // because the module block/size is always written in long format. Other
2195 // blocks are written in short format so the read_block method is used.
2196 unsigned Type, Size;
2199 if (Type != BytecodeFormat::ModuleBlockID) {
2200 error("Expected Module Block! Type:" + utostr(Type) + ", Size:"
2204 // It looks like the darwin ranlib program is broken, and adds trailing
2205 // garbage to the end of some bytecode files. This hack allows the bc
2206 // reader to ignore trailing garbage on bytecode files.
2207 if (At + Size < MemEnd)
2208 MemEnd = BlockEnd = At+Size;
2210 if (At + Size != MemEnd)
2211 error("Invalid Top Level Block Length! Type:" + utostr(Type)
2212 + ", Size:" + utostr(Size));
2214 // Parse the module contents
2215 this->ParseModule();
2217 // Check for missing functions
2219 error("Function expected, but bytecode stream ended!");
2221 // Tell the handler we're done with the module
2223 Handler->handleModuleEnd(ModuleID);
2225 // Tell the handler we're finished the parse
2226 if (Handler) Handler->handleFinish();
2232 //===----------------------------------------------------------------------===//
2233 //=== Default Implementations of Handler Methods
2234 //===----------------------------------------------------------------------===//
2236 BytecodeHandler::~BytecodeHandler() {}