//===-- llvmAsmParser.y - Parser for llvm assembly files --------*- C++ -*-===// // // The LLVM Compiler Infrastructure // // This file was developed by the LLVM research group and is distributed under // the University of Illinois Open Source License. See LICENSE.TXT for details. // //===----------------------------------------------------------------------===// // // This file implements the bison parser for LLVM assembly languages files. // //===----------------------------------------------------------------------===// %{ #include "ParserInternals.h" #include "llvm/Instructions.h" #include "llvm/Module.h" #include "llvm/SymbolTable.h" #include "llvm/Support/GetElementPtrTypeIterator.h" #include "llvm/ADT/STLExtras.h" #include #include #include #include int yyerror(const char *ErrorMsg); // Forward declarations to prevent "implicit int yylex(); // declaration" of xxx warnings. int yyparse(); namespace llvm { std::string CurFilename; } using namespace llvm; static Module *ParserResult; // DEBUG_UPREFS - Define this symbol if you want to enable debugging output // relating to upreferences in the input stream. // //#define DEBUG_UPREFS 1 #ifdef DEBUG_UPREFS #define UR_OUT(X) std::cerr << X #else #define UR_OUT(X) #endif #define YYERROR_VERBOSE 1 // HACK ALERT: This variable is used to implement the automatic conversion of // variable argument instructions from their old to new forms. When this // compatiblity "Feature" is removed, this should be too. // static BasicBlock *CurBB; static bool ObsoleteVarArgs; // This contains info used when building the body of a function. It is // destroyed when the function is completed. // typedef std::vector ValueList; // Numbered defs static void ResolveDefinitions(std::map &LateResolvers, std::map *FutureLateResolvers = 0); static struct PerModuleInfo { Module *CurrentModule; std::map Values; // Module level numbered definitions std::map LateResolveValues; std::vector Types; std::map LateResolveTypes; /// PlaceHolderInfo - When temporary placeholder objects are created, remember /// how they were referenced and one which line of the input they came from so /// that we can resolve them later and print error messages as appropriate. std::map > PlaceHolderInfo; // GlobalRefs - This maintains a mapping between 's and forward // references to global values. Global values may be referenced before they // are defined, and if so, the temporary object that they represent is held // here. This is used for forward references of GlobalValues. // typedef std::map, GlobalValue*> GlobalRefsType; GlobalRefsType GlobalRefs; void ModuleDone() { // If we could not resolve some functions at function compilation time // (calls to functions before they are defined), resolve them now... Types // are resolved when the constant pool has been completely parsed. // ResolveDefinitions(LateResolveValues); // Check to make sure that all global value forward references have been // resolved! // if (!GlobalRefs.empty()) { std::string UndefinedReferences = "Unresolved global references exist:\n"; for (GlobalRefsType::iterator I = GlobalRefs.begin(), E =GlobalRefs.end(); I != E; ++I) { UndefinedReferences += " " + I->first.first->getDescription() + " " + I->first.second.getName() + "\n"; } ThrowException(UndefinedReferences); } Values.clear(); // Clear out function local definitions Types.clear(); CurrentModule = 0; } // GetForwardRefForGlobal - Check to see if there is a forward reference // for this global. If so, remove it from the GlobalRefs map and return it. // If not, just return null. GlobalValue *GetForwardRefForGlobal(const PointerType *PTy, ValID ID) { // Check to see if there is a forward reference to this global variable... // if there is, eliminate it and patch the reference to use the new def'n. GlobalRefsType::iterator I = GlobalRefs.find(std::make_pair(PTy, ID)); GlobalValue *Ret = 0; if (I != GlobalRefs.end()) { Ret = I->second; GlobalRefs.erase(I); } return Ret; } } CurModule; static struct PerFunctionInfo { Function *CurrentFunction; // Pointer to current function being created std::map Values; // Keep track of #'d definitions std::map LateResolveValues; bool isDeclare; // Is this function a forward declararation? /// BBForwardRefs - When we see forward references to basic blocks, keep /// track of them here. std::map > BBForwardRefs; std::vector NumberedBlocks; unsigned NextBBNum; inline PerFunctionInfo() { CurrentFunction = 0; isDeclare = false; } inline void FunctionStart(Function *M) { CurrentFunction = M; NextBBNum = 0; } void FunctionDone() { NumberedBlocks.clear(); // Any forward referenced blocks left? if (!BBForwardRefs.empty()) ThrowException("Undefined reference to label " + BBForwardRefs.begin()->first->getName()); // Resolve all forward references now. ResolveDefinitions(LateResolveValues, &CurModule.LateResolveValues); Values.clear(); // Clear out function local definitions CurrentFunction = 0; isDeclare = false; } } CurFun; // Info for the current function... static bool inFunctionScope() { return CurFun.CurrentFunction != 0; } //===----------------------------------------------------------------------===// // Code to handle definitions of all the types //===----------------------------------------------------------------------===// static int InsertValue(Value *V, std::map &ValueTab = CurFun.Values) { if (V->hasName()) return -1; // Is this a numbered definition? // Yes, insert the value into the value table... ValueList &List = ValueTab[V->getType()]; List.push_back(V); return List.size()-1; } static const Type *getTypeVal(const ValID &D, bool DoNotImprovise = false) { switch (D.Type) { case ValID::NumberVal: // Is it a numbered definition? // Module constants occupy the lowest numbered slots... if ((unsigned)D.Num < CurModule.Types.size()) return CurModule.Types[(unsigned)D.Num]; break; case ValID::NameVal: // Is it a named definition? if (const Type *N = CurModule.CurrentModule->getTypeByName(D.Name)) { D.destroy(); // Free old strdup'd memory... return N; } break; default: ThrowException("Internal parser error: Invalid symbol type reference!"); } // If we reached here, we referenced either a symbol that we don't know about // or an id number that hasn't been read yet. We may be referencing something // forward, so just create an entry to be resolved later and get to it... // if (DoNotImprovise) return 0; // Do we just want a null to be returned? if (inFunctionScope()) { if (D.Type == ValID::NameVal) ThrowException("Reference to an undefined type: '" + D.getName() + "'"); else ThrowException("Reference to an undefined type: #" + itostr(D.Num)); } std::map::iterator I =CurModule.LateResolveTypes.find(D); if (I != CurModule.LateResolveTypes.end()) return I->second; Type *Typ = OpaqueType::get(); CurModule.LateResolveTypes.insert(std::make_pair(D, Typ)); return Typ; } static Value *lookupInSymbolTable(const Type *Ty, const std::string &Name) { SymbolTable &SymTab = inFunctionScope() ? CurFun.CurrentFunction->getSymbolTable() : CurModule.CurrentModule->getSymbolTable(); return SymTab.lookup(Ty, Name); } // getValNonImprovising - Look up the value specified by the provided type and // the provided ValID. If the value exists and has already been defined, return // it. Otherwise return null. // static Value *getValNonImprovising(const Type *Ty, const ValID &D) { if (isa(Ty)) ThrowException("Functions are not values and " "must be referenced as pointers"); switch (D.Type) { case ValID::NumberVal: { // Is it a numbered definition? unsigned Num = (unsigned)D.Num; // Module constants occupy the lowest numbered slots... std::map::iterator VI = CurModule.Values.find(Ty); if (VI != CurModule.Values.end()) { if (Num < VI->second.size()) return VI->second[Num]; Num -= VI->second.size(); } // Make sure that our type is within bounds VI = CurFun.Values.find(Ty); if (VI == CurFun.Values.end()) return 0; // Check that the number is within bounds... if (VI->second.size() <= Num) return 0; return VI->second[Num]; } case ValID::NameVal: { // Is it a named definition? Value *N = lookupInSymbolTable(Ty, std::string(D.Name)); if (N == 0) return 0; D.destroy(); // Free old strdup'd memory... return N; } // Check to make sure that "Ty" is an integral type, and that our // value will fit into the specified type... case ValID::ConstSIntVal: // Is it a constant pool reference?? if (!ConstantSInt::isValueValidForType(Ty, D.ConstPool64)) ThrowException("Signed integral constant '" + itostr(D.ConstPool64) + "' is invalid for type '" + Ty->getDescription() + "'!"); return ConstantSInt::get(Ty, D.ConstPool64); case ValID::ConstUIntVal: // Is it an unsigned const pool reference? if (!ConstantUInt::isValueValidForType(Ty, D.UConstPool64)) { if (!ConstantSInt::isValueValidForType(Ty, D.ConstPool64)) { ThrowException("Integral constant '" + utostr(D.UConstPool64) + "' is invalid or out of range!"); } else { // This is really a signed reference. Transmogrify. return ConstantSInt::get(Ty, D.ConstPool64); } } else { return ConstantUInt::get(Ty, D.UConstPool64); } case ValID::ConstFPVal: // Is it a floating point const pool reference? if (!ConstantFP::isValueValidForType(Ty, D.ConstPoolFP)) ThrowException("FP constant invalid for type!!"); return ConstantFP::get(Ty, D.ConstPoolFP); case ValID::ConstNullVal: // Is it a null value? if (!isa(Ty)) ThrowException("Cannot create a a non pointer null!"); return ConstantPointerNull::get(cast(Ty)); case ValID::ConstUndefVal: // Is it an undef value? return UndefValue::get(Ty); case ValID::ConstantVal: // Fully resolved constant? if (D.ConstantValue->getType() != Ty) ThrowException("Constant expression type different from required type!"); return D.ConstantValue; default: assert(0 && "Unhandled case!"); return 0; } // End of switch assert(0 && "Unhandled case!"); return 0; } // getVal - This function is identical to getValNonImprovising, except that if a // value is not already defined, it "improvises" by creating a placeholder var // that looks and acts just like the requested variable. When the value is // defined later, all uses of the placeholder variable are replaced with the // real thing. // static Value *getVal(const Type *Ty, const ValID &ID) { if (Ty == Type::LabelTy) ThrowException("Cannot use a basic block here"); // See if the value has already been defined. Value *V = getValNonImprovising(Ty, ID); if (V) return V; if (!Ty->isFirstClassType() && !isa(Ty)) ThrowException("Invalid use of a composite type!"); // If we reached here, we referenced either a symbol that we don't know about // or an id number that hasn't been read yet. We may be referencing something // forward, so just create an entry to be resolved later and get to it... // V = new Argument(Ty); // Remember where this forward reference came from. FIXME, shouldn't we try // to recycle these things?? CurModule.PlaceHolderInfo.insert(std::make_pair(V, std::make_pair(ID, llvmAsmlineno))); if (inFunctionScope()) InsertValue(V, CurFun.LateResolveValues); else InsertValue(V, CurModule.LateResolveValues); return V; } /// getBBVal - This is used for two purposes: /// * If isDefinition is true, a new basic block with the specified ID is being /// defined. /// * If isDefinition is true, this is a reference to a basic block, which may /// or may not be a forward reference. /// static BasicBlock *getBBVal(const ValID &ID, bool isDefinition = false) { assert(inFunctionScope() && "Can't get basic block at global scope!"); std::string Name; BasicBlock *BB = 0; switch (ID.Type) { default: ThrowException("Illegal label reference " + ID.getName()); case ValID::NumberVal: // Is it a numbered definition? if (unsigned(ID.Num) >= CurFun.NumberedBlocks.size()) CurFun.NumberedBlocks.resize(ID.Num+1); BB = CurFun.NumberedBlocks[ID.Num]; break; case ValID::NameVal: // Is it a named definition? Name = ID.Name; if (Value *N = CurFun.CurrentFunction-> getSymbolTable().lookup(Type::LabelTy, Name)) BB = cast(N); break; } // See if the block has already been defined. if (BB) { // If this is the definition of the block, make sure the existing value was // just a forward reference. If it was a forward reference, there will be // an entry for it in the PlaceHolderInfo map. if (isDefinition && !CurFun.BBForwardRefs.erase(BB)) // The existing value was a definition, not a forward reference. ThrowException("Redefinition of label " + ID.getName()); ID.destroy(); // Free strdup'd memory. return BB; } // Otherwise this block has not been seen before. BB = new BasicBlock("", CurFun.CurrentFunction); if (ID.Type == ValID::NameVal) { BB->setName(ID.Name); } else { CurFun.NumberedBlocks[ID.Num] = BB; } // If this is not a definition, keep track of it so we can use it as a forward // reference. if (!isDefinition) { // Remember where this forward reference came from. CurFun.BBForwardRefs[BB] = std::make_pair(ID, llvmAsmlineno); } else { // The forward declaration could have been inserted anywhere in the // function: insert it into the correct place now. CurFun.CurrentFunction->getBasicBlockList().remove(BB); CurFun.CurrentFunction->getBasicBlockList().push_back(BB); } ID.destroy(); return BB; } //===----------------------------------------------------------------------===// // Code to handle forward references in instructions //===----------------------------------------------------------------------===// // // This code handles the late binding needed with statements that reference // values not defined yet... for example, a forward branch, or the PHI node for // a loop body. // // This keeps a table (CurFun.LateResolveValues) of all such forward references // and back patchs after we are done. // // ResolveDefinitions - If we could not resolve some defs at parsing // time (forward branches, phi functions for loops, etc...) resolve the // defs now... // static void ResolveDefinitions(std::map &LateResolvers, std::map *FutureLateResolvers) { // Loop over LateResolveDefs fixing up stuff that couldn't be resolved for (std::map::iterator LRI = LateResolvers.begin(), E = LateResolvers.end(); LRI != E; ++LRI) { ValueList &List = LRI->second; while (!List.empty()) { Value *V = List.back(); List.pop_back(); std::map >::iterator PHI = CurModule.PlaceHolderInfo.find(V); assert(PHI != CurModule.PlaceHolderInfo.end() && "Placeholder error!"); ValID &DID = PHI->second.first; Value *TheRealValue = getValNonImprovising(LRI->first, DID); if (TheRealValue) { V->replaceAllUsesWith(TheRealValue); delete V; CurModule.PlaceHolderInfo.erase(PHI); } else if (FutureLateResolvers) { // Functions have their unresolved items forwarded to the module late // resolver table InsertValue(V, *FutureLateResolvers); } else { if (DID.Type == ValID::NameVal) ThrowException("Reference to an invalid definition: '" +DID.getName()+ "' of type '" + V->getType()->getDescription() + "'", PHI->second.second); else ThrowException("Reference to an invalid definition: #" + itostr(DID.Num) + " of type '" + V->getType()->getDescription() + "'", PHI->second.second); } } } LateResolvers.clear(); } // ResolveTypeTo - A brand new type was just declared. This means that (if // name is not null) things referencing Name can be resolved. Otherwise, things // refering to the number can be resolved. Do this now. // static void ResolveTypeTo(char *Name, const Type *ToTy) { ValID D; if (Name) D = ValID::create(Name); else D = ValID::create((int)CurModule.Types.size()); std::map::iterator I = CurModule.LateResolveTypes.find(D); if (I != CurModule.LateResolveTypes.end()) { ((DerivedType*)I->second.get())->refineAbstractTypeTo(ToTy); CurModule.LateResolveTypes.erase(I); } } // setValueName - Set the specified value to the name given. The name may be // null potentially, in which case this is a noop. The string passed in is // assumed to be a malloc'd string buffer, and is free'd by this function. // static void setValueName(Value *V, char *NameStr) { if (NameStr) { std::string Name(NameStr); // Copy string free(NameStr); // Free old string if (V->getType() == Type::VoidTy) ThrowException("Can't assign name '" + Name+"' to value with void type!"); assert(inFunctionScope() && "Must be in function scope!"); SymbolTable &ST = CurFun.CurrentFunction->getSymbolTable(); if (ST.lookup(V->getType(), Name)) ThrowException("Redefinition of value named '" + Name + "' in the '" + V->getType()->getDescription() + "' type plane!"); // Set the name. V->setName(Name); } } /// ParseGlobalVariable - Handle parsing of a global. If Initializer is null, /// this is a declaration, otherwise it is a definition. static void ParseGlobalVariable(char *NameStr,GlobalValue::LinkageTypes Linkage, bool isConstantGlobal, const Type *Ty, Constant *Initializer) { if (isa(Ty)) ThrowException("Cannot declare global vars of function type!"); const PointerType *PTy = PointerType::get(Ty); std::string Name; if (NameStr) { Name = NameStr; // Copy string free(NameStr); // Free old string } // See if this global value was forward referenced. If so, recycle the // object. ValID ID; if (!Name.empty()) { ID = ValID::create((char*)Name.c_str()); } else { ID = ValID::create((int)CurModule.Values[PTy].size()); } if (GlobalValue *FWGV = CurModule.GetForwardRefForGlobal(PTy, ID)) { // Move the global to the end of the list, from whereever it was // previously inserted. GlobalVariable *GV = cast(FWGV); CurModule.CurrentModule->getGlobalList().remove(GV); CurModule.CurrentModule->getGlobalList().push_back(GV); GV->setInitializer(Initializer); GV->setLinkage(Linkage); GV->setConstant(isConstantGlobal); InsertValue(GV, CurModule.Values); return; } // If this global has a name, check to see if there is already a definition // of this global in the module. If so, merge as appropriate. Note that // this is really just a hack around problems in the CFE. :( if (!Name.empty()) { // We are a simple redefinition of a value, check to see if it is defined // the same as the old one. if (GlobalVariable *EGV = CurModule.CurrentModule->getGlobalVariable(Name, Ty)) { // We are allowed to redefine a global variable in two circumstances: // 1. If at least one of the globals is uninitialized or // 2. If both initializers have the same value. // if (!EGV->hasInitializer() || !Initializer || EGV->getInitializer() == Initializer) { // Make sure the existing global version gets the initializer! Make // sure that it also gets marked const if the new version is. if (Initializer && !EGV->hasInitializer()) EGV->setInitializer(Initializer); if (isConstantGlobal) EGV->setConstant(true); EGV->setLinkage(Linkage); return; } ThrowException("Redefinition of global variable named '" + Name + "' in the '" + Ty->getDescription() + "' type plane!"); } } // Otherwise there is no existing GV to use, create one now. GlobalVariable *GV = new GlobalVariable(Ty, isConstantGlobal, Linkage, Initializer, Name, CurModule.CurrentModule); InsertValue(GV, CurModule.Values); } // setTypeName - Set the specified type to the name given. The name may be // null potentially, in which case this is a noop. The string passed in is // assumed to be a malloc'd string buffer, and is freed by this function. // // This function returns true if the type has already been defined, but is // allowed to be redefined in the specified context. If the name is a new name // for the type plane, it is inserted and false is returned. static bool setTypeName(const Type *T, char *NameStr) { assert(!inFunctionScope() && "Can't give types function-local names!"); if (NameStr == 0) return false; std::string Name(NameStr); // Copy string free(NameStr); // Free old string // We don't allow assigning names to void type if (T == Type::VoidTy) ThrowException("Can't assign name '" + Name + "' to the void type!"); // Set the type name, checking for conflicts as we do so. bool AlreadyExists = CurModule.CurrentModule->addTypeName(Name, T); if (AlreadyExists) { // Inserting a name that is already defined??? const Type *Existing = CurModule.CurrentModule->getTypeByName(Name); assert(Existing && "Conflict but no matching type?"); // There is only one case where this is allowed: when we are refining an // opaque type. In this case, Existing will be an opaque type. if (const OpaqueType *OpTy = dyn_cast(Existing)) { // We ARE replacing an opaque type! const_cast(OpTy)->refineAbstractTypeTo(T); return true; } // Otherwise, this is an attempt to redefine a type. That's okay if // the redefinition is identical to the original. This will be so if // Existing and T point to the same Type object. In this one case we // allow the equivalent redefinition. if (Existing == T) return true; // Yes, it's equal. // Any other kind of (non-equivalent) redefinition is an error. ThrowException("Redefinition of type named '" + Name + "' in the '" + T->getDescription() + "' type plane!"); } return false; } //===----------------------------------------------------------------------===// // Code for handling upreferences in type names... // // TypeContains - Returns true if Ty directly contains E in it. // static bool TypeContains(const Type *Ty, const Type *E) { return std::find(Ty->subtype_begin(), Ty->subtype_end(), E) != Ty->subtype_end(); } namespace { struct UpRefRecord { // NestingLevel - The number of nesting levels that need to be popped before // this type is resolved. unsigned NestingLevel; // LastContainedTy - This is the type at the current binding level for the // type. Every time we reduce the nesting level, this gets updated. const Type *LastContainedTy; // UpRefTy - This is the actual opaque type that the upreference is // represented with. OpaqueType *UpRefTy; UpRefRecord(unsigned NL, OpaqueType *URTy) : NestingLevel(NL), LastContainedTy(URTy), UpRefTy(URTy) {} }; } // UpRefs - A list of the outstanding upreferences that need to be resolved. static std::vector UpRefs; /// HandleUpRefs - Every time we finish a new layer of types, this function is /// called. It loops through the UpRefs vector, which is a list of the /// currently active types. For each type, if the up reference is contained in /// the newly completed type, we decrement the level count. When the level /// count reaches zero, the upreferenced type is the type that is passed in: /// thus we can complete the cycle. /// static PATypeHolder HandleUpRefs(const Type *ty) { if (!ty->isAbstract()) return ty; PATypeHolder Ty(ty); UR_OUT("Type '" << Ty->getDescription() << "' newly formed. Resolving upreferences.\n" << UpRefs.size() << " upreferences active!\n"); // If we find any resolvable upreferences (i.e., those whose NestingLevel goes // to zero), we resolve them all together before we resolve them to Ty. At // the end of the loop, if there is anything to resolve to Ty, it will be in // this variable. OpaqueType *TypeToResolve = 0; for (unsigned i = 0; i != UpRefs.size(); ++i) { UR_OUT(" UR#" << i << " - TypeContains(" << Ty->getDescription() << ", " << UpRefs[i].second->getDescription() << ") = " << (TypeContains(Ty, UpRefs[i].second) ? "true" : "false") << "\n"); if (TypeContains(Ty, UpRefs[i].LastContainedTy)) { // Decrement level of upreference unsigned Level = --UpRefs[i].NestingLevel; UpRefs[i].LastContainedTy = Ty; UR_OUT(" Uplevel Ref Level = " << Level << "\n"); if (Level == 0) { // Upreference should be resolved! if (!TypeToResolve) { TypeToResolve = UpRefs[i].UpRefTy; } else { UR_OUT(" * Resolving upreference for " << UpRefs[i].second->getDescription() << "\n"; std::string OldName = UpRefs[i].UpRefTy->getDescription()); UpRefs[i].UpRefTy->refineAbstractTypeTo(TypeToResolve); UR_OUT(" * Type '" << OldName << "' refined upreference to: " << (const void*)Ty << ", " << Ty->getDescription() << "\n"); } UpRefs.erase(UpRefs.begin()+i); // Remove from upreference list... --i; // Do not skip the next element... } } } if (TypeToResolve) { UR_OUT(" * Resolving upreference for " << UpRefs[i].second->getDescription() << "\n"; std::string OldName = TypeToResolve->getDescription()); TypeToResolve->refineAbstractTypeTo(Ty); } return Ty; } //===----------------------------------------------------------------------===// // RunVMAsmParser - Define an interface to this parser //===----------------------------------------------------------------------===// // Module *llvm::RunVMAsmParser(const std::string &Filename, FILE *F) { llvmAsmin = F; CurFilename = Filename; llvmAsmlineno = 1; // Reset the current line number... // Allocate a new module to read CurModule.CurrentModule = new Module(Filename); yyparse(); // Parse the file, potentially throwing exception Module *Result = ParserResult; llvmAsmin = stdin; // F is about to go away, don't use it anymore... ParserResult = 0; return Result; } %} %union { llvm::Module *ModuleVal; llvm::Function *FunctionVal; std::pair *ArgVal; llvm::BasicBlock *BasicBlockVal; llvm::TerminatorInst *TermInstVal; llvm::Instruction *InstVal; llvm::Constant *ConstVal; const llvm::Type *PrimType; llvm::PATypeHolder *TypeVal; llvm::Value *ValueVal; std::vector > *ArgList; std::vector *ValueList; std::list *TypeList; std::list > *PHIList; // Represent the RHS of PHI node std::vector > *JumpTable; std::vector *ConstVector; llvm::GlobalValue::LinkageTypes Linkage; int64_t SInt64Val; uint64_t UInt64Val; int SIntVal; unsigned UIntVal; double FPVal; bool BoolVal; char *StrVal; // This memory is strdup'd! llvm::ValID ValIDVal; // strdup'd memory maybe! llvm::Instruction::BinaryOps BinaryOpVal; llvm::Instruction::TermOps TermOpVal; llvm::Instruction::MemoryOps MemOpVal; llvm::Instruction::OtherOps OtherOpVal; llvm::Module::Endianness Endianness; } %type Module FunctionList %type Function FunctionProto FunctionHeader BasicBlockList %type BasicBlock InstructionList %type BBTerminatorInst %type Inst InstVal MemoryInst %type ConstVal ConstExpr %type ConstVector %type ArgList ArgListH %type ArgVal %type PHIList %type ValueRefList ValueRefListE // For call param lists %type IndexList // For GEP derived indices %type TypeListI ArgTypeListI %type JumpTable %type GlobalType // GLOBAL or CONSTANT? %type OptVolatile // 'volatile' or not %type OptLinkage %type BigOrLittle // ValueRef - Unresolved reference to a definition or BB %type ValueRef ConstValueRef SymbolicValueRef %type ResolvedVal // pair // Tokens and types for handling constant integer values // // ESINT64VAL - A negative number within long long range %token ESINT64VAL // EUINT64VAL - A positive number within uns. long long range %token EUINT64VAL %type EINT64VAL %token SINTVAL // Signed 32 bit ints... %token UINTVAL // Unsigned 32 bit ints... %type INTVAL %token FPVAL // Float or Double constant // Built in types... %type Types TypesV UpRTypes UpRTypesV %type SIntType UIntType IntType FPType PrimType // Classifications %token VOID BOOL SBYTE UBYTE SHORT USHORT INT UINT LONG ULONG %token FLOAT DOUBLE TYPE LABEL %token VAR_ID LABELSTR STRINGCONSTANT %type Name OptName OptAssign %token IMPLEMENTATION ZEROINITIALIZER TRUETOK FALSETOK BEGINTOK ENDTOK %token DECLARE GLOBAL CONSTANT VOLATILE %token TO DOTDOTDOT NULL_TOK UNDEF CONST INTERNAL LINKONCE WEAK APPENDING %token OPAQUE NOT EXTERNAL TARGET TRIPLE ENDIAN POINTERSIZE LITTLE BIG %token DEPLIBS // Basic Block Terminating Operators %token RET BR SWITCH INVOKE UNWIND UNREACHABLE // Binary Operators %type ArithmeticOps LogicalOps SetCondOps // Binops Subcatagories %token ADD SUB MUL DIV REM AND OR XOR %token SETLE SETGE SETLT SETGT SETEQ SETNE // Binary Comarators // Memory Instructions %token MALLOC ALLOCA FREE LOAD STORE GETELEMENTPTR // Other Operators %type ShiftOps %token PHI_TOK CALL CAST SELECT SHL SHR VAARG VANEXT %start Module %% // Handle constant integer size restriction and conversion... // INTVAL : SINTVAL; INTVAL : UINTVAL { if ($1 > (uint32_t)INT32_MAX) // Outside of my range! ThrowException("Value too large for type!"); $$ = (int32_t)$1; }; EINT64VAL : ESINT64VAL; // These have same type and can't cause problems... EINT64VAL : EUINT64VAL { if ($1 > (uint64_t)INT64_MAX) // Outside of my range! ThrowException("Value too large for type!"); $$ = (int64_t)$1; }; // Operations that are notably excluded from this list include: // RET, BR, & SWITCH because they end basic blocks and are treated specially. // ArithmeticOps: ADD | SUB | MUL | DIV | REM; LogicalOps : AND | OR | XOR; SetCondOps : SETLE | SETGE | SETLT | SETGT | SETEQ | SETNE; ShiftOps : SHL | SHR; // These are some types that allow classification if we only want a particular // thing... for example, only a signed, unsigned, or integral type. SIntType : LONG | INT | SHORT | SBYTE; UIntType : ULONG | UINT | USHORT | UBYTE; IntType : SIntType | UIntType; FPType : FLOAT | DOUBLE; // OptAssign - Value producing statements have an optional assignment component OptAssign : Name '=' { $$ = $1; } | /*empty*/ { $$ = 0; }; OptLinkage : INTERNAL { $$ = GlobalValue::InternalLinkage; } | LINKONCE { $$ = GlobalValue::LinkOnceLinkage; } | WEAK { $$ = GlobalValue::WeakLinkage; } | APPENDING { $$ = GlobalValue::AppendingLinkage; } | /*empty*/ { $$ = GlobalValue::ExternalLinkage; }; //===----------------------------------------------------------------------===// // Types includes all predefined types... except void, because it can only be // used in specific contexts (function returning void for example). To have // access to it, a user must explicitly use TypesV. // // TypesV includes all of 'Types', but it also includes the void type. TypesV : Types | VOID { $$ = new PATypeHolder($1); }; UpRTypesV : UpRTypes | VOID { $$ = new PATypeHolder($1); }; Types : UpRTypes { if (!UpRefs.empty()) ThrowException("Invalid upreference in type: " + (*$1)->getDescription()); $$ = $1; }; // Derived types are added later... // PrimType : BOOL | SBYTE | UBYTE | SHORT | USHORT | INT | UINT ; PrimType : LONG | ULONG | FLOAT | DOUBLE | TYPE | LABEL; UpRTypes : OPAQUE { $$ = new PATypeHolder(OpaqueType::get()); } | PrimType { $$ = new PATypeHolder($1); }; UpRTypes : SymbolicValueRef { // Named types are also simple types... $$ = new PATypeHolder(getTypeVal($1)); }; // Include derived types in the Types production. // UpRTypes : '\\' EUINT64VAL { // Type UpReference if ($2 > (uint64_t)~0U) ThrowException("Value out of range!"); OpaqueType *OT = OpaqueType::get(); // Use temporary placeholder UpRefs.push_back(UpRefRecord((unsigned)$2, OT)); // Add to vector... $$ = new PATypeHolder(OT); UR_OUT("New Upreference!\n"); } | UpRTypesV '(' ArgTypeListI ')' { // Function derived type? std::vector Params; for (std::list::iterator I = $3->begin(), E = $3->end(); I != E; ++I) Params.push_back(*I); bool isVarArg = Params.size() && Params.back() == Type::VoidTy; if (isVarArg) Params.pop_back(); $$ = new PATypeHolder(HandleUpRefs(FunctionType::get(*$1,Params,isVarArg))); delete $3; // Delete the argument list delete $1; // Delete the return type handle } | '[' EUINT64VAL 'x' UpRTypes ']' { // Sized array type? $$ = new PATypeHolder(HandleUpRefs(ArrayType::get(*$4, (unsigned)$2))); delete $4; } | '<' EUINT64VAL 'x' UpRTypes '>' { // Packed array type? const llvm::Type* ElemTy = $4->get(); if ((unsigned)$2 != $2) { ThrowException("Unsigned result not equal to signed result"); } if(!ElemTy->isPrimitiveType()) { ThrowException("Elemental type of a PackedType must be primitive"); } $$ = new PATypeHolder(HandleUpRefs(PackedType::get(*$4, (unsigned)$2))); delete $4; } | '{' TypeListI '}' { // Structure type? std::vector Elements; for (std::list::iterator I = $2->begin(), E = $2->end(); I != E; ++I) Elements.push_back(*I); $$ = new PATypeHolder(HandleUpRefs(StructType::get(Elements))); delete $2; } | '{' '}' { // Empty structure type? $$ = new PATypeHolder(StructType::get(std::vector())); } | UpRTypes '*' { // Pointer type? $$ = new PATypeHolder(HandleUpRefs(PointerType::get(*$1))); delete $1; }; // TypeList - Used for struct declarations and as a basis for function type // declaration type lists // TypeListI : UpRTypes { $$ = new std::list(); $$->push_back(*$1); delete $1; } | TypeListI ',' UpRTypes { ($$=$1)->push_back(*$3); delete $3; }; // ArgTypeList - List of types for a function type declaration... ArgTypeListI : TypeListI | TypeListI ',' DOTDOTDOT { ($$=$1)->push_back(Type::VoidTy); } | DOTDOTDOT { ($$ = new std::list())->push_back(Type::VoidTy); } | /*empty*/ { $$ = new std::list(); }; // ConstVal - The various declarations that go into the constant pool. This // production is used ONLY to represent constants that show up AFTER a 'const', // 'constant' or 'global' token at global scope. Constants that can be inlined // into other expressions (such as integers and constexprs) are handled by the // ResolvedVal, ValueRef and ConstValueRef productions. // ConstVal: Types '[' ConstVector ']' { // Nonempty unsized arr const ArrayType *ATy = dyn_cast($1->get()); if (ATy == 0) ThrowException("Cannot make array constant with type: '" + (*$1)->getDescription() + "'!"); const Type *ETy = ATy->getElementType(); int NumElements = ATy->getNumElements(); // Verify that we have the correct size... if (NumElements != -1 && NumElements != (int)$3->size()) ThrowException("Type mismatch: constant sized array initialized with " + utostr($3->size()) + " arguments, but has size of " + itostr(NumElements) + "!"); // Verify all elements are correct type! for (unsigned i = 0; i < $3->size(); i++) { if (ETy != (*$3)[i]->getType()) ThrowException("Element #" + utostr(i) + " is not of type '" + ETy->getDescription() +"' as required!\nIt is of type '"+ (*$3)[i]->getType()->getDescription() + "'."); } $$ = ConstantArray::get(ATy, *$3); delete $1; delete $3; } | Types '[' ']' { const ArrayType *ATy = dyn_cast($1->get()); if (ATy == 0) ThrowException("Cannot make array constant with type: '" + (*$1)->getDescription() + "'!"); int NumElements = ATy->getNumElements(); if (NumElements != -1 && NumElements != 0) ThrowException("Type mismatch: constant sized array initialized with 0" " arguments, but has size of " + itostr(NumElements) +"!"); $$ = ConstantArray::get(ATy, std::vector()); delete $1; } | Types 'c' STRINGCONSTANT { const ArrayType *ATy = dyn_cast($1->get()); if (ATy == 0) ThrowException("Cannot make array constant with type: '" + (*$1)->getDescription() + "'!"); int NumElements = ATy->getNumElements(); const Type *ETy = ATy->getElementType(); char *EndStr = UnEscapeLexed($3, true); if (NumElements != -1 && NumElements != (EndStr-$3)) ThrowException("Can't build string constant of size " + itostr((int)(EndStr-$3)) + " when array has size " + itostr(NumElements) + "!"); std::vector Vals; if (ETy == Type::SByteTy) { for (char *C = $3; C != EndStr; ++C) Vals.push_back(ConstantSInt::get(ETy, *C)); } else if (ETy == Type::UByteTy) { for (char *C = $3; C != EndStr; ++C) Vals.push_back(ConstantUInt::get(ETy, (unsigned char)*C)); } else { free($3); ThrowException("Cannot build string arrays of non byte sized elements!"); } free($3); $$ = ConstantArray::get(ATy, Vals); delete $1; } | Types '<' ConstVector '>' { // Nonempty unsized arr const PackedType *PTy = dyn_cast($1->get()); if (PTy == 0) ThrowException("Cannot make packed constant with type: '" + (*$1)->getDescription() + "'!"); const Type *ETy = PTy->getElementType(); int NumElements = PTy->getNumElements(); // Verify that we have the correct size... if (NumElements != -1 && NumElements != (int)$3->size()) ThrowException("Type mismatch: constant sized packed initialized with " + utostr($3->size()) + " arguments, but has size of " + itostr(NumElements) + "!"); // Verify all elements are correct type! for (unsigned i = 0; i < $3->size(); i++) { if (ETy != (*$3)[i]->getType()) ThrowException("Element #" + utostr(i) + " is not of type '" + ETy->getDescription() +"' as required!\nIt is of type '"+ (*$3)[i]->getType()->getDescription() + "'."); } $$ = ConstantPacked::get(PTy, *$3); delete $1; delete $3; } | Types '{' ConstVector '}' { const StructType *STy = dyn_cast($1->get()); if (STy == 0) ThrowException("Cannot make struct constant with type: '" + (*$1)->getDescription() + "'!"); if ($3->size() != STy->getNumContainedTypes()) ThrowException("Illegal number of initializers for structure type!"); // Check to ensure that constants are compatible with the type initializer! for (unsigned i = 0, e = $3->size(); i != e; ++i) if ((*$3)[i]->getType() != STy->getElementType(i)) ThrowException("Expected type '" + STy->getElementType(i)->getDescription() + "' for element #" + utostr(i) + " of structure initializer!"); $$ = ConstantStruct::get(STy, *$3); delete $1; delete $3; } | Types '{' '}' { const StructType *STy = dyn_cast($1->get()); if (STy == 0) ThrowException("Cannot make struct constant with type: '" + (*$1)->getDescription() + "'!"); if (STy->getNumContainedTypes() != 0) ThrowException("Illegal number of initializers for structure type!"); $$ = ConstantStruct::get(STy, std::vector()); delete $1; } | Types NULL_TOK { const PointerType *PTy = dyn_cast($1->get()); if (PTy == 0) ThrowException("Cannot make null pointer constant with type: '" + (*$1)->getDescription() + "'!"); $$ = ConstantPointerNull::get(PTy); delete $1; } | Types UNDEF { $$ = UndefValue::get($1->get()); delete $1; } | Types SymbolicValueRef { const PointerType *Ty = dyn_cast($1->get()); if (Ty == 0) ThrowException("Global const reference must be a pointer type!"); // ConstExprs can exist in the body of a function, thus creating // GlobalValues whenever they refer to a variable. Because we are in // the context of a function, getValNonImprovising will search the functions // symbol table instead of the module symbol table for the global symbol, // which throws things all off. To get around this, we just tell // getValNonImprovising that we are at global scope here. // Function *SavedCurFn = CurFun.CurrentFunction; CurFun.CurrentFunction = 0; Value *V = getValNonImprovising(Ty, $2); CurFun.CurrentFunction = SavedCurFn; // If this is an initializer for a constant pointer, which is referencing a // (currently) undefined variable, create a stub now that shall be replaced // in the future with the right type of variable. // if (V == 0) { assert(isa(Ty) && "Globals may only be used as pointers!"); const PointerType *PT = cast(Ty); // First check to see if the forward references value is already created! PerModuleInfo::GlobalRefsType::iterator I = CurModule.GlobalRefs.find(std::make_pair(PT, $2)); if (I != CurModule.GlobalRefs.end()) { V = I->second; // Placeholder already exists, use it... $2.destroy(); } else { std::string Name; if ($2.Type == ValID::NameVal) Name = $2.Name; // Create the forward referenced global. GlobalValue *GV; if (const FunctionType *FTy = dyn_cast(PT->getElementType())) { GV = new Function(FTy, GlobalValue::ExternalLinkage, Name, CurModule.CurrentModule); } else { GV = new GlobalVariable(PT->getElementType(), false, GlobalValue::ExternalLinkage, 0, Name, CurModule.CurrentModule); } // Keep track of the fact that we have a forward ref to recycle it CurModule.GlobalRefs.insert(std::make_pair(std::make_pair(PT, $2), GV)); V = GV; } } $$ = cast(V); delete $1; // Free the type handle } | Types ConstExpr { if ($1->get() != $2->getType()) ThrowException("Mismatched types for constant expression!"); $$ = $2; delete $1; } | Types ZEROINITIALIZER { const Type *Ty = $1->get(); if (isa(Ty) || Ty == Type::LabelTy || isa(Ty)) ThrowException("Cannot create a null initialized value of this type!"); $$ = Constant::getNullValue(Ty); delete $1; }; ConstVal : SIntType EINT64VAL { // integral constants if (!ConstantSInt::isValueValidForType($1, $2)) ThrowException("Constant value doesn't fit in type!"); $$ = ConstantSInt::get($1, $2); } | UIntType EUINT64VAL { // integral constants if (!ConstantUInt::isValueValidForType($1, $2)) ThrowException("Constant value doesn't fit in type!"); $$ = ConstantUInt::get($1, $2); } | BOOL TRUETOK { // Boolean constants $$ = ConstantBool::True; } | BOOL FALSETOK { // Boolean constants $$ = ConstantBool::False; } | FPType FPVAL { // Float & Double constants if (!ConstantFP::isValueValidForType($1, $2)) ThrowException("Floating point constant invalid for type!!"); $$ = ConstantFP::get($1, $2); }; ConstExpr: CAST '(' ConstVal TO Types ')' { if (!$3->getType()->isFirstClassType()) ThrowException("cast constant expression from a non-primitive type: '" + $3->getType()->getDescription() + "'!"); if (!$5->get()->isFirstClassType()) ThrowException("cast constant expression to a non-primitive type: '" + $5->get()->getDescription() + "'!"); $$ = ConstantExpr::getCast($3, $5->get()); delete $5; } | GETELEMENTPTR '(' ConstVal IndexList ')' { if (!isa($3->getType())) ThrowException("GetElementPtr requires a pointer operand!"); // LLVM 1.2 and earlier used ubyte struct indices. Convert any ubyte struct // indices to uint struct indices for compatibility. generic_gep_type_iterator::iterator> GTI = gep_type_begin($3->getType(), $4->begin(), $4->end()), GTE = gep_type_end($3->getType(), $4->begin(), $4->end()); for (unsigned i = 0, e = $4->size(); i != e && GTI != GTE; ++i, ++GTI) if (isa(*GTI)) // Only change struct indices if (ConstantUInt *CUI = dyn_cast((*$4)[i])) if (CUI->getType() == Type::UByteTy) (*$4)[i] = ConstantExpr::getCast(CUI, Type::UIntTy); const Type *IdxTy = GetElementPtrInst::getIndexedType($3->getType(), *$4, true); if (!IdxTy) ThrowException("Index list invalid for constant getelementptr!"); std::vector IdxVec; for (unsigned i = 0, e = $4->size(); i != e; ++i) if (Constant *C = dyn_cast((*$4)[i])) IdxVec.push_back(C); else ThrowException("Indices to constant getelementptr must be constants!"); delete $4; $$ = ConstantExpr::getGetElementPtr($3, IdxVec); } | SELECT '(' ConstVal ',' ConstVal ',' ConstVal ')' { if ($3->getType() != Type::BoolTy) ThrowException("Select condition must be of boolean type!"); if ($5->getType() != $7->getType()) ThrowException("Select operand types must match!"); $$ = ConstantExpr::getSelect($3, $5, $7); } | ArithmeticOps '(' ConstVal ',' ConstVal ')' { if ($3->getType() != $5->getType()) ThrowException("Binary operator types must match!"); // HACK: llvm 1.3 and earlier used to emit invalid pointer constant exprs. // To retain backward compatibility with these early compilers, we emit a // cast to the appropriate integer type automatically if we are in the // broken case. See PR424 for more information. if (!isa($3->getType())) { $$ = ConstantExpr::get($1, $3, $5); } else { const Type *IntPtrTy = 0; switch (CurModule.CurrentModule->getPointerSize()) { case Module::Pointer32: IntPtrTy = Type::IntTy; break; case Module::Pointer64: IntPtrTy = Type::LongTy; break; default: ThrowException("invalid pointer binary constant expr!"); } $$ = ConstantExpr::get($1, ConstantExpr::getCast($3, IntPtrTy), ConstantExpr::getCast($5, IntPtrTy)); $$ = ConstantExpr::getCast($$, $3->getType()); } } | LogicalOps '(' ConstVal ',' ConstVal ')' { if ($3->getType() != $5->getType()) ThrowException("Logical operator types must match!"); if (!$3->getType()->isIntegral()) ThrowException("Logical operands must have integral types!"); $$ = ConstantExpr::get($1, $3, $5); } | SetCondOps '(' ConstVal ',' ConstVal ')' { if ($3->getType() != $5->getType()) ThrowException("setcc operand types must match!"); $$ = ConstantExpr::get($1, $3, $5); } | ShiftOps '(' ConstVal ',' ConstVal ')' { if ($5->getType() != Type::UByteTy) ThrowException("Shift count for shift constant must be unsigned byte!"); if (!$3->getType()->isInteger()) ThrowException("Shift constant expression requires integer operand!"); $$ = ConstantExpr::get($1, $3, $5); }; // ConstVector - A list of comma separated constants. ConstVector : ConstVector ',' ConstVal { ($$ = $1)->push_back($3); } | ConstVal { $$ = new std::vector(); $$->push_back($1); }; // GlobalType - Match either GLOBAL or CONSTANT for global declarations... GlobalType : GLOBAL { $$ = false; } | CONSTANT { $$ = true; }; //===----------------------------------------------------------------------===// // Rules to match Modules //===----------------------------------------------------------------------===// // Module rule: Capture the result of parsing the whole file into a result // variable... // Module : FunctionList { $$ = ParserResult = $1; CurModule.ModuleDone(); }; // FunctionList - A list of functions, preceeded by a constant pool. // FunctionList : FunctionList Function { $$ = $1; CurFun.FunctionDone(); } | FunctionList FunctionProto { $$ = $1; } | FunctionList IMPLEMENTATION { $$ = $1; } | ConstPool { $$ = CurModule.CurrentModule; // Emit an error if there are any unresolved types left. if (!CurModule.LateResolveTypes.empty()) { const ValID &DID = CurModule.LateResolveTypes.begin()->first; if (DID.Type == ValID::NameVal) ThrowException("Reference to an undefined type: '"+DID.getName() + "'"); else ThrowException("Reference to an undefined type: #" + itostr(DID.Num)); } }; // ConstPool - Constants with optional names assigned to them. ConstPool : ConstPool OptAssign TYPE TypesV { // Eagerly resolve types. This is not an optimization, this is a // requirement that is due to the fact that we could have this: // // %list = type { %list * } // %list = type { %list * } ; repeated type decl // // If types are not resolved eagerly, then the two types will not be // determined to be the same type! // ResolveTypeTo($2, *$4); if (!setTypeName(*$4, $2) && !$2) { // If this is a named type that is not a redefinition, add it to the slot // table. CurModule.Types.push_back(*$4); } delete $4; } | ConstPool FunctionProto { // Function prototypes can be in const pool } | ConstPool OptAssign OptLinkage GlobalType ConstVal { if ($5 == 0) ThrowException("Global value initializer is not a constant!"); ParseGlobalVariable($2, $3, $4, $5->getType(), $5); } | ConstPool OptAssign EXTERNAL GlobalType Types { ParseGlobalVariable($2, GlobalValue::ExternalLinkage, $4, *$5, 0); delete $5; } | ConstPool TARGET TargetDefinition { } | ConstPool DEPLIBS '=' LibrariesDefinition { } | /* empty: end of list */ { }; BigOrLittle : BIG { $$ = Module::BigEndian; }; BigOrLittle : LITTLE { $$ = Module::LittleEndian; }; TargetDefinition : ENDIAN '=' BigOrLittle { CurModule.CurrentModule->setEndianness($3); } | POINTERSIZE '=' EUINT64VAL { if ($3 == 32) CurModule.CurrentModule->setPointerSize(Module::Pointer32); else if ($3 == 64) CurModule.CurrentModule->setPointerSize(Module::Pointer64); else ThrowException("Invalid pointer size: '" + utostr($3) + "'!"); } | TRIPLE '=' STRINGCONSTANT { CurModule.CurrentModule->setTargetTriple($3); free($3); }; LibrariesDefinition : '[' LibList ']'; LibList : LibList ',' STRINGCONSTANT { CurModule.CurrentModule->addLibrary($3); free($3); } | STRINGCONSTANT { CurModule.CurrentModule->addLibrary($1); free($1); } | /* empty: end of list */ { } ; //===----------------------------------------------------------------------===// // Rules to match Function Headers //===----------------------------------------------------------------------===// Name : VAR_ID | STRINGCONSTANT; OptName : Name | /*empty*/ { $$ = 0; }; ArgVal : Types OptName { if (*$1 == Type::VoidTy) ThrowException("void typed arguments are invalid!"); $$ = new std::pair($1, $2); }; ArgListH : ArgListH ',' ArgVal { $$ = $1; $1->push_back(*$3); delete $3; } | ArgVal { $$ = new std::vector >(); $$->push_back(*$1); delete $1; }; ArgList : ArgListH { $$ = $1; } | ArgListH ',' DOTDOTDOT { $$ = $1; $$->push_back(std::pair(new PATypeHolder(Type::VoidTy), 0)); } | DOTDOTDOT { $$ = new std::vector >(); $$->push_back(std::make_pair(new PATypeHolder(Type::VoidTy), (char*)0)); } | /* empty */ { $$ = 0; }; FunctionHeaderH : TypesV Name '(' ArgList ')' { UnEscapeLexed($2); std::string FunctionName($2); free($2); // Free strdup'd memory! if (!(*$1)->isFirstClassType() && *$1 != Type::VoidTy) ThrowException("LLVM functions cannot return aggregate types!"); std::vector ParamTypeList; if ($4) { // If there are arguments... for (std::vector >::iterator I = $4->begin(); I != $4->end(); ++I) ParamTypeList.push_back(I->first->get()); } bool isVarArg = ParamTypeList.size() && ParamTypeList.back() == Type::VoidTy; if (isVarArg) ParamTypeList.pop_back(); const FunctionType *FT = FunctionType::get(*$1, ParamTypeList, isVarArg); const PointerType *PFT = PointerType::get(FT); delete $1; ValID ID; if (!FunctionName.empty()) { ID = ValID::create((char*)FunctionName.c_str()); } else { ID = ValID::create((int)CurModule.Values[PFT].size()); } Function *Fn = 0; // See if this function was forward referenced. If so, recycle the object. if (GlobalValue *FWRef = CurModule.GetForwardRefForGlobal(PFT, ID)) { // Move the function to the end of the list, from whereever it was // previously inserted. Fn = cast(FWRef); CurModule.CurrentModule->getFunctionList().remove(Fn); CurModule.CurrentModule->getFunctionList().push_back(Fn); } else if (!FunctionName.empty() && // Merge with an earlier prototype? (Fn = CurModule.CurrentModule->getFunction(FunctionName, FT))) { // If this is the case, either we need to be a forward decl, or it needs // to be. if (!CurFun.isDeclare && !Fn->isExternal()) ThrowException("Redefinition of function '" + FunctionName + "'!"); // Make sure to strip off any argument names so we can't get conflicts. if (Fn->isExternal()) for (Function::arg_iterator AI = Fn->arg_begin(), AE = Fn->arg_end(); AI != AE; ++AI) AI->setName(""); } else { // Not already defined? Fn = new Function(FT, GlobalValue::ExternalLinkage, FunctionName, CurModule.CurrentModule); InsertValue(Fn, CurModule.Values); } CurFun.FunctionStart(Fn); // Add all of the arguments we parsed to the function... if ($4) { // Is null if empty... if (isVarArg) { // Nuke the last entry assert($4->back().first->get() == Type::VoidTy && $4->back().second == 0&& "Not a varargs marker!"); delete $4->back().first; $4->pop_back(); // Delete the last entry } Function::arg_iterator ArgIt = Fn->arg_begin(); for (std::vector >::iterator I =$4->begin(); I != $4->end(); ++I, ++ArgIt) { delete I->first; // Delete the typeholder... setValueName(ArgIt, I->second); // Insert arg into symtab... InsertValue(ArgIt); } delete $4; // We're now done with the argument list } }; BEGIN : BEGINTOK | '{'; // Allow BEGIN or '{' to start a function FunctionHeader : OptLinkage FunctionHeaderH BEGIN { $$ = CurFun.CurrentFunction; // Make sure that we keep track of the linkage type even if there was a // previous "declare". $$->setLinkage($1); }; END : ENDTOK | '}'; // Allow end of '}' to end a function Function : BasicBlockList END { $$ = $1; }; FunctionProto : DECLARE { CurFun.isDeclare = true; } FunctionHeaderH { $$ = CurFun.CurrentFunction; CurFun.FunctionDone(); }; //===----------------------------------------------------------------------===// // Rules to match Basic Blocks //===----------------------------------------------------------------------===// ConstValueRef : ESINT64VAL { // A reference to a direct constant $$ = ValID::create($1); } | EUINT64VAL { $$ = ValID::create($1); } | FPVAL { // Perhaps it's an FP constant? $$ = ValID::create($1); } | TRUETOK { $$ = ValID::create(ConstantBool::True); } | FALSETOK { $$ = ValID::create(ConstantBool::False); } | NULL_TOK { $$ = ValID::createNull(); } | UNDEF { $$ = ValID::createUndef(); } | '<' ConstVector '>' { // Nonempty unsized packed vector const Type *ETy = (*$2)[0]->getType(); int NumElements = $2->size(); PackedType* pt = PackedType::get(ETy, NumElements); PATypeHolder* PTy = new PATypeHolder( HandleUpRefs( PackedType::get( ETy, NumElements) ) ); // Verify all elements are correct type! for (unsigned i = 0; i < $2->size(); i++) { if (ETy != (*$2)[i]->getType()) ThrowException("Element #" + utostr(i) + " is not of type '" + ETy->getDescription() +"' as required!\nIt is of type '" + (*$2)[i]->getType()->getDescription() + "'."); } $$ = ValID::create(ConstantPacked::get(pt, *$2)); delete PTy; delete $2; } | ConstExpr { $$ = ValID::create($1); }; // SymbolicValueRef - Reference to one of two ways of symbolically refering to // another value. // SymbolicValueRef : INTVAL { // Is it an integer reference...? $$ = ValID::create($1); } | Name { // Is it a named reference...? $$ = ValID::create($1); }; // ValueRef - A reference to a definition... either constant or symbolic ValueRef : SymbolicValueRef | ConstValueRef; // ResolvedVal - a pair. This is used only in cases where the // type immediately preceeds the value reference, and allows complex constant // pool references (for things like: 'ret [2 x int] [ int 12, int 42]') ResolvedVal : Types ValueRef { $$ = getVal(*$1, $2); delete $1; }; BasicBlockList : BasicBlockList BasicBlock { $$ = $1; } | FunctionHeader BasicBlock { // Do not allow functions with 0 basic blocks $$ = $1; }; // Basic blocks are terminated by branching instructions: // br, br/cc, switch, ret // BasicBlock : InstructionList OptAssign BBTerminatorInst { setValueName($3, $2); InsertValue($3); $1->getInstList().push_back($3); InsertValue($1); $$ = $1; }; InstructionList : InstructionList Inst { $1->getInstList().push_back($2); $$ = $1; } | /* empty */ { $$ = CurBB = getBBVal(ValID::create((int)CurFun.NextBBNum++), true); // Make sure to move the basic block to the correct location in the // function, instead of leaving it inserted wherever it was first // referenced. CurFun.CurrentFunction->getBasicBlockList().remove(CurBB); CurFun.CurrentFunction->getBasicBlockList().push_back(CurBB); } | LABELSTR { $$ = CurBB = getBBVal(ValID::create($1), true); // Make sure to move the basic block to the correct location in the // function, instead of leaving it inserted wherever it was first // referenced. CurFun.CurrentFunction->getBasicBlockList().remove(CurBB); CurFun.CurrentFunction->getBasicBlockList().push_back(CurBB); }; BBTerminatorInst : RET ResolvedVal { // Return with a result... $$ = new ReturnInst($2); } | RET VOID { // Return with no result... $$ = new ReturnInst(); } | BR LABEL ValueRef { // Unconditional Branch... $$ = new BranchInst(getBBVal($3)); } // Conditional Branch... | BR BOOL ValueRef ',' LABEL ValueRef ',' LABEL ValueRef { $$ = new BranchInst(getBBVal($6), getBBVal($9), getVal(Type::BoolTy, $3)); } | SWITCH IntType ValueRef ',' LABEL ValueRef '[' JumpTable ']' { SwitchInst *S = new SwitchInst(getVal($2, $3), getBBVal($6), $8->size()); $$ = S; std::vector >::iterator I = $8->begin(), E = $8->end(); for (; I != E; ++I) { if (ConstantInt *CI = dyn_cast(I->first)) S->addCase(CI, I->second); else ThrowException("Switch case is constant, but not a simple integer!"); } delete $8; } | SWITCH IntType ValueRef ',' LABEL ValueRef '[' ']' { SwitchInst *S = new SwitchInst(getVal($2, $3), getBBVal($6), 0); $$ = S; } | INVOKE TypesV ValueRef '(' ValueRefListE ')' TO LABEL ValueRef UNWIND LABEL ValueRef { const PointerType *PFTy; const FunctionType *Ty; if (!(PFTy = dyn_cast($2->get())) || !(Ty = dyn_cast(PFTy->getElementType()))) { // Pull out the types of all of the arguments... std::vector ParamTypes; if ($5) { for (std::vector::iterator I = $5->begin(), E = $5->end(); I != E; ++I) ParamTypes.push_back((*I)->getType()); } bool isVarArg = ParamTypes.size() && ParamTypes.back() == Type::VoidTy; if (isVarArg) ParamTypes.pop_back(); Ty = FunctionType::get($2->get(), ParamTypes, isVarArg); PFTy = PointerType::get(Ty); } Value *V = getVal(PFTy, $3); // Get the function we're calling... BasicBlock *Normal = getBBVal($9); BasicBlock *Except = getBBVal($12); // Create the call node... if (!$5) { // Has no arguments? $$ = new InvokeInst(V, Normal, Except, std::vector()); } else { // Has arguments? // Loop through FunctionType's arguments and ensure they are specified // correctly! // FunctionType::param_iterator I = Ty->param_begin(); FunctionType::param_iterator E = Ty->param_end(); std::vector::iterator ArgI = $5->begin(), ArgE = $5->end(); for (; ArgI != ArgE && I != E; ++ArgI, ++I) if ((*ArgI)->getType() != *I) ThrowException("Parameter " +(*ArgI)->getName()+ " is not of type '" + (*I)->getDescription() + "'!"); if (I != E || (ArgI != ArgE && !Ty->isVarArg())) ThrowException("Invalid number of parameters detected!"); $$ = new InvokeInst(V, Normal, Except, *$5); } delete $2; delete $5; } | UNWIND { $$ = new UnwindInst(); } | UNREACHABLE { $$ = new UnreachableInst(); }; JumpTable : JumpTable IntType ConstValueRef ',' LABEL ValueRef { $$ = $1; Constant *V = cast(getValNonImprovising($2, $3)); if (V == 0) ThrowException("May only switch on a constant pool value!"); $$->push_back(std::make_pair(V, getBBVal($6))); } | IntType ConstValueRef ',' LABEL ValueRef { $$ = new std::vector >(); Constant *V = cast(getValNonImprovising($1, $2)); if (V == 0) ThrowException("May only switch on a constant pool value!"); $$->push_back(std::make_pair(V, getBBVal($5))); }; Inst : OptAssign InstVal { // Is this definition named?? if so, assign the name... setValueName($2, $1); InsertValue($2); $$ = $2; }; PHIList : Types '[' ValueRef ',' ValueRef ']' { // Used for PHI nodes $$ = new std::list >(); $$->push_back(std::make_pair(getVal(*$1, $3), getBBVal($5))); delete $1; } | PHIList ',' '[' ValueRef ',' ValueRef ']' { $$ = $1; $1->push_back(std::make_pair(getVal($1->front().first->getType(), $4), getBBVal($6))); }; ValueRefList : ResolvedVal { // Used for call statements, and memory insts... $$ = new std::vector(); $$->push_back($1); } | ValueRefList ',' ResolvedVal { $$ = $1; $1->push_back($3); }; // ValueRefListE - Just like ValueRefList, except that it may also be empty! ValueRefListE : ValueRefList | /*empty*/ { $$ = 0; }; InstVal : ArithmeticOps Types ValueRef ',' ValueRef { if (!(*$2)->isInteger() && !(*$2)->isFloatingPoint() && !isa((*$2).get())) ThrowException( "Arithmetic operator requires integer, FP, or packed operands!"); if(isa((*$2).get()) && $1 == Instruction::Rem) { ThrowException( "Rem not supported on packed types!"); } $$ = BinaryOperator::create($1, getVal(*$2, $3), getVal(*$2, $5)); if ($$ == 0) ThrowException("binary operator returned null!"); delete $2; } | LogicalOps Types ValueRef ',' ValueRef { if (!(*$2)->isIntegral()) ThrowException("Logical operator requires integral operands!"); $$ = BinaryOperator::create($1, getVal(*$2, $3), getVal(*$2, $5)); if ($$ == 0) ThrowException("binary operator returned null!"); delete $2; } | SetCondOps Types ValueRef ',' ValueRef { if(isa((*$2).get())) { ThrowException( "PackedTypes currently not supported in setcc instructions!"); } $$ = new SetCondInst($1, getVal(*$2, $3), getVal(*$2, $5)); if ($$ == 0) ThrowException("binary operator returned null!"); delete $2; } | NOT ResolvedVal { std::cerr << "WARNING: Use of eliminated 'not' instruction:" << " Replacing with 'xor'.\n"; Value *Ones = ConstantIntegral::getAllOnesValue($2->getType()); if (Ones == 0) ThrowException("Expected integral type for not instruction!"); $$ = BinaryOperator::create(Instruction::Xor, $2, Ones); if ($$ == 0) ThrowException("Could not create a xor instruction!"); } | ShiftOps ResolvedVal ',' ResolvedVal { if ($4->getType() != Type::UByteTy) ThrowException("Shift amount must be ubyte!"); if (!$2->getType()->isInteger()) ThrowException("Shift constant expression requires integer operand!"); $$ = new ShiftInst($1, $2, $4); } | CAST ResolvedVal TO Types { if (!$4->get()->isFirstClassType()) ThrowException("cast instruction to a non-primitive type: '" + $4->get()->getDescription() + "'!"); $$ = new CastInst($2, *$4); delete $4; } | SELECT ResolvedVal ',' ResolvedVal ',' ResolvedVal { if ($2->getType() != Type::BoolTy) ThrowException("select condition must be boolean!"); if ($4->getType() != $6->getType()) ThrowException("select value types should match!"); $$ = new SelectInst($2, $4, $6); } | VAARG ResolvedVal ',' Types { $$ = new VAArgInst($2, *$4); delete $4; } | VANEXT ResolvedVal ',' Types { $$ = new VANextInst($2, *$4); delete $4; } | PHI_TOK PHIList { const Type *Ty = $2->front().first->getType(); if (!Ty->isFirstClassType()) ThrowException("PHI node operands must be of first class type!"); $$ = new PHINode(Ty); ((PHINode*)$$)->reserveOperandSpace($2->size()); while ($2->begin() != $2->end()) { if ($2->front().first->getType() != Ty) ThrowException("All elements of a PHI node must be of the same type!"); cast($$)->addIncoming($2->front().first, $2->front().second); $2->pop_front(); } delete $2; // Free the list... } | CALL TypesV ValueRef '(' ValueRefListE ')' { const PointerType *PFTy; const FunctionType *Ty; if (!(PFTy = dyn_cast($2->get())) || !(Ty = dyn_cast(PFTy->getElementType()))) { // Pull out the types of all of the arguments... std::vector ParamTypes; if ($5) { for (std::vector::iterator I = $5->begin(), E = $5->end(); I != E; ++I) ParamTypes.push_back((*I)->getType()); } bool isVarArg = ParamTypes.size() && ParamTypes.back() == Type::VoidTy; if (isVarArg) ParamTypes.pop_back(); if (!(*$2)->isFirstClassType() && *$2 != Type::VoidTy) ThrowException("LLVM functions cannot return aggregate types!"); Ty = FunctionType::get($2->get(), ParamTypes, isVarArg); PFTy = PointerType::get(Ty); } Value *V = getVal(PFTy, $3); // Get the function we're calling... // Create the call node... if (!$5) { // Has no arguments? // Make sure no arguments is a good thing! if (Ty->getNumParams() != 0) ThrowException("No arguments passed to a function that " "expects arguments!"); $$ = new CallInst(V, std::vector()); } else { // Has arguments? // Loop through FunctionType's arguments and ensure they are specified // correctly! // FunctionType::param_iterator I = Ty->param_begin(); FunctionType::param_iterator E = Ty->param_end(); std::vector::iterator ArgI = $5->begin(), ArgE = $5->end(); for (; ArgI != ArgE && I != E; ++ArgI, ++I) if ((*ArgI)->getType() != *I) ThrowException("Parameter " +(*ArgI)->getName()+ " is not of type '" + (*I)->getDescription() + "'!"); if (I != E || (ArgI != ArgE && !Ty->isVarArg())) ThrowException("Invalid number of parameters detected!"); $$ = new CallInst(V, *$5); } delete $2; delete $5; } | MemoryInst { $$ = $1; }; // IndexList - List of indices for GEP based instructions... IndexList : ',' ValueRefList { $$ = $2; } | /* empty */ { $$ = new std::vector(); }; OptVolatile : VOLATILE { $$ = true; } | /* empty */ { $$ = false; }; MemoryInst : MALLOC Types { $$ = new MallocInst(*$2); delete $2; } | MALLOC Types ',' UINT ValueRef { $$ = new MallocInst(*$2, getVal($4, $5)); delete $2; } | ALLOCA Types { $$ = new AllocaInst(*$2); delete $2; } | ALLOCA Types ',' UINT ValueRef { $$ = new AllocaInst(*$2, getVal($4, $5)); delete $2; } | FREE ResolvedVal { if (!isa($2->getType())) ThrowException("Trying to free nonpointer type " + $2->getType()->getDescription() + "!"); $$ = new FreeInst($2); } | OptVolatile LOAD Types ValueRef { if (!isa($3->get())) ThrowException("Can't load from nonpointer type: " + (*$3)->getDescription()); if (!cast($3->get())->getElementType()->isFirstClassType()) ThrowException("Can't load from pointer of non-first-class type: " + (*$3)->getDescription()); $$ = new LoadInst(getVal(*$3, $4), "", $1); delete $3; } | OptVolatile STORE ResolvedVal ',' Types ValueRef { const PointerType *PT = dyn_cast($5->get()); if (!PT) ThrowException("Can't store to a nonpointer type: " + (*$5)->getDescription()); const Type *ElTy = PT->getElementType(); if (ElTy != $3->getType()) ThrowException("Can't store '" + $3->getType()->getDescription() + "' into space of type '" + ElTy->getDescription() + "'!"); $$ = new StoreInst($3, getVal(*$5, $6), $1); delete $5; } | GETELEMENTPTR Types ValueRef IndexList { if (!isa($2->get())) ThrowException("getelementptr insn requires pointer operand!"); // LLVM 1.2 and earlier used ubyte struct indices. Convert any ubyte struct // indices to uint struct indices for compatibility. generic_gep_type_iterator::iterator> GTI = gep_type_begin($2->get(), $4->begin(), $4->end()), GTE = gep_type_end($2->get(), $4->begin(), $4->end()); for (unsigned i = 0, e = $4->size(); i != e && GTI != GTE; ++i, ++GTI) if (isa(*GTI)) // Only change struct indices if (ConstantUInt *CUI = dyn_cast((*$4)[i])) if (CUI->getType() == Type::UByteTy) (*$4)[i] = ConstantExpr::getCast(CUI, Type::UIntTy); if (!GetElementPtrInst::getIndexedType(*$2, *$4, true)) ThrowException("Invalid getelementptr indices for type '" + (*$2)->getDescription()+ "'!"); $$ = new GetElementPtrInst(getVal(*$2, $3), *$4); delete $2; delete $4; }; %% int yyerror(const char *ErrorMsg) { std::string where = std::string((CurFilename == "-") ? std::string("") : CurFilename) + ":" + utostr((unsigned) llvmAsmlineno) + ": "; std::string errMsg = std::string(ErrorMsg) + "\n" + where + " while reading "; if (yychar == YYEMPTY || yychar == 0) errMsg += "end-of-file."; else errMsg += "token: '" + std::string(llvmAsmtext, llvmAsmleng) + "'"; ThrowException(errMsg); return 0; }