llvmAsmParser.y

//===-- llvmAsmParser.y - Parser for llvm assembly files ---------*- C++ -*--=//
//
//  This file implements the bison parser for LLVM assembly languages files.
//
//===------------------------------------------------------------------------=//

%{
#include "ParserInternals.h"
#include "llvm/SymbolTable.h"
#include "llvm/Module.h"
#include "llvm/iTerminators.h"
#include "llvm/iMemory.h"
#include "llvm/iOperators.h"
#include "llvm/iPHINode.h"
#include "Support/STLExtras.h"
#include "Support/DepthFirstIterator.h"
#include <list>
#include <utility>
#include <algorithm>

int yyerror(const char *ErrorMsg); // Forward declarations to prevent "implicit
int yylex();                       // declaration" of xxx warnings.
int yyparse();

static Module *ParserResult;
std::string CurFilename;

// 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
// load/store instructions with indexes into a load/store + getelementptr pair
// of instructions.  When this compatiblity "Feature" is removed, this should be
// too.
//
static BasicBlock *CurBB;


// This contains info used when building the body of a function.  It is
// destroyed when the function is completed.
//
typedef std::vector<Value *> ValueList;           // Numbered defs
static void ResolveDefinitions(std::vector<ValueList> &LateResolvers,
                               std::vector<ValueList> *FutureLateResolvers = 0);

static struct PerModuleInfo {
  Module *CurrentModule;
  std::vector<ValueList>    Values;     // Module level numbered definitions
  std::vector<ValueList>    LateResolveValues;
  std::vector<PATypeHolder> Types;
  std::map<ValID, PATypeHolder> LateResolveTypes;

  // GlobalRefs - This maintains a mapping between <Type, ValID>'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 ConstantPointerRefs.
  //
  typedef std::map<std::pair<const PointerType *,
                             ValID>, GlobalVariable*> 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;
  }


  // DeclareNewGlobalValue - Called every time a new GV has been defined.  This
  // is used to remove things from the forward declaration map, resolving them
  // to the correct thing as needed.
  //
  void DeclareNewGlobalValue(GlobalValue *GV, ValID D) {
    // 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(GV->getType(), D));

    if (I != GlobalRefs.end()) {
      GlobalVariable *OldGV = I->second;   // Get the placeholder...
      I->first.second.destroy();  // Free string memory if necessary
      
      // Loop over all of the uses of the GlobalValue.  The only thing they are
      // allowed to be is ConstantPointerRef's.
      assert(OldGV->use_size() == 1 && "Only one reference should exist!");
      User *U = OldGV->use_back();  // Must be a ConstantPointerRef...
      ConstantPointerRef *CPR = cast<ConstantPointerRef>(U);
        
      // Change the const pool reference to point to the real global variable
      // now.  This should drop a use from the OldGV.
      CPR->mutateReferences(OldGV, GV);
      assert(OldGV->use_empty() && "All uses should be gone now!");
      
      // Remove OldGV from the module...
      CurrentModule->getGlobalList().remove(OldGV);
      delete OldGV;                        // Delete the old placeholder
      
      // Remove the map entry for the global now that it has been created...
      GlobalRefs.erase(I);
    }
  }

} CurModule;

static struct PerFunctionInfo {
  Function *CurrentFunction;     // Pointer to current function being created

  std::vector<ValueList> Values;      // Keep track of numbered definitions
  std::vector<ValueList> LateResolveValues;
  std::vector<PATypeHolder> Types;
  std::map<ValID, PATypeHolder> LateResolveTypes;
  bool isDeclare;                // Is this function a forward declararation?

  inline PerFunctionInfo() {
    CurrentFunction = 0;
    isDeclare = false;
  }

  inline ~PerFunctionInfo() {}

  inline void FunctionStart(Function *M) {
    CurrentFunction = M;
  }

  void FunctionDone() {
    // If we could not resolve some blocks at parsing time (forward branches)
    // resolve the branches now...
    ResolveDefinitions(LateResolveValues, &CurModule.LateResolveValues);

    // Make sure to resolve any constant expr references that might exist within
    // the function we just declared itself.
    ValID FID;
    if (CurrentFunction->hasName()) {
      FID = ValID::create((char*)CurrentFunction->getName().c_str());
    } else {
      unsigned Slot = CurrentFunction->getType()->getUniqueID();
      assert(CurModule.Values.size() > Slot && "Function not inserted?");
      // Figure out which slot number if is...
      for (unsigned i = 0; ; ++i) {
        assert(i < CurModule.Values[Slot].size() && "Function not found!");
        if (CurModule.Values[Slot][i] == CurrentFunction) {
          FID = ValID::create((int)i);
          break;
        }
      }
    }
    CurModule.DeclareNewGlobalValue(CurrentFunction, FID);

    Values.clear();         // Clear out function local definitions
    Types.clear();
    CurrentFunction = 0;
    isDeclare = false;
  }
} CurMeth;  // Info for the current function...

static bool inFunctionScope() { return CurMeth.CurrentFunction != 0; }


//===----------------------------------------------------------------------===//
//               Code to handle definitions of all the types
//===----------------------------------------------------------------------===//

static int InsertValue(Value *D,
                       std::vector<ValueList> &ValueTab = CurMeth.Values) {
  if (D->hasName()) return -1;           // Is this a numbered definition?

  // Yes, insert the value into the value table...
  unsigned type = D->getType()->getUniqueID();
  if (ValueTab.size() <= type)
    ValueTab.resize(type+1, ValueList());
  //printf("Values[%d][%d] = %d\n", type, ValueTab[type].size(), D);
  ValueTab[type].push_back(D);
  return ValueTab[type].size()-1;
}

// TODO: FIXME when Type are not const
static void InsertType(const Type *Ty, std::vector<PATypeHolder> &Types) {
  Types.push_back(Ty);
}

static const Type *getTypeVal(const ValID &D, bool DoNotImprovise = false) {
  switch (D.Type) {
  case ValID::NumberVal: {                 // Is it a numbered definition?
    unsigned Num = (unsigned)D.Num;

    // Module constants occupy the lowest numbered slots...
    if (Num < CurModule.Types.size()) 
      return CurModule.Types[Num];

    Num -= CurModule.Types.size();

    // Check that the number is within bounds...
    if (Num <= CurMeth.Types.size())
      return CurMeth.Types[Num];
    break;
  }
  case ValID::NameVal: {                // Is it a named definition?
    std::string Name(D.Name);
    SymbolTable *SymTab = 0;
    Value *N = 0;
    if (inFunctionScope()) {
      SymTab = &CurMeth.CurrentFunction->getSymbolTable();
      N = SymTab->lookup(Type::TypeTy, Name);
    }

    if (N == 0) {
      // Symbol table doesn't automatically chain yet... because the function
      // hasn't been added to the module...
      //
      SymTab = &CurModule.CurrentModule->getSymbolTable();
      N = SymTab->lookup(Type::TypeTy, Name);
      if (N == 0) break;
    }

    D.destroy();  // Free old strdup'd memory...
    return cast<Type>(N);
  }
  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?

  std::map<ValID, PATypeHolder> &LateResolver = inFunctionScope() ? 
    CurMeth.LateResolveTypes : CurModule.LateResolveTypes;
  
  std::map<ValID, PATypeHolder>::iterator I = LateResolver.find(D);
  if (I != LateResolver.end()) {
    return I->second;
  }

  Type *Typ = OpaqueType::get();
  LateResolver.insert(std::make_pair(D, Typ));
  return Typ;
}

static Value *lookupInSymbolTable(const Type *Ty, const std::string &Name) {
  SymbolTable &SymTab = 
    inFunctionScope() ? CurMeth.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<FunctionType>(Ty))
    ThrowException("Functions are not values and "
                   "must be referenced as pointers");

  switch (D.Type) {
  case ValID::NumberVal: {                 // Is it a numbered definition?
    unsigned type = Ty->getUniqueID();
    unsigned Num = (unsigned)D.Num;

    // Module constants occupy the lowest numbered slots...
    if (type < CurModule.Values.size()) {
      if (Num < CurModule.Values[type].size()) 
        return CurModule.Values[type][Num];

      Num -= CurModule.Values[type].size();
    }

    // Make sure that our type is within bounds
    if (CurMeth.Values.size() <= type) return 0;

    // Check that the number is within bounds...
    if (CurMeth.Values[type].size() <= Num) return 0;
  
    return CurMeth.Values[type][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<PointerType>(Ty))
      ThrowException("Cannot create a a non pointer null!");
    return ConstantPointerNull::get(cast<PointerType>(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 &D) {
  assert(Ty != Type::TypeTy && "Should use getTypeVal for types!");

  // See if the value has already been defined...
  Value *V = getValNonImprovising(Ty, D);
  if (V) return V;

  // 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...
  //
  Value *d = 0;
  switch (Ty->getPrimitiveID()) {
  case Type::LabelTyID:  d = new   BBPlaceHolder(Ty, D); break;
  default:               d = new ValuePlaceHolder(Ty, D); break;
  }

  assert(d != 0 && "How did we not make something?");
  if (inFunctionScope())
    InsertValue(d, CurMeth.LateResolveValues);
  else 
    InsertValue(d, CurModule.LateResolveValues);
  return d;
}


//===----------------------------------------------------------------------===//
//              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 (CurMeth.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::vector<ValueList> &LateResolvers,
                               std::vector<ValueList> *FutureLateResolvers) {
  // Loop over LateResolveDefs fixing up stuff that couldn't be resolved
  for (unsigned ty = 0; ty < LateResolvers.size(); ty++) {
    while (!LateResolvers[ty].empty()) {
      Value *V = LateResolvers[ty].back();
      assert(!isa<Type>(V) && "Types should be in LateResolveTypes!");

      LateResolvers[ty].pop_back();
      ValID &DID = getValIDFromPlaceHolder(V);

      Value *TheRealValue = getValNonImprovising(Type::getUniqueIDType(ty),DID);
      if (TheRealValue) {
        V->replaceAllUsesWith(TheRealValue);
        delete V;
      } 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() + "'",
			 getLineNumFromPlaceHolder(V));
	else
	  ThrowException("Reference to an invalid definition: #" +
			 itostr(DID.Num) + " of type '" + 
			 V->getType()->getDescription() + "'",
			 getLineNumFromPlaceHolder(V));
      }
    }
  }

  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) {
  std::vector<PATypeHolder> &Types = inFunctionScope() ? 
     CurMeth.Types : CurModule.Types;

   ValID D;
   if (Name) D = ValID::create(Name);
   else      D = ValID::create((int)Types.size());

   std::map<ValID, PATypeHolder> &LateResolver = inFunctionScope() ? 
     CurMeth.LateResolveTypes : CurModule.LateResolveTypes;
  
   std::map<ValID, PATypeHolder>::iterator I = LateResolver.find(D);
   if (I != LateResolver.end()) {
     ((DerivedType*)I->second.get())->refineAbstractTypeTo(ToTy);
     LateResolver.erase(I);
   }
}

// ResolveTypes - At this point, all types should be resolved.  Any that aren't
// are errors.
//
static void ResolveTypes(std::map<ValID, PATypeHolder> &LateResolveTypes) {
  if (!LateResolveTypes.empty()) {
    const ValID &DID = LateResolveTypes.begin()->first;

    if (DID.Type == ValID::NameVal)
      ThrowException("Reference to an invalid type: '" +DID.getName() + "'");
    else
      ThrowException("Reference to an invalid type: #" + itostr(DID.Num));
  }
}


// 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 freed by this function.
//
// This function returns true if the value has already been defined, but is
// allowed to be redefined in the specified context.  If the name is a new name
// for the typeplane, false is returned.
//
static bool setValueName(Value *V, char *NameStr) {
  if (NameStr == 0) return false;
  
  std::string Name(NameStr);      // Copy string
  free(NameStr);                  // Free old string

  if (V->getType() == Type::VoidTy) 
    ThrowException("Can't assign name '" + Name + 
		   "' to a null valued instruction!");

  SymbolTable &ST = inFunctionScope() ? 
    CurMeth.CurrentFunction->getSymbolTable() : 
    CurModule.CurrentModule->getSymbolTable();

  Value *Existing = ST.lookup(V->getType(), Name);
  if (Existing) {    // Inserting a name that is already defined???
    // 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 Type *Ty = dyn_cast<Type>(Existing)) {
      if (const OpaqueType *OpTy = dyn_cast<OpaqueType>(Ty)) {
	// We ARE replacing an opaque type!
	((OpaqueType*)OpTy)->refineAbstractTypeTo(cast<Type>(V));
	return true;
      }
    }

    // Otherwise, we are a simple redefinition of a value, check to see if it
    // is defined the same as the old one...
    if (const Type *Ty = dyn_cast<Type>(Existing)) {
      if (Ty == cast<Type>(V)) return true;  // Yes, it's equal.
      // std::cerr << "Type: " << Ty->getDescription() << " != "
      //      << cast<Type>(V)->getDescription() << "!\n";
    } else if (const Constant *C = dyn_cast<Constant>(Existing)) {
      if (C == V) return true;      // Constants are equal to themselves
    } else if (GlobalVariable *EGV = dyn_cast<GlobalVariable>(Existing)) {
      // 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 (GlobalVariable *GV = dyn_cast<GlobalVariable>(V)) {
        if (!EGV->hasInitializer() || !GV->hasInitializer() ||
             EGV->getInitializer() == GV->getInitializer()) {

          // Make sure the existing global version gets the initializer!  Make
          // sure that it also gets marked const if the new version is.
          if (GV->hasInitializer() && !EGV->hasInitializer())
            EGV->setInitializer(GV->getInitializer());
          if (GV->isConstant())
            EGV->setConstant(true);
          EGV->setLinkage(GV->getLinkage());
          
	  delete GV;     // Destroy the duplicate!
          return true;   // They are equivalent!
        }
      }
    }
    ThrowException("Redefinition of value named '" + Name + "' in the '" +
		   V->getType()->getDescription() + "' type plane!");
  }

  V->setName(Name, &ST);
  return false;
}


//===----------------------------------------------------------------------===//
// Code for handling upreferences in type names...
//

// TypeContains - Returns true if Ty contains E in it.
//
static bool TypeContains(const Type *Ty, const Type *E) {
  return find(df_begin(Ty), df_end(Ty), E) != df_end(Ty);
}


static std::vector<std::pair<unsigned, OpaqueType *> > UpRefs;

static PATypeHolder HandleUpRefs(const Type *ty) {
  PATypeHolder Ty(ty);
  UR_OUT("Type '" << ty->getDescription() << 
         "' newly formed.  Resolving upreferences.\n" <<
         UpRefs.size() << " upreferences active!\n");
  for (unsigned i = 0; i < UpRefs.size(); ) {
    UR_OUT("  UR#" << i << " - TypeContains(" << Ty->getDescription() << ", " 
	   << UpRefs[i].second->getDescription() << ") = " 
	   << (TypeContains(Ty, UpRefs[i].second) ? "true" : "false") << endl);
    if (TypeContains(Ty, UpRefs[i].second)) {
      unsigned Level = --UpRefs[i].first;   // Decrement level of upreference
      UR_OUT("  Uplevel Ref Level = " << Level << endl);
      if (Level == 0) {                     // Upreference should be resolved! 
	UR_OUT("  * Resolving upreference for "
               << UpRefs[i].second->getDescription() << endl;
	       std::string OldName = UpRefs[i].second->getDescription());
	UpRefs[i].second->refineAbstractTypeTo(Ty);
	UpRefs.erase(UpRefs.begin()+i);     // Remove from upreference list...
	UR_OUT("  * Type '" << OldName << "' refined upreference to: "
	       << (const void*)Ty << ", " << Ty->getDescription() << endl);
	continue;
      }
    }

    ++i;                                  // Otherwise, no resolve, move on...
  }
  // FIXME: TODO: this should return the updated type
  return Ty;
}


//===----------------------------------------------------------------------===//
//            RunVMAsmParser - Define an interface to this parser
//===----------------------------------------------------------------------===//
//
Module *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.
  Module *Result = ParserResult;
  llvmAsmin = stdin;    // F is about to go away, don't use it anymore...
  ParserResult = 0;

  return Result;
}

%}

%union {
  Module                           *ModuleVal;
  Function                         *FunctionVal;
  std::pair<PATypeHolder*, char*>  *ArgVal;
  BasicBlock                       *BasicBlockVal;
  TerminatorInst                   *TermInstVal;
  Instruction                      *InstVal;
  Constant                         *ConstVal;

  const Type                       *PrimType;
  PATypeHolder                     *TypeVal;
  Value                            *ValueVal;

  std::vector<std::pair<PATypeHolder*,char*> > *ArgList;
  std::vector<Value*>              *ValueList;
  std::list<PATypeHolder>          *TypeList;
  std::list<std::pair<Value*,
                      BasicBlock*> > *PHIList; // Represent the RHS of PHI node
  std::vector<std::pair<Constant*, BasicBlock*> > *JumpTable;
  std::vector<Constant*>           *ConstVector;

  GlobalValue::LinkageTypes         Linkage;
  int64_t                           SInt64Val;
  uint64_t                          UInt64Val;
  int                               SIntVal;
  unsigned                          UIntVal;
  double                            FPVal;
  bool                              BoolVal;

  char                             *StrVal;   // This memory is strdup'd!
  ValID                             ValIDVal; // strdup'd memory maybe!

  Instruction::BinaryOps            BinaryOpVal;
  Instruction::TermOps              TermOpVal;
  Instruction::MemoryOps            MemOpVal;
  Instruction::OtherOps             OtherOpVal;
  Module::Endianness                Endianness;
}

%type <ModuleVal>     Module FunctionList
%type <FunctionVal>   Function FunctionProto FunctionHeader BasicBlockList
%type <BasicBlockVal> BasicBlock InstructionList
%type <TermInstVal>   BBTerminatorInst
%type <InstVal>       Inst InstVal MemoryInst
%type <ConstVal>      ConstVal ConstExpr
%type <ConstVector>   ConstVector
%type <ArgList>       ArgList ArgListH
%type <ArgVal>        ArgVal
%type <PHIList>       PHIList
%type <ValueList>     ValueRefList ValueRefListE  // For call param lists
%type <ValueList>     IndexList                   // For GEP derived indices
%type <TypeList>      TypeListI ArgTypeListI
%type <JumpTable>     JumpTable
%type <BoolVal>       GlobalType                  // GLOBAL or CONSTANT?
%type <Linkage>       OptLinkage
%type <Endianness>    BigOrLittle

// ValueRef - Unresolved reference to a definition or BB
%type <ValIDVal>      ValueRef ConstValueRef SymbolicValueRef
%type <ValueVal>      ResolvedVal            // <type> <valref> pair
// Tokens and types for handling constant integer values
//
// ESINT64VAL - A negative number within long long range
%token <SInt64Val> ESINT64VAL

// EUINT64VAL - A positive number within uns. long long range
%token <UInt64Val> EUINT64VAL
%type  <SInt64Val> EINT64VAL

%token  <SIntVal>   SINTVAL   // Signed 32 bit ints...
%token  <UIntVal>   UINTVAL   // Unsigned 32 bit ints...
%type   <SIntVal>   INTVAL
%token  <FPVal>     FPVAL     // Float or Double constant

// Built in types...
%type  <TypeVal> Types TypesV UpRTypes UpRTypesV
%type  <PrimType> SIntType UIntType IntType FPType PrimType   // Classifications
%token <PrimType> VOID BOOL SBYTE UBYTE SHORT USHORT INT UINT LONG ULONG
%token <PrimType> FLOAT DOUBLE TYPE LABEL

%token <StrVal> VAR_ID LABELSTR STRINGCONSTANT
%type  <StrVal> Name OptName OptAssign


%token IMPLEMENTATION ZEROINITIALIZER TRUE FALSE BEGINTOK ENDTOK
%token  DECLARE GLOBAL CONSTANT
%token TO EXCEPT DOTDOTDOT NULL_TOK CONST INTERNAL LINKONCE APPENDING
%token OPAQUE NOT EXTERNAL TARGET ENDIAN POINTERSIZE LITTLE BIG

// Basic Block Terminating Operators 
%token <TermOpVal> RET BR SWITCH

// Binary Operators 
%type  <BinaryOpVal> BinaryOps  // all the binary operators
%type  <BinaryOpVal> ArithmeticOps LogicalOps SetCondOps // Binops Subcatagories
%token <BinaryOpVal> ADD SUB MUL DIV REM AND OR XOR
%token <BinaryOpVal> SETLE SETGE SETLT SETGT SETEQ SETNE  // Binary Comarators

// Memory Instructions
%token <MemOpVal> MALLOC ALLOCA FREE LOAD STORE GETELEMENTPTR

// Other Operators
%type  <OtherOpVal> ShiftOps
%token <OtherOpVal> PHI CALL INVOKE CAST SHL SHR VA_ARG

%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;
BinaryOps : ArithmeticOps | LogicalOps | SetCondOps;

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; } |
             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.size())
      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)INT64_MAX) ThrowException("Value out of range!");
    OpaqueType *OT = OpaqueType::get();        // Use temporary placeholder
    UpRefs.push_back(std::make_pair((unsigned)$2, OT));  // Add to vector...
    $$ = new PATypeHolder(OT);
    UR_OUT("New Upreference!\n");
  }
  | UpRTypesV '(' ArgTypeListI ')' {           // Function derived type?
    std::vector<const Type*> Params;
    mapto($3->begin(), $3->end(), std::back_inserter(Params), 
	  std::mem_fun_ref(&PATypeHandle::get));
    bool isVarArg = Params.size() && Params.back() == Type::VoidTy;
    if (isVarArg) Params.pop_back();

    $$ = new PATypeHolder(HandleUpRefs(FunctionType::get(*$1,Params,isVarArg)));
    delete $3;      // Delete the argument list
    delete $1;      // Delete the old type handle
  }
  | '[' EUINT64VAL 'x' UpRTypes ']' {          // Sized array type?
    $$ = new PATypeHolder(HandleUpRefs(ArrayType::get(*$4, (unsigned)$2)));
    delete $4;
  }
  | '{' TypeListI '}' {                        // Structure type?
    std::vector<const Type*> Elements;
    mapto($2->begin(), $2->end(), std::back_inserter(Elements), 
	std::mem_fun_ref(&PATypeHandle::get));

    $$ = new PATypeHolder(HandleUpRefs(StructType::get(Elements)));
    delete $2;
  }
  | '{' '}' {                                  // Empty structure type?
    $$ = new PATypeHolder(StructType::get(std::vector<const Type*>()));
  }
  | 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 {
    $$ = new std::list<PATypeHolder>();
    $$->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<PATypeHolder>())->push_back(Type::VoidTy);
  }
  | /*empty*/ {
    $$ = new std::list<PATypeHolder>();
  };

// 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<ArrayType>($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<ArrayType>($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<Constant*>());
    delete $1;
  }
  | Types 'c' STRINGCONSTANT {
    const ArrayType *ATy = dyn_cast<ArrayType>($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<Constant*> 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 '}' {
    const StructType *STy = dyn_cast<StructType>($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->getElementTypes()[i])
        ThrowException("Expected type '" +
                       STy->getElementTypes()[i]->getDescription() +
                       "' for element #" + utostr(i) +
                       " of structure initializer!");

    $$ = ConstantStruct::get(STy, *$3);
    delete $1; delete $3;
  }
  | Types '{' '}' {
    const StructType *STy = dyn_cast<StructType>($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<Constant*>());
    delete $1;
  }
  | Types NULL_TOK {
    const PointerType *PTy = dyn_cast<PointerType>($1->get());
    if (PTy == 0)
      ThrowException("Cannot make null pointer constant with type: '" + 
                     (*$1)->getDescription() + "'!");

    $$ = ConstantPointerNull::get(PTy);
    delete $1;
  }
  | Types SymbolicValueRef {
    const PointerType *Ty = dyn_cast<PointerType>($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
    // ConstantPointerRefs 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 = CurMeth.CurrentFunction;
    CurMeth.CurrentFunction = 0;

    Value *V = getValNonImprovising(Ty, $2);

    CurMeth.CurrentFunction = SavedCurFn;