It's not necessary to do rounding for alloca operations when the requested
alignment is equal to the stack alignment.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@40004 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/lib/ExecutionEngine/ExecutionEngine.cpp b/lib/ExecutionEngine/ExecutionEngine.cpp
new file mode 100644
index 0000000..bbddb95
--- /dev/null
+++ b/lib/ExecutionEngine/ExecutionEngine.cpp
@@ -0,0 +1,809 @@
+//===-- ExecutionEngine.cpp - Common Implementation shared by EEs ---------===//
+//
+//                     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 defines the common interface used by the various execution engine
+// subclasses.
+//
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "jit"
+#include "llvm/Constants.h"
+#include "llvm/DerivedTypes.h"
+#include "llvm/Module.h"
+#include "llvm/ModuleProvider.h"
+#include "llvm/ADT/Statistic.h"
+#include "llvm/ExecutionEngine/ExecutionEngine.h"
+#include "llvm/ExecutionEngine/GenericValue.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/MutexGuard.h"
+#include "llvm/System/DynamicLibrary.h"
+#include "llvm/Target/TargetData.h"
+#include <math.h>
+using namespace llvm;
+
+STATISTIC(NumInitBytes, "Number of bytes of global vars initialized");
+STATISTIC(NumGlobals  , "Number of global vars initialized");
+
+ExecutionEngine::EECtorFn ExecutionEngine::JITCtor = 0;
+ExecutionEngine::EECtorFn ExecutionEngine::InterpCtor = 0;
+
+ExecutionEngine::ExecutionEngine(ModuleProvider *P) {
+  LazyCompilationDisabled = false;
+  Modules.push_back(P);
+  assert(P && "ModuleProvider is null?");
+}
+
+ExecutionEngine::ExecutionEngine(Module *M) {
+  LazyCompilationDisabled = false;
+  assert(M && "Module is null?");
+  Modules.push_back(new ExistingModuleProvider(M));
+}
+
+ExecutionEngine::~ExecutionEngine() {
+  clearAllGlobalMappings();
+  for (unsigned i = 0, e = Modules.size(); i != e; ++i)
+    delete Modules[i];
+}
+
+/// FindFunctionNamed - Search all of the active modules to find the one that
+/// defines FnName.  This is very slow operation and shouldn't be used for
+/// general code.
+Function *ExecutionEngine::FindFunctionNamed(const char *FnName) {
+  for (unsigned i = 0, e = Modules.size(); i != e; ++i) {
+    if (Function *F = Modules[i]->getModule()->getFunction(FnName))
+      return F;
+  }
+  return 0;
+}
+
+
+/// addGlobalMapping - Tell the execution engine that the specified global is
+/// at the specified location.  This is used internally as functions are JIT'd
+/// and as global variables are laid out in memory.  It can and should also be
+/// used by clients of the EE that want to have an LLVM global overlay
+/// existing data in memory.
+void ExecutionEngine::addGlobalMapping(const GlobalValue *GV, void *Addr) {
+  MutexGuard locked(lock);
+  
+  void *&CurVal = state.getGlobalAddressMap(locked)[GV];
+  assert((CurVal == 0 || Addr == 0) && "GlobalMapping already established!");
+  CurVal = Addr;
+  
+  // If we are using the reverse mapping, add it too
+  if (!state.getGlobalAddressReverseMap(locked).empty()) {
+    const GlobalValue *&V = state.getGlobalAddressReverseMap(locked)[Addr];
+    assert((V == 0 || GV == 0) && "GlobalMapping already established!");
+    V = GV;
+  }
+}
+
+/// clearAllGlobalMappings - Clear all global mappings and start over again
+/// use in dynamic compilation scenarios when you want to move globals
+void ExecutionEngine::clearAllGlobalMappings() {
+  MutexGuard locked(lock);
+  
+  state.getGlobalAddressMap(locked).clear();
+  state.getGlobalAddressReverseMap(locked).clear();
+}
+
+/// updateGlobalMapping - Replace an existing mapping for GV with a new
+/// address.  This updates both maps as required.  If "Addr" is null, the
+/// entry for the global is removed from the mappings.
+void ExecutionEngine::updateGlobalMapping(const GlobalValue *GV, void *Addr) {
+  MutexGuard locked(lock);
+  
+  // Deleting from the mapping?
+  if (Addr == 0) {
+    state.getGlobalAddressMap(locked).erase(GV);
+    if (!state.getGlobalAddressReverseMap(locked).empty())
+      state.getGlobalAddressReverseMap(locked).erase(Addr);
+    return;
+  }
+  
+  void *&CurVal = state.getGlobalAddressMap(locked)[GV];
+  if (CurVal && !state.getGlobalAddressReverseMap(locked).empty())
+    state.getGlobalAddressReverseMap(locked).erase(CurVal);
+  CurVal = Addr;
+  
+  // If we are using the reverse mapping, add it too
+  if (!state.getGlobalAddressReverseMap(locked).empty()) {
+    const GlobalValue *&V = state.getGlobalAddressReverseMap(locked)[Addr];
+    assert((V == 0 || GV == 0) && "GlobalMapping already established!");
+    V = GV;
+  }
+}
+
+/// getPointerToGlobalIfAvailable - This returns the address of the specified
+/// global value if it is has already been codegen'd, otherwise it returns null.
+///
+void *ExecutionEngine::getPointerToGlobalIfAvailable(const GlobalValue *GV) {
+  MutexGuard locked(lock);
+  
+  std::map<const GlobalValue*, void*>::iterator I =
+  state.getGlobalAddressMap(locked).find(GV);
+  return I != state.getGlobalAddressMap(locked).end() ? I->second : 0;
+}
+
+/// getGlobalValueAtAddress - Return the LLVM global value object that starts
+/// at the specified address.
+///
+const GlobalValue *ExecutionEngine::getGlobalValueAtAddress(void *Addr) {
+  MutexGuard locked(lock);
+
+  // If we haven't computed the reverse mapping yet, do so first.
+  if (state.getGlobalAddressReverseMap(locked).empty()) {
+    for (std::map<const GlobalValue*, void *>::iterator
+         I = state.getGlobalAddressMap(locked).begin(),
+         E = state.getGlobalAddressMap(locked).end(); I != E; ++I)
+      state.getGlobalAddressReverseMap(locked).insert(std::make_pair(I->second,
+                                                                     I->first));
+  }
+
+  std::map<void *, const GlobalValue*>::iterator I =
+    state.getGlobalAddressReverseMap(locked).find(Addr);
+  return I != state.getGlobalAddressReverseMap(locked).end() ? I->second : 0;
+}
+
+// CreateArgv - Turn a vector of strings into a nice argv style array of
+// pointers to null terminated strings.
+//
+static void *CreateArgv(ExecutionEngine *EE,
+                        const std::vector<std::string> &InputArgv) {
+  unsigned PtrSize = EE->getTargetData()->getPointerSize();
+  char *Result = new char[(InputArgv.size()+1)*PtrSize];
+
+  DOUT << "ARGV = " << (void*)Result << "\n";
+  const Type *SBytePtr = PointerType::get(Type::Int8Ty);
+
+  for (unsigned i = 0; i != InputArgv.size(); ++i) {
+    unsigned Size = InputArgv[i].size()+1;
+    char *Dest = new char[Size];
+    DOUT << "ARGV[" << i << "] = " << (void*)Dest << "\n";
+
+    std::copy(InputArgv[i].begin(), InputArgv[i].end(), Dest);
+    Dest[Size-1] = 0;
+
+    // Endian safe: Result[i] = (PointerTy)Dest;
+    EE->StoreValueToMemory(PTOGV(Dest), (GenericValue*)(Result+i*PtrSize),
+                           SBytePtr);
+  }
+
+  // Null terminate it
+  EE->StoreValueToMemory(PTOGV(0),
+                         (GenericValue*)(Result+InputArgv.size()*PtrSize),
+                         SBytePtr);
+  return Result;
+}
+
+
+/// runStaticConstructorsDestructors - This method is used to execute all of
+/// the static constructors or destructors for a program, depending on the
+/// value of isDtors.
+void ExecutionEngine::runStaticConstructorsDestructors(bool isDtors) {
+  const char *Name = isDtors ? "llvm.global_dtors" : "llvm.global_ctors";
+  
+  // Execute global ctors/dtors for each module in the program.
+  for (unsigned m = 0, e = Modules.size(); m != e; ++m) {
+    GlobalVariable *GV = Modules[m]->getModule()->getNamedGlobal(Name);
+
+    // If this global has internal linkage, or if it has a use, then it must be
+    // an old-style (llvmgcc3) static ctor with __main linked in and in use.  If
+    // this is the case, don't execute any of the global ctors, __main will do
+    // it.
+    if (!GV || GV->isDeclaration() || GV->hasInternalLinkage()) continue;
+  
+    // Should be an array of '{ int, void ()* }' structs.  The first value is
+    // the init priority, which we ignore.
+    ConstantArray *InitList = dyn_cast<ConstantArray>(GV->getInitializer());
+    if (!InitList) continue;
+    for (unsigned i = 0, e = InitList->getNumOperands(); i != e; ++i)
+      if (ConstantStruct *CS = 
+          dyn_cast<ConstantStruct>(InitList->getOperand(i))) {
+        if (CS->getNumOperands() != 2) break; // Not array of 2-element structs.
+      
+        Constant *FP = CS->getOperand(1);
+        if (FP->isNullValue())
+          break;  // Found a null terminator, exit.
+      
+        if (ConstantExpr *CE = dyn_cast<ConstantExpr>(FP))
+          if (CE->isCast())
+            FP = CE->getOperand(0);
+        if (Function *F = dyn_cast<Function>(FP)) {
+          // Execute the ctor/dtor function!
+          runFunction(F, std::vector<GenericValue>());
+        }
+      }
+  }
+}
+
+/// runFunctionAsMain - This is a helper function which wraps runFunction to
+/// handle the common task of starting up main with the specified argc, argv,
+/// and envp parameters.
+int ExecutionEngine::runFunctionAsMain(Function *Fn,
+                                       const std::vector<std::string> &argv,
+                                       const char * const * envp) {
+  std::vector<GenericValue> GVArgs;
+  GenericValue GVArgc;
+  GVArgc.IntVal = APInt(32, argv.size());
+
+  // Check main() type
+  unsigned NumArgs = Fn->getFunctionType()->getNumParams();
+  const FunctionType *FTy = Fn->getFunctionType();
+  const Type* PPInt8Ty = PointerType::get(PointerType::get(Type::Int8Ty));
+  switch (NumArgs) {
+  case 3:
+   if (FTy->getParamType(2) != PPInt8Ty) {
+     cerr << "Invalid type for third argument of main() supplied\n";
+     abort();
+   }
+   // FALLS THROUGH
+  case 2:
+   if (FTy->getParamType(1) != PPInt8Ty) {
+     cerr << "Invalid type for second argument of main() supplied\n";
+     abort();
+   }
+   // FALLS THROUGH
+  case 1:
+   if (FTy->getParamType(0) != Type::Int32Ty) {
+     cerr << "Invalid type for first argument of main() supplied\n";
+     abort();
+   }
+   // FALLS THROUGH
+  case 0:
+   if (FTy->getReturnType() != Type::Int32Ty &&
+       FTy->getReturnType() != Type::VoidTy) {
+     cerr << "Invalid return type of main() supplied\n";
+     abort();
+   }
+   break;
+  default:
+   cerr << "Invalid number of arguments of main() supplied\n";
+   abort();
+  }
+  
+  if (NumArgs) {
+    GVArgs.push_back(GVArgc); // Arg #0 = argc.
+    if (NumArgs > 1) {
+      GVArgs.push_back(PTOGV(CreateArgv(this, argv))); // Arg #1 = argv.
+      assert(((char **)GVTOP(GVArgs[1]))[0] &&
+             "argv[0] was null after CreateArgv");
+      if (NumArgs > 2) {
+        std::vector<std::string> EnvVars;
+        for (unsigned i = 0; envp[i]; ++i)
+          EnvVars.push_back(envp[i]);
+        GVArgs.push_back(PTOGV(CreateArgv(this, EnvVars))); // Arg #2 = envp.
+      }
+    }
+  }
+  return runFunction(Fn, GVArgs).IntVal.getZExtValue();
+}
+
+/// If possible, create a JIT, unless the caller specifically requests an
+/// Interpreter or there's an error. If even an Interpreter cannot be created,
+/// NULL is returned.
+///
+ExecutionEngine *ExecutionEngine::create(ModuleProvider *MP,
+                                         bool ForceInterpreter,
+                                         std::string *ErrorStr) {
+  ExecutionEngine *EE = 0;
+
+  // Unless the interpreter was explicitly selected, try making a JIT.
+  if (!ForceInterpreter && JITCtor)
+    EE = JITCtor(MP, ErrorStr);
+
+  // If we can't make a JIT, make an interpreter instead.
+  if (EE == 0 && InterpCtor)
+    EE = InterpCtor(MP, ErrorStr);
+
+  if (EE) {
+    // Make sure we can resolve symbols in the program as well. The zero arg
+    // to the function tells DynamicLibrary to load the program, not a library.
+    try {
+      sys::DynamicLibrary::LoadLibraryPermanently(0);
+    } catch (...) {
+    }
+  }
+
+  return EE;
+}
+
+/// getPointerToGlobal - This returns the address of the specified global
+/// value.  This may involve code generation if it's a function.
+///
+void *ExecutionEngine::getPointerToGlobal(const GlobalValue *GV) {
+  if (Function *F = const_cast<Function*>(dyn_cast<Function>(GV)))
+    return getPointerToFunction(F);
+
+  MutexGuard locked(lock);
+  void *p = state.getGlobalAddressMap(locked)[GV];
+  if (p)
+    return p;
+
+  // Global variable might have been added since interpreter started.
+  if (GlobalVariable *GVar =
+          const_cast<GlobalVariable *>(dyn_cast<GlobalVariable>(GV)))
+    EmitGlobalVariable(GVar);
+  else
+    assert(0 && "Global hasn't had an address allocated yet!");
+  return state.getGlobalAddressMap(locked)[GV];
+}
+
+/// This function converts a Constant* into a GenericValue. The interesting 
+/// part is if C is a ConstantExpr.
+/// @brief Get a GenericValue for a Constnat*
+GenericValue ExecutionEngine::getConstantValue(const Constant *C) {
+  // If its undefined, return the garbage.
+  if (isa<UndefValue>(C)) 
+    return GenericValue();
+
+  // If the value is a ConstantExpr
+  if (const ConstantExpr *CE = dyn_cast<ConstantExpr>(C)) {
+    Constant *Op0 = CE->getOperand(0);
+    switch (CE->getOpcode()) {
+    case Instruction::GetElementPtr: {
+      // Compute the index 
+      GenericValue Result = getConstantValue(Op0);
+      SmallVector<Value*, 8> Indices(CE->op_begin()+1, CE->op_end());
+      uint64_t Offset =
+        TD->getIndexedOffset(Op0->getType(), &Indices[0], Indices.size());
+
+      char* tmp = (char*) Result.PointerVal;
+      Result = PTOGV(tmp + Offset);
+      return Result;
+    }
+    case Instruction::Trunc: {
+      GenericValue GV = getConstantValue(Op0);
+      uint32_t BitWidth = cast<IntegerType>(CE->getType())->getBitWidth();
+      GV.IntVal = GV.IntVal.trunc(BitWidth);
+      return GV;
+    }
+    case Instruction::ZExt: {
+      GenericValue GV = getConstantValue(Op0);
+      uint32_t BitWidth = cast<IntegerType>(CE->getType())->getBitWidth();
+      GV.IntVal = GV.IntVal.zext(BitWidth);
+      return GV;
+    }
+    case Instruction::SExt: {
+      GenericValue GV = getConstantValue(Op0);
+      uint32_t BitWidth = cast<IntegerType>(CE->getType())->getBitWidth();
+      GV.IntVal = GV.IntVal.sext(BitWidth);
+      return GV;
+    }
+    case Instruction::FPTrunc: {
+      GenericValue GV = getConstantValue(Op0);
+      GV.FloatVal = float(GV.DoubleVal);
+      return GV;
+    }
+    case Instruction::FPExt:{
+      GenericValue GV = getConstantValue(Op0);
+      GV.DoubleVal = double(GV.FloatVal);
+      return GV;
+    }
+    case Instruction::UIToFP: {
+      GenericValue GV = getConstantValue(Op0);
+      if (CE->getType() == Type::FloatTy)
+        GV.FloatVal = float(GV.IntVal.roundToDouble());
+      else
+        GV.DoubleVal = GV.IntVal.roundToDouble();
+      return GV;
+    }
+    case Instruction::SIToFP: {
+      GenericValue GV = getConstantValue(Op0);
+      if (CE->getType() == Type::FloatTy)
+        GV.FloatVal = float(GV.IntVal.signedRoundToDouble());
+      else
+        GV.DoubleVal = GV.IntVal.signedRoundToDouble();
+      return GV;
+    }
+    case Instruction::FPToUI: // double->APInt conversion handles sign
+    case Instruction::FPToSI: {
+      GenericValue GV = getConstantValue(Op0);
+      uint32_t BitWidth = cast<IntegerType>(CE->getType())->getBitWidth();
+      if (Op0->getType() == Type::FloatTy)
+        GV.IntVal = APIntOps::RoundFloatToAPInt(GV.FloatVal, BitWidth);
+      else
+        GV.IntVal = APIntOps::RoundDoubleToAPInt(GV.DoubleVal, BitWidth);
+      return GV;
+    }
+    case Instruction::PtrToInt: {
+      GenericValue GV = getConstantValue(Op0);
+      uint32_t PtrWidth = TD->getPointerSizeInBits();
+      GV.IntVal = APInt(PtrWidth, uintptr_t(GV.PointerVal));
+      return GV;
+    }
+    case Instruction::IntToPtr: {
+      GenericValue GV = getConstantValue(Op0);
+      uint32_t PtrWidth = TD->getPointerSizeInBits();
+      if (PtrWidth != GV.IntVal.getBitWidth())
+        GV.IntVal = GV.IntVal.zextOrTrunc(PtrWidth);
+      assert(GV.IntVal.getBitWidth() <= 64 && "Bad pointer width");
+      GV.PointerVal = PointerTy(uintptr_t(GV.IntVal.getZExtValue()));
+      return GV;
+    }
+    case Instruction::BitCast: {
+      GenericValue GV = getConstantValue(Op0);
+      const Type* DestTy = CE->getType();
+      switch (Op0->getType()->getTypeID()) {
+        default: assert(0 && "Invalid bitcast operand");
+        case Type::IntegerTyID:
+          assert(DestTy->isFloatingPoint() && "invalid bitcast");
+          if (DestTy == Type::FloatTy)
+            GV.FloatVal = GV.IntVal.bitsToFloat();
+          else if (DestTy == Type::DoubleTy)
+            GV.DoubleVal = GV.IntVal.bitsToDouble();
+          break;
+        case Type::FloatTyID: 
+          assert(DestTy == Type::Int32Ty && "Invalid bitcast");
+          GV.IntVal.floatToBits(GV.FloatVal);
+          break;
+        case Type::DoubleTyID:
+          assert(DestTy == Type::Int64Ty && "Invalid bitcast");
+          GV.IntVal.doubleToBits(GV.DoubleVal);
+          break;
+        case Type::PointerTyID:
+          assert(isa<PointerType>(DestTy) && "Invalid bitcast");
+          break; // getConstantValue(Op0)  above already converted it
+      }
+      return GV;
+    }
+    case Instruction::Add:
+    case Instruction::Sub:
+    case Instruction::Mul:
+    case Instruction::UDiv:
+    case Instruction::SDiv:
+    case Instruction::URem:
+    case Instruction::SRem:
+    case Instruction::And:
+    case Instruction::Or:
+    case Instruction::Xor: {
+      GenericValue LHS = getConstantValue(Op0);
+      GenericValue RHS = getConstantValue(CE->getOperand(1));
+      GenericValue GV;
+      switch (CE->getOperand(0)->getType()->getTypeID()) {
+      default: assert(0 && "Bad add type!"); abort();
+      case Type::IntegerTyID:
+        switch (CE->getOpcode()) {
+          default: assert(0 && "Invalid integer opcode");
+          case Instruction::Add: GV.IntVal = LHS.IntVal + RHS.IntVal; break;
+          case Instruction::Sub: GV.IntVal = LHS.IntVal - RHS.IntVal; break;
+          case Instruction::Mul: GV.IntVal = LHS.IntVal * RHS.IntVal; break;
+          case Instruction::UDiv:GV.IntVal = LHS.IntVal.udiv(RHS.IntVal); break;
+          case Instruction::SDiv:GV.IntVal = LHS.IntVal.sdiv(RHS.IntVal); break;
+          case Instruction::URem:GV.IntVal = LHS.IntVal.urem(RHS.IntVal); break;
+          case Instruction::SRem:GV.IntVal = LHS.IntVal.srem(RHS.IntVal); break;
+          case Instruction::And: GV.IntVal = LHS.IntVal & RHS.IntVal; break;
+          case Instruction::Or:  GV.IntVal = LHS.IntVal | RHS.IntVal; break;
+          case Instruction::Xor: GV.IntVal = LHS.IntVal ^ RHS.IntVal; break;
+        }
+        break;
+      case Type::FloatTyID:
+        switch (CE->getOpcode()) {
+          default: assert(0 && "Invalid float opcode"); abort();
+          case Instruction::Add:  
+            GV.FloatVal = LHS.FloatVal + RHS.FloatVal; break;
+          case Instruction::Sub:  
+            GV.FloatVal = LHS.FloatVal - RHS.FloatVal; break;
+          case Instruction::Mul:  
+            GV.FloatVal = LHS.FloatVal * RHS.FloatVal; break;
+          case Instruction::FDiv: 
+            GV.FloatVal = LHS.FloatVal / RHS.FloatVal; break;
+          case Instruction::FRem: 
+            GV.FloatVal = ::fmodf(LHS.FloatVal,RHS.FloatVal); break;
+        }
+        break;
+      case Type::DoubleTyID:
+        switch (CE->getOpcode()) {
+          default: assert(0 && "Invalid double opcode"); abort();
+          case Instruction::Add:  
+            GV.DoubleVal = LHS.DoubleVal + RHS.DoubleVal; break;
+          case Instruction::Sub:  
+            GV.DoubleVal = LHS.DoubleVal - RHS.DoubleVal; break;
+          case Instruction::Mul:  
+            GV.DoubleVal = LHS.DoubleVal * RHS.DoubleVal; break;
+          case Instruction::FDiv: 
+            GV.DoubleVal = LHS.DoubleVal / RHS.DoubleVal; break;
+          case Instruction::FRem: 
+            GV.DoubleVal = ::fmod(LHS.DoubleVal,RHS.DoubleVal); break;
+        }
+        break;
+      }
+      return GV;
+    }
+    default:
+      break;
+    }
+    cerr << "ConstantExpr not handled: " << *CE << "\n";
+    abort();
+  }
+
+  GenericValue Result;
+  switch (C->getType()->getTypeID()) {
+  case Type::FloatTyID: 
+    Result.FloatVal = (float)cast<ConstantFP>(C)->getValue(); 
+    break;
+  case Type::DoubleTyID:
+    Result.DoubleVal = (double)cast<ConstantFP>(C)->getValue(); 
+    break;
+  case Type::IntegerTyID:
+    Result.IntVal = cast<ConstantInt>(C)->getValue();
+    break;
+  case Type::PointerTyID:
+    if (isa<ConstantPointerNull>(C))
+      Result.PointerVal = 0;
+    else if (const Function *F = dyn_cast<Function>(C))
+      Result = PTOGV(getPointerToFunctionOrStub(const_cast<Function*>(F)));
+    else if (const GlobalVariable* GV = dyn_cast<GlobalVariable>(C))
+      Result = PTOGV(getOrEmitGlobalVariable(const_cast<GlobalVariable*>(GV)));
+    else
+      assert(0 && "Unknown constant pointer type!");
+    break;
+  default:
+    cerr << "ERROR: Constant unimplemented for type: " << *C->getType() << "\n";
+    abort();
+  }
+  return Result;
+}
+
+/// StoreValueToMemory - Stores the data in Val of type Ty at address Ptr.  Ptr
+/// is the address of the memory at which to store Val, cast to GenericValue *.
+/// It is not a pointer to a GenericValue containing the address at which to
+/// store Val.
+///
+void ExecutionEngine::StoreValueToMemory(const GenericValue &Val, GenericValue *Ptr,
+                                         const Type *Ty) {
+  switch (Ty->getTypeID()) {
+  case Type::IntegerTyID: {
+    unsigned BitWidth = cast<IntegerType>(Ty)->getBitWidth();
+    GenericValue TmpVal = Val;
+    if (BitWidth <= 8)
+      *((uint8_t*)Ptr) = uint8_t(Val.IntVal.getZExtValue());
+    else if (BitWidth <= 16) {
+      *((uint16_t*)Ptr) = uint16_t(Val.IntVal.getZExtValue());
+    } else if (BitWidth <= 32) {
+      *((uint32_t*)Ptr) = uint32_t(Val.IntVal.getZExtValue());
+    } else if (BitWidth <= 64) {
+      *((uint64_t*)Ptr) = uint64_t(Val.IntVal.getZExtValue());
+    } else {
+      uint64_t *Dest = (uint64_t*)Ptr;
+      const uint64_t *Src = Val.IntVal.getRawData();
+      for (uint32_t i = 0; i < Val.IntVal.getNumWords(); ++i)
+        Dest[i] = Src[i];
+    }
+    break;
+  }
+  case Type::FloatTyID:
+    *((float*)Ptr) = Val.FloatVal;
+    break;
+  case Type::DoubleTyID:
+    *((double*)Ptr) = Val.DoubleVal;
+    break;
+  case Type::PointerTyID: 
+    *((PointerTy*)Ptr) = Val.PointerVal;
+    break;
+  default:
+    cerr << "Cannot store value of type " << *Ty << "!\n";
+  }
+}
+
+/// FIXME: document
+///
+void ExecutionEngine::LoadValueFromMemory(GenericValue &Result, 
+                                                  GenericValue *Ptr,
+                                                  const Type *Ty) {
+  switch (Ty->getTypeID()) {
+  case Type::IntegerTyID: {
+    unsigned BitWidth = cast<IntegerType>(Ty)->getBitWidth();
+    if (BitWidth <= 8)
+      Result.IntVal = APInt(BitWidth, *((uint8_t*)Ptr));
+    else if (BitWidth <= 16) {
+      Result.IntVal = APInt(BitWidth, *((uint16_t*)Ptr));
+    } else if (BitWidth <= 32) {
+      Result.IntVal = APInt(BitWidth, *((uint32_t*)Ptr));
+    } else if (BitWidth <= 64) {
+      Result.IntVal = APInt(BitWidth, *((uint64_t*)Ptr));
+    } else
+      Result.IntVal = APInt(BitWidth, (BitWidth+63)/64, (uint64_t*)Ptr);
+    break;
+  }
+  case Type::FloatTyID:
+    Result.FloatVal = *((float*)Ptr);
+    break;
+  case Type::DoubleTyID:
+    Result.DoubleVal = *((double*)Ptr); 
+    break;
+  case Type::PointerTyID: 
+    Result.PointerVal = *((PointerTy*)Ptr);
+    break;
+  default:
+    cerr << "Cannot load value of type " << *Ty << "!\n";
+    abort();
+  }
+}
+
+// InitializeMemory - Recursive function to apply a Constant value into the
+// specified memory location...
+//
+void ExecutionEngine::InitializeMemory(const Constant *Init, void *Addr) {
+  if (isa<UndefValue>(Init)) {
+    return;
+  } else if (const ConstantVector *CP = dyn_cast<ConstantVector>(Init)) {
+    unsigned ElementSize =
+      getTargetData()->getTypeSize(CP->getType()->getElementType());
+    for (unsigned i = 0, e = CP->getNumOperands(); i != e; ++i)
+      InitializeMemory(CP->getOperand(i), (char*)Addr+i*ElementSize);
+    return;
+  } else if (Init->getType()->isFirstClassType()) {
+    GenericValue Val = getConstantValue(Init);
+    StoreValueToMemory(Val, (GenericValue*)Addr, Init->getType());
+    return;
+  } else if (isa<ConstantAggregateZero>(Init)) {
+    memset(Addr, 0, (size_t)getTargetData()->getTypeSize(Init->getType()));
+    return;
+  }
+
+  switch (Init->getType()->getTypeID()) {
+  case Type::ArrayTyID: {
+    const ConstantArray *CPA = cast<ConstantArray>(Init);
+    unsigned ElementSize =
+      getTargetData()->getTypeSize(CPA->getType()->getElementType());
+    for (unsigned i = 0, e = CPA->getNumOperands(); i != e; ++i)
+      InitializeMemory(CPA->getOperand(i), (char*)Addr+i*ElementSize);
+    return;
+  }
+
+  case Type::StructTyID: {
+    const ConstantStruct *CPS = cast<ConstantStruct>(Init);
+    const StructLayout *SL =
+      getTargetData()->getStructLayout(cast<StructType>(CPS->getType()));
+    for (unsigned i = 0, e = CPS->getNumOperands(); i != e; ++i)
+      InitializeMemory(CPS->getOperand(i), (char*)Addr+SL->getElementOffset(i));
+    return;
+  }
+
+  default:
+    cerr << "Bad Type: " << *Init->getType() << "\n";
+    assert(0 && "Unknown constant type to initialize memory with!");
+  }
+}
+
+/// EmitGlobals - Emit all of the global variables to memory, storing their
+/// addresses into GlobalAddress.  This must make sure to copy the contents of
+/// their initializers into the memory.
+///
+void ExecutionEngine::emitGlobals() {
+  const TargetData *TD = getTargetData();
+
+  // Loop over all of the global variables in the program, allocating the memory
+  // to hold them.  If there is more than one module, do a prepass over globals
+  // to figure out how the different modules should link together.
+  //
+  std::map<std::pair<std::string, const Type*>,
+           const GlobalValue*> LinkedGlobalsMap;
+
+  if (Modules.size() != 1) {
+    for (unsigned m = 0, e = Modules.size(); m != e; ++m) {
+      Module &M = *Modules[m]->getModule();
+      for (Module::const_global_iterator I = M.global_begin(),
+           E = M.global_end(); I != E; ++I) {
+        const GlobalValue *GV = I;
+        if (GV->hasInternalLinkage() || GV->isDeclaration() ||
+            GV->hasAppendingLinkage() || !GV->hasName())
+          continue;// Ignore external globals and globals with internal linkage.
+          
+        const GlobalValue *&GVEntry = 
+          LinkedGlobalsMap[std::make_pair(GV->getName(), GV->getType())];
+
+        // If this is the first time we've seen this global, it is the canonical
+        // version.
+        if (!GVEntry) {
+          GVEntry = GV;
+          continue;
+        }
+        
+        // If the existing global is strong, never replace it.
+        if (GVEntry->hasExternalLinkage() ||
+            GVEntry->hasDLLImportLinkage() ||
+            GVEntry->hasDLLExportLinkage())
+          continue;
+        
+        // Otherwise, we know it's linkonce/weak, replace it if this is a strong
+        // symbol.
+        if (GV->hasExternalLinkage() || GVEntry->hasExternalWeakLinkage())
+          GVEntry = GV;
+      }
+    }
+  }
+  
+  std::vector<const GlobalValue*> NonCanonicalGlobals;
+  for (unsigned m = 0, e = Modules.size(); m != e; ++m) {
+    Module &M = *Modules[m]->getModule();
+    for (Module::const_global_iterator I = M.global_begin(), E = M.global_end();
+         I != E; ++I) {
+      // In the multi-module case, see what this global maps to.
+      if (!LinkedGlobalsMap.empty()) {
+        if (const GlobalValue *GVEntry = 
+              LinkedGlobalsMap[std::make_pair(I->getName(), I->getType())]) {
+          // If something else is the canonical global, ignore this one.
+          if (GVEntry != &*I) {
+            NonCanonicalGlobals.push_back(I);
+            continue;
+          }
+        }
+      }
+      
+      if (!I->isDeclaration()) {
+        // Get the type of the global.
+        const Type *Ty = I->getType()->getElementType();
+
+        // Allocate some memory for it!
+        unsigned Size = TD->getTypeSize(Ty);
+        addGlobalMapping(I, new char[Size]);
+      } else {
+        // External variable reference. Try to use the dynamic loader to
+        // get a pointer to it.
+        if (void *SymAddr =
+            sys::DynamicLibrary::SearchForAddressOfSymbol(I->getName().c_str()))
+          addGlobalMapping(I, SymAddr);
+        else {
+          cerr << "Could not resolve external global address: "
+               << I->getName() << "\n";
+          abort();
+        }
+      }
+    }
+    
+    // If there are multiple modules, map the non-canonical globals to their
+    // canonical location.
+    if (!NonCanonicalGlobals.empty()) {
+      for (unsigned i = 0, e = NonCanonicalGlobals.size(); i != e; ++i) {
+        const GlobalValue *GV = NonCanonicalGlobals[i];
+        const GlobalValue *CGV =
+          LinkedGlobalsMap[std::make_pair(GV->getName(), GV->getType())];
+        void *Ptr = getPointerToGlobalIfAvailable(CGV);
+        assert(Ptr && "Canonical global wasn't codegen'd!");
+        addGlobalMapping(GV, getPointerToGlobalIfAvailable(CGV));
+      }
+    }
+    
+    // Now that all of the globals are set up in memory, loop through them all 
+    // and initialize their contents.
+    for (Module::const_global_iterator I = M.global_begin(), E = M.global_end();
+         I != E; ++I) {
+      if (!I->isDeclaration()) {
+        if (!LinkedGlobalsMap.empty()) {
+          if (const GlobalValue *GVEntry = 
+                LinkedGlobalsMap[std::make_pair(I->getName(), I->getType())])
+            if (GVEntry != &*I)  // Not the canonical variable.
+              continue;
+        }
+        EmitGlobalVariable(I);
+      }
+    }
+  }
+}
+
+// EmitGlobalVariable - This method emits the specified global variable to the
+// address specified in GlobalAddresses, or allocates new memory if it's not
+// already in the map.
+void ExecutionEngine::EmitGlobalVariable(const GlobalVariable *GV) {
+  void *GA = getPointerToGlobalIfAvailable(GV);
+  DOUT << "Global '" << GV->getName() << "' -> " << GA << "\n";
+
+  const Type *ElTy = GV->getType()->getElementType();
+  size_t GVSize = (size_t)getTargetData()->getTypeSize(ElTy);
+  if (GA == 0) {
+    // If it's not already specified, allocate memory for the global.
+    GA = new char[GVSize];
+    addGlobalMapping(GV, GA);
+  }
+
+  InitializeMemory(GV->getInitializer(), GA);
+  NumInitBytes += (unsigned)GVSize;
+  ++NumGlobals;
+}
diff --git a/lib/ExecutionEngine/Interpreter/Execution.cpp b/lib/ExecutionEngine/Interpreter/Execution.cpp
new file mode 100644
index 0000000..281f774
--- /dev/null
+++ b/lib/ExecutionEngine/Interpreter/Execution.cpp
@@ -0,0 +1,1374 @@
+//===-- Execution.cpp - Implement code to simulate the program ------------===//
+//
+//                     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 contains the actual instruction interpreter.
+//
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "interpreter"
+#include "Interpreter.h"
+#include "llvm/Constants.h"
+#include "llvm/DerivedTypes.h"
+#include "llvm/Instructions.h"
+#include "llvm/ParameterAttributes.h"
+#include "llvm/CodeGen/IntrinsicLowering.h"
+#include "llvm/Support/GetElementPtrTypeIterator.h"
+#include "llvm/ADT/APInt.h"
+#include "llvm/ADT/Statistic.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/MathExtras.h"
+#include <cmath>
+using namespace llvm;
+
+STATISTIC(NumDynamicInsts, "Number of dynamic instructions executed");
+static Interpreter *TheEE = 0;
+
+//===----------------------------------------------------------------------===//
+//                     Various Helper Functions
+//===----------------------------------------------------------------------===//
+
+static inline uint64_t doSignExtension(uint64_t Val, const IntegerType* ITy) {
+  // Determine if the value is signed or not
+  bool isSigned = (Val & (1 << (ITy->getBitWidth()-1))) != 0;
+  // If its signed, extend the sign bits
+  if (isSigned)
+    Val |= ~ITy->getBitMask();
+  return Val;
+}
+
+static void SetValue(Value *V, GenericValue Val, ExecutionContext &SF) {
+  SF.Values[V] = Val;
+}
+
+void Interpreter::initializeExecutionEngine() {
+  TheEE = this;
+}
+
+//===----------------------------------------------------------------------===//
+//                    Binary Instruction Implementations
+//===----------------------------------------------------------------------===//
+
+#define IMPLEMENT_BINARY_OPERATOR(OP, TY) \
+   case Type::TY##TyID: \
+     Dest.TY##Val = Src1.TY##Val OP Src2.TY##Val; \
+     break
+
+#define IMPLEMENT_INTEGER_BINOP1(OP, TY) \
+   case Type::IntegerTyID: { \
+     Dest.IntVal = Src1.IntVal OP Src2.IntVal; \
+     break; \
+   }
+
+
+static void executeAddInst(GenericValue &Dest, GenericValue Src1, 
+                           GenericValue Src2, const Type *Ty) {
+  switch (Ty->getTypeID()) {
+    IMPLEMENT_INTEGER_BINOP1(+, Ty);
+    IMPLEMENT_BINARY_OPERATOR(+, Float);
+    IMPLEMENT_BINARY_OPERATOR(+, Double);
+  default:
+    cerr << "Unhandled type for Add instruction: " << *Ty << "\n";
+    abort();
+  }
+}
+
+static void executeSubInst(GenericValue &Dest, GenericValue Src1, 
+                           GenericValue Src2, const Type *Ty) {
+  switch (Ty->getTypeID()) {
+    IMPLEMENT_INTEGER_BINOP1(-, Ty);
+    IMPLEMENT_BINARY_OPERATOR(-, Float);
+    IMPLEMENT_BINARY_OPERATOR(-, Double);
+  default:
+    cerr << "Unhandled type for Sub instruction: " << *Ty << "\n";
+    abort();
+  }
+}
+
+static void executeMulInst(GenericValue &Dest, GenericValue Src1, 
+                           GenericValue Src2, const Type *Ty) {
+  switch (Ty->getTypeID()) {
+    IMPLEMENT_INTEGER_BINOP1(*, Ty);
+    IMPLEMENT_BINARY_OPERATOR(*, Float);
+    IMPLEMENT_BINARY_OPERATOR(*, Double);
+  default:
+    cerr << "Unhandled type for Mul instruction: " << *Ty << "\n";
+    abort();
+  }
+}
+
+static void executeFDivInst(GenericValue &Dest, GenericValue Src1, 
+                            GenericValue Src2, const Type *Ty) {
+  switch (Ty->getTypeID()) {
+    IMPLEMENT_BINARY_OPERATOR(/, Float);
+    IMPLEMENT_BINARY_OPERATOR(/, Double);
+  default:
+    cerr << "Unhandled type for FDiv instruction: " << *Ty << "\n";
+    abort();
+  }
+}
+
+static void executeFRemInst(GenericValue &Dest, GenericValue Src1, 
+                            GenericValue Src2, const Type *Ty) {
+  switch (Ty->getTypeID()) {
+  case Type::FloatTyID:
+    Dest.FloatVal = fmod(Src1.FloatVal, Src2.FloatVal);
+    break;
+  case Type::DoubleTyID:
+    Dest.DoubleVal = fmod(Src1.DoubleVal, Src2.DoubleVal);
+    break;
+  default:
+    cerr << "Unhandled type for Rem instruction: " << *Ty << "\n";
+    abort();
+  }
+}
+
+#define IMPLEMENT_INTEGER_ICMP(OP, TY) \
+   case Type::IntegerTyID:  \
+      Dest.IntVal = APInt(1,Src1.IntVal.OP(Src2.IntVal)); \
+      break;
+
+// Handle pointers specially because they must be compared with only as much
+// width as the host has.  We _do not_ want to be comparing 64 bit values when
+// running on a 32-bit target, otherwise the upper 32 bits might mess up
+// comparisons if they contain garbage.
+#define IMPLEMENT_POINTER_ICMP(OP) \
+   case Type::PointerTyID: \
+      Dest.IntVal = APInt(1,(void*)(intptr_t)Src1.PointerVal OP \
+                            (void*)(intptr_t)Src2.PointerVal); \
+      break;
+
+static GenericValue executeICMP_EQ(GenericValue Src1, GenericValue Src2,
+                                   const Type *Ty) {
+  GenericValue Dest;
+  switch (Ty->getTypeID()) {
+    IMPLEMENT_INTEGER_ICMP(eq,Ty);
+    IMPLEMENT_POINTER_ICMP(==);
+  default:
+    cerr << "Unhandled type for ICMP_EQ predicate: " << *Ty << "\n";
+    abort();
+  }
+  return Dest;
+}
+
+static GenericValue executeICMP_NE(GenericValue Src1, GenericValue Src2,
+                                   const Type *Ty) {
+  GenericValue Dest;
+  switch (Ty->getTypeID()) {
+    IMPLEMENT_INTEGER_ICMP(ne,Ty);
+    IMPLEMENT_POINTER_ICMP(!=);
+  default:
+    cerr << "Unhandled type for ICMP_NE predicate: " << *Ty << "\n";
+    abort();
+  }
+  return Dest;
+}
+
+static GenericValue executeICMP_ULT(GenericValue Src1, GenericValue Src2,
+                                    const Type *Ty) {
+  GenericValue Dest;
+  switch (Ty->getTypeID()) {
+    IMPLEMENT_INTEGER_ICMP(ult,Ty);
+    IMPLEMENT_POINTER_ICMP(<);
+  default:
+    cerr << "Unhandled type for ICMP_ULT predicate: " << *Ty << "\n";
+    abort();
+  }
+  return Dest;
+}
+
+static GenericValue executeICMP_SLT(GenericValue Src1, GenericValue Src2,
+                                    const Type *Ty) {
+  GenericValue Dest;
+  switch (Ty->getTypeID()) {
+    IMPLEMENT_INTEGER_ICMP(slt,Ty);
+    IMPLEMENT_POINTER_ICMP(<);
+  default:
+    cerr << "Unhandled type for ICMP_SLT predicate: " << *Ty << "\n";
+    abort();
+  }
+  return Dest;
+}
+
+static GenericValue executeICMP_UGT(GenericValue Src1, GenericValue Src2,
+                                    const Type *Ty) {
+  GenericValue Dest;
+  switch (Ty->getTypeID()) {
+    IMPLEMENT_INTEGER_ICMP(ugt,Ty);
+    IMPLEMENT_POINTER_ICMP(>);
+  default:
+    cerr << "Unhandled type for ICMP_UGT predicate: " << *Ty << "\n";
+    abort();
+  }
+  return Dest;
+}
+
+static GenericValue executeICMP_SGT(GenericValue Src1, GenericValue Src2,
+                                    const Type *Ty) {
+  GenericValue Dest;
+  switch (Ty->getTypeID()) {
+    IMPLEMENT_INTEGER_ICMP(sgt,Ty);
+    IMPLEMENT_POINTER_ICMP(>);
+  default:
+    cerr << "Unhandled type for ICMP_SGT predicate: " << *Ty << "\n";
+    abort();
+  }
+  return Dest;
+}
+
+static GenericValue executeICMP_ULE(GenericValue Src1, GenericValue Src2,
+                                    const Type *Ty) {
+  GenericValue Dest;
+  switch (Ty->getTypeID()) {
+    IMPLEMENT_INTEGER_ICMP(ule,Ty);
+    IMPLEMENT_POINTER_ICMP(<=);
+  default:
+    cerr << "Unhandled type for ICMP_ULE predicate: " << *Ty << "\n";
+    abort();
+  }
+  return Dest;
+}
+
+static GenericValue executeICMP_SLE(GenericValue Src1, GenericValue Src2,
+                                    const Type *Ty) {
+  GenericValue Dest;
+  switch (Ty->getTypeID()) {
+    IMPLEMENT_INTEGER_ICMP(sle,Ty);
+    IMPLEMENT_POINTER_ICMP(<=);
+  default:
+    cerr << "Unhandled type for ICMP_SLE predicate: " << *Ty << "\n";
+    abort();
+  }
+  return Dest;
+}
+
+static GenericValue executeICMP_UGE(GenericValue Src1, GenericValue Src2,
+                                    const Type *Ty) {
+  GenericValue Dest;
+  switch (Ty->getTypeID()) {
+    IMPLEMENT_INTEGER_ICMP(uge,Ty);
+    IMPLEMENT_POINTER_ICMP(>=);
+  default:
+    cerr << "Unhandled type for ICMP_UGE predicate: " << *Ty << "\n";
+    abort();
+  }
+  return Dest;
+}
+
+static GenericValue executeICMP_SGE(GenericValue Src1, GenericValue Src2,
+                                    const Type *Ty) {
+  GenericValue Dest;
+  switch (Ty->getTypeID()) {
+    IMPLEMENT_INTEGER_ICMP(sge,Ty);
+    IMPLEMENT_POINTER_ICMP(>=);
+  default:
+    cerr << "Unhandled type for ICMP_SGE predicate: " << *Ty << "\n";
+    abort();
+  }
+  return Dest;
+}
+
+void Interpreter::visitICmpInst(ICmpInst &I) {
+  ExecutionContext &SF = ECStack.back();
+  const Type *Ty    = I.getOperand(0)->getType();
+  GenericValue Src1 = getOperandValue(I.getOperand(0), SF);
+  GenericValue Src2 = getOperandValue(I.getOperand(1), SF);
+  GenericValue R;   // Result
+  
+  switch (I.getPredicate()) {
+  case ICmpInst::ICMP_EQ:  R = executeICMP_EQ(Src1,  Src2, Ty); break;
+  case ICmpInst::ICMP_NE:  R = executeICMP_NE(Src1,  Src2, Ty); break;
+  case ICmpInst::ICMP_ULT: R = executeICMP_ULT(Src1, Src2, Ty); break;
+  case ICmpInst::ICMP_SLT: R = executeICMP_SLT(Src1, Src2, Ty); break;
+  case ICmpInst::ICMP_UGT: R = executeICMP_UGT(Src1, Src2, Ty); break;
+  case ICmpInst::ICMP_SGT: R = executeICMP_SGT(Src1, Src2, Ty); break;
+  case ICmpInst::ICMP_ULE: R = executeICMP_ULE(Src1, Src2, Ty); break;
+  case ICmpInst::ICMP_SLE: R = executeICMP_SLE(Src1, Src2, Ty); break;
+  case ICmpInst::ICMP_UGE: R = executeICMP_UGE(Src1, Src2, Ty); break;
+  case ICmpInst::ICMP_SGE: R = executeICMP_SGE(Src1, Src2, Ty); break;
+  default:
+    cerr << "Don't know how to handle this ICmp predicate!\n-->" << I;
+    abort();
+  }
+ 
+  SetValue(&I, R, SF);
+}
+
+#define IMPLEMENT_FCMP(OP, TY) \
+   case Type::TY##TyID: \
+     Dest.IntVal = APInt(1,Src1.TY##Val OP Src2.TY##Val); \
+     break
+
+static GenericValue executeFCMP_OEQ(GenericValue Src1, GenericValue Src2,
+                                   const Type *Ty) {
+  GenericValue Dest;
+  switch (Ty->getTypeID()) {
+    IMPLEMENT_FCMP(==, Float);
+    IMPLEMENT_FCMP(==, Double);
+  default:
+    cerr << "Unhandled type for FCmp EQ instruction: " << *Ty << "\n";
+    abort();
+  }
+  return Dest;
+}
+
+static GenericValue executeFCMP_ONE(GenericValue Src1, GenericValue Src2,
+                                   const Type *Ty) {
+  GenericValue Dest;
+  switch (Ty->getTypeID()) {
+    IMPLEMENT_FCMP(!=, Float);
+    IMPLEMENT_FCMP(!=, Double);
+
+  default:
+    cerr << "Unhandled type for FCmp NE instruction: " << *Ty << "\n";
+    abort();
+  }
+  return Dest;
+}
+
+static GenericValue executeFCMP_OLE(GenericValue Src1, GenericValue Src2,
+                                   const Type *Ty) {
+  GenericValue Dest;
+  switch (Ty->getTypeID()) {
+    IMPLEMENT_FCMP(<=, Float);
+    IMPLEMENT_FCMP(<=, Double);
+  default:
+    cerr << "Unhandled type for FCmp LE instruction: " << *Ty << "\n";
+    abort();
+  }
+  return Dest;
+}
+
+static GenericValue executeFCMP_OGE(GenericValue Src1, GenericValue Src2,
+                                   const Type *Ty) {
+  GenericValue Dest;
+  switch (Ty->getTypeID()) {
+    IMPLEMENT_FCMP(>=, Float);
+    IMPLEMENT_FCMP(>=, Double);
+  default:
+    cerr << "Unhandled type for FCmp GE instruction: " << *Ty << "\n";
+    abort();
+  }
+  return Dest;
+}
+
+static GenericValue executeFCMP_OLT(GenericValue Src1, GenericValue Src2,
+                                   const Type *Ty) {
+  GenericValue Dest;
+  switch (Ty->getTypeID()) {
+    IMPLEMENT_FCMP(<, Float);
+    IMPLEMENT_FCMP(<, Double);
+  default:
+    cerr << "Unhandled type for FCmp LT instruction: " << *Ty << "\n";
+    abort();
+  }
+  return Dest;
+}
+
+static GenericValue executeFCMP_OGT(GenericValue Src1, GenericValue Src2,
+                                     const Type *Ty) {
+  GenericValue Dest;
+  switch (Ty->getTypeID()) {
+    IMPLEMENT_FCMP(>, Float);
+    IMPLEMENT_FCMP(>, Double);
+  default:
+    cerr << "Unhandled type for FCmp GT instruction: " << *Ty << "\n";
+    abort();
+  }
+  return Dest;
+}
+
+#define IMPLEMENT_UNORDERED(TY, X,Y) \
+   if (TY == Type::FloatTy) \
+     if (X.FloatVal != X.FloatVal || Y.FloatVal != Y.FloatVal) { \
+       Dest.IntVal = APInt(1,true); \
+       return Dest; \
+     } \
+   else if (X.DoubleVal != X.DoubleVal || Y.DoubleVal != Y.DoubleVal) { \
+     Dest.IntVal = APInt(1,true); \
+     return Dest; \
+   }
+
+
+static GenericValue executeFCMP_UEQ(GenericValue Src1, GenericValue Src2,
+                                   const Type *Ty) {
+  GenericValue Dest;
+  IMPLEMENT_UNORDERED(Ty, Src1, Src2)
+  return executeFCMP_OEQ(Src1, Src2, Ty);
+}
+
+static GenericValue executeFCMP_UNE(GenericValue Src1, GenericValue Src2,
+                                   const Type *Ty) {
+  GenericValue Dest;
+  IMPLEMENT_UNORDERED(Ty, Src1, Src2)
+  return executeFCMP_ONE(Src1, Src2, Ty);
+}
+
+static GenericValue executeFCMP_ULE(GenericValue Src1, GenericValue Src2,
+                                   const Type *Ty) {
+  GenericValue Dest;
+  IMPLEMENT_UNORDERED(Ty, Src1, Src2)
+  return executeFCMP_OLE(Src1, Src2, Ty);
+}
+
+static GenericValue executeFCMP_UGE(GenericValue Src1, GenericValue Src2,
+                                   const Type *Ty) {
+  GenericValue Dest;
+  IMPLEMENT_UNORDERED(Ty, Src1, Src2)
+  return executeFCMP_OGE(Src1, Src2, Ty);
+}
+
+static GenericValue executeFCMP_ULT(GenericValue Src1, GenericValue Src2,
+                                   const Type *Ty) {
+  GenericValue Dest;
+  IMPLEMENT_UNORDERED(Ty, Src1, Src2)
+  return executeFCMP_OLT(Src1, Src2, Ty);
+}
+
+static GenericValue executeFCMP_UGT(GenericValue Src1, GenericValue Src2,
+                                     const Type *Ty) {
+  GenericValue Dest;
+  IMPLEMENT_UNORDERED(Ty, Src1, Src2)
+  return executeFCMP_OGT(Src1, Src2, Ty);
+}
+
+static GenericValue executeFCMP_ORD(GenericValue Src1, GenericValue Src2,
+                                     const Type *Ty) {
+  GenericValue Dest;
+  if (Ty == Type::FloatTy)
+    Dest.IntVal = APInt(1,(Src1.FloatVal == Src1.FloatVal && 
+                           Src2.FloatVal == Src2.FloatVal));
+  else
+    Dest.IntVal = APInt(1,(Src1.DoubleVal == Src1.DoubleVal && 
+                           Src2.DoubleVal == Src2.DoubleVal));
+  return Dest;
+}
+
+static GenericValue executeFCMP_UNO(GenericValue Src1, GenericValue Src2,
+                                     const Type *Ty) {
+  GenericValue Dest;
+  if (Ty == Type::FloatTy)
+    Dest.IntVal = APInt(1,(Src1.FloatVal != Src1.FloatVal || 
+                           Src2.FloatVal != Src2.FloatVal));
+  else
+    Dest.IntVal = APInt(1,(Src1.DoubleVal != Src1.DoubleVal || 
+                           Src2.DoubleVal != Src2.DoubleVal));
+  return Dest;
+}
+
+void Interpreter::visitFCmpInst(FCmpInst &I) {
+  ExecutionContext &SF = ECStack.back();
+  const Type *Ty    = I.getOperand(0)->getType();
+  GenericValue Src1 = getOperandValue(I.getOperand(0), SF);
+  GenericValue Src2 = getOperandValue(I.getOperand(1), SF);
+  GenericValue R;   // Result
+  
+  switch (I.getPredicate()) {
+  case FCmpInst::FCMP_FALSE: R.IntVal = APInt(1,false); break;
+  case FCmpInst::FCMP_TRUE:  R.IntVal = APInt(1,true); break;
+  case FCmpInst::FCMP_ORD:   R = executeFCMP_ORD(Src1, Src2, Ty); break;
+  case FCmpInst::FCMP_UNO:   R = executeFCMP_UNO(Src1, Src2, Ty); break;
+  case FCmpInst::FCMP_UEQ:   R = executeFCMP_UEQ(Src1, Src2, Ty); break;
+  case FCmpInst::FCMP_OEQ:   R = executeFCMP_OEQ(Src1, Src2, Ty); break;
+  case FCmpInst::FCMP_UNE:   R = executeFCMP_UNE(Src1, Src2, Ty); break;
+  case FCmpInst::FCMP_ONE:   R = executeFCMP_ONE(Src1, Src2, Ty); break;
+  case FCmpInst::FCMP_ULT:   R = executeFCMP_ULT(Src1, Src2, Ty); break;
+  case FCmpInst::FCMP_OLT:   R = executeFCMP_OLT(Src1, Src2, Ty); break;
+  case FCmpInst::FCMP_UGT:   R = executeFCMP_UGT(Src1, Src2, Ty); break;
+  case FCmpInst::FCMP_OGT:   R = executeFCMP_OGT(Src1, Src2, Ty); break;
+  case FCmpInst::FCMP_ULE:   R = executeFCMP_ULE(Src1, Src2, Ty); break;
+  case FCmpInst::FCMP_OLE:   R = executeFCMP_OLE(Src1, Src2, Ty); break;
+  case FCmpInst::FCMP_UGE:   R = executeFCMP_UGE(Src1, Src2, Ty); break;
+  case FCmpInst::FCMP_OGE:   R = executeFCMP_OGE(Src1, Src2, Ty); break;
+  default:
+    cerr << "Don't know how to handle this FCmp predicate!\n-->" << I;
+    abort();
+  }
+ 
+  SetValue(&I, R, SF);
+}
+
+static GenericValue executeCmpInst(unsigned predicate, GenericValue Src1, 
+                                   GenericValue Src2, const Type *Ty) {
+  GenericValue Result;
+  switch (predicate) {
+  case ICmpInst::ICMP_EQ:    return executeICMP_EQ(Src1, Src2, Ty);
+  case ICmpInst::ICMP_NE:    return executeICMP_NE(Src1, Src2, Ty);
+  case ICmpInst::ICMP_UGT:   return executeICMP_UGT(Src1, Src2, Ty);
+  case ICmpInst::ICMP_SGT:   return executeICMP_SGT(Src1, Src2, Ty);
+  case ICmpInst::ICMP_ULT:   return executeICMP_ULT(Src1, Src2, Ty);
+  case ICmpInst::ICMP_SLT:   return executeICMP_SLT(Src1, Src2, Ty);
+  case ICmpInst::ICMP_UGE:   return executeICMP_UGE(Src1, Src2, Ty);
+  case ICmpInst::ICMP_SGE:   return executeICMP_SGE(Src1, Src2, Ty);
+  case ICmpInst::ICMP_ULE:   return executeICMP_ULE(Src1, Src2, Ty);
+  case ICmpInst::ICMP_SLE:   return executeICMP_SLE(Src1, Src2, Ty);
+  case FCmpInst::FCMP_ORD:   return executeFCMP_ORD(Src1, Src2, Ty);
+  case FCmpInst::FCMP_UNO:   return executeFCMP_UNO(Src1, Src2, Ty);
+  case FCmpInst::FCMP_OEQ:   return executeFCMP_OEQ(Src1, Src2, Ty);
+  case FCmpInst::FCMP_UEQ:   return executeFCMP_UEQ(Src1, Src2, Ty);
+  case FCmpInst::FCMP_ONE:   return executeFCMP_ONE(Src1, Src2, Ty);
+  case FCmpInst::FCMP_UNE:   return executeFCMP_UNE(Src1, Src2, Ty);
+  case FCmpInst::FCMP_OLT:   return executeFCMP_OLT(Src1, Src2, Ty);
+  case FCmpInst::FCMP_ULT:   return executeFCMP_ULT(Src1, Src2, Ty);
+  case FCmpInst::FCMP_OGT:   return executeFCMP_OGT(Src1, Src2, Ty);
+  case FCmpInst::FCMP_UGT:   return executeFCMP_UGT(Src1, Src2, Ty);
+  case FCmpInst::FCMP_OLE:   return executeFCMP_OLE(Src1, Src2, Ty);
+  case FCmpInst::FCMP_ULE:   return executeFCMP_ULE(Src1, Src2, Ty);
+  case FCmpInst::FCMP_OGE:   return executeFCMP_OGE(Src1, Src2, Ty);
+  case FCmpInst::FCMP_UGE:   return executeFCMP_UGE(Src1, Src2, Ty);
+  case FCmpInst::FCMP_FALSE: { 
+    GenericValue Result;
+    Result.IntVal = APInt(1, false);
+    return Result;
+  }
+  case FCmpInst::FCMP_TRUE: {
+    GenericValue Result;
+    Result.IntVal = APInt(1, true);
+    return Result;
+  }
+  default:
+    cerr << "Unhandled Cmp predicate\n";
+    abort();
+  }
+}
+
+void Interpreter::visitBinaryOperator(BinaryOperator &I) {
+  ExecutionContext &SF = ECStack.back();
+  const Type *Ty    = I.getOperand(0)->getType();
+  GenericValue Src1 = getOperandValue(I.getOperand(0), SF);
+  GenericValue Src2 = getOperandValue(I.getOperand(1), SF);
+  GenericValue R;   // Result
+
+  switch (I.getOpcode()) {
+  case Instruction::Add:   executeAddInst  (R, Src1, Src2, Ty); break;
+  case Instruction::Sub:   executeSubInst  (R, Src1, Src2, Ty); break;
+  case Instruction::Mul:   executeMulInst  (R, Src1, Src2, Ty); break;
+  case Instruction::FDiv:  executeFDivInst (R, Src1, Src2, Ty); break;
+  case Instruction::FRem:  executeFRemInst (R, Src1, Src2, Ty); break;
+  case Instruction::UDiv:  R.IntVal = Src1.IntVal.udiv(Src2.IntVal); break;
+  case Instruction::SDiv:  R.IntVal = Src1.IntVal.sdiv(Src2.IntVal); break;
+  case Instruction::URem:  R.IntVal = Src1.IntVal.urem(Src2.IntVal); break;
+  case Instruction::SRem:  R.IntVal = Src1.IntVal.srem(Src2.IntVal); break;
+  case Instruction::And:   R.IntVal = Src1.IntVal & Src2.IntVal; break;
+  case Instruction::Or:    R.IntVal = Src1.IntVal | Src2.IntVal; break;
+  case Instruction::Xor:   R.IntVal = Src1.IntVal ^ Src2.IntVal; break;
+  default:
+    cerr << "Don't know how to handle this binary operator!\n-->" << I;
+    abort();
+  }
+
+  SetValue(&I, R, SF);
+}
+
+static GenericValue executeSelectInst(GenericValue Src1, GenericValue Src2,
+                                      GenericValue Src3) {
+  return Src1.IntVal == 0 ? Src3 : Src2;
+}
+
+void Interpreter::visitSelectInst(SelectInst &I) {
+  ExecutionContext &SF = ECStack.back();
+  GenericValue Src1 = getOperandValue(I.getOperand(0), SF);
+  GenericValue Src2 = getOperandValue(I.getOperand(1), SF);
+  GenericValue Src3 = getOperandValue(I.getOperand(2), SF);
+  GenericValue R = executeSelectInst(Src1, Src2, Src3);
+  SetValue(&I, R, SF);
+}
+
+
+//===----------------------------------------------------------------------===//
+//                     Terminator Instruction Implementations
+//===----------------------------------------------------------------------===//
+
+void Interpreter::exitCalled(GenericValue GV) {
+  // runAtExitHandlers() assumes there are no stack frames, but
+  // if exit() was called, then it had a stack frame. Blow away
+  // the stack before interpreting atexit handlers.
+  ECStack.clear ();
+  runAtExitHandlers ();
+  exit (GV.IntVal.zextOrTrunc(32).getZExtValue());
+}
+
+/// Pop the last stack frame off of ECStack and then copy the result
+/// back into the result variable if we are not returning void. The
+/// result variable may be the ExitValue, or the Value of the calling
+/// CallInst if there was a previous stack frame. This method may
+/// invalidate any ECStack iterators you have. This method also takes
+/// care of switching to the normal destination BB, if we are returning
+/// from an invoke.
+///
+void Interpreter::popStackAndReturnValueToCaller (const Type *RetTy,
+                                                  GenericValue Result) {
+  // Pop the current stack frame.
+  ECStack.pop_back();
+
+  if (ECStack.empty()) {  // Finished main.  Put result into exit code...
+    if (RetTy && RetTy->isInteger()) {          // Nonvoid return type?
+      ExitValue = Result;   // Capture the exit value of the program
+    } else {
+      memset(&ExitValue.Untyped, 0, sizeof(ExitValue.Untyped));
+    }
+  } else {
+    // If we have a previous stack frame, and we have a previous call,
+    // fill in the return value...
+    ExecutionContext &CallingSF = ECStack.back();
+    if (Instruction *I = CallingSF.Caller.getInstruction()) {
+      if (CallingSF.Caller.getType() != Type::VoidTy)      // Save result...
+        SetValue(I, Result, CallingSF);
+      if (InvokeInst *II = dyn_cast<InvokeInst> (I))
+        SwitchToNewBasicBlock (II->getNormalDest (), CallingSF);
+      CallingSF.Caller = CallSite();          // We returned from the call...
+    }
+  }
+}
+
+void Interpreter::visitReturnInst(ReturnInst &I) {
+  ExecutionContext &SF = ECStack.back();
+  const Type *RetTy = Type::VoidTy;
+  GenericValue Result;
+
+  // Save away the return value... (if we are not 'ret void')
+  if (I.getNumOperands()) {
+    RetTy  = I.getReturnValue()->getType();
+    Result = getOperandValue(I.getReturnValue(), SF);
+  }
+
+  popStackAndReturnValueToCaller(RetTy, Result);
+}
+
+void Interpreter::visitUnwindInst(UnwindInst &I) {
+  // Unwind stack
+  Instruction *Inst;
+  do {
+    ECStack.pop_back ();
+    if (ECStack.empty ())
+      abort ();
+    Inst = ECStack.back ().Caller.getInstruction ();
+  } while (!(Inst && isa<InvokeInst> (Inst)));
+
+  // Return from invoke
+  ExecutionContext &InvokingSF = ECStack.back ();
+  InvokingSF.Caller = CallSite ();
+
+  // Go to exceptional destination BB of invoke instruction
+  SwitchToNewBasicBlock(cast<InvokeInst>(Inst)->getUnwindDest(), InvokingSF);
+}
+
+void Interpreter::visitUnreachableInst(UnreachableInst &I) {
+  cerr << "ERROR: Program executed an 'unreachable' instruction!\n";
+  abort();
+}
+
+void Interpreter::visitBranchInst(BranchInst &I) {
+  ExecutionContext &SF = ECStack.back();
+  BasicBlock *Dest;
+
+  Dest = I.getSuccessor(0);          // Uncond branches have a fixed dest...
+  if (!I.isUnconditional()) {
+    Value *Cond = I.getCondition();
+    if (getOperandValue(Cond, SF).IntVal == 0) // If false cond...
+      Dest = I.getSuccessor(1);
+  }
+  SwitchToNewBasicBlock(Dest, SF);
+}
+
+void Interpreter::visitSwitchInst(SwitchInst &I) {
+  ExecutionContext &SF = ECStack.back();
+  GenericValue CondVal = getOperandValue(I.getOperand(0), SF);
+  const Type *ElTy = I.getOperand(0)->getType();
+
+  // Check to see if any of the cases match...
+  BasicBlock *Dest = 0;
+  for (unsigned i = 2, e = I.getNumOperands(); i != e; i += 2)
+    if (executeICMP_EQ(CondVal, getOperandValue(I.getOperand(i), SF), ElTy)
+        .IntVal != 0) {
+      Dest = cast<BasicBlock>(I.getOperand(i+1));
+      break;
+    }
+
+  if (!Dest) Dest = I.getDefaultDest();   // No cases matched: use default
+  SwitchToNewBasicBlock(Dest, SF);
+}
+
+// SwitchToNewBasicBlock - This method is used to jump to a new basic block.
+// This function handles the actual updating of block and instruction iterators
+// as well as execution of all of the PHI nodes in the destination block.
+//
+// This method does this because all of the PHI nodes must be executed
+// atomically, reading their inputs before any of the results are updated.  Not
+// doing this can cause problems if the PHI nodes depend on other PHI nodes for
+// their inputs.  If the input PHI node is updated before it is read, incorrect
+// results can happen.  Thus we use a two phase approach.
+//
+void Interpreter::SwitchToNewBasicBlock(BasicBlock *Dest, ExecutionContext &SF){
+  BasicBlock *PrevBB = SF.CurBB;      // Remember where we came from...
+  SF.CurBB   = Dest;                  // Update CurBB to branch destination
+  SF.CurInst = SF.CurBB->begin();     // Update new instruction ptr...
+
+  if (!isa<PHINode>(SF.CurInst)) return;  // Nothing fancy to do
+
+  // Loop over all of the PHI nodes in the current block, reading their inputs.
+  std::vector<GenericValue> ResultValues;
+
+  for (; PHINode *PN = dyn_cast<PHINode>(SF.CurInst); ++SF.CurInst) {
+    // Search for the value corresponding to this previous bb...
+    int i = PN->getBasicBlockIndex(PrevBB);
+    assert(i != -1 && "PHINode doesn't contain entry for predecessor??");
+    Value *IncomingValue = PN->getIncomingValue(i);
+
+    // Save the incoming value for this PHI node...
+    ResultValues.push_back(getOperandValue(IncomingValue, SF));
+  }
+
+  // Now loop over all of the PHI nodes setting their values...
+  SF.CurInst = SF.CurBB->begin();
+  for (unsigned i = 0; isa<PHINode>(SF.CurInst); ++SF.CurInst, ++i) {
+    PHINode *PN = cast<PHINode>(SF.CurInst);
+    SetValue(PN, ResultValues[i], SF);
+  }
+}
+
+//===----------------------------------------------------------------------===//
+//                     Memory Instruction Implementations
+//===----------------------------------------------------------------------===//
+
+void Interpreter::visitAllocationInst(AllocationInst &I) {
+  ExecutionContext &SF = ECStack.back();
+
+  const Type *Ty = I.getType()->getElementType();  // Type to be allocated
+
+  // Get the number of elements being allocated by the array...
+  unsigned NumElements = 
+    getOperandValue(I.getOperand(0), SF).IntVal.getZExtValue();
+
+  unsigned TypeSize = (size_t)TD.getTypeSize(Ty);
+
+  unsigned MemToAlloc = NumElements * TypeSize;
+
+  // Allocate enough memory to hold the type...
+  void *Memory = malloc(MemToAlloc);
+
+  DOUT << "Allocated Type: " << *Ty << " (" << TypeSize << " bytes) x " 
+       << NumElements << " (Total: " << MemToAlloc << ") at "
+       << uintptr_t(Memory) << '\n';
+
+  GenericValue Result = PTOGV(Memory);
+  assert(Result.PointerVal != 0 && "Null pointer returned by malloc!");
+  SetValue(&I, Result, SF);
+
+  if (I.getOpcode() == Instruction::Alloca)
+    ECStack.back().Allocas.add(Memory);
+}
+
+void Interpreter::visitFreeInst(FreeInst &I) {
+  ExecutionContext &SF = ECStack.back();
+  assert(isa<PointerType>(I.getOperand(0)->getType()) && "Freeing nonptr?");
+  GenericValue Value = getOperandValue(I.getOperand(0), SF);
+  // TODO: Check to make sure memory is allocated
+  free(GVTOP(Value));   // Free memory
+}
+
+// getElementOffset - The workhorse for getelementptr.
+//
+GenericValue Interpreter::executeGEPOperation(Value *Ptr, gep_type_iterator I,
+                                              gep_type_iterator E,
+                                              ExecutionContext &SF) {
+  assert(isa<PointerType>(Ptr->getType()) &&
+         "Cannot getElementOffset of a nonpointer type!");
+
+  uint64_t Total = 0;
+
+  for (; I != E; ++I) {
+    if (const StructType *STy = dyn_cast<StructType>(*I)) {
+      const StructLayout *SLO = TD.getStructLayout(STy);
+
+      const ConstantInt *CPU = cast<ConstantInt>(I.getOperand());
+      unsigned Index = unsigned(CPU->getZExtValue());
+
+      Total += SLO->getElementOffset(Index);
+    } else {
+      const SequentialType *ST = cast<SequentialType>(*I);
+      // Get the index number for the array... which must be long type...
+      GenericValue IdxGV = getOperandValue(I.getOperand(), SF);
+
+      int64_t Idx;
+      unsigned BitWidth = 
+        cast<IntegerType>(I.getOperand()->getType())->getBitWidth();
+      if (BitWidth == 32)
+        Idx = (int64_t)(int32_t)IdxGV.IntVal.getZExtValue();
+      else if (BitWidth == 64)
+        Idx = (int64_t)IdxGV.IntVal.getZExtValue();
+      else 
+        assert(0 && "Invalid index type for getelementptr");
+      Total += TD.getTypeSize(ST->getElementType())*Idx;
+    }
+  }
+
+  GenericValue Result;
+  Result.PointerVal = ((char*)getOperandValue(Ptr, SF).PointerVal) + Total;
+  DOUT << "GEP Index " << Total << " bytes.\n";
+  return Result;
+}
+
+void Interpreter::visitGetElementPtrInst(GetElementPtrInst &I) {
+  ExecutionContext &SF = ECStack.back();
+  SetValue(&I, TheEE->executeGEPOperation(I.getPointerOperand(),
+                                   gep_type_begin(I), gep_type_end(I), SF), SF);
+}
+
+void Interpreter::visitLoadInst(LoadInst &I) {
+  ExecutionContext &SF = ECStack.back();
+  GenericValue SRC = getOperandValue(I.getPointerOperand(), SF);
+  GenericValue *Ptr = (GenericValue*)GVTOP(SRC);
+  GenericValue Result;
+  LoadValueFromMemory(Result, Ptr, I.getType());
+  SetValue(&I, Result, SF);
+}
+
+void Interpreter::visitStoreInst(StoreInst &I) {
+  ExecutionContext &SF = ECStack.back();
+  GenericValue Val = getOperandValue(I.getOperand(0), SF);
+  GenericValue SRC = getOperandValue(I.getPointerOperand(), SF);
+  StoreValueToMemory(Val, (GenericValue *)GVTOP(SRC),
+                     I.getOperand(0)->getType());
+}
+
+//===----------------------------------------------------------------------===//
+//                 Miscellaneous Instruction Implementations
+//===----------------------------------------------------------------------===//
+
+void Interpreter::visitCallSite(CallSite CS) {
+  ExecutionContext &SF = ECStack.back();
+
+  // Check to see if this is an intrinsic function call...
+  Function *F = CS.getCalledFunction();
+  if (F && F->isDeclaration ())
+    switch (F->getIntrinsicID()) {
+    case Intrinsic::not_intrinsic:
+      break;
+    case Intrinsic::vastart: { // va_start
+      GenericValue ArgIndex;
+      ArgIndex.UIntPairVal.first = ECStack.size() - 1;
+      ArgIndex.UIntPairVal.second = 0;
+      SetValue(CS.getInstruction(), ArgIndex, SF);
+      return;
+    }
+    case Intrinsic::vaend:    // va_end is a noop for the interpreter
+      return;
+    case Intrinsic::vacopy:   // va_copy: dest = src
+      SetValue(CS.getInstruction(), getOperandValue(*CS.arg_begin(), SF), SF);
+      return;
+    default:
+      // If it is an unknown intrinsic function, use the intrinsic lowering
+      // class to transform it into hopefully tasty LLVM code.
+      //
+      BasicBlock::iterator me(CS.getInstruction());
+      BasicBlock *Parent = CS.getInstruction()->getParent();
+      bool atBegin(Parent->begin() == me);
+      if (!atBegin)
+        --me;
+      IL->LowerIntrinsicCall(cast<CallInst>(CS.getInstruction()));
+
+      // Restore the CurInst pointer to the first instruction newly inserted, if
+      // any.
+      if (atBegin) {
+        SF.CurInst = Parent->begin();
+      } else {
+        SF.CurInst = me;
+        ++SF.CurInst;
+      }
+      return;
+    }
+
+
+  SF.Caller = CS;
+  std::vector<GenericValue> ArgVals;
+  const unsigned NumArgs = SF.Caller.arg_size();
+  ArgVals.reserve(NumArgs);
+  uint16_t pNum = 1;
+  for (CallSite::arg_iterator i = SF.Caller.arg_begin(),
+         e = SF.Caller.arg_end(); i != e; ++i, ++pNum) {
+    Value *V = *i;
+    ArgVals.push_back(getOperandValue(V, SF));
+    if (F) {
+     // Promote all integral types whose size is < sizeof(i32) into i32.  
+     // We do this by zero or sign extending the value as appropriate 
+     // according to the parameter attributes
+      const Type *Ty = V->getType();
+      if (Ty->isInteger() && (ArgVals.back().IntVal.getBitWidth() < 32))
+        if (const ParamAttrsList *PA = F->getParamAttrs())
+          if (PA->paramHasAttr(pNum, ParamAttr::ZExt))
+            ArgVals.back().IntVal = ArgVals.back().IntVal.zext(32);
+          else if (PA->paramHasAttr(pNum, ParamAttr::SExt))
+            ArgVals.back().IntVal = ArgVals.back().IntVal.sext(32);
+     }
+  }
+
+  // To handle indirect calls, we must get the pointer value from the argument
+  // and treat it as a function pointer.
+  GenericValue SRC = getOperandValue(SF.Caller.getCalledValue(), SF);
+  callFunction((Function*)GVTOP(SRC), ArgVals);
+}
+
+void Interpreter::visitShl(BinaryOperator &I) {
+  ExecutionContext &SF = ECStack.back();
+  GenericValue Src1 = getOperandValue(I.getOperand(0), SF);
+  GenericValue Src2 = getOperandValue(I.getOperand(1), SF);
+  GenericValue Dest;
+  Dest.IntVal = Src1.IntVal.shl(Src2.IntVal.getZExtValue());
+  SetValue(&I, Dest, SF);
+}
+
+void Interpreter::visitLShr(BinaryOperator &I) {
+  ExecutionContext &SF = ECStack.back();
+  GenericValue Src1 = getOperandValue(I.getOperand(0), SF);
+  GenericValue Src2 = getOperandValue(I.getOperand(1), SF);
+  GenericValue Dest;
+  Dest.IntVal =  Src1.IntVal.lshr(Src2.IntVal.getZExtValue());
+  SetValue(&I, Dest, SF);
+}
+
+void Interpreter::visitAShr(BinaryOperator &I) {
+  ExecutionContext &SF = ECStack.back();
+  GenericValue Src1 = getOperandValue(I.getOperand(0), SF);
+  GenericValue Src2 = getOperandValue(I.getOperand(1), SF);
+  GenericValue Dest; 
+  Dest.IntVal = Src1.IntVal.ashr(Src2.IntVal.getZExtValue());
+  SetValue(&I, Dest, SF);
+}
+
+GenericValue Interpreter::executeTruncInst(Value *SrcVal, const Type *DstTy,
+                                           ExecutionContext &SF) {
+  const Type *SrcTy = SrcVal->getType();
+  GenericValue Dest, Src = getOperandValue(SrcVal, SF);
+  const IntegerType *DITy = cast<IntegerType>(DstTy);
+  const IntegerType *SITy = cast<IntegerType>(SrcTy);
+  unsigned DBitWidth = DITy->getBitWidth();
+  unsigned SBitWidth = SITy->getBitWidth();
+  assert(SBitWidth > DBitWidth && "Invalid truncate");
+  Dest.IntVal = Src.IntVal.trunc(DBitWidth);
+  return Dest;
+}
+
+GenericValue Interpreter::executeSExtInst(Value *SrcVal, const Type *DstTy,
+                                          ExecutionContext &SF) {
+  const Type *SrcTy = SrcVal->getType();
+  GenericValue Dest, Src = getOperandValue(SrcVal, SF);
+  const IntegerType *DITy = cast<IntegerType>(DstTy);
+  const IntegerType *SITy = cast<IntegerType>(SrcTy);
+  unsigned DBitWidth = DITy->getBitWidth();
+  unsigned SBitWidth = SITy->getBitWidth();
+  assert(SBitWidth < DBitWidth && "Invalid sign extend");
+  Dest.IntVal = Src.IntVal.sext(DBitWidth);
+  return Dest;
+}
+
+GenericValue Interpreter::executeZExtInst(Value *SrcVal, const Type *DstTy,
+                                          ExecutionContext &SF) {
+  const Type *SrcTy = SrcVal->getType();
+  GenericValue Dest, Src = getOperandValue(SrcVal, SF);
+  const IntegerType *DITy = cast<IntegerType>(DstTy);
+  const IntegerType *SITy = cast<IntegerType>(SrcTy);
+  unsigned DBitWidth = DITy->getBitWidth();
+  unsigned SBitWidth = SITy->getBitWidth();
+  assert(SBitWidth < DBitWidth && "Invalid sign extend");
+  Dest.IntVal = Src.IntVal.zext(DBitWidth);
+  return Dest;
+}
+
+GenericValue Interpreter::executeFPTruncInst(Value *SrcVal, const Type *DstTy,
+                                             ExecutionContext &SF) {
+  const Type *SrcTy = SrcVal->getType();
+  GenericValue Dest, Src = getOperandValue(SrcVal, SF);
+  assert(SrcTy == Type::DoubleTy && DstTy == Type::FloatTy &&
+         "Invalid FPTrunc instruction");
+  Dest.FloatVal = (float) Src.DoubleVal;
+  return Dest;
+}
+
+GenericValue Interpreter::executeFPExtInst(Value *SrcVal, const Type *DstTy,
+                                           ExecutionContext &SF) {
+  const Type *SrcTy = SrcVal->getType();
+  GenericValue Dest, Src = getOperandValue(SrcVal, SF);
+  assert(SrcTy == Type::FloatTy && DstTy == Type::DoubleTy &&
+         "Invalid FPTrunc instruction");
+  Dest.DoubleVal = (double) Src.FloatVal;
+  return Dest;
+}
+
+GenericValue Interpreter::executeFPToUIInst(Value *SrcVal, const Type *DstTy,
+                                            ExecutionContext &SF) {
+  const Type *SrcTy = SrcVal->getType();
+  uint32_t DBitWidth = cast<IntegerType>(DstTy)->getBitWidth();
+  GenericValue Dest, Src = getOperandValue(SrcVal, SF);
+  assert(SrcTy->isFloatingPoint() && "Invalid FPToUI instruction");
+
+  if (SrcTy->getTypeID() == Type::FloatTyID)
+    Dest.IntVal = APIntOps::RoundFloatToAPInt(Src.FloatVal, DBitWidth);
+  else
+    Dest.IntVal = APIntOps::RoundDoubleToAPInt(Src.DoubleVal, DBitWidth);
+  return Dest;
+}
+
+GenericValue Interpreter::executeFPToSIInst(Value *SrcVal, const Type *DstTy,
+                                            ExecutionContext &SF) {
+  const Type *SrcTy = SrcVal->getType();
+  uint32_t DBitWidth = cast<IntegerType>(DstTy)->getBitWidth();
+  GenericValue Dest, Src = getOperandValue(SrcVal, SF);
+  assert(SrcTy->isFloatingPoint() && "Invalid FPToSI instruction");
+
+  if (SrcTy->getTypeID() == Type::FloatTyID)
+    Dest.IntVal = APIntOps::RoundFloatToAPInt(Src.FloatVal, DBitWidth);
+  else
+    Dest.IntVal = APIntOps::RoundDoubleToAPInt(Src.DoubleVal, DBitWidth);
+  return Dest;
+}
+
+GenericValue Interpreter::executeUIToFPInst(Value *SrcVal, const Type *DstTy,
+                                            ExecutionContext &SF) {
+  GenericValue Dest, Src = getOperandValue(SrcVal, SF);
+  assert(DstTy->isFloatingPoint() && "Invalid UIToFP instruction");
+
+  if (DstTy->getTypeID() == Type::FloatTyID)
+    Dest.FloatVal = APIntOps::RoundAPIntToFloat(Src.IntVal);
+  else
+    Dest.DoubleVal = APIntOps::RoundAPIntToDouble(Src.IntVal);
+  return Dest;
+}
+
+GenericValue Interpreter::executeSIToFPInst(Value *SrcVal, const Type *DstTy,
+                                            ExecutionContext &SF) {
+  GenericValue Dest, Src = getOperandValue(SrcVal, SF);
+  assert(DstTy->isFloatingPoint() && "Invalid SIToFP instruction");
+
+  if (DstTy->getTypeID() == Type::FloatTyID)
+    Dest.FloatVal = APIntOps::RoundSignedAPIntToFloat(Src.IntVal);
+  else
+    Dest.DoubleVal = APIntOps::RoundSignedAPIntToDouble(Src.IntVal);
+  return Dest;
+
+}
+
+GenericValue Interpreter::executePtrToIntInst(Value *SrcVal, const Type *DstTy,
+                                              ExecutionContext &SF) {
+  const Type *SrcTy = SrcVal->getType();
+  uint32_t DBitWidth = cast<IntegerType>(DstTy)->getBitWidth();
+  GenericValue Dest, Src = getOperandValue(SrcVal, SF);
+  assert(isa<PointerType>(SrcTy) && "Invalid PtrToInt instruction");
+
+  Dest.IntVal = APInt(DBitWidth, (intptr_t) Src.PointerVal);
+  return Dest;
+}
+
+GenericValue Interpreter::executeIntToPtrInst(Value *SrcVal, const Type *DstTy,
+                                              ExecutionContext &SF) {
+  GenericValue Dest, Src = getOperandValue(SrcVal, SF);
+  assert(isa<PointerType>(DstTy) && "Invalid PtrToInt instruction");
+
+  uint32_t PtrSize = TD.getPointerSizeInBits();
+  if (PtrSize != Src.IntVal.getBitWidth())
+    Src.IntVal = Src.IntVal.zextOrTrunc(PtrSize);
+
+  Dest.PointerVal = PointerTy(intptr_t(Src.IntVal.getZExtValue()));
+  return Dest;
+}
+
+GenericValue Interpreter::executeBitCastInst(Value *SrcVal, const Type *DstTy,
+                                             ExecutionContext &SF) {
+  
+  const Type *SrcTy = SrcVal->getType();
+  GenericValue Dest, Src = getOperandValue(SrcVal, SF);
+  if (isa<PointerType>(DstTy)) {
+    assert(isa<PointerType>(SrcTy) && "Invalid BitCast");
+    Dest.PointerVal = Src.PointerVal;
+  } else if (DstTy->isInteger()) {
+    if (SrcTy == Type::FloatTy) {
+      Dest.IntVal.zext(sizeof(Src.FloatVal) * 8);
+      Dest.IntVal.floatToBits(Src.FloatVal);
+    } else if (SrcTy == Type::DoubleTy) {
+      Dest.IntVal.zext(sizeof(Src.DoubleVal) * 8);
+      Dest.IntVal.doubleToBits(Src.DoubleVal);
+    } else if (SrcTy->isInteger()) {
+      Dest.IntVal = Src.IntVal;
+    } else 
+      assert(0 && "Invalid BitCast");
+  } else if (DstTy == Type::FloatTy) {
+    if (SrcTy->isInteger())
+      Dest.FloatVal = Src.IntVal.bitsToFloat();
+    else
+      Dest.FloatVal = Src.FloatVal;
+  } else if (DstTy == Type::DoubleTy) {
+    if (SrcTy->isInteger())
+      Dest.DoubleVal = Src.IntVal.bitsToDouble();
+    else
+      Dest.DoubleVal = Src.DoubleVal;
+  } else
+    assert(0 && "Invalid Bitcast");
+
+  return Dest;
+}
+
+void Interpreter::visitTruncInst(TruncInst &I) {
+  ExecutionContext &SF = ECStack.back();
+  SetValue(&I, executeTruncInst(I.getOperand(0), I.getType(), SF), SF);
+}
+
+void Interpreter::visitSExtInst(SExtInst &I) {
+  ExecutionContext &SF = ECStack.back();
+  SetValue(&I, executeSExtInst(I.getOperand(0), I.getType(), SF), SF);
+}
+
+void Interpreter::visitZExtInst(ZExtInst &I) {
+  ExecutionContext &SF = ECStack.back();
+  SetValue(&I, executeZExtInst(I.getOperand(0), I.getType(), SF), SF);
+}
+
+void Interpreter::visitFPTruncInst(FPTruncInst &I) {
+  ExecutionContext &SF = ECStack.back();
+  SetValue(&I, executeFPTruncInst(I.getOperand(0), I.getType(), SF), SF);
+}
+
+void Interpreter::visitFPExtInst(FPExtInst &I) {
+  ExecutionContext &SF = ECStack.back();
+  SetValue(&I, executeFPExtInst(I.getOperand(0), I.getType(), SF), SF);
+}
+
+void Interpreter::visitUIToFPInst(UIToFPInst &I) {
+  ExecutionContext &SF = ECStack.back();
+  SetValue(&I, executeUIToFPInst(I.getOperand(0), I.getType(), SF), SF);
+}
+
+void Interpreter::visitSIToFPInst(SIToFPInst &I) {
+  ExecutionContext &SF = ECStack.back();
+  SetValue(&I, executeSIToFPInst(I.getOperand(0), I.getType(), SF), SF);
+}
+
+void Interpreter::visitFPToUIInst(FPToUIInst &I) {
+  ExecutionContext &SF = ECStack.back();
+  SetValue(&I, executeFPToUIInst(I.getOperand(0), I.getType(), SF), SF);
+}
+
+void Interpreter::visitFPToSIInst(FPToSIInst &I) {
+  ExecutionContext &SF = ECStack.back();
+  SetValue(&I, executeFPToSIInst(I.getOperand(0), I.getType(), SF), SF);
+}
+
+void Interpreter::visitPtrToIntInst(PtrToIntInst &I) {
+  ExecutionContext &SF = ECStack.back();
+  SetValue(&I, executePtrToIntInst(I.getOperand(0), I.getType(), SF), SF);
+}
+
+void Interpreter::visitIntToPtrInst(IntToPtrInst &I) {
+  ExecutionContext &SF = ECStack.back();
+  SetValue(&I, executeIntToPtrInst(I.getOperand(0), I.getType(), SF), SF);
+}
+
+void Interpreter::visitBitCastInst(BitCastInst &I) {
+  ExecutionContext &SF = ECStack.back();
+  SetValue(&I, executeBitCastInst(I.getOperand(0), I.getType(), SF), SF);
+}
+
+#define IMPLEMENT_VAARG(TY) \
+   case Type::TY##TyID: Dest.TY##Val = Src.TY##Val; break
+
+void Interpreter::visitVAArgInst(VAArgInst &I) {
+  ExecutionContext &SF = ECStack.back();
+
+  // Get the incoming valist parameter.  LLI treats the valist as a
+  // (ec-stack-depth var-arg-index) pair.
+  GenericValue VAList = getOperandValue(I.getOperand(0), SF);
+  GenericValue Dest;
+  GenericValue Src = ECStack[VAList.UIntPairVal.first]
+                      .VarArgs[VAList.UIntPairVal.second];
+  const Type *Ty = I.getType();
+  switch (Ty->getTypeID()) {
+    case Type::IntegerTyID: Dest.IntVal = Src.IntVal;
+    IMPLEMENT_VAARG(Pointer);
+    IMPLEMENT_VAARG(Float);
+    IMPLEMENT_VAARG(Double);
+  default:
+    cerr << "Unhandled dest type for vaarg instruction: " << *Ty << "\n";
+    abort();
+  }
+
+  // Set the Value of this Instruction.
+  SetValue(&I, Dest, SF);
+
+  // Move the pointer to the next vararg.
+  ++VAList.UIntPairVal.second;
+}
+
+GenericValue Interpreter::getConstantExprValue (ConstantExpr *CE,
+                                                ExecutionContext &SF) {
+  switch (CE->getOpcode()) {
+  case Instruction::Trunc:   
+      return executeTruncInst(CE->getOperand(0), CE->getType(), SF);
+  case Instruction::ZExt:
+      return executeZExtInst(CE->getOperand(0), CE->getType(), SF);
+  case Instruction::SExt:
+      return executeSExtInst(CE->getOperand(0), CE->getType(), SF);
+  case Instruction::FPTrunc:
+      return executeFPTruncInst(CE->getOperand(0), CE->getType(), SF);
+  case Instruction::FPExt:
+      return executeFPExtInst(CE->getOperand(0), CE->getType(), SF);
+  case Instruction::UIToFP:
+      return executeUIToFPInst(CE->getOperand(0), CE->getType(), SF);
+  case Instruction::SIToFP:
+      return executeSIToFPInst(CE->getOperand(0), CE->getType(), SF);
+  case Instruction::FPToUI:
+      return executeFPToUIInst(CE->getOperand(0), CE->getType(), SF);
+  case Instruction::FPToSI:
+      return executeFPToSIInst(CE->getOperand(0), CE->getType(), SF);
+  case Instruction::PtrToInt:
+      return executePtrToIntInst(CE->getOperand(0), CE->getType(), SF);
+  case Instruction::IntToPtr:
+      return executeIntToPtrInst(CE->getOperand(0), CE->getType(), SF);
+  case Instruction::BitCast:
+      return executeBitCastInst(CE->getOperand(0), CE->getType(), SF);
+  case Instruction::GetElementPtr:
+    return executeGEPOperation(CE->getOperand(0), gep_type_begin(CE),
+                               gep_type_end(CE), SF);
+  case Instruction::FCmp:
+  case Instruction::ICmp:
+    return executeCmpInst(CE->getPredicate(),
+                          getOperandValue(CE->getOperand(0), SF),
+                          getOperandValue(CE->getOperand(1), SF),
+                          CE->getOperand(0)->getType());
+  case Instruction::Select:
+    return executeSelectInst(getOperandValue(CE->getOperand(0), SF),
+                             getOperandValue(CE->getOperand(1), SF),
+                             getOperandValue(CE->getOperand(2), SF));
+  default :
+    break;
+  }
+
+  // The cases below here require a GenericValue parameter for the result
+  // so we initialize one, compute it and then return it.
+  GenericValue Op0 = getOperandValue(CE->getOperand(0), SF);
+  GenericValue Op1 = getOperandValue(CE->getOperand(1), SF);
+  GenericValue Dest;
+  const Type * Ty = CE->getOperand(0)->getType();
+  switch (CE->getOpcode()) {
+  case Instruction::Add:  executeAddInst (Dest, Op0, Op1, Ty); break;
+  case Instruction::Sub:  executeSubInst (Dest, Op0, Op1, Ty); break;
+  case Instruction::Mul:  executeMulInst (Dest, Op0, Op1, Ty); break;
+  case Instruction::FDiv: executeFDivInst(Dest, Op0, Op1, Ty); break;
+  case Instruction::FRem: executeFRemInst(Dest, Op0, Op1, Ty); break;
+  case Instruction::SDiv: Dest.IntVal = Op0.IntVal.sdiv(Op1.IntVal); break;
+  case Instruction::UDiv: Dest.IntVal = Op0.IntVal.udiv(Op1.IntVal); break;
+  case Instruction::URem: Dest.IntVal = Op0.IntVal.urem(Op1.IntVal); break;
+  case Instruction::SRem: Dest.IntVal = Op0.IntVal.srem(Op1.IntVal); break;
+  case Instruction::And:  Dest.IntVal = Op0.IntVal.And(Op1.IntVal); break;
+  case Instruction::Or:   Dest.IntVal = Op0.IntVal.Or(Op1.IntVal); break;
+  case Instruction::Xor:  Dest.IntVal = Op0.IntVal.Xor(Op1.IntVal); break;
+  case Instruction::Shl:  
+    Dest.IntVal = Op0.IntVal.shl(Op1.IntVal.getZExtValue());
+    break;
+  case Instruction::LShr: 
+    Dest.IntVal = Op0.IntVal.lshr(Op1.IntVal.getZExtValue());
+    break;
+  case Instruction::AShr: 
+    Dest.IntVal = Op0.IntVal.ashr(Op1.IntVal.getZExtValue());
+    break;
+  default:
+    cerr << "Unhandled ConstantExpr: " << *CE << "\n";
+    abort();
+    return GenericValue();
+  }
+  return Dest;
+}
+
+GenericValue Interpreter::getOperandValue(Value *V, ExecutionContext &SF) {
+  if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V)) {
+    return getConstantExprValue(CE, SF);
+  } else if (Constant *CPV = dyn_cast<Constant>(V)) {
+    return getConstantValue(CPV);
+  } else if (GlobalValue *GV = dyn_cast<GlobalValue>(V)) {
+    return PTOGV(getPointerToGlobal(GV));
+  } else {
+    return SF.Values[V];
+  }
+}
+
+//===----------------------------------------------------------------------===//
+//                        Dispatch and Execution Code
+//===----------------------------------------------------------------------===//
+
+//===----------------------------------------------------------------------===//
+// callFunction - Execute the specified function...
+//
+void Interpreter::callFunction(Function *F,
+                               const std::vector<GenericValue> &ArgVals) {
+  assert((ECStack.empty() || ECStack.back().Caller.getInstruction() == 0 ||
+          ECStack.back().Caller.arg_size() == ArgVals.size()) &&
+         "Incorrect number of arguments passed into function call!");
+  // Make a new stack frame... and fill it in.
+  ECStack.push_back(ExecutionContext());
+  ExecutionContext &StackFrame = ECStack.back();
+  StackFrame.CurFunction = F;
+
+  // Special handling for external functions.
+  if (F->isDeclaration()) {
+    GenericValue Result = callExternalFunction (F, ArgVals);
+    // Simulate a 'ret' instruction of the appropriate type.
+    popStackAndReturnValueToCaller (F->getReturnType (), Result);
+    return;
+  }
+
+  // Get pointers to first LLVM BB & Instruction in function.
+  StackFrame.CurBB     = F->begin();
+  StackFrame.CurInst   = StackFrame.CurBB->begin();
+
+  // Run through the function arguments and initialize their values...
+  assert((ArgVals.size() == F->arg_size() ||
+         (ArgVals.size() > F->arg_size() && F->getFunctionType()->isVarArg()))&&
+         "Invalid number of values passed to function invocation!");
+
+  // Handle non-varargs arguments...
+  unsigned i = 0;
+  for (Function::arg_iterator AI = F->arg_begin(), E = F->arg_end(); 
+       AI != E; ++AI, ++i)
+    SetValue(AI, ArgVals[i], StackFrame);
+
+  // Handle varargs arguments...
+  StackFrame.VarArgs.assign(ArgVals.begin()+i, ArgVals.end());
+}
+
+static void PrintGenericValue(const GenericValue &Val, const Type* Ty) {
+  switch (Ty->getTypeID()) {
+    default: assert(0 && "Invalid GenericValue Type");
+    case Type::VoidTyID:    DOUT << "void"; break;
+    case Type::FloatTyID:   DOUT << "float " << Val.FloatVal; break;
+    case Type::DoubleTyID:  DOUT << "double " << Val.DoubleVal; break;
+    case Type::PointerTyID: DOUT << "void* " << intptr_t(Val.PointerVal); break;
+    case Type::IntegerTyID: 
+      DOUT << "i" << Val.IntVal.getBitWidth() << " " << Val.IntVal.toString(10)
+           << " (0x" << Val.IntVal.toString(16) << ")\n";
+      break;
+  }
+}
+
+void Interpreter::run() {
+  while (!ECStack.empty()) {
+    // Interpret a single instruction & increment the "PC".
+    ExecutionContext &SF = ECStack.back();  // Current stack frame
+    Instruction &I = *SF.CurInst++;         // Increment before execute
+
+    // Track the number of dynamic instructions executed.
+    ++NumDynamicInsts;
+
+    DOUT << "About to interpret: " << I;
+    visit(I);   // Dispatch to one of the visit* methods...
+#ifndef NDEBUG
+    if (!isa<CallInst>(I) && !isa<InvokeInst>(I) && 
+        I.getType() != Type::VoidTy) {
+      DOUT << "  --> ";
+      PrintGenericValue(SF.Values[&I], I.getType());
+    }
+#endif
+  }
+}
diff --git a/lib/ExecutionEngine/Interpreter/ExternalFunctions.cpp b/lib/ExecutionEngine/Interpreter/ExternalFunctions.cpp
new file mode 100644
index 0000000..14dcdf9
--- /dev/null
+++ b/lib/ExecutionEngine/Interpreter/ExternalFunctions.cpp
@@ -0,0 +1,757 @@
+//===-- ExternalFunctions.cpp - Implement External Functions --------------===//
+//
+//                     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 contains both code to deal with invoking "external" functions, but
+//  also contains code that implements "exported" external functions.
+//
+//  External functions in the interpreter are implemented by
+//  using the system's dynamic loader to look up the address of the function
+//  we want to invoke.  If a function is found, then one of the
+//  many lle_* wrapper functions in this file will translate its arguments from
+//  GenericValues to the types the function is actually expecting, before the
+//  function is called.
+//
+//===----------------------------------------------------------------------===//
+
+#include "Interpreter.h"
+#include "llvm/DerivedTypes.h"
+#include "llvm/Module.h"
+#include "llvm/Support/Streams.h"
+#include "llvm/System/DynamicLibrary.h"
+#include "llvm/Target/TargetData.h"
+#include <csignal>
+#include <map>
+#include <cmath>
+using std::vector;
+
+using namespace llvm;
+
+typedef GenericValue (*ExFunc)(FunctionType *, const vector<GenericValue> &);
+static std::map<const Function *, ExFunc> Functions;
+static std::map<std::string, ExFunc> FuncNames;
+
+static Interpreter *TheInterpreter;
+
+static char getTypeID(const Type *Ty) {
+  switch (Ty->getTypeID()) {
+  case Type::VoidTyID:    return 'V';
+  case Type::IntegerTyID:
+    switch (cast<IntegerType>(Ty)->getBitWidth()) {
+      case 1:  return 'o';
+      case 8:  return 'B';
+      case 16: return 'S';
+      case 32: return 'I';
+      case 64: return 'L';
+      default: return 'N';
+    }
+  case Type::FloatTyID:   return 'F';
+  case Type::DoubleTyID:  return 'D';
+  case Type::PointerTyID: return 'P';
+  case Type::FunctionTyID:return 'M';
+  case Type::StructTyID:  return 'T';
+  case Type::ArrayTyID:   return 'A';
+  case Type::OpaqueTyID:  return 'O';
+  default: return 'U';
+  }
+}
+
+static ExFunc lookupFunction(const Function *F) {
+  // Function not found, look it up... start by figuring out what the
+  // composite function name should be.
+  std::string ExtName = "lle_";
+  const FunctionType *FT = F->getFunctionType();
+  for (unsigned i = 0, e = FT->getNumContainedTypes(); i != e; ++i)
+    ExtName += getTypeID(FT->getContainedType(i));
+  ExtName += "_" + F->getName();
+
+  ExFunc FnPtr = FuncNames[ExtName];
+  if (FnPtr == 0)
+    FnPtr = FuncNames["lle_X_"+F->getName()];
+  if (FnPtr == 0)  // Try calling a generic function... if it exists...
+    FnPtr = (ExFunc)(intptr_t)sys::DynamicLibrary::SearchForAddressOfSymbol(
+            ("lle_X_"+F->getName()).c_str());
+  if (FnPtr == 0)
+    FnPtr = (ExFunc)(intptr_t)
+      sys::DynamicLibrary::SearchForAddressOfSymbol(F->getName());
+  if (FnPtr != 0)
+    Functions.insert(std::make_pair(F, FnPtr));  // Cache for later
+  return FnPtr;
+}
+
+GenericValue Interpreter::callExternalFunction(Function *F,
+                                     const std::vector<GenericValue> &ArgVals) {
+  TheInterpreter = this;
+
+  // Do a lookup to see if the function is in our cache... this should just be a
+  // deferred annotation!
+  std::map<const Function *, ExFunc>::iterator FI = Functions.find(F);
+  ExFunc Fn = (FI == Functions.end()) ? lookupFunction(F) : FI->second;
+  if (Fn == 0) {
+    cerr << "Tried to execute an unknown external function: "
+         << F->getType()->getDescription() << " " << F->getName() << "\n";
+    if (F->getName() == "__main")
+      return GenericValue();
+    abort();
+  }
+
+  // TODO: FIXME when types are not const!
+  GenericValue Result = Fn(const_cast<FunctionType*>(F->getFunctionType()),
+                           ArgVals);
+  return Result;
+}
+
+
+//===----------------------------------------------------------------------===//
+//  Functions "exported" to the running application...
+//
+extern "C" {  // Don't add C++ manglings to llvm mangling :)
+
+// void putchar(ubyte)
+GenericValue lle_X_putchar(FunctionType *FT, const vector<GenericValue> &Args){
+  cout << ((char)Args[0].IntVal.getZExtValue()) << std::flush;
+  return Args[0];
+}
+
+// void _IO_putc(int c, FILE* fp)
+GenericValue lle_X__IO_putc(FunctionType *FT, const vector<GenericValue> &Args){
+#ifdef __linux__
+  _IO_putc((char)Args[0].IntVal.getZExtValue(), (FILE*) Args[1].PointerVal);
+#else
+  assert(0 && "Can't call _IO_putc on this platform");
+#endif
+  return Args[0];
+}
+
+// void atexit(Function*)
+GenericValue lle_X_atexit(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 1);
+  TheInterpreter->addAtExitHandler((Function*)GVTOP(Args[0]));
+  GenericValue GV;
+  GV.IntVal = 0;
+  return GV;
+}
+
+// void exit(int)
+GenericValue lle_X_exit(FunctionType *FT, const vector<GenericValue> &Args) {
+  TheInterpreter->exitCalled(Args[0]);
+  return GenericValue();
+}
+
+// void abort(void)
+GenericValue lle_X_abort(FunctionType *FT, const vector<GenericValue> &Args) {
+  raise (SIGABRT);
+  return GenericValue();
+}
+
+// void *malloc(uint)
+GenericValue lle_X_malloc(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 1 && "Malloc expects one argument!");
+  assert(isa<PointerType>(FT->getReturnType()) && "malloc must return pointer");
+  return PTOGV(malloc(Args[0].IntVal.getZExtValue()));
+}
+
+// void *calloc(uint, uint)
+GenericValue lle_X_calloc(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 2 && "calloc expects two arguments!");
+  assert(isa<PointerType>(FT->getReturnType()) && "calloc must return pointer");
+  return PTOGV(calloc(Args[0].IntVal.getZExtValue(), 
+                      Args[1].IntVal.getZExtValue()));
+}
+
+// void *calloc(uint, uint)
+GenericValue lle_X_realloc(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 2 && "calloc expects two arguments!");
+  assert(isa<PointerType>(FT->getReturnType()) &&"realloc must return pointer");
+  return PTOGV(realloc(GVTOP(Args[0]), Args[1].IntVal.getZExtValue()));
+}
+
+// void free(void *)
+GenericValue lle_X_free(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 1);
+  free(GVTOP(Args[0]));
+  return GenericValue();
+}
+
+// int atoi(char *)
+GenericValue lle_X_atoi(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 1);
+  GenericValue GV;
+  GV.IntVal = APInt(32, atoi((char*)GVTOP(Args[0])));
+  return GV;
+}
+
+// double pow(double, double)
+GenericValue lle_X_pow(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 2);
+  GenericValue GV;
+  GV.DoubleVal = pow(Args[0].DoubleVal, Args[1].DoubleVal);
+  return GV;
+}
+
+// double exp(double)
+GenericValue lle_X_exp(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 1);
+  GenericValue GV;
+  GV.DoubleVal = exp(Args[0].DoubleVal);
+  return GV;
+}
+
+// double sqrt(double)
+GenericValue lle_X_sqrt(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 1);
+  GenericValue GV;
+  GV.DoubleVal = sqrt(Args[0].DoubleVal);
+  return GV;
+}
+
+// double log(double)
+GenericValue lle_X_log(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 1);
+  GenericValue GV;
+  GV.DoubleVal = log(Args[0].DoubleVal);
+  return GV;
+}
+
+// double floor(double)
+GenericValue lle_X_floor(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 1);
+  GenericValue GV;
+  GV.DoubleVal = floor(Args[0].DoubleVal);
+  return GV;
+}
+
+#ifdef HAVE_RAND48
+
+// double drand48()
+GenericValue lle_X_drand48(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 0);
+  GenericValue GV;
+  GV.DoubleVal = drand48();
+  return GV;
+}
+
+// long lrand48()
+GenericValue lle_X_lrand48(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 0);
+  GenericValue GV;
+  GV.Int32Val = lrand48();
+  return GV;
+}
+
+// void srand48(long)
+GenericValue lle_X_srand48(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 1);
+  srand48(Args[0].Int32Val);
+  return GenericValue();
+}
+
+#endif
+
+// int rand()
+GenericValue lle_X_rand(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 0);
+  GenericValue GV;
+  GV.IntVal = APInt(32, rand());
+  return GV;
+}
+
+// void srand(uint)
+GenericValue lle_X_srand(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 1);
+  srand(Args[0].IntVal.getZExtValue());
+  return GenericValue();
+}
+
+// int puts(const char*)
+GenericValue lle_X_puts(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 1);
+  GenericValue GV;
+  GV.IntVal = APInt(32, puts((char*)GVTOP(Args[0])));
+  return GV;
+}
+
+// int sprintf(sbyte *, sbyte *, ...) - a very rough implementation to make
+// output useful.
+GenericValue lle_X_sprintf(FunctionType *FT, const vector<GenericValue> &Args) {
+  char *OutputBuffer = (char *)GVTOP(Args[0]);
+  const char *FmtStr = (const char *)GVTOP(Args[1]);
+  unsigned ArgNo = 2;
+
+  // printf should return # chars printed.  This is completely incorrect, but
+  // close enough for now.
+  GenericValue GV; 
+  GV.IntVal = APInt(32, strlen(FmtStr));
+  while (1) {
+    switch (*FmtStr) {
+    case 0: return GV;             // Null terminator...
+    default:                       // Normal nonspecial character
+      sprintf(OutputBuffer++, "%c", *FmtStr++);
+      break;
+    case '\\': {                   // Handle escape codes
+      sprintf(OutputBuffer, "%c%c", *FmtStr, *(FmtStr+1));
+      FmtStr += 2; OutputBuffer += 2;
+      break;
+    }
+    case '%': {                    // Handle format specifiers
+      char FmtBuf[100] = "", Buffer[1000] = "";
+      char *FB = FmtBuf;
+      *FB++ = *FmtStr++;
+      char Last = *FB++ = *FmtStr++;
+      unsigned HowLong = 0;
+      while (Last != 'c' && Last != 'd' && Last != 'i' && Last != 'u' &&
+             Last != 'o' && Last != 'x' && Last != 'X' && Last != 'e' &&
+             Last != 'E' && Last != 'g' && Last != 'G' && Last != 'f' &&
+             Last != 'p' && Last != 's' && Last != '%') {
+        if (Last == 'l' || Last == 'L') HowLong++;  // Keep track of l's
+        Last = *FB++ = *FmtStr++;
+      }
+      *FB = 0;
+
+      switch (Last) {
+      case '%':
+        sprintf(Buffer, FmtBuf); break;
+      case 'c':
+        sprintf(Buffer, FmtBuf, uint32_t(Args[ArgNo++].IntVal.getZExtValue()));
+        break;
+      case 'd': case 'i':
+      case 'u': case 'o':
+      case 'x': case 'X':
+        if (HowLong >= 1) {
+          if (HowLong == 1 &&
+              TheInterpreter->getTargetData()->getPointerSizeInBits() == 64 &&
+              sizeof(long) < sizeof(int64_t)) {
+            // Make sure we use %lld with a 64 bit argument because we might be
+            // compiling LLI on a 32 bit compiler.
+            unsigned Size = strlen(FmtBuf);
+            FmtBuf[Size] = FmtBuf[Size-1];
+            FmtBuf[Size+1] = 0;
+            FmtBuf[Size-1] = 'l';
+          }
+          sprintf(Buffer, FmtBuf, Args[ArgNo++].IntVal.getZExtValue());
+        } else
+          sprintf(Buffer, FmtBuf,uint32_t(Args[ArgNo++].IntVal.getZExtValue()));
+        break;
+      case 'e': case 'E': case 'g': case 'G': case 'f':
+        sprintf(Buffer, FmtBuf, Args[ArgNo++].DoubleVal); break;
+      case 'p':
+        sprintf(Buffer, FmtBuf, (void*)GVTOP(Args[ArgNo++])); break;
+      case 's':
+        sprintf(Buffer, FmtBuf, (char*)GVTOP(Args[ArgNo++])); break;
+      default:  cerr << "<unknown printf code '" << *FmtStr << "'!>";
+        ArgNo++; break;
+      }
+      strcpy(OutputBuffer, Buffer);
+      OutputBuffer += strlen(Buffer);
+      }
+      break;
+    }
+  }
+  return GV;
+}
+
+// int printf(sbyte *, ...) - a very rough implementation to make output useful.
+GenericValue lle_X_printf(FunctionType *FT, const vector<GenericValue> &Args) {
+  char Buffer[10000];
+  vector<GenericValue> NewArgs;
+  NewArgs.push_back(PTOGV((void*)&Buffer[0]));
+  NewArgs.insert(NewArgs.end(), Args.begin(), Args.end());
+  GenericValue GV = lle_X_sprintf(FT, NewArgs);
+  cout << Buffer;
+  return GV;
+}
+
+static void ByteswapSCANFResults(const char *Fmt, void *Arg0, void *Arg1,
+                                 void *Arg2, void *Arg3, void *Arg4, void *Arg5,
+                                 void *Arg6, void *Arg7, void *Arg8) {
+  void *Args[] = { Arg0, Arg1, Arg2, Arg3, Arg4, Arg5, Arg6, Arg7, Arg8, 0 };
+
+  // Loop over the format string, munging read values as appropriate (performs
+  // byteswaps as necessary).
+  unsigned ArgNo = 0;
+  while (*Fmt) {
+    if (*Fmt++ == '%') {
+      // Read any flag characters that may be present...
+      bool Suppress = false;
+      bool Half = false;
+      bool Long = false;
+      bool LongLong = false;  // long long or long double
+
+      while (1) {
+        switch (*Fmt++) {
+        case '*': Suppress = true; break;
+        case 'a': /*Allocate = true;*/ break;  // We don't need to track this
+        case 'h': Half = true; break;
+        case 'l': Long = true; break;
+        case 'q':
+        case 'L': LongLong = true; break;
+        default:
+          if (Fmt[-1] > '9' || Fmt[-1] < '0')   // Ignore field width specs
+            goto Out;
+        }
+      }
+    Out:
+
+      // Read the conversion character
+      if (!Suppress && Fmt[-1] != '%') { // Nothing to do?
+        unsigned Size = 0;
+        const Type *Ty = 0;
+
+        switch (Fmt[-1]) {
+        case 'i': case 'o': case 'u': case 'x': case 'X': case 'n': case 'p':
+        case 'd':
+          if (Long || LongLong) {
+            Size = 8; Ty = Type::Int64Ty;
+          } else if (Half) {
+            Size = 4; Ty = Type::Int16Ty;
+          } else {
+            Size = 4; Ty = Type::Int32Ty;
+          }
+          break;
+
+        case 'e': case 'g': case 'E':
+        case 'f':
+          if (Long || LongLong) {
+            Size = 8; Ty = Type::DoubleTy;
+          } else {
+            Size = 4; Ty = Type::FloatTy;
+          }
+          break;
+
+        case 's': case 'c': case '[':  // No byteswap needed
+          Size = 1;
+          Ty = Type::Int8Ty;
+          break;
+
+        default: break;
+        }
+
+        if (Size) {
+          GenericValue GV;
+          void *Arg = Args[ArgNo++];
+          memcpy(&GV, Arg, Size);
+          TheInterpreter->StoreValueToMemory(GV, (GenericValue*)Arg, Ty);
+        }
+      }
+    }
+  }
+}
+
+// int sscanf(const char *format, ...);
+GenericValue lle_X_sscanf(FunctionType *FT, const vector<GenericValue> &args) {
+  assert(args.size() < 10 && "Only handle up to 10 args to sscanf right now!");
+
+  char *Args[10];
+  for (unsigned i = 0; i < args.size(); ++i)
+    Args[i] = (char*)GVTOP(args[i]);
+
+  GenericValue GV;
+  GV.IntVal = APInt(32, sscanf(Args[0], Args[1], Args[2], Args[3], Args[4],
+                        Args[5], Args[6], Args[7], Args[8], Args[9]));
+  ByteswapSCANFResults(Args[1], Args[2], Args[3], Args[4],
+                       Args[5], Args[6], Args[7], Args[8], Args[9], 0);
+  return GV;
+}
+
+// int scanf(const char *format, ...);
+GenericValue lle_X_scanf(FunctionType *FT, const vector<GenericValue> &args) {
+  assert(args.size() < 10 && "Only handle up to 10 args to scanf right now!");
+
+  char *Args[10];
+  for (unsigned i = 0; i < args.size(); ++i)
+    Args[i] = (char*)GVTOP(args[i]);
+
+  GenericValue GV;
+  GV.IntVal = APInt(32, scanf( Args[0], Args[1], Args[2], Args[3], Args[4],
+                        Args[5], Args[6], Args[7], Args[8], Args[9]));
+  ByteswapSCANFResults(Args[0], Args[1], Args[2], Args[3], Args[4],
+                       Args[5], Args[6], Args[7], Args[8], Args[9]);
+  return GV;
+}
+
+
+// int clock(void) - Profiling implementation
+GenericValue lle_i_clock(FunctionType *FT, const vector<GenericValue> &Args) {
+  extern unsigned int clock(void);
+  GenericValue GV; 
+  GV.IntVal = APInt(32, clock());
+  return GV;
+}
+
+
+//===----------------------------------------------------------------------===//
+// String Functions...
+//===----------------------------------------------------------------------===//
+
+// int strcmp(const char *S1, const char *S2);
+GenericValue lle_X_strcmp(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 2);
+  GenericValue Ret;
+  Ret.IntVal = APInt(32, strcmp((char*)GVTOP(Args[0]), (char*)GVTOP(Args[1])));
+  return Ret;
+}
+
+// char *strcat(char *Dest, const char *src);
+GenericValue lle_X_strcat(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 2);
+  assert(isa<PointerType>(FT->getReturnType()) &&"strcat must return pointer");
+  return PTOGV(strcat((char*)GVTOP(Args[0]), (char*)GVTOP(Args[1])));
+}
+
+// char *strcpy(char *Dest, const char *src);
+GenericValue lle_X_strcpy(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 2);
+  assert(isa<PointerType>(FT->getReturnType()) &&"strcpy must return pointer");
+  return PTOGV(strcpy((char*)GVTOP(Args[0]), (char*)GVTOP(Args[1])));
+}
+
+static GenericValue size_t_to_GV (size_t n) {
+  GenericValue Ret;
+  if (sizeof (size_t) == sizeof (uint64_t)) {
+    Ret.IntVal = APInt(64, n);
+  } else {
+    assert (sizeof (size_t) == sizeof (unsigned int));
+    Ret.IntVal = APInt(32, n);
+  }
+  return Ret;
+}
+
+static size_t GV_to_size_t (GenericValue GV) {
+  size_t count;
+  if (sizeof (size_t) == sizeof (uint64_t)) {
+    count = (size_t)GV.IntVal.getZExtValue();
+  } else {
+    assert (sizeof (size_t) == sizeof (unsigned int));
+    count = (size_t)GV.IntVal.getZExtValue();
+  }
+  return count;
+}
+
+// size_t strlen(const char *src);
+GenericValue lle_X_strlen(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 1);
+  size_t strlenResult = strlen ((char *) GVTOP (Args[0]));
+  return size_t_to_GV (strlenResult);
+}
+
+// char *strdup(const char *src);
+GenericValue lle_X_strdup(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 1);
+  assert(isa<PointerType>(FT->getReturnType()) && "strdup must return pointer");
+  return PTOGV(strdup((char*)GVTOP(Args[0])));
+}
+
+// char *__strdup(const char *src);
+GenericValue lle_X___strdup(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 1);
+  assert(isa<PointerType>(FT->getReturnType()) &&"_strdup must return pointer");
+  return PTOGV(strdup((char*)GVTOP(Args[0])));
+}
+
+// void *memset(void *S, int C, size_t N)
+GenericValue lle_X_memset(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 3);
+  size_t count = GV_to_size_t (Args[2]);
+  assert(isa<PointerType>(FT->getReturnType()) && "memset must return pointer");
+  return PTOGV(memset(GVTOP(Args[0]), uint32_t(Args[1].IntVal.getZExtValue()), 
+                      count));
+}
+
+// void *memcpy(void *Dest, void *src, size_t Size);
+GenericValue lle_X_memcpy(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 3);
+  assert(isa<PointerType>(FT->getReturnType()) && "memcpy must return pointer");
+  size_t count = GV_to_size_t (Args[2]);
+  return PTOGV(memcpy((char*)GVTOP(Args[0]), (char*)GVTOP(Args[1]), count));
+}
+
+//===----------------------------------------------------------------------===//
+// IO Functions...
+//===----------------------------------------------------------------------===//
+
+// getFILE - Turn a pointer in the host address space into a legit pointer in
+// the interpreter address space.  This is an identity transformation.
+#define getFILE(ptr) ((FILE*)ptr)
+
+// FILE *fopen(const char *filename, const char *mode);
+GenericValue lle_X_fopen(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 2);
+  assert(isa<PointerType>(FT->getReturnType()) && "fopen must return pointer");
+  return PTOGV(fopen((const char *)GVTOP(Args[0]),
+                     (const char *)GVTOP(Args[1])));
+}
+
+// int fclose(FILE *F);
+GenericValue lle_X_fclose(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 1);
+  GenericValue GV;
+  GV.IntVal = APInt(32, fclose(getFILE(GVTOP(Args[0]))));
+  return GV;
+}
+
+// int feof(FILE *stream);
+GenericValue lle_X_feof(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 1);
+  GenericValue GV;
+
+  GV.IntVal = APInt(32, feof(getFILE(GVTOP(Args[0]))));
+  return GV;
+}
+
+// size_t fread(void *ptr, size_t size, size_t nitems, FILE *stream);
+GenericValue lle_X_fread(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 4);
+  size_t result;
+
+  result = fread((void*)GVTOP(Args[0]), GV_to_size_t (Args[1]),
+                 GV_to_size_t (Args[2]), getFILE(GVTOP(Args[3])));
+  return size_t_to_GV (result);
+}
+
+// size_t fwrite(const void *ptr, size_t size, size_t nitems, FILE *stream);
+GenericValue lle_X_fwrite(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 4);
+  size_t result;
+
+  result = fwrite((void*)GVTOP(Args[0]), GV_to_size_t (Args[1]),
+                  GV_to_size_t (Args[2]), getFILE(GVTOP(Args[3])));
+  return size_t_to_GV (result);
+}
+
+// char *fgets(char *s, int n, FILE *stream);
+GenericValue lle_X_fgets(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 3);
+  return GVTOP(fgets((char*)GVTOP(Args[0]), Args[1].IntVal.getZExtValue(),
+                     getFILE(GVTOP(Args[2]))));
+}
+
+// FILE *freopen(const char *path, const char *mode, FILE *stream);
+GenericValue lle_X_freopen(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 3);
+  assert(isa<PointerType>(FT->getReturnType()) &&"freopen must return pointer");
+  return PTOGV(freopen((char*)GVTOP(Args[0]), (char*)GVTOP(Args[1]),
+                       getFILE(GVTOP(Args[2]))));
+}
+
+// int fflush(FILE *stream);
+GenericValue lle_X_fflush(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 1);
+  GenericValue GV;
+  GV.IntVal = APInt(32, fflush(getFILE(GVTOP(Args[0]))));
+  return GV;
+}
+
+// int getc(FILE *stream);
+GenericValue lle_X_getc(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 1);
+  GenericValue GV;
+  GV.IntVal = APInt(32, getc(getFILE(GVTOP(Args[0]))));
+  return GV;
+}
+
+// int _IO_getc(FILE *stream);
+GenericValue lle_X__IO_getc(FunctionType *F, const vector<GenericValue> &Args) {
+  return lle_X_getc(F, Args);
+}
+
+// int fputc(int C, FILE *stream);
+GenericValue lle_X_fputc(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 2);
+  GenericValue GV;
+  GV.IntVal = APInt(32, fputc(Args[0].IntVal.getZExtValue(), 
+                              getFILE(GVTOP(Args[1]))));
+  return GV;
+}
+
+// int ungetc(int C, FILE *stream);
+GenericValue lle_X_ungetc(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 2);
+  GenericValue GV;
+  GV.IntVal = APInt(32, ungetc(Args[0].IntVal.getZExtValue(), 
+                               getFILE(GVTOP(Args[1]))));
+  return GV;
+}
+
+// int ferror (FILE *stream);
+GenericValue lle_X_ferror(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() == 1);
+  GenericValue GV;
+  GV.IntVal = APInt(32, ferror (getFILE(GVTOP(Args[0]))));
+  return GV;
+}
+
+// int fprintf(FILE *,sbyte *, ...) - a very rough implementation to make output
+// useful.
+GenericValue lle_X_fprintf(FunctionType *FT, const vector<GenericValue> &Args) {
+  assert(Args.size() >= 2);
+  char Buffer[10000];
+  vector<GenericValue> NewArgs;
+  NewArgs.push_back(PTOGV(Buffer));
+  NewArgs.insert(NewArgs.end(), Args.begin()+1, Args.end());
+  GenericValue GV = lle_X_sprintf(FT, NewArgs);
+
+  fputs(Buffer, getFILE(GVTOP(Args[0])));
+  return GV;
+}
+
+} // End extern "C"
+
+
+void Interpreter::initializeExternalFunctions() {
+  FuncNames["lle_X_putchar"]      = lle_X_putchar;
+  FuncNames["lle_X__IO_putc"]     = lle_X__IO_putc;
+  FuncNames["lle_X_exit"]         = lle_X_exit;
+  FuncNames["lle_X_abort"]        = lle_X_abort;
+  FuncNames["lle_X_malloc"]       = lle_X_malloc;
+  FuncNames["lle_X_calloc"]       = lle_X_calloc;
+  FuncNames["lle_X_realloc"]      = lle_X_realloc;
+  FuncNames["lle_X_free"]         = lle_X_free;
+  FuncNames["lle_X_atoi"]         = lle_X_atoi;
+  FuncNames["lle_X_pow"]          = lle_X_pow;
+  FuncNames["lle_X_exp"]          = lle_X_exp;
+  FuncNames["lle_X_log"]          = lle_X_log;
+  FuncNames["lle_X_floor"]        = lle_X_floor;
+  FuncNames["lle_X_srand"]        = lle_X_srand;
+  FuncNames["lle_X_rand"]         = lle_X_rand;
+#ifdef HAVE_RAND48
+  FuncNames["lle_X_drand48"]      = lle_X_drand48;
+  FuncNames["lle_X_srand48"]      = lle_X_srand48;
+  FuncNames["lle_X_lrand48"]      = lle_X_lrand48;
+#endif
+  FuncNames["lle_X_sqrt"]         = lle_X_sqrt;
+  FuncNames["lle_X_puts"]         = lle_X_puts;
+  FuncNames["lle_X_printf"]       = lle_X_printf;
+  FuncNames["lle_X_sprintf"]      = lle_X_sprintf;
+  FuncNames["lle_X_sscanf"]       = lle_X_sscanf;
+  FuncNames["lle_X_scanf"]        = lle_X_scanf;
+  FuncNames["lle_i_clock"]        = lle_i_clock;
+
+  FuncNames["lle_X_strcmp"]       = lle_X_strcmp;
+  FuncNames["lle_X_strcat"]       = lle_X_strcat;
+  FuncNames["lle_X_strcpy"]       = lle_X_strcpy;
+  FuncNames["lle_X_strlen"]       = lle_X_strlen;
+  FuncNames["lle_X___strdup"]     = lle_X___strdup;
+  FuncNames["lle_X_memset"]       = lle_X_memset;
+  FuncNames["lle_X_memcpy"]       = lle_X_memcpy;
+
+  FuncNames["lle_X_fopen"]        = lle_X_fopen;
+  FuncNames["lle_X_fclose"]       = lle_X_fclose;
+  FuncNames["lle_X_feof"]         = lle_X_feof;
+  FuncNames["lle_X_fread"]        = lle_X_fread;
+  FuncNames["lle_X_fwrite"]       = lle_X_fwrite;
+  FuncNames["lle_X_fgets"]        = lle_X_fgets;
+  FuncNames["lle_X_fflush"]       = lle_X_fflush;
+  FuncNames["lle_X_fgetc"]        = lle_X_getc;
+  FuncNames["lle_X_getc"]         = lle_X_getc;
+  FuncNames["lle_X__IO_getc"]     = lle_X__IO_getc;
+  FuncNames["lle_X_fputc"]        = lle_X_fputc;
+  FuncNames["lle_X_ungetc"]       = lle_X_ungetc;
+  FuncNames["lle_X_fprintf"]      = lle_X_fprintf;
+  FuncNames["lle_X_freopen"]      = lle_X_freopen;
+}
+
diff --git a/lib/ExecutionEngine/Interpreter/Interpreter.cpp b/lib/ExecutionEngine/Interpreter/Interpreter.cpp
new file mode 100644
index 0000000..3a156bf
--- /dev/null
+++ b/lib/ExecutionEngine/Interpreter/Interpreter.cpp
@@ -0,0 +1,106 @@
+//===- Interpreter.cpp - Top-Level LLVM Interpreter Implementation --------===//
+//
+//                     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 top-level functionality for the LLVM interpreter.
+// This interpreter is designed to be a very simple, portable, inefficient
+// interpreter.
+//
+//===----------------------------------------------------------------------===//
+
+#include "Interpreter.h"
+#include "llvm/CodeGen/IntrinsicLowering.h"
+#include "llvm/DerivedTypes.h"
+#include "llvm/Module.h"
+#include "llvm/ModuleProvider.h"
+using namespace llvm;
+
+static struct RegisterInterp {
+  RegisterInterp() { Interpreter::Register(); }
+} InterpRegistrator;
+
+namespace llvm {
+  void LinkInInterpreter() {
+  }
+}
+
+/// create - Create a new interpreter object.  This can never fail.
+///
+ExecutionEngine *Interpreter::create(ModuleProvider *MP, std::string* ErrStr) {
+  // Tell this ModuleProvide to materialize and release the module
+  Module *M = MP->releaseModule(ErrStr);
+  if (!M)
+    // We got an error, just return 0
+    return 0;
+
+  // This is a bit nasty, but the ExecutionEngine won't be able to delete the
+  // module due to use/def issues if we don't delete this MP here. Below we
+  // construct a new Interpreter with the Module we just got. This creates a
+  // new ExistingModuleProvider in the EE instance. Consequently, MP is left
+  // dangling and it contains references into the module which cause problems
+  // when the module is deleted via the ExistingModuleProvide via EE.
+  delete MP;
+  
+  return new Interpreter(M);
+}
+
+//===----------------------------------------------------------------------===//
+// Interpreter ctor - Initialize stuff
+//
+Interpreter::Interpreter(Module *M) : ExecutionEngine(M), TD(M) {
+      
+  memset(&ExitValue.Untyped, 0, sizeof(ExitValue.Untyped));
+  setTargetData(&TD);
+  // Initialize the "backend"
+  initializeExecutionEngine();
+  initializeExternalFunctions();
+  emitGlobals();
+
+  IL = new IntrinsicLowering(TD);
+}
+
+Interpreter::~Interpreter() {
+  delete IL;
+}
+
+void Interpreter::runAtExitHandlers () {
+  while (!AtExitHandlers.empty()) {
+    callFunction(AtExitHandlers.back(), std::vector<GenericValue>());
+    AtExitHandlers.pop_back();
+    run();
+  }
+}
+
+/// run - Start execution with the specified function and arguments.
+///
+GenericValue
+Interpreter::runFunction(Function *F,
+                         const std::vector<GenericValue> &ArgValues) {
+  assert (F && "Function *F was null at entry to run()");
+
+  // Try extra hard not to pass extra args to a function that isn't
+  // expecting them.  C programmers frequently bend the rules and
+  // declare main() with fewer parameters than it actually gets
+  // passed, and the interpreter barfs if you pass a function more
+  // parameters than it is declared to take. This does not attempt to
+  // take into account gratuitous differences in declared types,
+  // though.
+  std::vector<GenericValue> ActualArgs;
+  const unsigned ArgCount = F->getFunctionType()->getNumParams();
+  for (unsigned i = 0; i < ArgCount; ++i)
+    ActualArgs.push_back(ArgValues[i]);
+
+  // Set up the function call.
+  callFunction(F, ActualArgs);
+
+  // Start executing the function.
+  run();
+
+  return ExitValue;
+}
+
diff --git a/lib/ExecutionEngine/Interpreter/Interpreter.h b/lib/ExecutionEngine/Interpreter/Interpreter.h
new file mode 100644
index 0000000..3238850
--- /dev/null
+++ b/lib/ExecutionEngine/Interpreter/Interpreter.h
@@ -0,0 +1,241 @@
+//===-- Interpreter.h ------------------------------------------*- 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 header file defines the interpreter structure
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLI_INTERPRETER_H
+#define LLI_INTERPRETER_H
+
+#include "llvm/Function.h"
+#include "llvm/ExecutionEngine/ExecutionEngine.h"
+#include "llvm/ExecutionEngine/GenericValue.h"
+#include "llvm/ADT/APInt.h"
+#include "llvm/Support/InstVisitor.h"
+#include "llvm/Support/CallSite.h"
+#include "llvm/Target/TargetData.h"
+#include "llvm/Support/DataTypes.h"
+
+namespace llvm {
+
+class IntrinsicLowering;
+struct FunctionInfo;
+template<typename T> class generic_gep_type_iterator;
+class ConstantExpr;
+typedef generic_gep_type_iterator<User::const_op_iterator> gep_type_iterator;
+
+
+// AllocaHolder - Object to track all of the blocks of memory allocated by
+// alloca.  When the function returns, this object is popped off the execution
+// stack, which causes the dtor to be run, which frees all the alloca'd memory.
+//
+class AllocaHolder {
+  friend class AllocaHolderHandle;
+  std::vector<void*> Allocations;
+  unsigned RefCnt;
+public:
+  AllocaHolder() : RefCnt(0) {}
+  void add(void *mem) { Allocations.push_back(mem); }
+  ~AllocaHolder() {
+    for (unsigned i = 0; i < Allocations.size(); ++i)
+      free(Allocations[i]);
+  }
+};
+
+// AllocaHolderHandle gives AllocaHolder value semantics so we can stick it into
+// a vector...
+//
+class AllocaHolderHandle {
+  AllocaHolder *H;
+public:
+  AllocaHolderHandle() : H(new AllocaHolder()) { H->RefCnt++; }
+  AllocaHolderHandle(const AllocaHolderHandle &AH) : H(AH.H) { H->RefCnt++; }
+  ~AllocaHolderHandle() { if (--H->RefCnt == 0) delete H; }
+
+  void add(void *mem) { H->add(mem); }
+};
+
+typedef std::vector<GenericValue> ValuePlaneTy;
+
+// ExecutionContext struct - This struct represents one stack frame currently
+// executing.
+//
+struct ExecutionContext {
+  Function             *CurFunction;// The currently executing function
+  BasicBlock           *CurBB;      // The currently executing BB
+  BasicBlock::iterator  CurInst;    // The next instruction to execute
+  std::map<Value *, GenericValue> Values; // LLVM values used in this invocation
+  std::vector<GenericValue>  VarArgs; // Values passed through an ellipsis
+  CallSite             Caller;     // Holds the call that called subframes.
+                                   // NULL if main func or debugger invoked fn
+  AllocaHolderHandle    Allocas;    // Track memory allocated by alloca
+};
+
+// Interpreter - This class represents the entirety of the interpreter.
+//
+class Interpreter : public ExecutionEngine, public InstVisitor<Interpreter> {
+  GenericValue ExitValue;          // The return value of the called function
+  TargetData TD;
+  IntrinsicLowering *IL;
+
+  // The runtime stack of executing code.  The top of the stack is the current
+  // function record.
+  std::vector<ExecutionContext> ECStack;
+
+  // AtExitHandlers - List of functions to call when the program exits,
+  // registered with the atexit() library function.
+  std::vector<Function*> AtExitHandlers;
+
+public:
+  Interpreter(Module *M);
+  ~Interpreter();
+
+  /// runAtExitHandlers - Run any functions registered by the program's calls to
+  /// atexit(3), which we intercept and store in AtExitHandlers.
+  ///
+  void runAtExitHandlers();
+
+  static void Register() {
+    InterpCtor = create;
+  }
+  
+  /// create - Create an interpreter ExecutionEngine. This can never fail.
+  ///
+  static ExecutionEngine *create(ModuleProvider *M, std::string *ErrorStr = 0);
+
+  /// run - Start execution with the specified function and arguments.
+  ///
+  virtual GenericValue runFunction(Function *F,
+                                   const std::vector<GenericValue> &ArgValues);
+
+  /// recompileAndRelinkFunction - For the interpreter, functions are always
+  /// up-to-date.
+  ///
+  virtual void *recompileAndRelinkFunction(Function *F) {
+    return getPointerToFunction(F);
+  }
+
+  /// freeMachineCodeForFunction - The interpreter does not generate any code.
+  ///
+  void freeMachineCodeForFunction(Function *F) { }
+
+  // Methods used to execute code:
+  // Place a call on the stack
+  void callFunction(Function *F, const std::vector<GenericValue> &ArgVals);
+  void run();                // Execute instructions until nothing left to do
+
+  // Opcode Implementations
+  void visitReturnInst(ReturnInst &I);
+  void visitBranchInst(BranchInst &I);
+  void visitSwitchInst(SwitchInst &I);
+
+  void visitBinaryOperator(BinaryOperator &I);
+  void visitICmpInst(ICmpInst &I);
+  void visitFCmpInst(FCmpInst &I);
+  void visitAllocationInst(AllocationInst &I);
+  void visitFreeInst(FreeInst &I);
+  void visitLoadInst(LoadInst &I);
+  void visitStoreInst(StoreInst &I);
+  void visitGetElementPtrInst(GetElementPtrInst &I);
+  void visitPHINode(PHINode &PN) { assert(0 && "PHI nodes already handled!"); }
+  void visitTruncInst(TruncInst &I);
+  void visitZExtInst(ZExtInst &I);
+  void visitSExtInst(SExtInst &I);
+  void visitFPTruncInst(FPTruncInst &I);
+  void visitFPExtInst(FPExtInst &I);
+  void visitUIToFPInst(UIToFPInst &I);
+  void visitSIToFPInst(SIToFPInst &I);
+  void visitFPToUIInst(FPToUIInst &I);
+  void visitFPToSIInst(FPToSIInst &I);
+  void visitPtrToIntInst(PtrToIntInst &I);
+  void visitIntToPtrInst(IntToPtrInst &I);
+  void visitBitCastInst(BitCastInst &I);
+  void visitSelectInst(SelectInst &I);
+
+
+  void visitCallSite(CallSite CS);
+  void visitCallInst(CallInst &I) { visitCallSite (CallSite (&I)); }
+  void visitInvokeInst(InvokeInst &I) { visitCallSite (CallSite (&I)); }
+  void visitUnwindInst(UnwindInst &I);
+  void visitUnreachableInst(UnreachableInst &I);
+
+  void visitShl(BinaryOperator &I);
+  void visitLShr(BinaryOperator &I);
+  void visitAShr(BinaryOperator &I);
+
+  void visitVAArgInst(VAArgInst &I);
+  void visitInstruction(Instruction &I) {
+    cerr << I;
+    assert(0 && "Instruction not interpretable yet!");
+  }
+
+  GenericValue callExternalFunction(Function *F,
+                                    const std::vector<GenericValue> &ArgVals);
+  void exitCalled(GenericValue GV);
+
+  void addAtExitHandler(Function *F) {
+    AtExitHandlers.push_back(F);
+  }
+
+  GenericValue *getFirstVarArg () {
+    return &(ECStack.back ().VarArgs[0]);
+  }
+
+  //FIXME: private:
+public:
+  GenericValue executeGEPOperation(Value *Ptr, gep_type_iterator I,
+                                   gep_type_iterator E, ExecutionContext &SF);
+
+private:  // Helper functions
+  // SwitchToNewBasicBlock - Start execution in a new basic block and run any
+  // PHI nodes in the top of the block.  This is used for intraprocedural
+  // control flow.
+  //
+  void SwitchToNewBasicBlock(BasicBlock *Dest, ExecutionContext &SF);
+
+  void *getPointerToFunction(Function *F) { return (void*)F; }
+
+  void initializeExecutionEngine();
+  void initializeExternalFunctions();
+  GenericValue getConstantExprValue(ConstantExpr *CE, ExecutionContext &SF);
+  GenericValue getOperandValue(Value *V, ExecutionContext &SF);
+  GenericValue executeTruncInst(Value *SrcVal, const Type *DstTy,
+                                ExecutionContext &SF);
+  GenericValue executeSExtInst(Value *SrcVal, const Type *DstTy,
+                               ExecutionContext &SF);
+  GenericValue executeZExtInst(Value *SrcVal, const Type *DstTy,
+                               ExecutionContext &SF);
+  GenericValue executeFPTruncInst(Value *SrcVal, const Type *DstTy,
+                                  ExecutionContext &SF);
+  GenericValue executeFPExtInst(Value *SrcVal, const Type *DstTy,
+                                ExecutionContext &SF);
+  GenericValue executeFPToUIInst(Value *SrcVal, const Type *DstTy,
+                                 ExecutionContext &SF);
+  GenericValue executeFPToSIInst(Value *SrcVal, const Type *DstTy,
+                                 ExecutionContext &SF);
+  GenericValue executeUIToFPInst(Value *SrcVal, const Type *DstTy,
+                                 ExecutionContext &SF);
+  GenericValue executeSIToFPInst(Value *SrcVal, const Type *DstTy,
+                                 ExecutionContext &SF);
+  GenericValue executePtrToIntInst(Value *SrcVal, const Type *DstTy,
+                                   ExecutionContext &SF);
+  GenericValue executeIntToPtrInst(Value *SrcVal, const Type *DstTy,
+                                   ExecutionContext &SF);
+  GenericValue executeBitCastInst(Value *SrcVal, const Type *DstTy,
+                                  ExecutionContext &SF);
+  GenericValue executeCastOperation(Instruction::CastOps opcode, Value *SrcVal, 
+                                    const Type *Ty, ExecutionContext &SF);
+  void popStackAndReturnValueToCaller(const Type *RetTy, GenericValue Result);
+
+};
+
+} // End llvm namespace
+
+#endif
diff --git a/lib/ExecutionEngine/Interpreter/Makefile b/lib/ExecutionEngine/Interpreter/Makefile
new file mode 100644
index 0000000..93f74ec
--- /dev/null
+++ b/lib/ExecutionEngine/Interpreter/Makefile
@@ -0,0 +1,12 @@
+##===- lib/ExecutionEngine/Interpreter/Makefile ------------*- Makefile -*-===##
+# 
+#                     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.
+# 
+##===----------------------------------------------------------------------===##
+LEVEL = ../../..
+LIBRARYNAME = LLVMInterpreter
+
+include $(LEVEL)/Makefile.common
diff --git a/lib/ExecutionEngine/JIT/Intercept.cpp b/lib/ExecutionEngine/JIT/Intercept.cpp
new file mode 100644
index 0000000..f370e5b
--- /dev/null
+++ b/lib/ExecutionEngine/JIT/Intercept.cpp
@@ -0,0 +1,121 @@
+//===-- Intercept.cpp - System function interception routines -------------===//
+//
+//                     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.
+//
+//===----------------------------------------------------------------------===//
+//
+// If a function call occurs to an external function, the JIT is designed to use
+// the dynamic loader interface to find a function to call.  This is useful for
+// calling system calls and library functions that are not available in LLVM.
+// Some system calls, however, need to be handled specially.  For this reason,
+// we intercept some of them here and use our own stubs to handle them.
+//
+//===----------------------------------------------------------------------===//
+
+#include "JIT.h"
+#include "llvm/System/DynamicLibrary.h"
+#include "llvm/Config/config.h"
+using namespace llvm;
+
+// AtExitHandlers - List of functions to call when the program exits,
+// registered with the atexit() library function.
+static std::vector<void (*)()> AtExitHandlers;
+
+/// runAtExitHandlers - Run any functions registered by the program's
+/// calls to atexit(3), which we intercept and store in
+/// AtExitHandlers.
+///
+static void runAtExitHandlers() {
+  while (!AtExitHandlers.empty()) {
+    void (*Fn)() = AtExitHandlers.back();
+    AtExitHandlers.pop_back();
+    Fn();
+  }
+}
+
+//===----------------------------------------------------------------------===//
+// Function stubs that are invoked instead of certain library calls
+//===----------------------------------------------------------------------===//
+
+// Force the following functions to be linked in to anything that uses the
+// JIT. This is a hack designed to work around the all-too-clever Glibc
+// strategy of making these functions work differently when inlined vs. when
+// not inlined, and hiding their real definitions in a separate archive file
+// that the dynamic linker can't see. For more info, search for
+// 'libc_nonshared.a' on Google, or read http://llvm.org/PR274.
+#if defined(__linux__)
+#if defined(HAVE_SYS_STAT_H)
+#include <sys/stat.h>
+#endif
+void *FunctionPointers[] = {
+  (void *)(intptr_t) stat,
+  (void *)(intptr_t) fstat,
+  (void *)(intptr_t) lstat,
+  (void *)(intptr_t) stat64,
+  (void *)(intptr_t) fstat64,
+  (void *)(intptr_t) lstat64,
+  (void *)(intptr_t) atexit,
+  (void *)(intptr_t) mknod
+};
+#endif // __linux__
+
+// __mainFunc - If the program does not have a linked in __main function, allow
+// it to run, but print a warning.
+static void __mainFunc() {
+  fprintf(stderr, "WARNING: Program called __main but was not linked to "
+          "libcrtend.a.\nThis probably won't hurt anything unless the "
+          "program is written in C++.\n");
+}
+
+// jit_exit - Used to intercept the "exit" library call.
+static void jit_exit(int Status) {
+  runAtExitHandlers();   // Run atexit handlers...
+  exit(Status);
+}
+
+// jit_atexit - Used to intercept the "atexit" library call.
+static int jit_atexit(void (*Fn)(void)) {
+  AtExitHandlers.push_back(Fn);    // Take note of atexit handler...
+  return 0;  // Always successful
+}
+
+//===----------------------------------------------------------------------===//
+//
+/// getPointerToNamedFunction - This method returns the address of the specified
+/// function by using the dynamic loader interface.  As such it is only useful
+/// for resolving library symbols, not code generated symbols.
+///
+void *JIT::getPointerToNamedFunction(const std::string &Name) {
+  // Check to see if this is one of the functions we want to intercept.  Note,
+  // we cast to intptr_t here to silence a -pedantic warning that complains
+  // about casting a function pointer to a normal pointer.
+  if (Name == "exit") return (void*)(intptr_t)&jit_exit;
+  if (Name == "atexit") return (void*)(intptr_t)&jit_atexit;
+
+  // If the program does not have a linked in __main function, allow it to run,
+  // but print a warning.
+  if (Name == "__main") return (void*)(intptr_t)&__mainFunc;
+
+  const char *NameStr = Name.c_str();
+  // If this is an asm specifier, skip the sentinal.
+  if (NameStr[0] == 1) ++NameStr;
+  
+  // If it's an external function, look it up in the process image...
+  void *Ptr = sys::DynamicLibrary::SearchForAddressOfSymbol(NameStr);
+  if (Ptr) return Ptr;
+  
+  // If it wasn't found and if it starts with an underscore ('_') character, and
+  // has an asm specifier, try again without the underscore.
+  if (Name[0] == 1 && NameStr[0] == '_') {
+    Ptr = sys::DynamicLibrary::SearchForAddressOfSymbol(NameStr+1);
+    if (Ptr) return Ptr;
+  }
+
+  cerr << "ERROR: Program used external function '" << Name
+       << "' which could not be resolved!\n";
+  abort();
+  return 0;
+}
diff --git a/lib/ExecutionEngine/JIT/JIT.cpp b/lib/ExecutionEngine/JIT/JIT.cpp
new file mode 100644
index 0000000..603f8ec
--- /dev/null
+++ b/lib/ExecutionEngine/JIT/JIT.cpp
@@ -0,0 +1,369 @@
+//===-- JIT.cpp - LLVM Just in Time Compiler ------------------------------===//
+//
+//                     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 tool implements a just-in-time compiler for LLVM, allowing direct
+// execution of LLVM bitcode in an efficient manner.
+//
+//===----------------------------------------------------------------------===//
+
+#include "JIT.h"
+#include "llvm/Constants.h"
+#include "llvm/DerivedTypes.h"
+#include "llvm/Function.h"
+#include "llvm/GlobalVariable.h"
+#include "llvm/Instructions.h"
+#include "llvm/ModuleProvider.h"
+#include "llvm/CodeGen/MachineCodeEmitter.h"
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/ExecutionEngine/GenericValue.h"
+#include "llvm/Support/MutexGuard.h"
+#include "llvm/System/DynamicLibrary.h"
+#include "llvm/Target/TargetData.h"
+#include "llvm/Target/TargetMachine.h"
+#include "llvm/Target/TargetJITInfo.h"
+using namespace llvm;
+
+#ifdef __APPLE__ 
+#include <AvailabilityMacros.h>
+#if defined(MAC_OS_X_VERSION_10_4) && \
+    ((MAC_OS_X_VERSION_MIN_REQUIRED > MAC_OS_X_VERSION_10_4) || \
+     (MAC_OS_X_VERSION_MIN_REQUIRED == MAC_OS_X_VERSION_10_4 && \
+      __APPLE_CC__ >= 5330))
+// __dso_handle is resolved by Mac OS X dynamic linker.
+extern void *__dso_handle __attribute__ ((__visibility__ ("hidden")));
+#endif
+#endif
+
+static struct RegisterJIT {
+  RegisterJIT() { JIT::Register(); }
+} JITRegistrator;
+
+namespace llvm {
+  void LinkInJIT() {
+  }
+}
+
+JIT::JIT(ModuleProvider *MP, TargetMachine &tm, TargetJITInfo &tji)
+  : ExecutionEngine(MP), TM(tm), TJI(tji), jitstate(MP) {
+  setTargetData(TM.getTargetData());
+
+  // Initialize MCE
+  MCE = createEmitter(*this);
+
+  // Add target data
+  MutexGuard locked(lock);
+  FunctionPassManager &PM = jitstate.getPM(locked);
+  PM.add(new TargetData(*TM.getTargetData()));
+
+  // Turn the machine code intermediate representation into bytes in memory that
+  // may be executed.
+  if (TM.addPassesToEmitMachineCode(PM, *MCE, false /*fast*/)) {
+    cerr << "Target does not support machine code emission!\n";
+    abort();
+  }
+  
+  // Initialize passes.
+  PM.doInitialization();
+}
+
+JIT::~JIT() {
+  delete MCE;
+  delete &TM;
+}
+
+/// run - Start execution with the specified function and arguments.
+///
+GenericValue JIT::runFunction(Function *F,
+                              const std::vector<GenericValue> &ArgValues) {
+  assert(F && "Function *F was null at entry to run()");
+
+  void *FPtr = getPointerToFunction(F);
+  assert(FPtr && "Pointer to fn's code was null after getPointerToFunction");
+  const FunctionType *FTy = F->getFunctionType();
+  const Type *RetTy = FTy->getReturnType();
+
+  assert((FTy->getNumParams() <= ArgValues.size() || FTy->isVarArg()) &&
+         "Too many arguments passed into function!");
+  assert(FTy->getNumParams() == ArgValues.size() &&
+         "This doesn't support passing arguments through varargs (yet)!");
+
+  // Handle some common cases first.  These cases correspond to common `main'
+  // prototypes.
+  if (RetTy == Type::Int32Ty || RetTy == Type::Int32Ty || RetTy == Type::VoidTy) {
+    switch (ArgValues.size()) {
+    case 3:
+      if ((FTy->getParamType(0) == Type::Int32Ty ||
+           FTy->getParamType(0) == Type::Int32Ty) &&
+          isa<PointerType>(FTy->getParamType(1)) &&
+          isa<PointerType>(FTy->getParamType(2))) {
+        int (*PF)(int, char **, const char **) =
+          (int(*)(int, char **, const char **))(intptr_t)FPtr;
+
+        // Call the function.
+        GenericValue rv;
+        rv.IntVal = APInt(32, PF(ArgValues[0].IntVal.getZExtValue(), 
+                                 (char **)GVTOP(ArgValues[1]),
+                                 (const char **)GVTOP(ArgValues[2])));
+        return rv;
+      }
+      break;
+    case 2:
+      if ((FTy->getParamType(0) == Type::Int32Ty ||
+           FTy->getParamType(0) == Type::Int32Ty) &&
+          isa<PointerType>(FTy->getParamType(1))) {
+        int (*PF)(int, char **) = (int(*)(int, char **))(intptr_t)FPtr;
+
+        // Call the function.
+        GenericValue rv;
+        rv.IntVal = APInt(32, PF(ArgValues[0].IntVal.getZExtValue(), 
+                                 (char **)GVTOP(ArgValues[1])));
+        return rv;
+      }
+      break;
+    case 1:
+      if (FTy->getNumParams() == 1 &&
+          (FTy->getParamType(0) == Type::Int32Ty ||
+           FTy->getParamType(0) == Type::Int32Ty)) {
+        GenericValue rv;
+        int (*PF)(int) = (int(*)(int))(intptr_t)FPtr;
+        rv.IntVal = APInt(32, PF(ArgValues[0].IntVal.getZExtValue()));
+        return rv;
+      }
+      break;
+    }
+  }
+
+  // Handle cases where no arguments are passed first.
+  if (ArgValues.empty()) {
+    GenericValue rv;
+    switch (RetTy->getTypeID()) {
+    default: assert(0 && "Unknown return type for function call!");
+    case Type::IntegerTyID: {
+      unsigned BitWidth = cast<IntegerType>(RetTy)->getBitWidth();
+      if (BitWidth == 1)
+        rv.IntVal = APInt(BitWidth, ((bool(*)())(intptr_t)FPtr)());
+      else if (BitWidth <= 8)
+        rv.IntVal = APInt(BitWidth, ((char(*)())(intptr_t)FPtr)());
+      else if (BitWidth <= 16)
+        rv.IntVal = APInt(BitWidth, ((short(*)())(intptr_t)FPtr)());
+      else if (BitWidth <= 32)
+        rv.IntVal = APInt(BitWidth, ((int(*)())(intptr_t)FPtr)());
+      else if (BitWidth <= 64)
+        rv.IntVal = APInt(BitWidth, ((int64_t(*)())(intptr_t)FPtr)());
+      else 
+        assert(0 && "Integer types > 64 bits not supported");
+      return rv;
+    }
+    case Type::VoidTyID:
+      rv.IntVal = APInt(32, ((int(*)())(intptr_t)FPtr)());
+      return rv;
+    case Type::FloatTyID:
+      rv.FloatVal = ((float(*)())(intptr_t)FPtr)();
+      return rv;
+    case Type::DoubleTyID:
+      rv.DoubleVal = ((double(*)())(intptr_t)FPtr)();
+      return rv;
+    case Type::PointerTyID:
+      return PTOGV(((void*(*)())(intptr_t)FPtr)());
+    }
+  }
+
+  // Okay, this is not one of our quick and easy cases.  Because we don't have a
+  // full FFI, we have to codegen a nullary stub function that just calls the
+  // function we are interested in, passing in constants for all of the
+  // arguments.  Make this function and return.
+
+  // First, create the function.
+  FunctionType *STy=FunctionType::get(RetTy, std::vector<const Type*>(), false);
+  Function *Stub = new Function(STy, Function::InternalLinkage, "",
+                                F->getParent());
+
+  // Insert a basic block.
+  BasicBlock *StubBB = new BasicBlock("", Stub);
+
+  // Convert all of the GenericValue arguments over to constants.  Note that we
+  // currently don't support varargs.
+  SmallVector<Value*, 8> Args;
+  for (unsigned i = 0, e = ArgValues.size(); i != e; ++i) {
+    Constant *C = 0;
+    const Type *ArgTy = FTy->getParamType(i);
+    const GenericValue &AV = ArgValues[i];
+    switch (ArgTy->getTypeID()) {
+    default: assert(0 && "Unknown argument type for function call!");
+    case Type::IntegerTyID: C = ConstantInt::get(AV.IntVal); break;
+    case Type::FloatTyID:   C = ConstantFP ::get(ArgTy, AV.FloatVal);  break;
+    case Type::DoubleTyID:  C = ConstantFP ::get(ArgTy, AV.DoubleVal); break;
+    case Type::PointerTyID:
+      void *ArgPtr = GVTOP(AV);
+      if (sizeof(void*) == 4) {
+        C = ConstantInt::get(Type::Int32Ty, (int)(intptr_t)ArgPtr);
+      } else {
+        C = ConstantInt::get(Type::Int64Ty, (intptr_t)ArgPtr);
+      }
+      C = ConstantExpr::getIntToPtr(C, ArgTy);  // Cast the integer to pointer
+      break;
+    }
+    Args.push_back(C);
+  }
+
+  CallInst *TheCall = new CallInst(F, &Args[0], Args.size(), "", StubBB);
+  TheCall->setTailCall();
+  if (TheCall->getType() != Type::VoidTy)
+    new ReturnInst(TheCall, StubBB);             // Return result of the call.
+  else
+    new ReturnInst(StubBB);                      // Just return void.
+
+  // Finally, return the value returned by our nullary stub function.
+  return runFunction(Stub, std::vector<GenericValue>());
+}
+
+/// runJITOnFunction - Run the FunctionPassManager full of
+/// just-in-time compilation passes on F, hopefully filling in
+/// GlobalAddress[F] with the address of F's machine code.
+///
+void JIT::runJITOnFunction(Function *F) {
+  static bool isAlreadyCodeGenerating = false;
+  assert(!isAlreadyCodeGenerating && "Error: Recursive compilation detected!");
+
+  MutexGuard locked(lock);
+
+  // JIT the function
+  isAlreadyCodeGenerating = true;
+  jitstate.getPM(locked).run(*F);
+  isAlreadyCodeGenerating = false;
+
+  // If the function referred to a global variable that had not yet been
+  // emitted, it allocates memory for the global, but doesn't emit it yet.  Emit
+  // all of these globals now.
+  while (!jitstate.getPendingGlobals(locked).empty()) {
+    const GlobalVariable *GV = jitstate.getPendingGlobals(locked).back();
+    jitstate.getPendingGlobals(locked).pop_back();
+    EmitGlobalVariable(GV);
+  }
+}
+
+/// getPointerToFunction - This method is used to get the address of the
+/// specified function, compiling it if neccesary.
+///
+void *JIT::getPointerToFunction(Function *F) {
+  MutexGuard locked(lock);
+
+  if (void *Addr = getPointerToGlobalIfAvailable(F))
+    return Addr;   // Check if function already code gen'd
+
+  // Make sure we read in the function if it exists in this Module.
+  if (F->hasNotBeenReadFromBitcode()) {
+    // Determine the module provider this function is provided by.
+    Module *M = F->getParent();
+    ModuleProvider *MP = 0;
+    for (unsigned i = 0, e = Modules.size(); i != e; ++i) {
+      if (Modules[i]->getModule() == M) {
+        MP = Modules[i];
+        break;
+      }
+    }
+    assert(MP && "Function isn't in a module we know about!");
+    
+    std::string ErrorMsg;
+    if (MP->materializeFunction(F, &ErrorMsg)) {
+      cerr << "Error reading function '" << F->getName()
+           << "' from bitcode file: " << ErrorMsg << "\n";
+      abort();
+    }
+  }
+
+  if (F->isDeclaration()) {
+    void *Addr = getPointerToNamedFunction(F->getName());
+    addGlobalMapping(F, Addr);
+    return Addr;
+  }
+
+  runJITOnFunction(F);
+
+  void *Addr = getPointerToGlobalIfAvailable(F);
+  assert(Addr && "Code generation didn't add function to GlobalAddress table!");
+  return Addr;
+}
+
+/// getOrEmitGlobalVariable - Return the address of the specified global
+/// variable, possibly emitting it to memory if needed.  This is used by the
+/// Emitter.
+void *JIT::getOrEmitGlobalVariable(const GlobalVariable *GV) {
+  MutexGuard locked(lock);
+
+  void *Ptr = getPointerToGlobalIfAvailable(GV);
+  if (Ptr) return Ptr;
+
+  // If the global is external, just remember the address.
+  if (GV->isDeclaration()) {
+#if defined(__APPLE__) && defined(MAC_OS_X_VERSION_10_4) && \
+    ((MAC_OS_X_VERSION_MIN_REQUIRED > MAC_OS_X_VERSION_10_4) || \
+     (MAC_OS_X_VERSION_MIN_REQUIRED == MAC_OS_X_VERSION_10_4 && \
+      __APPLE_CC__ >= 5330))
+    // Apple gcc defaults to -fuse-cxa-atexit (i.e. calls __cxa_atexit instead
+    // of atexit). It passes the address of linker generated symbol __dso_handle
+    // to the function.
+    // This configuration change happened at version 5330.
+    if (GV->getName() == "__dso_handle")
+      return (void*)&__dso_handle;
+#endif
+    Ptr = sys::DynamicLibrary::SearchForAddressOfSymbol(GV->getName().c_str());
+    if (Ptr == 0) {
+      cerr << "Could not resolve external global address: "
+           << GV->getName() << "\n";
+      abort();
+    }
+  } else {
+    // If the global hasn't been emitted to memory yet, allocate space.  We will
+    // actually initialize the global after current function has finished
+    // compilation.
+    const Type *GlobalType = GV->getType()->getElementType();
+    size_t S = getTargetData()->getTypeSize(GlobalType);
+    size_t A = getTargetData()->getPrefTypeAlignment(GlobalType);
+    if (A <= 8) {
+      Ptr = malloc(S);
+    } else {
+      // Allocate S+A bytes of memory, then use an aligned pointer within that
+      // space.
+      Ptr = malloc(S+A);
+      unsigned MisAligned = ((intptr_t)Ptr & (A-1));
+      Ptr = (char*)Ptr + (MisAligned ? (A-MisAligned) : 0);
+    }
+    jitstate.getPendingGlobals(locked).push_back(GV);
+  }
+  addGlobalMapping(GV, Ptr);
+  return Ptr;
+}
+
+
+/// recompileAndRelinkFunction - This method is used to force a function
+/// which has already been compiled, to be compiled again, possibly
+/// after it has been modified. Then the entry to the old copy is overwritten
+/// with a branch to the new copy. If there was no old copy, this acts
+/// just like JIT::getPointerToFunction().
+///
+void *JIT::recompileAndRelinkFunction(Function *F) {
+  void *OldAddr = getPointerToGlobalIfAvailable(F);
+
+  // If it's not already compiled there is no reason to patch it up.
+  if (OldAddr == 0) { return getPointerToFunction(F); }
+
+  // Delete the old function mapping.
+  addGlobalMapping(F, 0);
+
+  // Recodegen the function
+  runJITOnFunction(F);
+
+  // Update state, forward the old function to the new function.
+  void *Addr = getPointerToGlobalIfAvailable(F);
+  assert(Addr && "Code generation didn't add function to GlobalAddress table!");
+  TJI.replaceMachineCodeForFunction(OldAddr, Addr);
+  return Addr;
+}
+
diff --git a/lib/ExecutionEngine/JIT/JIT.h b/lib/ExecutionEngine/JIT/JIT.h
new file mode 100644
index 0000000..5a3d661
--- /dev/null
+++ b/lib/ExecutionEngine/JIT/JIT.h
@@ -0,0 +1,130 @@
+//===-- JIT.h - Class definition for the JIT --------------------*- 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 defines the top-level JIT data structure.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef JIT_H
+#define JIT_H
+
+#include "llvm/ExecutionEngine/ExecutionEngine.h"
+#include "llvm/PassManager.h"
+#include <map>
+
+namespace llvm {
+
+class Function;
+class GlobalValue;
+class Constant;
+class TargetMachine;
+class TargetJITInfo;
+class MachineCodeEmitter;
+
+class JITState {
+private:
+  FunctionPassManager PM;  // Passes to compile a function
+
+  /// PendingGlobals - Global variables which have had memory allocated for them
+  /// while a function was code generated, but which have not been initialized
+  /// yet.
+  std::vector<const GlobalVariable*> PendingGlobals;
+
+public:
+  JITState(ModuleProvider *MP) : PM(MP) {}
+
+  FunctionPassManager &getPM(const MutexGuard &L) {
+    return PM;
+  }
+
+  std::vector<const GlobalVariable*> &getPendingGlobals(const MutexGuard &L) {
+    return PendingGlobals;
+  }
+};
+
+
+class JIT : public ExecutionEngine {
+  TargetMachine &TM;       // The current target we are compiling to
+  TargetJITInfo &TJI;      // The JITInfo for the target we are compiling to
+  MachineCodeEmitter *MCE; // MCE object
+
+  JITState jitstate;
+
+  JIT(ModuleProvider *MP, TargetMachine &tm, TargetJITInfo &tji);
+public:
+  ~JIT();
+
+  static void Register() {
+    JITCtor = create;
+  }
+  
+  /// getJITInfo - Return the target JIT information structure.
+  ///
+  TargetJITInfo &getJITInfo() const { return TJI; }
+
+  /// create - Create an return a new JIT compiler if there is one available
+  /// for the current target.  Otherwise, return null.
+  ///
+  static ExecutionEngine *create(ModuleProvider *MP, std::string* = 0);
+
+  /// run - Start execution with the specified function and arguments.
+  ///
+  virtual GenericValue runFunction(Function *F,
+                                   const std::vector<GenericValue> &ArgValues);
+
+  /// getPointerToNamedFunction - This method returns the address of the
+  /// specified function by using the dlsym function call.  As such it is only
+  /// useful for resolving library symbols, not code generated symbols.
+  ///
+  void *getPointerToNamedFunction(const std::string &Name);
+
+  // CompilationCallback - Invoked the first time that a call site is found,
+  // which causes lazy compilation of the target function.
+  //
+  static void CompilationCallback();
+
+  /// getPointerToFunction - This returns the address of the specified function,
+  /// compiling it if necessary.
+  ///
+  void *getPointerToFunction(Function *F);
+
+  /// getOrEmitGlobalVariable - Return the address of the specified global
+  /// variable, possibly emitting it to memory if needed.  This is used by the
+  /// Emitter.
+  void *getOrEmitGlobalVariable(const GlobalVariable *GV);
+
+  /// getPointerToFunctionOrStub - If the specified function has been
+  /// code-gen'd, return a pointer to the function.  If not, compile it, or use
+  /// a stub to implement lazy compilation if available.
+  ///
+  void *getPointerToFunctionOrStub(Function *F);
+
+  /// recompileAndRelinkFunction - This method is used to force a function
+  /// which has already been compiled, to be compiled again, possibly
+  /// after it has been modified. Then the entry to the old copy is overwritten
+  /// with a branch to the new copy. If there was no old copy, this acts
+  /// just like JIT::getPointerToFunction().
+  ///
+  void *recompileAndRelinkFunction(Function *F);
+
+  /// freeMachineCodeForFunction - deallocate memory used to code-generate this
+  /// Function.
+  ///
+  void freeMachineCodeForFunction(Function *F);
+
+  /// getCodeEmitter - Return the code emitter this JIT is emitting into.
+  MachineCodeEmitter *getCodeEmitter() const { return MCE; }
+private:
+  static MachineCodeEmitter *createEmitter(JIT &J);
+  void runJITOnFunction (Function *F);
+};
+
+} // End llvm namespace
+
+#endif
diff --git a/lib/ExecutionEngine/JIT/JITEmitter.cpp b/lib/ExecutionEngine/JIT/JITEmitter.cpp
new file mode 100644
index 0000000..484af48
--- /dev/null
+++ b/lib/ExecutionEngine/JIT/JITEmitter.cpp
@@ -0,0 +1,1067 @@
+//===-- JITEmitter.cpp - Write machine code to executable memory ----------===//
+//
+//                     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 defines a MachineCodeEmitter object that is used by the JIT to
+// write machine code to memory and remember where relocatable values are.
+//
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "jit"
+#include "JIT.h"
+#include "llvm/Constant.h"
+#include "llvm/Module.h"
+#include "llvm/Type.h"
+#include "llvm/CodeGen/MachineCodeEmitter.h"
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineConstantPool.h"
+#include "llvm/CodeGen/MachineJumpTableInfo.h"
+#include "llvm/CodeGen/MachineRelocation.h"
+#include "llvm/ExecutionEngine/GenericValue.h"
+#include "llvm/Target/TargetData.h"
+#include "llvm/Target/TargetJITInfo.h"
+#include "llvm/Target/TargetMachine.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/MutexGuard.h"
+#include "llvm/System/Disassembler.h"
+#include "llvm/ADT/Statistic.h"
+#include "llvm/System/Memory.h"
+#include <algorithm>
+using namespace llvm;
+
+STATISTIC(NumBytes, "Number of bytes of machine code compiled");
+STATISTIC(NumRelos, "Number of relocations applied");
+static JIT *TheJIT = 0;
+
+//===----------------------------------------------------------------------===//
+// JITMemoryManager code.
+//
+namespace {
+  /// MemoryRangeHeader - For a range of memory, this is the header that we put
+  /// on the block of memory.  It is carefully crafted to be one word of memory.
+  /// Allocated blocks have just this header, free'd blocks have FreeRangeHeader
+  /// which starts with this.
+  struct FreeRangeHeader;
+  struct MemoryRangeHeader {
+    /// ThisAllocated - This is true if this block is currently allocated.  If
+    /// not, this can be converted to a FreeRangeHeader.
+    intptr_t ThisAllocated : 1;
+    
+    /// PrevAllocated - Keep track of whether the block immediately before us is
+    /// allocated.  If not, the word immediately before this header is the size
+    /// of the previous block.
+    intptr_t PrevAllocated : 1;
+    
+    /// BlockSize - This is the size in bytes of this memory block,
+    /// including this header.
+    uintptr_t BlockSize : (sizeof(intptr_t)*8 - 2);
+    
+
+    /// getBlockAfter - Return the memory block immediately after this one.
+    ///
+    MemoryRangeHeader &getBlockAfter() const {
+      return *(MemoryRangeHeader*)((char*)this+BlockSize);
+    }
+    
+    /// getFreeBlockBefore - If the block before this one is free, return it,
+    /// otherwise return null.
+    FreeRangeHeader *getFreeBlockBefore() const {
+      if (PrevAllocated) return 0;
+      intptr_t PrevSize = ((intptr_t *)this)[-1];
+      return (FreeRangeHeader*)((char*)this-PrevSize);
+    }
+    
+    /// FreeBlock - Turn an allocated block into a free block, adjusting
+    /// bits in the object headers, and adding an end of region memory block.
+    FreeRangeHeader *FreeBlock(FreeRangeHeader *FreeList);
+    
+    /// TrimAllocationToSize - If this allocated block is significantly larger
+    /// than NewSize, split it into two pieces (where the former is NewSize
+    /// bytes, including the header), and add the new block to the free list.
+    FreeRangeHeader *TrimAllocationToSize(FreeRangeHeader *FreeList, 
+                                          uint64_t NewSize);
+  };
+
+  /// FreeRangeHeader - For a memory block that isn't already allocated, this
+  /// keeps track of the current block and has a pointer to the next free block.
+  /// Free blocks are kept on a circularly linked list.
+  struct FreeRangeHeader : public MemoryRangeHeader {
+    FreeRangeHeader *Prev;
+    FreeRangeHeader *Next;
+    
+    /// getMinBlockSize - Get the minimum size for a memory block.  Blocks
+    /// smaller than this size cannot be created.
+    static unsigned getMinBlockSize() {
+      return sizeof(FreeRangeHeader)+sizeof(intptr_t);
+    }
+    
+    /// SetEndOfBlockSizeMarker - The word at the end of every free block is
+    /// known to be the size of the free block.  Set it for this block.
+    void SetEndOfBlockSizeMarker() {
+      void *EndOfBlock = (char*)this + BlockSize;
+      ((intptr_t *)EndOfBlock)[-1] = BlockSize;
+    }
+
+    FreeRangeHeader *RemoveFromFreeList() {
+      assert(Next->Prev == this && Prev->Next == this && "Freelist broken!");
+      Next->Prev = Prev;
+      return Prev->Next = Next;
+    }
+    
+    void AddToFreeList(FreeRangeHeader *FreeList) {
+      Next = FreeList;
+      Prev = FreeList->Prev;
+      Prev->Next = this;
+      Next->Prev = this;
+    }
+
+    /// GrowBlock - The block after this block just got deallocated.  Merge it
+    /// into the current block.
+    void GrowBlock(uintptr_t NewSize);
+    
+    /// AllocateBlock - Mark this entire block allocated, updating freelists
+    /// etc.  This returns a pointer to the circular free-list.
+    FreeRangeHeader *AllocateBlock();
+  };
+}
+
+
+/// AllocateBlock - Mark this entire block allocated, updating freelists
+/// etc.  This returns a pointer to the circular free-list.
+FreeRangeHeader *FreeRangeHeader::AllocateBlock() {
+  assert(!ThisAllocated && !getBlockAfter().PrevAllocated &&
+         "Cannot allocate an allocated block!");
+  // Mark this block allocated.
+  ThisAllocated = 1;
+  getBlockAfter().PrevAllocated = 1;
+ 
+  // Remove it from the free list.
+  return RemoveFromFreeList();
+}
+
+/// FreeBlock - Turn an allocated block into a free block, adjusting
+/// bits in the object headers, and adding an end of region memory block.
+/// If possible, coalesce this block with neighboring blocks.  Return the
+/// FreeRangeHeader to allocate from.
+FreeRangeHeader *MemoryRangeHeader::FreeBlock(FreeRangeHeader *FreeList) {
+  MemoryRangeHeader *FollowingBlock = &getBlockAfter();
+  assert(ThisAllocated && "This block is already allocated!");
+  assert(FollowingBlock->PrevAllocated && "Flags out of sync!");
+  
+  FreeRangeHeader *FreeListToReturn = FreeList;
+  
+  // If the block after this one is free, merge it into this block.
+  if (!FollowingBlock->ThisAllocated) {
+    FreeRangeHeader &FollowingFreeBlock = *(FreeRangeHeader *)FollowingBlock;
+    // "FreeList" always needs to be a valid free block.  If we're about to
+    // coalesce with it, update our notion of what the free list is.
+    if (&FollowingFreeBlock == FreeList) {
+      FreeList = FollowingFreeBlock.Next;
+      FreeListToReturn = 0;
+      assert(&FollowingFreeBlock != FreeList && "No tombstone block?");
+    }
+    FollowingFreeBlock.RemoveFromFreeList();
+    
+    // Include the following block into this one.
+    BlockSize += FollowingFreeBlock.BlockSize;
+    FollowingBlock = &FollowingFreeBlock.getBlockAfter();
+    
+    // Tell the block after the block we are coalescing that this block is
+    // allocated.
+    FollowingBlock->PrevAllocated = 1;
+  }
+  
+  assert(FollowingBlock->ThisAllocated && "Missed coalescing?");
+  
+  if (FreeRangeHeader *PrevFreeBlock = getFreeBlockBefore()) {
+    PrevFreeBlock->GrowBlock(PrevFreeBlock->BlockSize + BlockSize);
+    return FreeListToReturn ? FreeListToReturn : PrevFreeBlock;
+  }
+
+  // Otherwise, mark this block free.
+  FreeRangeHeader &FreeBlock = *(FreeRangeHeader*)this;
+  FollowingBlock->PrevAllocated = 0;
+  FreeBlock.ThisAllocated = 0;
+
+  // Link this into the linked list of free blocks.
+  FreeBlock.AddToFreeList(FreeList);
+
+  // Add a marker at the end of the block, indicating the size of this free
+  // block.
+  FreeBlock.SetEndOfBlockSizeMarker();
+  return FreeListToReturn ? FreeListToReturn : &FreeBlock;
+}
+
+/// GrowBlock - The block after this block just got deallocated.  Merge it
+/// into the current block.
+void FreeRangeHeader::GrowBlock(uintptr_t NewSize) {
+  assert(NewSize > BlockSize && "Not growing block?");
+  BlockSize = NewSize;
+  SetEndOfBlockSizeMarker();
+  getBlockAfter().PrevAllocated = 0;
+}
+
+/// TrimAllocationToSize - If this allocated block is significantly larger
+/// than NewSize, split it into two pieces (where the former is NewSize
+/// bytes, including the header), and add the new block to the free list.
+FreeRangeHeader *MemoryRangeHeader::
+TrimAllocationToSize(FreeRangeHeader *FreeList, uint64_t NewSize) {
+  assert(ThisAllocated && getBlockAfter().PrevAllocated &&
+         "Cannot deallocate part of an allocated block!");
+
+  // Round up size for alignment of header.
+  unsigned HeaderAlign = __alignof(FreeRangeHeader);
+  NewSize = (NewSize+ (HeaderAlign-1)) & ~(HeaderAlign-1);
+  
+  // Size is now the size of the block we will remove from the start of the
+  // current block.
+  assert(NewSize <= BlockSize &&
+         "Allocating more space from this block than exists!");
+  
+  // If splitting this block will cause the remainder to be too small, do not
+  // split the block.
+  if (BlockSize <= NewSize+FreeRangeHeader::getMinBlockSize())
+    return FreeList;
+  
+  // Otherwise, we splice the required number of bytes out of this block, form
+  // a new block immediately after it, then mark this block allocated.
+  MemoryRangeHeader &FormerNextBlock = getBlockAfter();
+  
+  // Change the size of this block.
+  BlockSize = NewSize;
+  
+  // Get the new block we just sliced out and turn it into a free block.
+  FreeRangeHeader &NewNextBlock = (FreeRangeHeader &)getBlockAfter();
+  NewNextBlock.BlockSize = (char*)&FormerNextBlock - (char*)&NewNextBlock;
+  NewNextBlock.ThisAllocated = 0;
+  NewNextBlock.PrevAllocated = 1;
+  NewNextBlock.SetEndOfBlockSizeMarker();
+  FormerNextBlock.PrevAllocated = 0;
+  NewNextBlock.AddToFreeList(FreeList);
+  return &NewNextBlock;
+}
+
+ 
+namespace {  
+  /// JITMemoryManager - Manage memory for the JIT code generation in a logical,
+  /// sane way.  This splits a large block of MAP_NORESERVE'd memory into two
+  /// sections, one for function stubs, one for the functions themselves.  We
+  /// have to do this because we may need to emit a function stub while in the
+  /// middle of emitting a function, and we don't know how large the function we
+  /// are emitting is.  This never bothers to release the memory, because when
+  /// we are ready to destroy the JIT, the program exits.
+  class JITMemoryManager {
+    std::vector<sys::MemoryBlock> Blocks; // Memory blocks allocated by the JIT
+    FreeRangeHeader *FreeMemoryList;      // Circular list of free blocks.
+    
+    // When emitting code into a memory block, this is the block.
+    MemoryRangeHeader *CurBlock;
+    
+    unsigned char *CurStubPtr, *StubBase;
+    unsigned char *GOTBase;      // Target Specific reserved memory
+
+    // Centralize memory block allocation.
+    sys::MemoryBlock getNewMemoryBlock(unsigned size);
+    
+    std::map<const Function*, MemoryRangeHeader*> FunctionBlocks;
+  public:
+    JITMemoryManager(bool useGOT);
+    ~JITMemoryManager();
+
+    inline unsigned char *allocateStub(unsigned StubSize, unsigned Alignment);
+    
+    /// startFunctionBody - When a function starts, allocate a block of free
+    /// executable memory, returning a pointer to it and its actual size.
+    unsigned char *startFunctionBody(uintptr_t &ActualSize) {
+      CurBlock = FreeMemoryList;
+      
+      // Allocate the entire memory block.
+      FreeMemoryList = FreeMemoryList->AllocateBlock();
+      ActualSize = CurBlock->BlockSize-sizeof(MemoryRangeHeader);
+      return (unsigned char *)(CurBlock+1);
+    }
+    
+    /// endFunctionBody - The function F is now allocated, and takes the memory
+    /// in the range [FunctionStart,FunctionEnd).
+    void endFunctionBody(const Function *F, unsigned char *FunctionStart,
+                         unsigned char *FunctionEnd) {
+      assert(FunctionEnd > FunctionStart);
+      assert(FunctionStart == (unsigned char *)(CurBlock+1) &&
+             "Mismatched function start/end!");
+      
+      uintptr_t BlockSize = FunctionEnd - (unsigned char *)CurBlock;
+      FunctionBlocks[F] = CurBlock;
+
+      // Release the memory at the end of this block that isn't needed.
+      FreeMemoryList =CurBlock->TrimAllocationToSize(FreeMemoryList, BlockSize);
+    }
+    
+    unsigned char *getGOTBase() const {
+      return GOTBase;
+    }
+    bool isManagingGOT() const {
+      return GOTBase != NULL;
+    }
+    
+    /// deallocateMemForFunction - Deallocate all memory for the specified
+    /// function body.
+    void deallocateMemForFunction(const Function *F) {
+      std::map<const Function*, MemoryRangeHeader*>::iterator
+        I = FunctionBlocks.find(F);
+      if (I == FunctionBlocks.end()) return;
+      
+      // Find the block that is allocated for this function.
+      MemoryRangeHeader *MemRange = I->second;
+      assert(MemRange->ThisAllocated && "Block isn't allocated!");
+      
+      // Fill the buffer with garbage!
+      DEBUG(memset(MemRange+1, 0xCD, MemRange->BlockSize-sizeof(*MemRange)));
+      
+      // Free the memory.
+      FreeMemoryList = MemRange->FreeBlock(FreeMemoryList);
+      
+      // Finally, remove this entry from FunctionBlocks.
+      FunctionBlocks.erase(I);
+    }
+  };
+}
+
+JITMemoryManager::JITMemoryManager(bool useGOT) {
+  // Allocate a 16M block of memory for functions.
+  sys::MemoryBlock MemBlock = getNewMemoryBlock(16 << 20);
+
+  unsigned char *MemBase = reinterpret_cast<unsigned char*>(MemBlock.base());
+
+  // Allocate stubs backwards from the base, allocate functions forward
+  // from the base.
+  StubBase   = MemBase;
+  CurStubPtr = MemBase + 512*1024; // Use 512k for stubs, working backwards.
+  
+  // We set up the memory chunk with 4 mem regions, like this:
+  //  [ START
+  //    [ Free      #0 ] -> Large space to allocate functions from.
+  //    [ Allocated #1 ] -> Tiny space to separate regions.
+  //    [ Free      #2 ] -> Tiny space so there is always at least 1 free block.
+  //    [ Allocated #3 ] -> Tiny space to prevent looking past end of block.
+  //  END ]
+  //
+  // The last three blocks are never deallocated or touched.
+  
+  // Add MemoryRangeHeader to the end of the memory region, indicating that
+  // the space after the block of memory is allocated.  This is block #3.
+  MemoryRangeHeader *Mem3 = (MemoryRangeHeader*)(MemBase+MemBlock.size())-1;
+  Mem3->ThisAllocated = 1;
+  Mem3->PrevAllocated = 0;
+  Mem3->BlockSize     = 0;
+  
+  /// Add a tiny free region so that the free list always has one entry.
+  FreeRangeHeader *Mem2 = 
+    (FreeRangeHeader *)(((char*)Mem3)-FreeRangeHeader::getMinBlockSize());
+  Mem2->ThisAllocated = 0;
+  Mem2->PrevAllocated = 1;
+  Mem2->BlockSize     = FreeRangeHeader::getMinBlockSize();
+  Mem2->SetEndOfBlockSizeMarker();
+  Mem2->Prev = Mem2;   // Mem2 *is* the free list for now.
+  Mem2->Next = Mem2;
+
+  /// Add a tiny allocated region so that Mem2 is never coalesced away.
+  MemoryRangeHeader *Mem1 = (MemoryRangeHeader*)Mem2-1;
+  Mem1->ThisAllocated = 1;
+  Mem1->PrevAllocated = 0;
+  Mem1->BlockSize     = (char*)Mem2 - (char*)Mem1;
+  
+  // Add a FreeRangeHeader to the start of the function body region, indicating
+  // that the space is free.  Mark the previous block allocated so we never look
+  // at it.
+  FreeRangeHeader *Mem0 = (FreeRangeHeader*)CurStubPtr;
+  Mem0->ThisAllocated = 0;
+  Mem0->PrevAllocated = 1;
+  Mem0->BlockSize = (char*)Mem1-(char*)Mem0;
+  Mem0->SetEndOfBlockSizeMarker();
+  Mem0->AddToFreeList(Mem2);
+  
+  // Start out with the freelist pointing to Mem0.
+  FreeMemoryList = Mem0;
+
+  // Allocate the GOT.
+  GOTBase = NULL;
+  if (useGOT) GOTBase = new unsigned char[sizeof(void*) * 8192];
+}
+
+JITMemoryManager::~JITMemoryManager() {
+  for (unsigned i = 0, e = Blocks.size(); i != e; ++i)
+    sys::Memory::ReleaseRWX(Blocks[i]);
+  
+  delete[] GOTBase;
+  Blocks.clear();
+}
+
+unsigned char *JITMemoryManager::allocateStub(unsigned StubSize,
+                                              unsigned Alignment) {
+  CurStubPtr -= StubSize;
+  CurStubPtr = (unsigned char*)(((intptr_t)CurStubPtr) &
+                                ~(intptr_t)(Alignment-1));
+  if (CurStubPtr < StubBase) {
+    // FIXME: allocate a new block
+    cerr << "JIT ran out of memory for function stubs!\n";
+    abort();
+  }
+  return CurStubPtr;
+}
+
+sys::MemoryBlock JITMemoryManager::getNewMemoryBlock(unsigned size) {
+  // Allocate a new block close to the last one.
+  const sys::MemoryBlock *BOld = Blocks.empty() ? 0 : &Blocks.front();
+  std::string ErrMsg;
+  sys::MemoryBlock B = sys::Memory::AllocateRWX(size, BOld, &ErrMsg);
+  if (B.base() == 0) {
+    cerr << "Allocation failed when allocating new memory in the JIT\n";
+    cerr << ErrMsg << "\n";
+    abort();
+  }
+  Blocks.push_back(B);
+  return B;
+}
+
+//===----------------------------------------------------------------------===//
+// JIT lazy compilation code.
+//
+namespace {
+  class JITResolverState {
+  private:
+    /// FunctionToStubMap - Keep track of the stub created for a particular
+    /// function so that we can reuse them if necessary.
+    std::map<Function*, void*> FunctionToStubMap;
+
+    /// StubToFunctionMap - Keep track of the function that each stub
+    /// corresponds to.
+    std::map<void*, Function*> StubToFunctionMap;
+
+  public:
+    std::map<Function*, void*>& getFunctionToStubMap(const MutexGuard& locked) {
+      assert(locked.holds(TheJIT->lock));
+      return FunctionToStubMap;
+    }
+
+    std::map<void*, Function*>& getStubToFunctionMap(const MutexGuard& locked) {
+      assert(locked.holds(TheJIT->lock));
+      return StubToFunctionMap;
+    }
+  };
+
+  /// JITResolver - Keep track of, and resolve, call sites for functions that
+  /// have not yet been compiled.
+  class JITResolver {
+    /// LazyResolverFn - The target lazy resolver function that we actually
+    /// rewrite instructions to use.
+    TargetJITInfo::LazyResolverFn LazyResolverFn;
+
+    JITResolverState state;
+
+    /// ExternalFnToStubMap - This is the equivalent of FunctionToStubMap for
+    /// external functions.
+    std::map<void*, void*> ExternalFnToStubMap;
+
+    //map addresses to indexes in the GOT
+    std::map<void*, unsigned> revGOTMap;
+    unsigned nextGOTIndex;
+
+    static JITResolver *TheJITResolver;
+  public:
+    JITResolver(JIT &jit) : nextGOTIndex(0) {
+      TheJIT = &jit;
+
+      LazyResolverFn = jit.getJITInfo().getLazyResolverFunction(JITCompilerFn);
+      assert(TheJITResolver == 0 && "Multiple JIT resolvers?");
+      TheJITResolver = this;
+    }
+    
+    ~JITResolver() {
+      TheJITResolver = 0;
+    }
+
+    /// getFunctionStub - This returns a pointer to a function stub, creating
+    /// one on demand as needed.
+    void *getFunctionStub(Function *F);
+
+    /// getExternalFunctionStub - Return a stub for the function at the
+    /// specified address, created lazily on demand.
+    void *getExternalFunctionStub(void *FnAddr);
+
+    /// AddCallbackAtLocation - If the target is capable of rewriting an
+    /// instruction without the use of a stub, record the location of the use so
+    /// we know which function is being used at the location.
+    void *AddCallbackAtLocation(Function *F, void *Location) {
+      MutexGuard locked(TheJIT->lock);
+      /// Get the target-specific JIT resolver function.
+      state.getStubToFunctionMap(locked)[Location] = F;
+      return (void*)(intptr_t)LazyResolverFn;
+    }
+
+    /// getGOTIndexForAddress - Return a new or existing index in the GOT for
+    /// and address.  This function only manages slots, it does not manage the
+    /// contents of the slots or the memory associated with the GOT.
+    unsigned getGOTIndexForAddr(void* addr);
+
+    /// JITCompilerFn - This function is called to resolve a stub to a compiled
+    /// address.  If the LLVM Function corresponding to the stub has not yet
+    /// been compiled, this function compiles it first.
+    static void *JITCompilerFn(void *Stub);
+  };
+}
+
+JITResolver *JITResolver::TheJITResolver = 0;
+
+#if (defined(__POWERPC__) || defined (__ppc__) || defined(_POWER)) && \
+    defined(__APPLE__)
+extern "C" void sys_icache_invalidate(const void *Addr, size_t len);
+#endif
+
+/// synchronizeICache - On some targets, the JIT emitted code must be
+/// explicitly refetched to ensure correct execution.
+static void synchronizeICache(const void *Addr, size_t len) {
+#if (defined(__POWERPC__) || defined (__ppc__) || defined(_POWER)) && \
+    defined(__APPLE__)
+  sys_icache_invalidate(Addr, len);
+#endif
+}
+
+/// getFunctionStub - This returns a pointer to a function stub, creating
+/// one on demand as needed.
+void *JITResolver::getFunctionStub(Function *F) {
+  MutexGuard locked(TheJIT->lock);
+
+  // If we already have a stub for this function, recycle it.
+  void *&Stub = state.getFunctionToStubMap(locked)[F];
+  if (Stub) return Stub;
+
+  // Call the lazy resolver function unless we already KNOW it is an external
+  // function, in which case we just skip the lazy resolution step.
+  void *Actual = (void*)(intptr_t)LazyResolverFn;
+  if (F->isDeclaration() && !F->hasNotBeenReadFromBitcode())
+    Actual = TheJIT->getPointerToFunction(F);
+
+  // Otherwise, codegen a new stub.  For now, the stub will call the lazy
+  // resolver function.
+  Stub = TheJIT->getJITInfo().emitFunctionStub(Actual,
+                                               *TheJIT->getCodeEmitter());
+
+  if (Actual != (void*)(intptr_t)LazyResolverFn) {
+    // If we are getting the stub for an external function, we really want the
+    // address of the stub in the GlobalAddressMap for the JIT, not the address
+    // of the external function.
+    TheJIT->updateGlobalMapping(F, Stub);
+  }
+
+  // Invalidate the icache if necessary.
+  synchronizeICache(Stub, TheJIT->getCodeEmitter()->getCurrentPCValue() -
+                          (intptr_t)Stub);
+
+  DOUT << "JIT: Stub emitted at [" << Stub << "] for function '"
+       << F->getName() << "'\n";
+
+  // Finally, keep track of the stub-to-Function mapping so that the
+  // JITCompilerFn knows which function to compile!
+  state.getStubToFunctionMap(locked)[Stub] = F;
+  return Stub;
+}
+
+/// getExternalFunctionStub - Return a stub for the function at the
+/// specified address, created lazily on demand.
+void *JITResolver::getExternalFunctionStub(void *FnAddr) {
+  // If we already have a stub for this function, recycle it.
+  void *&Stub = ExternalFnToStubMap[FnAddr];
+  if (Stub) return Stub;
+
+  Stub = TheJIT->getJITInfo().emitFunctionStub(FnAddr,
+                                               *TheJIT->getCodeEmitter());
+
+  // Invalidate the icache if necessary.
+  synchronizeICache(Stub, TheJIT->getCodeEmitter()->getCurrentPCValue() -
+                    (intptr_t)Stub);
+
+  DOUT << "JIT: Stub emitted at [" << Stub
+       << "] for external function at '" << FnAddr << "'\n";
+  return Stub;
+}
+
+unsigned JITResolver::getGOTIndexForAddr(void* addr) {
+  unsigned idx = revGOTMap[addr];
+  if (!idx) {
+    idx = ++nextGOTIndex;
+    revGOTMap[addr] = idx;
+    DOUT << "Adding GOT entry " << idx
+         << " for addr " << addr << "\n";
+    //    ((void**)MemMgr.getGOTBase())[idx] = addr;
+  }
+  return idx;
+}
+
+/// JITCompilerFn - This function is called when a lazy compilation stub has
+/// been entered.  It looks up which function this stub corresponds to, compiles
+/// it if necessary, then returns the resultant function pointer.
+void *JITResolver::JITCompilerFn(void *Stub) {
+  JITResolver &JR = *TheJITResolver;
+
+  MutexGuard locked(TheJIT->lock);
+
+  // The address given to us for the stub may not be exactly right, it might be
+  // a little bit after the stub.  As such, use upper_bound to find it.
+  std::map<void*, Function*>::iterator I =
+    JR.state.getStubToFunctionMap(locked).upper_bound(Stub);
+  assert(I != JR.state.getStubToFunctionMap(locked).begin() &&
+         "This is not a known stub!");
+  Function *F = (--I)->second;
+
+  // If we have already code generated the function, just return the address.
+  void *Result = TheJIT->getPointerToGlobalIfAvailable(F);
+  
+  if (!Result) {
+    // Otherwise we don't have it, do lazy compilation now.
+    
+    // If lazy compilation is disabled, emit a useful error message and abort.
+    if (TheJIT->isLazyCompilationDisabled()) {
+      cerr << "LLVM JIT requested to do lazy compilation of function '"
+      << F->getName() << "' when lazy compiles are disabled!\n";
+      abort();
+    }
+  
+    // We might like to remove the stub from the StubToFunction map.
+    // We can't do that! Multiple threads could be stuck, waiting to acquire the
+    // lock above. As soon as the 1st function finishes compiling the function,
+    // the next one will be released, and needs to be able to find the function
+    // it needs to call.
+    //JR.state.getStubToFunctionMap(locked).erase(I);
+
+    DOUT << "JIT: Lazily resolving function '" << F->getName()
+         << "' In stub ptr = " << Stub << " actual ptr = "
+         << I->first << "\n";
+
+    Result = TheJIT->getPointerToFunction(F);
+  }
+
+  // We don't need to reuse this stub in the future, as F is now compiled.
+  JR.state.getFunctionToStubMap(locked).erase(F);
+
+  // FIXME: We could rewrite all references to this stub if we knew them.
+
+  // What we will do is set the compiled function address to map to the
+  // same GOT entry as the stub so that later clients may update the GOT
+  // if they see it still using the stub address.
+  // Note: this is done so the Resolver doesn't have to manage GOT memory
+  // Do this without allocating map space if the target isn't using a GOT
+  if(JR.revGOTMap.find(Stub) != JR.revGOTMap.end())
+    JR.revGOTMap[Result] = JR.revGOTMap[Stub];
+
+  return Result;
+}
+
+
+//===----------------------------------------------------------------------===//
+// JITEmitter code.
+//
+namespace {
+  /// JITEmitter - The JIT implementation of the MachineCodeEmitter, which is
+  /// used to output functions to memory for execution.
+  class JITEmitter : public MachineCodeEmitter {
+    JITMemoryManager MemMgr;
+
+    // When outputting a function stub in the context of some other function, we
+    // save BufferBegin/BufferEnd/CurBufferPtr here.
+    unsigned char *SavedBufferBegin, *SavedBufferEnd, *SavedCurBufferPtr;
+
+    /// Relocations - These are the relocations that the function needs, as
+    /// emitted.
+    std::vector<MachineRelocation> Relocations;
+    
+    /// MBBLocations - This vector is a mapping from MBB ID's to their address.
+    /// It is filled in by the StartMachineBasicBlock callback and queried by
+    /// the getMachineBasicBlockAddress callback.
+    std::vector<intptr_t> MBBLocations;
+
+    /// ConstantPool - The constant pool for the current function.
+    ///
+    MachineConstantPool *ConstantPool;
+
+    /// ConstantPoolBase - A pointer to the first entry in the constant pool.
+    ///
+    void *ConstantPoolBase;
+
+    /// JumpTable - The jump tables for the current function.
+    ///
+    MachineJumpTableInfo *JumpTable;
+    
+    /// JumpTableBase - A pointer to the first entry in the jump table.
+    ///
+    void *JumpTableBase;
+    
+    /// Resolver - This contains info about the currently resolved functions.
+    JITResolver Resolver;
+  public:
+    JITEmitter(JIT &jit)
+       : MemMgr(jit.getJITInfo().needsGOT()), Resolver(jit) {
+      if (MemMgr.isManagingGOT()) DOUT << "JIT is managing a GOT\n";
+    }
+    
+    JITResolver &getJITResolver() { return Resolver; }
+
+    virtual void startFunction(MachineFunction &F);
+    virtual bool finishFunction(MachineFunction &F);
+    
+    void emitConstantPool(MachineConstantPool *MCP);
+    void initJumpTableInfo(MachineJumpTableInfo *MJTI);
+    void emitJumpTableInfo(MachineJumpTableInfo *MJTI);
+    
+    virtual void startFunctionStub(unsigned StubSize, unsigned Alignment = 1);
+    virtual void* finishFunctionStub(const Function *F);
+
+    virtual void addRelocation(const MachineRelocation &MR) {
+      Relocations.push_back(MR);
+    }
+    
+    virtual void StartMachineBasicBlock(MachineBasicBlock *MBB) {
+      if (MBBLocations.size() <= (unsigned)MBB->getNumber())
+        MBBLocations.resize((MBB->getNumber()+1)*2);
+      MBBLocations[MBB->getNumber()] = getCurrentPCValue();
+    }
+
+    virtual intptr_t getConstantPoolEntryAddress(unsigned Entry) const;
+    virtual intptr_t getJumpTableEntryAddress(unsigned Entry) const;
+    
+    virtual intptr_t getMachineBasicBlockAddress(MachineBasicBlock *MBB) const {
+      assert(MBBLocations.size() > (unsigned)MBB->getNumber() && 
+             MBBLocations[MBB->getNumber()] && "MBB not emitted!");
+      return MBBLocations[MBB->getNumber()];
+    }
+
+    /// deallocateMemForFunction - Deallocate all memory for the specified
+    /// function body.
+    void deallocateMemForFunction(Function *F) {
+      MemMgr.deallocateMemForFunction(F);
+    }
+  private:
+    void *getPointerToGlobal(GlobalValue *GV, void *Reference, bool NoNeedStub);
+  };
+}
+
+void *JITEmitter::getPointerToGlobal(GlobalValue *V, void *Reference,
+                                     bool DoesntNeedStub) {
+  if (GlobalVariable *GV = dyn_cast<GlobalVariable>(V)) {
+    /// FIXME: If we straightened things out, this could actually emit the
+    /// global immediately instead of queuing it for codegen later!
+    return TheJIT->getOrEmitGlobalVariable(GV);
+  }
+
+  // If we have already compiled the function, return a pointer to its body.
+  Function *F = cast<Function>(V);
+  void *ResultPtr = TheJIT->getPointerToGlobalIfAvailable(F);
+  if (ResultPtr) return ResultPtr;
+
+  if (F->isDeclaration() && !F->hasNotBeenReadFromBitcode()) {
+    // If this is an external function pointer, we can force the JIT to
+    // 'compile' it, which really just adds it to the map.
+    if (DoesntNeedStub)
+      return TheJIT->getPointerToFunction(F);
+
+    return Resolver.getFunctionStub(F);
+  }
+
+  // Okay, the function has not been compiled yet, if the target callback
+  // mechanism is capable of rewriting the instruction directly, prefer to do
+  // that instead of emitting a stub.
+  if (DoesntNeedStub)
+    return Resolver.AddCallbackAtLocation(F, Reference);
+
+  // Otherwise, we have to emit a lazy resolving stub.
+  return Resolver.getFunctionStub(F);
+}
+
+void JITEmitter::startFunction(MachineFunction &F) {
+  uintptr_t ActualSize;
+  BufferBegin = CurBufferPtr = MemMgr.startFunctionBody(ActualSize);
+  BufferEnd = BufferBegin+ActualSize;
+  
+  // Ensure the constant pool/jump table info is at least 4-byte aligned.
+  emitAlignment(16);
+
+  emitConstantPool(F.getConstantPool());
+  initJumpTableInfo(F.getJumpTableInfo());
+
+  // About to start emitting the machine code for the function.
+  emitAlignment(std::max(F.getFunction()->getAlignment(), 8U));
+  TheJIT->updateGlobalMapping(F.getFunction(), CurBufferPtr);
+
+  MBBLocations.clear();
+}
+
+bool JITEmitter::finishFunction(MachineFunction &F) {
+  if (CurBufferPtr == BufferEnd) {
+    // FIXME: Allocate more space, then try again.
+    cerr << "JIT: Ran out of space for generated machine code!\n";
+    abort();
+  }
+  
+  emitJumpTableInfo(F.getJumpTableInfo());
+  
+  // FnStart is the start of the text, not the start of the constant pool and
+  // other per-function data.
+  unsigned char *FnStart =
+    (unsigned char *)TheJIT->getPointerToGlobalIfAvailable(F.getFunction());
+  unsigned char *FnEnd   = CurBufferPtr;
+  
+  MemMgr.endFunctionBody(F.getFunction(), BufferBegin, FnEnd);
+  NumBytes += FnEnd-FnStart;
+
+  if (!Relocations.empty()) {
+    NumRelos += Relocations.size();
+
+    // Resolve the relocations to concrete pointers.
+    for (unsigned i = 0, e = Relocations.size(); i != e; ++i) {
+      MachineRelocation &MR = Relocations[i];
+      void *ResultPtr;
+      if (MR.isString()) {
+        ResultPtr = TheJIT->getPointerToNamedFunction(MR.getString());
+
+        // If the target REALLY wants a stub for this function, emit it now.
+        if (!MR.doesntNeedFunctionStub())
+          ResultPtr = Resolver.getExternalFunctionStub(ResultPtr);
+      } else if (MR.isGlobalValue()) {
+        ResultPtr = getPointerToGlobal(MR.getGlobalValue(),
+                                       BufferBegin+MR.getMachineCodeOffset(),
+                                       MR.doesntNeedFunctionStub());
+      } else if (MR.isBasicBlock()) {
+        ResultPtr = (void*)getMachineBasicBlockAddress(MR.getBasicBlock());
+      } else if (MR.isConstantPoolIndex()) {
+        ResultPtr=(void*)getConstantPoolEntryAddress(MR.getConstantPoolIndex());
+      } else {
+        assert(MR.isJumpTableIndex());
+        ResultPtr=(void*)getJumpTableEntryAddress(MR.getJumpTableIndex());
+      }
+
+      MR.setResultPointer(ResultPtr);
+
+      // if we are managing the GOT and the relocation wants an index,
+      // give it one
+      if (MemMgr.isManagingGOT() && MR.isGOTRelative()) {
+        unsigned idx = Resolver.getGOTIndexForAddr(ResultPtr);
+        MR.setGOTIndex(idx);
+        if (((void**)MemMgr.getGOTBase())[idx] != ResultPtr) {
+          DOUT << "GOT was out of date for " << ResultPtr
+               << " pointing at " << ((void**)MemMgr.getGOTBase())[idx]
+               << "\n";
+          ((void**)MemMgr.getGOTBase())[idx] = ResultPtr;
+        }
+      }
+    }
+
+    TheJIT->getJITInfo().relocate(BufferBegin, &Relocations[0],
+                                  Relocations.size(), MemMgr.getGOTBase());
+  }
+
+  // Update the GOT entry for F to point to the new code.
+  if (MemMgr.isManagingGOT()) {
+    unsigned idx = Resolver.getGOTIndexForAddr((void*)BufferBegin);
+    if (((void**)MemMgr.getGOTBase())[idx] != (void*)BufferBegin) {
+      DOUT << "GOT was out of date for " << (void*)BufferBegin
+           << " pointing at " << ((void**)MemMgr.getGOTBase())[idx] << "\n";
+      ((void**)MemMgr.getGOTBase())[idx] = (void*)BufferBegin;
+    }
+  }
+
+  // Invalidate the icache if necessary.
+  synchronizeICache(FnStart, FnEnd-FnStart);
+
+  DOUT << "JIT: Finished CodeGen of [" << (void*)FnStart
+       << "] Function: " << F.getFunction()->getName()
+       << ": " << (FnEnd-FnStart) << " bytes of text, "
+       << Relocations.size() << " relocations\n";
+  Relocations.clear();
+
+#ifndef NDEBUG
+  if (sys::hasDisassembler())
+    DOUT << "Disassembled code:\n"
+         << sys::disassembleBuffer(FnStart, FnEnd-FnStart, (uintptr_t)FnStart);
+#endif
+  
+  return false;
+}
+
+void JITEmitter::emitConstantPool(MachineConstantPool *MCP) {
+  const std::vector<MachineConstantPoolEntry> &Constants = MCP->getConstants();
+  if (Constants.empty()) return;
+
+  MachineConstantPoolEntry CPE = Constants.back();
+  unsigned Size = CPE.Offset;
+  const Type *Ty = CPE.isMachineConstantPoolEntry()
+    ? CPE.Val.MachineCPVal->getType() : CPE.Val.ConstVal->getType();
+  Size += TheJIT->getTargetData()->getTypeSize(Ty);
+
+  ConstantPoolBase = allocateSpace(Size, 1 << MCP->getConstantPoolAlignment());
+  ConstantPool = MCP;
+
+  if (ConstantPoolBase == 0) return;  // Buffer overflow.
+
+  // Initialize the memory for all of the constant pool entries.
+  for (unsigned i = 0, e = Constants.size(); i != e; ++i) {
+    void *CAddr = (char*)ConstantPoolBase+Constants[i].Offset;
+    if (Constants[i].isMachineConstantPoolEntry()) {
+      // FIXME: add support to lower machine constant pool values into bytes!
+      cerr << "Initialize memory with machine specific constant pool entry"
+           << " has not been implemented!\n";
+      abort();
+    }
+    TheJIT->InitializeMemory(Constants[i].Val.ConstVal, CAddr);
+  }
+}
+
+void JITEmitter::initJumpTableInfo(MachineJumpTableInfo *MJTI) {
+  const std::vector<MachineJumpTableEntry> &JT = MJTI->getJumpTables();
+  if (JT.empty()) return;
+  
+  unsigned NumEntries = 0;
+  for (unsigned i = 0, e = JT.size(); i != e; ++i)
+    NumEntries += JT[i].MBBs.size();
+
+  unsigned EntrySize = MJTI->getEntrySize();
+
+  // Just allocate space for all the jump tables now.  We will fix up the actual
+  // MBB entries in the tables after we emit the code for each block, since then
+  // we will know the final locations of the MBBs in memory.
+  JumpTable = MJTI;
+  JumpTableBase = allocateSpace(NumEntries * EntrySize, MJTI->getAlignment());
+}
+
+void JITEmitter::emitJumpTableInfo(MachineJumpTableInfo *MJTI) {
+  const std::vector<MachineJumpTableEntry> &JT = MJTI->getJumpTables();
+  if (JT.empty() || JumpTableBase == 0) return;
+  
+  if (TargetMachine::getRelocationModel() == Reloc::PIC_) {
+    assert(MJTI->getEntrySize() == 4 && "Cross JIT'ing?");
+    // For each jump table, place the offset from the beginning of the table
+    // to the target address.
+    int *SlotPtr = (int*)JumpTableBase;
+
+    for (unsigned i = 0, e = JT.size(); i != e; ++i) {
+      const std::vector<MachineBasicBlock*> &MBBs = JT[i].MBBs;
+      // Store the offset of the basic block for this jump table slot in the
+      // memory we allocated for the jump table in 'initJumpTableInfo'
+      intptr_t Base = (intptr_t)SlotPtr;
+      for (unsigned mi = 0, me = MBBs.size(); mi != me; ++mi)
+        *SlotPtr++ = (intptr_t)getMachineBasicBlockAddress(MBBs[mi]) - Base;
+    }
+  } else {
+    assert(MJTI->getEntrySize() == sizeof(void*) && "Cross JIT'ing?");
+    
+    // For each jump table, map each target in the jump table to the address of 
+    // an emitted MachineBasicBlock.
+    intptr_t *SlotPtr = (intptr_t*)JumpTableBase;
+
+    for (unsigned i = 0, e = JT.size(); i != e; ++i) {
+      const std::vector<MachineBasicBlock*> &MBBs = JT[i].MBBs;
+      // Store the address of the basic block for this jump table slot in the
+      // memory we allocated for the jump table in 'initJumpTableInfo'
+      for (unsigned mi = 0, me = MBBs.size(); mi != me; ++mi)
+        *SlotPtr++ = getMachineBasicBlockAddress(MBBs[mi]);
+    }
+  }
+}
+
+void JITEmitter::startFunctionStub(unsigned StubSize, unsigned Alignment) {
+  SavedBufferBegin = BufferBegin;
+  SavedBufferEnd = BufferEnd;
+  SavedCurBufferPtr = CurBufferPtr;
+  
+  BufferBegin = CurBufferPtr = MemMgr.allocateStub(StubSize, Alignment);
+  BufferEnd = BufferBegin+StubSize+1;
+}
+
+void *JITEmitter::finishFunctionStub(const Function *F) {
+  NumBytes += getCurrentPCOffset();
+  std::swap(SavedBufferBegin, BufferBegin);
+  BufferEnd = SavedBufferEnd;
+  CurBufferPtr = SavedCurBufferPtr;
+  return SavedBufferBegin;
+}
+
+// getConstantPoolEntryAddress - Return the address of the 'ConstantNum' entry
+// in the constant pool that was last emitted with the 'emitConstantPool'
+// method.
+//
+intptr_t JITEmitter::getConstantPoolEntryAddress(unsigned ConstantNum) const {
+  assert(ConstantNum < ConstantPool->getConstants().size() &&
+         "Invalid ConstantPoolIndex!");
+  return (intptr_t)ConstantPoolBase +
+         ConstantPool->getConstants()[ConstantNum].Offset;
+}
+
+// getJumpTableEntryAddress - Return the address of the JumpTable with index
+// 'Index' in the jumpp table that was last initialized with 'initJumpTableInfo'
+//
+intptr_t JITEmitter::getJumpTableEntryAddress(unsigned Index) const {
+  const std::vector<MachineJumpTableEntry> &JT = JumpTable->getJumpTables();
+  assert(Index < JT.size() && "Invalid jump table index!");
+  
+  unsigned Offset = 0;
+  unsigned EntrySize = JumpTable->getEntrySize();
+  
+  for (unsigned i = 0; i < Index; ++i)
+    Offset += JT[i].MBBs.size();
+  
+   Offset *= EntrySize;
+  
+  return (intptr_t)((char *)JumpTableBase + Offset);
+}
+
+//===----------------------------------------------------------------------===//
+//  Public interface to this file
+//===----------------------------------------------------------------------===//
+
+MachineCodeEmitter *JIT::createEmitter(JIT &jit) {
+  return new JITEmitter(jit);
+}
+
+// getPointerToNamedFunction - This function is used as a global wrapper to
+// JIT::getPointerToNamedFunction for the purpose of resolving symbols when
+// bugpoint is debugging the JIT. In that scenario, we are loading an .so and
+// need to resolve function(s) that are being mis-codegenerated, so we need to
+// resolve their addresses at runtime, and this is the way to do it.
+extern "C" {
+  void *getPointerToNamedFunction(const char *Name) {
+    if (Function *F = TheJIT->FindFunctionNamed(Name))
+      return TheJIT->getPointerToFunction(F);
+    return TheJIT->getPointerToNamedFunction(Name);
+  }
+}
+
+// getPointerToFunctionOrStub - If the specified function has been
+// code-gen'd, return a pointer to the function.  If not, compile it, or use
+// a stub to implement lazy compilation if available.
+//
+void *JIT::getPointerToFunctionOrStub(Function *F) {
+  // If we have already code generated the function, just return the address.
+  if (void *Addr = getPointerToGlobalIfAvailable(F))
+    return Addr;
+  
+  // Get a stub if the target supports it.
+  assert(dynamic_cast<JITEmitter*>(MCE) && "Unexpected MCE?");
+  JITEmitter *JE = static_cast<JITEmitter*>(getCodeEmitter());
+  return JE->getJITResolver().getFunctionStub(F);
+}
+
+/// freeMachineCodeForFunction - release machine code memory for given Function.
+///
+void JIT::freeMachineCodeForFunction(Function *F) {
+  // Delete translation for this from the ExecutionEngine, so it will get
+  // retranslated next time it is used.
+  updateGlobalMapping(F, 0);
+
+  // Free the actual memory for the function body and related stuff.
+  assert(dynamic_cast<JITEmitter*>(MCE) && "Unexpected MCE?");
+  static_cast<JITEmitter*>(MCE)->deallocateMemForFunction(F);
+}
+
diff --git a/lib/ExecutionEngine/JIT/Makefile b/lib/ExecutionEngine/JIT/Makefile
new file mode 100644
index 0000000..ebbdc3f
--- /dev/null
+++ b/lib/ExecutionEngine/JIT/Makefile
@@ -0,0 +1,37 @@
+##===- lib/ExecutionEngine/JIT/Makefile --------------------*- Makefile -*-===##
+# 
+#                     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.
+# 
+##===----------------------------------------------------------------------===##
+LEVEL = ../../..
+LIBRARYNAME = LLVMJIT
+
+# Get the $(ARCH) setting
+include $(LEVEL)/Makefile.config
+
+# Enable the X86 JIT if compiling on X86
+ifeq ($(ARCH), x86)
+  ENABLE_X86_JIT = 1
+endif
+
+# This flag can also be used on the command line to force inclusion
+# of the X86 JIT on non-X86 hosts
+ifdef ENABLE_X86_JIT
+  CPPFLAGS += -DENABLE_X86_JIT
+endif
+
+# Enable the Sparc JIT if compiling on Sparc
+ifeq ($(ARCH), Sparc)
+  ENABLE_SPARC_JIT = 1
+endif
+
+# This flag can also be used on the command line to force inclusion
+# of the Sparc JIT on non-Sparc hosts
+ifdef ENABLE_SPARC_JIT
+  CPPFLAGS += -DENABLE_SPARC_JIT
+endif
+
+include $(LEVEL)/Makefile.common
diff --git a/lib/ExecutionEngine/JIT/TargetSelect.cpp b/lib/ExecutionEngine/JIT/TargetSelect.cpp
new file mode 100644
index 0000000..bf968af
--- /dev/null
+++ b/lib/ExecutionEngine/JIT/TargetSelect.cpp
@@ -0,0 +1,78 @@
+//===-- TargetSelect.cpp - Target Chooser Code ----------------------------===//
+//
+//                     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 just asks the TargetMachineRegistry for the appropriate JIT to use, and
+// allows the user to specify a specific one on the commandline with -march=x.
+//
+//===----------------------------------------------------------------------===//
+
+#include "JIT.h"
+#include "llvm/Module.h"
+#include "llvm/ModuleProvider.h"
+#include "llvm/Target/SubtargetFeature.h"
+#include "llvm/Target/TargetMachine.h"
+#include "llvm/Target/TargetMachineRegistry.h"
+using namespace llvm;
+
+static cl::opt<const TargetMachineRegistry::Entry*, false, TargetNameParser>
+MArch("march", cl::desc("Architecture to generate assembly for:"));
+
+static cl::opt<std::string>
+MCPU("mcpu", 
+  cl::desc("Target a specific cpu type (-mcpu=help for details)"),
+  cl::value_desc("cpu-name"),
+  cl::init(""));
+
+static cl::list<std::string>
+MAttrs("mattr", 
+  cl::CommaSeparated,
+  cl::desc("Target specific attributes (-mattr=help for details)"),
+  cl::value_desc("a1,+a2,-a3,..."));
+
+/// create - Create an return a new JIT compiler if there is one available
+/// for the current target.  Otherwise, return null.
+///
+ExecutionEngine *JIT::create(ModuleProvider *MP, std::string *ErrorStr) {
+  const TargetMachineRegistry::Entry *TheArch = MArch;
+  if (TheArch == 0) {
+    std::string Error;
+    TheArch = TargetMachineRegistry::getClosestTargetForJIT(Error);
+    if (TheArch == 0) {
+      if (ErrorStr)
+        *ErrorStr = Error;
+      return 0;
+    }
+  } else if (TheArch->JITMatchQualityFn() == 0) {
+    cerr << "WARNING: This target JIT is not designed for the host you are"
+         << " running.  If bad things happen, please choose a different "
+         << "-march switch.\n";
+  }
+
+  // Package up features to be passed to target/subtarget
+  std::string FeaturesStr;
+  if (MCPU.size() || MAttrs.size()) {
+    SubtargetFeatures Features;
+    Features.setCPU(MCPU);
+    for (unsigned i = 0; i != MAttrs.size(); ++i)
+      Features.AddFeature(MAttrs[i]);
+    FeaturesStr = Features.getString();
+  }
+
+  // Allocate a target...
+  TargetMachine *Target = TheArch->CtorFn(*MP->getModule(), FeaturesStr);
+  assert(Target && "Could not allocate target machine!");
+
+  // If the target supports JIT code generation, return a new JIT now.
+  if (TargetJITInfo *TJ = Target->getJITInfo())
+    return new JIT(MP, *Target, *TJ);
+
+  if (ErrorStr)
+    *ErrorStr = "target does not support JIT code generation";
+  return 0;
+}
diff --git a/lib/ExecutionEngine/Makefile b/lib/ExecutionEngine/Makefile
new file mode 100644
index 0000000..c4998fc
--- /dev/null
+++ b/lib/ExecutionEngine/Makefile
@@ -0,0 +1,13 @@
+##===- lib/ExecutionEngine/Makefile ------------------------*- Makefile -*-===##
+# 
+#                     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.
+# 
+##===----------------------------------------------------------------------===##
+LEVEL = ../..
+LIBRARYNAME = LLVMExecutionEngine
+PARALLEL_DIRS = Interpreter JIT
+
+include $(LEVEL)/Makefile.common