[SystemZ] Use zeroing form of RISBG for some AND sequences
[oota-llvm.git] / lib / ExecutionEngine / Interpreter / Execution.cpp
index 97411d3943c26f6f647f11e05e06639354176794..b95a9e867c103f2cbf3a42cdaea609cec7adf52b 100644 (file)
@@ -2,8 +2,8 @@
 //
 //                     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 is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
 //
 //===----------------------------------------------------------------------===//
 //
 
 #define DEBUG_TYPE "interpreter"
 #include "Interpreter.h"
-#include "llvm/Constants.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Instructions.h"
-#include "llvm/CodeGen/IntrinsicLowering.h"
-#include "llvm/Support/GetElementPtrTypeIterator.h"
+#include "llvm/ADT/APInt.h"
 #include "llvm/ADT/Statistic.h"
+#include "llvm/CodeGen/IntrinsicLowering.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/GetElementPtrTypeIterator.h"
+#include "llvm/Support/MathExtras.h"
+#include <algorithm>
 #include <cmath>
 using namespace llvm;
 
 STATISTIC(NumDynamicInsts, "Number of dynamic instructions executed");
-static Interpreter *TheEE = 0;
 
+static cl::opt<bool> PrintVolatile("interpreter-print-volatile", cl::Hidden,
+          cl::desc("make the interpreter print every volatile load and store"));
 
 //===----------------------------------------------------------------------===//
-//                     Value Manipulation code
+//                     Various Helper Functions
 //===----------------------------------------------------------------------===//
 
-static GenericValue executeAddInst(GenericValue Src1, GenericValue Src2,
-                                   const Type *Ty);
-static GenericValue executeSubInst(GenericValue Src1, GenericValue Src2,
-                                   const Type *Ty);
-static GenericValue executeMulInst(GenericValue Src1, GenericValue Src2,
-                                   const Type *Ty);
-static GenericValue executeUDivInst(GenericValue Src1, GenericValue Src2,
-                                    const Type *Ty);
-static GenericValue executeSDivInst(GenericValue Src1, GenericValue Src2,
-                                    const Type *Ty);
-static GenericValue executeFDivInst(GenericValue Src1, GenericValue Src2,
-                                    const Type *Ty);
-static GenericValue executeURemInst(GenericValue Src1, GenericValue Src2,
-                                    const Type *Ty);
-static GenericValue executeSRemInst(GenericValue Src1, GenericValue Src2,
-                                    const Type *Ty);
-static GenericValue executeFRemInst(GenericValue Src1, GenericValue Src2,
-                                    const Type *Ty);
-static GenericValue executeAndInst(GenericValue Src1, GenericValue Src2,
-                                   const Type *Ty);
-static GenericValue executeOrInst(GenericValue Src1, GenericValue Src2,
-                                   const Type *Ty);
-static GenericValue executeXorInst(GenericValue Src1, GenericValue Src2,
-                                   const Type *Ty);
-static GenericValue executeCmpInst(unsigned predicate, GenericValue Src1, 
-                                   GenericValue Src2, const Type *Ty);
-static GenericValue executeShlInst(GenericValue Src1, GenericValue Src2,
-                                   const Type *Ty);
-static GenericValue executeLShrInst(GenericValue Src1, GenericValue Src2,
-                                    const Type *Ty);
-static GenericValue executeAShrInst(GenericValue Src1, GenericValue Src2,
-                                    const Type *Ty);
-static GenericValue executeSelectInst(GenericValue Src1, GenericValue Src2,
-                                      GenericValue Src3);
-
-GenericValue Interpreter::getConstantExprValue (ConstantExpr *CE,
-                                                ExecutionContext &SF) {
-  switch (CE->getOpcode()) {
-  case Instruction::Trunc:
-  case Instruction::ZExt:
-  case Instruction::SExt:
-  case Instruction::FPTrunc:
-  case Instruction::FPExt:
-  case Instruction::UIToFP:
-  case Instruction::SIToFP:
-  case Instruction::FPToUI:
-  case Instruction::FPToSI:
-  case Instruction::PtrToInt:
-  case Instruction::IntToPtr:
-  case Instruction::BitCast:
-    return executeCastOperation(Instruction::CastOps(CE->getOpcode()), 
-                                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::Add:
-    return executeAddInst(getOperandValue(CE->getOperand(0), SF),
-                          getOperandValue(CE->getOperand(1), SF),
-                          CE->getOperand(0)->getType());
-  case Instruction::Sub:
-    return executeSubInst(getOperandValue(CE->getOperand(0), SF),
-                          getOperandValue(CE->getOperand(1), SF),
-                          CE->getOperand(0)->getType());
-  case Instruction::Mul:
-    return executeMulInst(getOperandValue(CE->getOperand(0), SF),
-                          getOperandValue(CE->getOperand(1), SF),
-                          CE->getOperand(0)->getType());
-  case Instruction::SDiv:
-    return executeSDivInst(getOperandValue(CE->getOperand(0), SF),
-                           getOperandValue(CE->getOperand(1), SF),
-                           CE->getOperand(0)->getType());
-  case Instruction::UDiv:
-    return executeUDivInst(getOperandValue(CE->getOperand(0), SF),
-                           getOperandValue(CE->getOperand(1), SF),
-                           CE->getOperand(0)->getType());
-  case Instruction::FDiv:
-    return executeFDivInst(getOperandValue(CE->getOperand(0), SF),
-                           getOperandValue(CE->getOperand(1), SF),
-                           CE->getOperand(0)->getType());
-  case Instruction::URem:
-    return executeURemInst(getOperandValue(CE->getOperand(0), SF),
-                          getOperandValue(CE->getOperand(1), SF),
-                          CE->getOperand(0)->getType());
-  case Instruction::SRem:
-    return executeSRemInst(getOperandValue(CE->getOperand(0), SF),
-                          getOperandValue(CE->getOperand(1), SF),
-                          CE->getOperand(0)->getType());
-  case Instruction::FRem:
-    return executeFRemInst(getOperandValue(CE->getOperand(0), SF),
-                           getOperandValue(CE->getOperand(1), SF),
-                           CE->getOperand(0)->getType());
-  case Instruction::And:
-    return executeAndInst(getOperandValue(CE->getOperand(0), SF),
-                          getOperandValue(CE->getOperand(1), SF),
-                          CE->getOperand(0)->getType());
-  case Instruction::Or:
-    return executeOrInst(getOperandValue(CE->getOperand(0), SF),
-                         getOperandValue(CE->getOperand(1), SF),
-                         CE->getOperand(0)->getType());
-  case Instruction::Xor:
-    return executeXorInst(getOperandValue(CE->getOperand(0), SF),
-                          getOperandValue(CE->getOperand(1), SF),
-                          CE->getOperand(0)->getType());
-  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::Shl:
-    return executeShlInst(getOperandValue(CE->getOperand(0), SF),
-                          getOperandValue(CE->getOperand(1), SF),
-                          CE->getOperand(0)->getType());
-  case Instruction::LShr:
-    return executeLShrInst(getOperandValue(CE->getOperand(0), SF),
-                           getOperandValue(CE->getOperand(1), SF),
-                           CE->getOperand(0)->getType());
-  case Instruction::AShr:
-    return executeAShrInst(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:
-    cerr << "Unhandled ConstantExpr: " << *CE << "\n";
-    abort();
-    return GenericValue();
-  }
-}
-
-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];
-  }
-}
-
 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
+   case Type::TY##TyID: \
+     Dest.TY##Val = Src1.TY##Val OP Src2.TY##Val; \
+     break
 
-static GenericValue executeAddInst(GenericValue Src1, GenericValue Src2,
-                                   const Type *Ty) {
-  GenericValue Dest;
+static void executeFAddInst(GenericValue &Dest, GenericValue Src1,
+                            GenericValue Src2, Type *Ty) {
   switch (Ty->getTypeID()) {
-    IMPLEMENT_BINARY_OPERATOR(+, Int8);
-    IMPLEMENT_BINARY_OPERATOR(+, Int16);
-    IMPLEMENT_BINARY_OPERATOR(+, Int32);
-    IMPLEMENT_BINARY_OPERATOR(+, Int64);
     IMPLEMENT_BINARY_OPERATOR(+, Float);
     IMPLEMENT_BINARY_OPERATOR(+, Double);
   default:
-    cerr << "Unhandled type for Add instruction: " << *Ty << "\n";
-    abort();
+    dbgs() << "Unhandled type for FAdd instruction: " << *Ty << "\n";
+    llvm_unreachable(0);
   }
-  return Dest;
 }
 
-static GenericValue executeSubInst(GenericValue Src1, GenericValue Src2,
-                                   const Type *Ty) {
-  GenericValue Dest;
+static void executeFSubInst(GenericValue &Dest, GenericValue Src1,
+                            GenericValue Src2, Type *Ty) {
   switch (Ty->getTypeID()) {
-    IMPLEMENT_BINARY_OPERATOR(-, Int8);
-    IMPLEMENT_BINARY_OPERATOR(-, Int16);
-    IMPLEMENT_BINARY_OPERATOR(-, Int32);
-    IMPLEMENT_BINARY_OPERATOR(-, Int64);
     IMPLEMENT_BINARY_OPERATOR(-, Float);
     IMPLEMENT_BINARY_OPERATOR(-, Double);
   default:
-    cerr << "Unhandled type for Sub instruction: " << *Ty << "\n";
-    abort();
+    dbgs() << "Unhandled type for FSub instruction: " << *Ty << "\n";
+    llvm_unreachable(0);
   }
-  return Dest;
 }
 
-static GenericValue executeMulInst(GenericValue Src1, GenericValue Src2,
-                                   const Type *Ty) {
-  GenericValue Dest;
+static void executeFMulInst(GenericValue &Dest, GenericValue Src1,
+                            GenericValue Src2, Type *Ty) {
   switch (Ty->getTypeID()) {
-    IMPLEMENT_BINARY_OPERATOR(*, Int8);
-    IMPLEMENT_BINARY_OPERATOR(*, Int16);
-    IMPLEMENT_BINARY_OPERATOR(*, Int32);
-    IMPLEMENT_BINARY_OPERATOR(*, Int64);
     IMPLEMENT_BINARY_OPERATOR(*, Float);
     IMPLEMENT_BINARY_OPERATOR(*, Double);
   default:
-    cerr << "Unhandled type for Mul instruction: " << *Ty << "\n";
-    abort();
+    dbgs() << "Unhandled type for FMul instruction: " << *Ty << "\n";
+    llvm_unreachable(0);
   }
-  return Dest;
 }
 
-#define IMPLEMENT_SIGNLESS_BINOP(OP, TY, CAST) \
-   case Type::TY##TyID: Dest.TY##Val = \
-    ((CAST)Src1.TY##Val) OP ((CAST)Src2.TY##Val); break
-
-static GenericValue executeUDivInst(GenericValue Src1, GenericValue Src2,
-                                   const Type *Ty) {
-  GenericValue Dest;
-  switch (Ty->getTypeID()) {
-    IMPLEMENT_SIGNLESS_BINOP(/, Int8,  uint8_t);
-    IMPLEMENT_SIGNLESS_BINOP(/, Int16, uint16_t);
-    IMPLEMENT_SIGNLESS_BINOP(/, Int32, uint32_t);
-    IMPLEMENT_SIGNLESS_BINOP(/, Int64, uint64_t);
-  default:
-    cerr << "Unhandled type for UDiv instruction: " << *Ty << "\n";
-    abort();
-  }
-  return Dest;
-}
-
-static GenericValue executeSDivInst(GenericValue Src1, GenericValue Src2,
-                                   const Type *Ty) {
-  GenericValue Dest;
-  switch (Ty->getTypeID()) {
-    IMPLEMENT_SIGNLESS_BINOP(/, Int8,  int8_t);
-    IMPLEMENT_SIGNLESS_BINOP(/, Int16, int16_t);
-    IMPLEMENT_SIGNLESS_BINOP(/, Int32, int32_t);
-    IMPLEMENT_SIGNLESS_BINOP(/, Int64, int64_t);
-  default:
-    cerr << "Unhandled type for SDiv instruction: " << *Ty << "\n";
-    abort();
-  }
-  return Dest;
-}
-
-static GenericValue executeFDivInst(GenericValue Src1, GenericValue Src2,
-                                   const Type *Ty) {
-  GenericValue Dest;
+static void executeFDivInst(GenericValue &Dest, GenericValue Src1, 
+                            GenericValue Src2, Type *Ty) {
   switch (Ty->getTypeID()) {
     IMPLEMENT_BINARY_OPERATOR(/, Float);
     IMPLEMENT_BINARY_OPERATOR(/, Double);
   default:
-    cerr << "Unhandled type for Div instruction: " << *Ty << "\n";
-    abort();
-  }
-  return Dest;
-}
-
-static GenericValue executeURemInst(GenericValue Src1, GenericValue Src2,
-                                   const Type *Ty) {
-  GenericValue Dest;
-  switch (Ty->getTypeID()) {
-    IMPLEMENT_SIGNLESS_BINOP(%, Int8,  uint8_t);
-    IMPLEMENT_SIGNLESS_BINOP(%, Int16, uint16_t);
-    IMPLEMENT_SIGNLESS_BINOP(%, Int32, uint32_t);
-    IMPLEMENT_SIGNLESS_BINOP(%, Int64, uint64_t );
-  default:
-    cerr << "Unhandled type for URem instruction: " << *Ty << "\n";
-    abort();
+    dbgs() << "Unhandled type for FDiv instruction: " << *Ty << "\n";
+    llvm_unreachable(0);
   }
-  return Dest;
 }
 
-static GenericValue executeSRemInst(GenericValue Src1, GenericValue Src2,
-                                   const Type *Ty) {
-  GenericValue Dest;
-  switch (Ty->getTypeID()) {
-    IMPLEMENT_SIGNLESS_BINOP(%, Int8,  int8_t);
-    IMPLEMENT_SIGNLESS_BINOP(%, Int16, int16_t);
-    IMPLEMENT_SIGNLESS_BINOP(%, Int32, int32_t);
-    IMPLEMENT_SIGNLESS_BINOP(%, Int64, int64_t);
-  default:
-    cerr << "Unhandled type for Rem instruction: " << *Ty << "\n";
-    abort();
-  }
-  return Dest;
-}
-
-static GenericValue executeFRemInst(GenericValue Src1, GenericValue Src2,
-                                   const Type *Ty) {
-  GenericValue Dest;
+static void executeFRemInst(GenericValue &Dest, GenericValue Src1, 
+                            GenericValue Src2, Type *Ty) {
   switch (Ty->getTypeID()) {
   case Type::FloatTyID:
     Dest.FloatVal = fmod(Src1.FloatVal, Src2.FloatVal);
@@ -329,243 +104,185 @@ static GenericValue executeFRemInst(GenericValue Src1, GenericValue Src2,
     Dest.DoubleVal = fmod(Src1.DoubleVal, Src2.DoubleVal);
     break;
   default:
-    cerr << "Unhandled type for Rem instruction: " << *Ty << "\n";
-    abort();
+    dbgs() << "Unhandled type for Rem instruction: " << *Ty << "\n";
+    llvm_unreachable(0);
   }
-  return Dest;
-}
-
-static GenericValue executeAndInst(GenericValue Src1, GenericValue Src2,
-                                   const Type *Ty) {
-  GenericValue Dest;
-  switch (Ty->getTypeID()) {
-    IMPLEMENT_BINARY_OPERATOR(&, Bool);
-    IMPLEMENT_BINARY_OPERATOR(&, Int8);
-    IMPLEMENT_BINARY_OPERATOR(&, Int16);
-    IMPLEMENT_BINARY_OPERATOR(&, Int32);
-    IMPLEMENT_BINARY_OPERATOR(&, Int64);
-  default:
-    cerr << "Unhandled type for And instruction: " << *Ty << "\n";
-    abort();
-  }
-  return Dest;
 }
 
-static GenericValue executeOrInst(GenericValue Src1, GenericValue Src2,
-                                  const Type *Ty) {
-  GenericValue Dest;
-  switch (Ty->getTypeID()) {
-    IMPLEMENT_BINARY_OPERATOR(|, Bool);
-    IMPLEMENT_BINARY_OPERATOR(|, Int8);
-    IMPLEMENT_BINARY_OPERATOR(|, Int16);
-    IMPLEMENT_BINARY_OPERATOR(|, Int32);
-    IMPLEMENT_BINARY_OPERATOR(|, Int64);
-  default:
-    cerr << "Unhandled type for Or instruction: " << *Ty << "\n";
-    abort();
-  }
-  return Dest;
-}
-
-static GenericValue executeXorInst(GenericValue Src1, GenericValue Src2,
-                                   const Type *Ty) {
-  GenericValue Dest;
-  switch (Ty->getTypeID()) {
-    IMPLEMENT_BINARY_OPERATOR(^, Bool);
-    IMPLEMENT_BINARY_OPERATOR(^, Int8);
-    IMPLEMENT_BINARY_OPERATOR(^, Int16);
-    IMPLEMENT_BINARY_OPERATOR(^, Int32);
-    IMPLEMENT_BINARY_OPERATOR(^, Int64);
-  default:
-    cerr << "Unhandled type for Xor instruction: " << *Ty << "\n";
-    abort();
-  }
-  return Dest;
-}
+#define IMPLEMENT_INTEGER_ICMP(OP, TY) \
+   case Type::IntegerTyID:  \
+      Dest.IntVal = APInt(1,Src1.IntVal.OP(Src2.IntVal)); \
+      break;
 
-#define IMPLEMENT_ICMP(OP, TY, CAST) \
-   case Type::TY##TyID: Dest.BoolVal = \
-     ((CAST)Src1.TY##Val) OP ((CAST)Src2.TY##Val); break
+#define IMPLEMENT_VECTOR_INTEGER_ICMP(OP, TY)                        \
+  case Type::VectorTyID: {                                           \
+    assert(Src1.AggregateVal.size() == Src2.AggregateVal.size());    \
+    Dest.AggregateVal.resize( Src1.AggregateVal.size() );            \
+    for( uint32_t _i=0;_i<Src1.AggregateVal.size();_i++)             \
+      Dest.AggregateVal[_i].IntVal = APInt(1,                        \
+      Src1.AggregateVal[_i].IntVal.OP(Src2.AggregateVal[_i].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_POINTERCMP(OP) \
+#define IMPLEMENT_POINTER_ICMP(OP) \
    case Type::PointerTyID: \
-        Dest.BoolVal = (void*)(intptr_t)Src1.PointerVal OP \
-                       (void*)(intptr_t)Src2.PointerVal; break
+      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) {
+                                   Type *Ty) {
   GenericValue Dest;
   switch (Ty->getTypeID()) {
-    IMPLEMENT_ICMP(==, Int8,  uint8_t);
-    IMPLEMENT_ICMP(==, Int16, uint16_t);
-    IMPLEMENT_ICMP(==, Int32, uint32_t);
-    IMPLEMENT_ICMP(==, Int64, uint64_t);
-    IMPLEMENT_POINTERCMP(==);
+    IMPLEMENT_INTEGER_ICMP(eq,Ty);
+    IMPLEMENT_VECTOR_INTEGER_ICMP(eq,Ty);
+    IMPLEMENT_POINTER_ICMP(==);
   default:
-    cerr << "Unhandled type for ICMP_EQ predicate: " << *Ty << "\n";
-    abort();
+    dbgs() << "Unhandled type for ICMP_EQ predicate: " << *Ty << "\n";
+    llvm_unreachable(0);
   }
   return Dest;
 }
 
 static GenericValue executeICMP_NE(GenericValue Src1, GenericValue Src2,
-                                   const Type *Ty) {
+                                   Type *Ty) {
   GenericValue Dest;
   switch (Ty->getTypeID()) {
-    IMPLEMENT_ICMP(!=, Int8,  uint8_t);
-    IMPLEMENT_ICMP(!=, Int16, uint16_t);
-    IMPLEMENT_ICMP(!=, Int32, uint32_t);
-    IMPLEMENT_ICMP(!=, Int64, uint64_t);
-    IMPLEMENT_POINTERCMP(!=);
+    IMPLEMENT_INTEGER_ICMP(ne,Ty);
+    IMPLEMENT_VECTOR_INTEGER_ICMP(ne,Ty);
+    IMPLEMENT_POINTER_ICMP(!=);
   default:
-    cerr << "Unhandled type for ICMP_NE predicate: " << *Ty << "\n";
-    abort();
+    dbgs() << "Unhandled type for ICMP_NE predicate: " << *Ty << "\n";
+    llvm_unreachable(0);
   }
   return Dest;
 }
 
 static GenericValue executeICMP_ULT(GenericValue Src1, GenericValue Src2,
-                                    const Type *Ty) {
+                                    Type *Ty) {
   GenericValue Dest;
   switch (Ty->getTypeID()) {
-    IMPLEMENT_ICMP(<, Int8,  uint8_t);
-    IMPLEMENT_ICMP(<, Int16, uint16_t);
-    IMPLEMENT_ICMP(<, Int32, uint32_t);
-    IMPLEMENT_ICMP(<, Int64, uint64_t);
-    IMPLEMENT_POINTERCMP(<);
+    IMPLEMENT_INTEGER_ICMP(ult,Ty);
+    IMPLEMENT_VECTOR_INTEGER_ICMP(ult,Ty);
+    IMPLEMENT_POINTER_ICMP(<);
   default:
-    cerr << "Unhandled type for ICMP_ULT predicate: " << *Ty << "\n";
-    abort();
+    dbgs() << "Unhandled type for ICMP_ULT predicate: " << *Ty << "\n";
+    llvm_unreachable(0);
   }
   return Dest;
 }
 
 static GenericValue executeICMP_SLT(GenericValue Src1, GenericValue Src2,
-                                    const Type *Ty) {
+                                    Type *Ty) {
   GenericValue Dest;
   switch (Ty->getTypeID()) {
-    IMPLEMENT_ICMP(<, Int8,  int8_t);
-    IMPLEMENT_ICMP(<, Int16, int16_t);
-    IMPLEMENT_ICMP(<, Int32, int32_t);
-    IMPLEMENT_ICMP(<, Int64, int64_t);
-    IMPLEMENT_POINTERCMP(<);
+    IMPLEMENT_INTEGER_ICMP(slt,Ty);
+    IMPLEMENT_VECTOR_INTEGER_ICMP(slt,Ty);
+    IMPLEMENT_POINTER_ICMP(<);
   default:
-    cerr << "Unhandled type for ICMP_SLT predicate: " << *Ty << "\n";
-    abort();
+    dbgs() << "Unhandled type for ICMP_SLT predicate: " << *Ty << "\n";
+    llvm_unreachable(0);
   }
   return Dest;
 }
 
 static GenericValue executeICMP_UGT(GenericValue Src1, GenericValue Src2,
-                                    const Type *Ty) {
+                                    Type *Ty) {
   GenericValue Dest;
   switch (Ty->getTypeID()) {
-    IMPLEMENT_ICMP(>, Int8,  uint8_t);
-    IMPLEMENT_ICMP(>, Int16, uint16_t);
-    IMPLEMENT_ICMP(>, Int32, uint32_t);
-    IMPLEMENT_ICMP(>, Int64, uint64_t);
-    IMPLEMENT_POINTERCMP(>);
+    IMPLEMENT_INTEGER_ICMP(ugt,Ty);
+    IMPLEMENT_VECTOR_INTEGER_ICMP(ugt,Ty);
+    IMPLEMENT_POINTER_ICMP(>);
   default:
-    cerr << "Unhandled type for ICMP_UGT predicate: " << *Ty << "\n";
-    abort();
+    dbgs() << "Unhandled type for ICMP_UGT predicate: " << *Ty << "\n";
+    llvm_unreachable(0);
   }
   return Dest;
 }
 
 static GenericValue executeICMP_SGT(GenericValue Src1, GenericValue Src2,
-                                    const Type *Ty) {
+                                    Type *Ty) {
   GenericValue Dest;
   switch (Ty->getTypeID()) {
-    IMPLEMENT_ICMP(>, Int8,  int8_t);
-    IMPLEMENT_ICMP(>, Int16, int16_t);
-    IMPLEMENT_ICMP(>, Int32, int32_t);
-    IMPLEMENT_ICMP(>, Int64, int64_t);
-    IMPLEMENT_POINTERCMP(>);
+    IMPLEMENT_INTEGER_ICMP(sgt,Ty);
+    IMPLEMENT_VECTOR_INTEGER_ICMP(sgt,Ty);
+    IMPLEMENT_POINTER_ICMP(>);
   default:
-    cerr << "Unhandled type for ICMP_SGT predicate: " << *Ty << "\n";
-    abort();
+    dbgs() << "Unhandled type for ICMP_SGT predicate: " << *Ty << "\n";
+    llvm_unreachable(0);
   }
   return Dest;
 }
 
 static GenericValue executeICMP_ULE(GenericValue Src1, GenericValue Src2,
-                                    const Type *Ty) {
+                                    Type *Ty) {
   GenericValue Dest;
   switch (Ty->getTypeID()) {
-    IMPLEMENT_ICMP(<=, Int8,  uint8_t);
-    IMPLEMENT_ICMP(<=, Int16, uint16_t);
-    IMPLEMENT_ICMP(<=, Int32, uint32_t);
-    IMPLEMENT_ICMP(<=, Int64, uint64_t);
-    IMPLEMENT_POINTERCMP(<=);
+    IMPLEMENT_INTEGER_ICMP(ule,Ty);
+    IMPLEMENT_VECTOR_INTEGER_ICMP(ule,Ty);
+    IMPLEMENT_POINTER_ICMP(<=);
   default:
-    cerr << "Unhandled type for ICMP_ULE predicate: " << *Ty << "\n";
-    abort();
+    dbgs() << "Unhandled type for ICMP_ULE predicate: " << *Ty << "\n";
+    llvm_unreachable(0);
   }
   return Dest;
 }
 
 static GenericValue executeICMP_SLE(GenericValue Src1, GenericValue Src2,
-                                    const Type *Ty) {
+                                    Type *Ty) {
   GenericValue Dest;
   switch (Ty->getTypeID()) {
-    IMPLEMENT_ICMP(<=, Int8,  int8_t);
-    IMPLEMENT_ICMP(<=, Int16, int16_t);
-    IMPLEMENT_ICMP(<=, Int32, int32_t);
-    IMPLEMENT_ICMP(<=, Int64, int64_t);
-    IMPLEMENT_POINTERCMP(<=);
+    IMPLEMENT_INTEGER_ICMP(sle,Ty);
+    IMPLEMENT_VECTOR_INTEGER_ICMP(sle,Ty);
+    IMPLEMENT_POINTER_ICMP(<=);
   default:
-    cerr << "Unhandled type for ICMP_SLE predicate: " << *Ty << "\n";
-    abort();
+    dbgs() << "Unhandled type for ICMP_SLE predicate: " << *Ty << "\n";
+    llvm_unreachable(0);
   }
   return Dest;
 }
 
 static GenericValue executeICMP_UGE(GenericValue Src1, GenericValue Src2,
-                                    const Type *Ty) {
+                                    Type *Ty) {
   GenericValue Dest;
   switch (Ty->getTypeID()) {
-    IMPLEMENT_ICMP(>=, Int8,  uint8_t);
-    IMPLEMENT_ICMP(>=, Int16, uint16_t);
-    IMPLEMENT_ICMP(>=, Int32, uint32_t);
-    IMPLEMENT_ICMP(>=, Int64, uint64_t);
-    IMPLEMENT_POINTERCMP(>=);
+    IMPLEMENT_INTEGER_ICMP(uge,Ty);
+    IMPLEMENT_VECTOR_INTEGER_ICMP(uge,Ty);
+    IMPLEMENT_POINTER_ICMP(>=);
   default:
-    cerr << "Unhandled type for ICMP_UGE predicate: " << *Ty << "\n";
-    abort();
+    dbgs() << "Unhandled type for ICMP_UGE predicate: " << *Ty << "\n";
+    llvm_unreachable(0);
   }
   return Dest;
 }
 
 static GenericValue executeICMP_SGE(GenericValue Src1, GenericValue Src2,
-                                    const Type *Ty) {
+                                    Type *Ty) {
   GenericValue Dest;
   switch (Ty->getTypeID()) {
-    IMPLEMENT_ICMP(>=, Int8,  int8_t);
-    IMPLEMENT_ICMP(>=, Int16, int16_t);
-    IMPLEMENT_ICMP(>=, Int32, int32_t);
-    IMPLEMENT_ICMP(>=, Int64, int64_t);
-    IMPLEMENT_POINTERCMP(>=);
+    IMPLEMENT_INTEGER_ICMP(sge,Ty);
+    IMPLEMENT_VECTOR_INTEGER_ICMP(sge,Ty);
+    IMPLEMENT_POINTER_ICMP(>=);
   default:
-    cerr << "Unhandled type for ICMP_SGE predicate: " << *Ty << "\n";
-    abort();
+    dbgs() << "Unhandled type for ICMP_SGE predicate: " << *Ty << "\n";
+    llvm_unreachable(0);
   }
   return Dest;
 }
 
 void Interpreter::visitICmpInst(ICmpInst &I) {
   ExecutionContext &SF = ECStack.back();
-  const Type *Ty    = I.getOperand(0)->getType();
+  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_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;
@@ -575,129 +292,356 @@ void Interpreter::visitICmpInst(ICmpInst &I) {
   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();
+    dbgs() << "Don't know how to handle this ICmp predicate!\n-->" << I;
+    llvm_unreachable(0);
   }
  
   SetValue(&I, R, SF);
 }
 
 #define IMPLEMENT_FCMP(OP, TY) \
-   case Type::TY##TyID: Dest.BoolVal = Src1.TY##Val OP Src2.TY##Val; break
+   case Type::TY##TyID: \
+     Dest.IntVal = APInt(1,Src1.TY##Val OP Src2.TY##Val); \
+     break
 
-static GenericValue executeFCMP_EQ(GenericValue Src1, GenericValue Src2,
-                                   const Type *Ty) {
+#define IMPLEMENT_VECTOR_FCMP_T(OP, TY)                             \
+  assert(Src1.AggregateVal.size() == Src2.AggregateVal.size());     \
+  Dest.AggregateVal.resize( Src1.AggregateVal.size() );             \
+  for( uint32_t _i=0;_i<Src1.AggregateVal.size();_i++)              \
+    Dest.AggregateVal[_i].IntVal = APInt(1,                         \
+    Src1.AggregateVal[_i].TY##Val OP Src2.AggregateVal[_i].TY##Val);\
+  break;
+
+#define IMPLEMENT_VECTOR_FCMP(OP)                                   \
+  case Type::VectorTyID:                                            \
+    if(dyn_cast<VectorType>(Ty)->getElementType()->isFloatTy()) {   \
+      IMPLEMENT_VECTOR_FCMP_T(OP, Float);                           \
+    } else {                                                        \
+        IMPLEMENT_VECTOR_FCMP_T(OP, Double);                        \
+    }
+
+static GenericValue executeFCMP_OEQ(GenericValue Src1, GenericValue Src2,
+                                   Type *Ty) {
   GenericValue Dest;
   switch (Ty->getTypeID()) {
     IMPLEMENT_FCMP(==, Float);
     IMPLEMENT_FCMP(==, Double);
+    IMPLEMENT_VECTOR_FCMP(==);
   default:
-    cerr << "Unhandled type for SetEQ instruction: " << *Ty << "\n";
-    abort();
+    dbgs() << "Unhandled type for FCmp EQ instruction: " << *Ty << "\n";
+    llvm_unreachable(0);
   }
   return Dest;
 }
 
-static GenericValue executeFCMP_NE(GenericValue Src1, GenericValue Src2,
-                                   const Type *Ty) {
+#define IMPLEMENT_SCALAR_NANS(TY, X,Y)                                      \
+  if (TY->isFloatTy()) {                                                    \
+    if (X.FloatVal != X.FloatVal || Y.FloatVal != Y.FloatVal) {             \
+      Dest.IntVal = APInt(1,false);                                         \
+      return Dest;                                                          \
+    }                                                                       \
+  } else {                                                                  \
+    if (X.DoubleVal != X.DoubleVal || Y.DoubleVal != Y.DoubleVal) {         \
+      Dest.IntVal = APInt(1,false);                                         \
+      return Dest;                                                          \
+    }                                                                       \
+  }
+
+#define MASK_VECTOR_NANS_T(X,Y, TZ, FLAG)                                   \
+  assert(X.AggregateVal.size() == Y.AggregateVal.size());                   \
+  Dest.AggregateVal.resize( X.AggregateVal.size() );                        \
+  for( uint32_t _i=0;_i<X.AggregateVal.size();_i++) {                       \
+    if (X.AggregateVal[_i].TZ##Val != X.AggregateVal[_i].TZ##Val ||         \
+        Y.AggregateVal[_i].TZ##Val != Y.AggregateVal[_i].TZ##Val)           \
+      Dest.AggregateVal[_i].IntVal = APInt(1,FLAG);                         \
+    else  {                                                                 \
+      Dest.AggregateVal[_i].IntVal = APInt(1,!FLAG);                        \
+    }                                                                       \
+  }
+
+#define MASK_VECTOR_NANS(TY, X,Y, FLAG)                                     \
+  if (TY->isVectorTy()) {                                                   \
+    if (dyn_cast<VectorType>(TY)->getElementType()->isFloatTy()) {          \
+      MASK_VECTOR_NANS_T(X, Y, Float, FLAG)                                 \
+    } else {                                                                \
+      MASK_VECTOR_NANS_T(X, Y, Double, FLAG)                                \
+    }                                                                       \
+  }                                                                         \
+
+
+
+static GenericValue executeFCMP_ONE(GenericValue Src1, GenericValue Src2,
+                                    Type *Ty)
+{
   GenericValue Dest;
+  // if input is scalar value and Src1 or Src2 is NaN return false
+  IMPLEMENT_SCALAR_NANS(Ty, Src1, Src2)
+  // if vector input detect NaNs and fill mask
+  MASK_VECTOR_NANS(Ty, Src1, Src2, false)
+  GenericValue DestMask = Dest;
   switch (Ty->getTypeID()) {
     IMPLEMENT_FCMP(!=, Float);
     IMPLEMENT_FCMP(!=, Double);
-
-  default:
-    cerr << "Unhandled type for SetNE instruction: " << *Ty << "\n";
-    abort();
+    IMPLEMENT_VECTOR_FCMP(!=);
+    default:
+      dbgs() << "Unhandled type for FCmp NE instruction: " << *Ty << "\n";
+      llvm_unreachable(0);
   }
+  // in vector case mask out NaN elements
+  if (Ty->isVectorTy())
+    for( size_t _i=0; _i<Src1.AggregateVal.size(); _i++)
+      if (DestMask.AggregateVal[_i].IntVal == false)
+        Dest.AggregateVal[_i].IntVal = APInt(1,false);
+
   return Dest;
 }
 
-static GenericValue executeFCMP_LE(GenericValue Src1, GenericValue Src2,
-                                   const Type *Ty) {
+static GenericValue executeFCMP_OLE(GenericValue Src1, GenericValue Src2,
+                                   Type *Ty) {
   GenericValue Dest;
   switch (Ty->getTypeID()) {
     IMPLEMENT_FCMP(<=, Float);
     IMPLEMENT_FCMP(<=, Double);
+    IMPLEMENT_VECTOR_FCMP(<=);
   default:
-    cerr << "Unhandled type for SetLE instruction: " << *Ty << "\n";
-    abort();
+    dbgs() << "Unhandled type for FCmp LE instruction: " << *Ty << "\n";
+    llvm_unreachable(0);
   }
   return Dest;
 }
 
-static GenericValue executeFCMP_GE(GenericValue Src1, GenericValue Src2,
-                                   const Type *Ty) {
+static GenericValue executeFCMP_OGE(GenericValue Src1, GenericValue Src2,
+                                   Type *Ty) {
   GenericValue Dest;
   switch (Ty->getTypeID()) {
     IMPLEMENT_FCMP(>=, Float);
     IMPLEMENT_FCMP(>=, Double);
+    IMPLEMENT_VECTOR_FCMP(>=);
   default:
-    cerr << "Unhandled type for SetGE instruction: " << *Ty << "\n";
-    abort();
+    dbgs() << "Unhandled type for FCmp GE instruction: " << *Ty << "\n";
+    llvm_unreachable(0);
   }
   return Dest;
 }
 
-static GenericValue executeFCMP_LT(GenericValue Src1, GenericValue Src2,
-                                   const Type *Ty) {
+static GenericValue executeFCMP_OLT(GenericValue Src1, GenericValue Src2,
+                                   Type *Ty) {
   GenericValue Dest;
   switch (Ty->getTypeID()) {
     IMPLEMENT_FCMP(<, Float);
     IMPLEMENT_FCMP(<, Double);
+    IMPLEMENT_VECTOR_FCMP(<);
   default:
-    cerr << "Unhandled type for SetLT instruction: " << *Ty << "\n";
-    abort();
+    dbgs() << "Unhandled type for FCmp LT instruction: " << *Ty << "\n";
+    llvm_unreachable(0);
   }
   return Dest;
 }
 
-static GenericValue executeFCMP_GT(GenericValue Src1, GenericValue Src2,
-                                     const Type *Ty) {
+static GenericValue executeFCMP_OGT(GenericValue Src1, GenericValue Src2,
+                                     Type *Ty) {
   GenericValue Dest;
   switch (Ty->getTypeID()) {
     IMPLEMENT_FCMP(>, Float);
     IMPLEMENT_FCMP(>, Double);
+    IMPLEMENT_VECTOR_FCMP(>);
   default:
-    cerr << "Unhandled type for SetGT instruction: " << *Ty << "\n";
-    abort();
+    dbgs() << "Unhandled type for FCmp GT instruction: " << *Ty << "\n";
+    llvm_unreachable(0);
   }
   return Dest;
 }
 
+#define IMPLEMENT_UNORDERED(TY, X,Y)                                     \
+  if (TY->isFloatTy()) {                                                 \
+    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;                                                         \
+  }
+
+#define IMPLEMENT_VECTOR_UNORDERED(TY, X,Y, _FUNC)                       \
+  if (TY->isVectorTy()) {                                                \
+    GenericValue DestMask = Dest;                                        \
+    Dest = _FUNC(Src1, Src2, Ty);                                        \
+      for( size_t _i=0; _i<Src1.AggregateVal.size(); _i++)               \
+        if (DestMask.AggregateVal[_i].IntVal == true)                    \
+          Dest.AggregateVal[_i].IntVal = APInt(1,true);                  \
+      return Dest;                                                       \
+  }
+
+static GenericValue executeFCMP_UEQ(GenericValue Src1, GenericValue Src2,
+                                   Type *Ty) {
+  GenericValue Dest;
+  IMPLEMENT_UNORDERED(Ty, Src1, Src2)
+  MASK_VECTOR_NANS(Ty, Src1, Src2, true)
+  IMPLEMENT_VECTOR_UNORDERED(Ty, Src1, Src2, executeFCMP_OEQ)
+  return executeFCMP_OEQ(Src1, Src2, Ty);
+
+}
+
+static GenericValue executeFCMP_UNE(GenericValue Src1, GenericValue Src2,
+                                   Type *Ty) {
+  GenericValue Dest;
+  IMPLEMENT_UNORDERED(Ty, Src1, Src2)
+  MASK_VECTOR_NANS(Ty, Src1, Src2, true)
+  IMPLEMENT_VECTOR_UNORDERED(Ty, Src1, Src2, executeFCMP_ONE)
+  return executeFCMP_ONE(Src1, Src2, Ty);
+}
+
+static GenericValue executeFCMP_ULE(GenericValue Src1, GenericValue Src2,
+                                   Type *Ty) {
+  GenericValue Dest;
+  IMPLEMENT_UNORDERED(Ty, Src1, Src2)
+  MASK_VECTOR_NANS(Ty, Src1, Src2, true)
+  IMPLEMENT_VECTOR_UNORDERED(Ty, Src1, Src2, executeFCMP_OLE)
+  return executeFCMP_OLE(Src1, Src2, Ty);
+}
+
+static GenericValue executeFCMP_UGE(GenericValue Src1, GenericValue Src2,
+                                   Type *Ty) {
+  GenericValue Dest;
+  IMPLEMENT_UNORDERED(Ty, Src1, Src2)
+  MASK_VECTOR_NANS(Ty, Src1, Src2, true)
+  IMPLEMENT_VECTOR_UNORDERED(Ty, Src1, Src2, executeFCMP_OGE)
+  return executeFCMP_OGE(Src1, Src2, Ty);
+}
+
+static GenericValue executeFCMP_ULT(GenericValue Src1, GenericValue Src2,
+                                   Type *Ty) {
+  GenericValue Dest;
+  IMPLEMENT_UNORDERED(Ty, Src1, Src2)
+  MASK_VECTOR_NANS(Ty, Src1, Src2, true)
+  IMPLEMENT_VECTOR_UNORDERED(Ty, Src1, Src2, executeFCMP_OLT)
+  return executeFCMP_OLT(Src1, Src2, Ty);
+}
+
+static GenericValue executeFCMP_UGT(GenericValue Src1, GenericValue Src2,
+                                     Type *Ty) {
+  GenericValue Dest;
+  IMPLEMENT_UNORDERED(Ty, Src1, Src2)
+  MASK_VECTOR_NANS(Ty, Src1, Src2, true)
+  IMPLEMENT_VECTOR_UNORDERED(Ty, Src1, Src2, executeFCMP_OGT)
+  return executeFCMP_OGT(Src1, Src2, Ty);
+}
+
+static GenericValue executeFCMP_ORD(GenericValue Src1, GenericValue Src2,
+                                     Type *Ty) {
+  GenericValue Dest;
+  if(Ty->isVectorTy()) {
+    assert(Src1.AggregateVal.size() == Src2.AggregateVal.size());
+    Dest.AggregateVal.resize( Src1.AggregateVal.size() );
+    if(dyn_cast<VectorType>(Ty)->getElementType()->isFloatTy()) {
+      for( size_t _i=0;_i<Src1.AggregateVal.size();_i++)
+        Dest.AggregateVal[_i].IntVal = APInt(1,
+        ( (Src1.AggregateVal[_i].FloatVal ==
+        Src1.AggregateVal[_i].FloatVal) &&
+        (Src2.AggregateVal[_i].FloatVal ==
+        Src2.AggregateVal[_i].FloatVal)));
+    } else {
+      for( size_t _i=0;_i<Src1.AggregateVal.size();_i++)
+        Dest.AggregateVal[_i].IntVal = APInt(1,
+        ( (Src1.AggregateVal[_i].DoubleVal ==
+        Src1.AggregateVal[_i].DoubleVal) &&
+        (Src2.AggregateVal[_i].DoubleVal ==
+        Src2.AggregateVal[_i].DoubleVal)));
+    }
+  } else if (Ty->isFloatTy())
+    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,
+                                     Type *Ty) {
+  GenericValue Dest;
+  if(Ty->isVectorTy()) {
+    assert(Src1.AggregateVal.size() == Src2.AggregateVal.size());
+    Dest.AggregateVal.resize( Src1.AggregateVal.size() );
+    if(dyn_cast<VectorType>(Ty)->getElementType()->isFloatTy()) {
+      for( size_t _i=0;_i<Src1.AggregateVal.size();_i++)
+        Dest.AggregateVal[_i].IntVal = APInt(1,
+        ( (Src1.AggregateVal[_i].FloatVal !=
+           Src1.AggregateVal[_i].FloatVal) ||
+          (Src2.AggregateVal[_i].FloatVal !=
+           Src2.AggregateVal[_i].FloatVal)));
+      } else {
+        for( size_t _i=0;_i<Src1.AggregateVal.size();_i++)
+          Dest.AggregateVal[_i].IntVal = APInt(1,
+          ( (Src1.AggregateVal[_i].DoubleVal !=
+             Src1.AggregateVal[_i].DoubleVal) ||
+            (Src2.AggregateVal[_i].DoubleVal !=
+             Src2.AggregateVal[_i].DoubleVal)));
+      }
+  } else if (Ty->isFloatTy())
+    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_BOOL(GenericValue Src1, GenericValue Src2,
+                                    const Type *Ty, const bool val) {
+  GenericValue Dest;
+    if(Ty->isVectorTy()) {
+      assert(Src1.AggregateVal.size() == Src2.AggregateVal.size());
+      Dest.AggregateVal.resize( Src1.AggregateVal.size() );
+      for( size_t _i=0; _i<Src1.AggregateVal.size(); _i++)
+        Dest.AggregateVal[_i].IntVal = APInt(1,val);
+    } else {
+      Dest.IntVal = APInt(1, val);
+    }
+
+    return Dest;
+}
+
 void Interpreter::visitFCmpInst(FCmpInst &I) {
   ExecutionContext &SF = ECStack.back();
-  const Type *Ty    = I.getOperand(0)->getType();
+  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.BoolVal = false;
-  case FCmpInst::FCMP_ORD:   R = executeFCMP_EQ(Src1, Src2, Ty); break; ///???
-  case FCmpInst::FCMP_UNO:   R = executeFCMP_NE(Src1, Src2, Ty); break; ///???
-  case FCmpInst::FCMP_OEQ:
-  case FCmpInst::FCMP_UEQ:   R = executeFCMP_EQ(Src1, Src2, Ty);  break;
-  case FCmpInst::FCMP_ONE:
-  case FCmpInst::FCMP_UNE:   R = executeFCMP_NE(Src1, Src2, Ty);  break;
-  case FCmpInst::FCMP_OLT:
-  case FCmpInst::FCMP_ULT:   R = executeFCMP_LT(Src1, Src2, Ty); break;
-  case FCmpInst::FCMP_OGT:
-  case FCmpInst::FCMP_UGT:   R = executeFCMP_GT(Src1, Src2, Ty); break;
-  case FCmpInst::FCMP_OLE:
-  case FCmpInst::FCMP_ULE:   R = executeFCMP_LE(Src1, Src2, Ty); break;
-  case FCmpInst::FCMP_OGE:
-  case FCmpInst::FCMP_UGE:   R = executeFCMP_GE(Src1, Src2, Ty); break;
-  case FCmpInst::FCMP_TRUE:  R.BoolVal = true;
   default:
-    cerr << "Don't know how to handle this FCmp predicate!\n-->" << I;
-    abort();
+    dbgs() << "Don't know how to handle this FCmp predicate!\n-->" << I;
+    llvm_unreachable(0);
+  break;
+  case FCmpInst::FCMP_FALSE: R = executeFCMP_BOOL(Src1, Src2, Ty, false); 
+  break;
+  case FCmpInst::FCMP_TRUE:  R = executeFCMP_BOOL(Src1, Src2, Ty, 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;
   }
  
   SetValue(&I, R, SF);
 }
 
 static GenericValue executeCmpInst(unsigned predicate, GenericValue Src1, 
-                                   GenericValue Src2, const Type *Ty) {
+                                   GenericValue Src2, Type *Ty) {
   GenericValue Result;
   switch (predicate) {
   case ICmpInst::ICMP_EQ:    return executeICMP_EQ(Src1, Src2, Ty);
@@ -710,67 +654,140 @@ static GenericValue executeCmpInst(unsigned predicate, GenericValue Src1,
   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_EQ(Src1, Src2, Ty); break; 
-  case FCmpInst::FCMP_UNO:   return executeFCMP_NE(Src1, Src2, Ty); break; 
-  case FCmpInst::FCMP_OEQ:
-  case FCmpInst::FCMP_UEQ:   return executeFCMP_EQ(Src1, Src2, Ty);  break;
-  case FCmpInst::FCMP_ONE:
-  case FCmpInst::FCMP_UNE:   return executeFCMP_NE(Src1, Src2, Ty);  break;
-  case FCmpInst::FCMP_OLT:
-  case FCmpInst::FCMP_ULT:   return executeFCMP_LT(Src1, Src2, Ty); break;
-  case FCmpInst::FCMP_OGT:
-  case FCmpInst::FCMP_UGT:   return executeFCMP_GT(Src1, Src2, Ty); break;
-  case FCmpInst::FCMP_OLE:
-  case FCmpInst::FCMP_ULE:   return executeFCMP_LE(Src1, Src2, Ty); break;
-  case FCmpInst::FCMP_OGE:
-  case FCmpInst::FCMP_UGE:   return executeFCMP_GE(Src1, Src2, Ty); break;
-  case FCmpInst::FCMP_FALSE: { 
-    GenericValue Result;
-    Result.BoolVal = false; 
-    return Result;
-  }
-  case FCmpInst::FCMP_TRUE: {
-    GenericValue Result;
-    Result.BoolVal = true;
-    return Result;
-  }
+  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: return executeFCMP_BOOL(Src1, Src2, Ty, false);
+  case FCmpInst::FCMP_TRUE:  return executeFCMP_BOOL(Src1, Src2, Ty, true);
   default:
-    cerr << "Unhandled Cmp predicate\n";
-    abort();
+    dbgs() << "Unhandled Cmp predicate\n";
+    llvm_unreachable(0);
   }
 }
 
 void Interpreter::visitBinaryOperator(BinaryOperator &I) {
   ExecutionContext &SF = ECStack.back();
-  const Type *Ty    = I.getOperand(0)->getType();
+  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:   R = executeAddInst  (Src1, Src2, Ty); break;
-  case Instruction::Sub:   R = executeSubInst  (Src1, Src2, Ty); break;
-  case Instruction::Mul:   R = executeMulInst  (Src1, Src2, Ty); break;
-  case Instruction::UDiv:  R = executeUDivInst (Src1, Src2, Ty); break;
-  case Instruction::SDiv:  R = executeSDivInst (Src1, Src2, Ty); break;
-  case Instruction::FDiv:  R = executeFDivInst (Src1, Src2, Ty); break;
-  case Instruction::URem:  R = executeURemInst (Src1, Src2, Ty); break;
-  case Instruction::SRem:  R = executeSRemInst (Src1, Src2, Ty); break;
-  case Instruction::FRem:  R = executeFRemInst (Src1, Src2, Ty); break;
-  case Instruction::And:   R = executeAndInst  (Src1, Src2, Ty); break;
-  case Instruction::Or:    R = executeOrInst   (Src1, Src2, Ty); break;
-  case Instruction::Xor:   R = executeXorInst  (Src1, Src2, Ty); break;
-  default:
-    cerr << "Don't know how to handle this binary operator!\n-->" << I;
-    abort();
+  // First process vector operation
+  if (Ty->isVectorTy()) {
+    assert(Src1.AggregateVal.size() == Src2.AggregateVal.size());
+    R.AggregateVal.resize(Src1.AggregateVal.size());
+
+    // Macros to execute binary operation 'OP' over integer vectors
+#define INTEGER_VECTOR_OPERATION(OP)                               \
+    for (unsigned i = 0; i < R.AggregateVal.size(); ++i)           \
+      R.AggregateVal[i].IntVal =                                   \
+      Src1.AggregateVal[i].IntVal OP Src2.AggregateVal[i].IntVal;
+
+    // Additional macros to execute binary operations udiv/sdiv/urem/srem since
+    // they have different notation.
+#define INTEGER_VECTOR_FUNCTION(OP)                                \
+    for (unsigned i = 0; i < R.AggregateVal.size(); ++i)           \
+      R.AggregateVal[i].IntVal =                                   \
+      Src1.AggregateVal[i].IntVal.OP(Src2.AggregateVal[i].IntVal);
+
+    // Macros to execute binary operation 'OP' over floating point type TY
+    // (float or double) vectors
+#define FLOAT_VECTOR_FUNCTION(OP, TY)                               \
+      for (unsigned i = 0; i < R.AggregateVal.size(); ++i)          \
+        R.AggregateVal[i].TY =                                      \
+        Src1.AggregateVal[i].TY OP Src2.AggregateVal[i].TY;
+
+    // Macros to choose appropriate TY: float or double and run operation
+    // execution
+#define FLOAT_VECTOR_OP(OP) {                                         \
+  if (dyn_cast<VectorType>(Ty)->getElementType()->isFloatTy())        \
+    FLOAT_VECTOR_FUNCTION(OP, FloatVal)                               \
+  else {                                                              \
+    if (dyn_cast<VectorType>(Ty)->getElementType()->isDoubleTy())     \
+      FLOAT_VECTOR_FUNCTION(OP, DoubleVal)                            \
+    else {                                                            \
+      dbgs() << "Unhandled type for OP instruction: " << *Ty << "\n"; \
+      llvm_unreachable(0);                                            \
+    }                                                                 \
+  }                                                                   \
+}
+
+    switch(I.getOpcode()){
+    default:
+      dbgs() << "Don't know how to handle this binary operator!\n-->" << I;
+      llvm_unreachable(0);
+      break;
+    case Instruction::Add:   INTEGER_VECTOR_OPERATION(+) break;
+    case Instruction::Sub:   INTEGER_VECTOR_OPERATION(-) break;
+    case Instruction::Mul:   INTEGER_VECTOR_OPERATION(*) break;
+    case Instruction::UDiv:  INTEGER_VECTOR_FUNCTION(udiv) break;
+    case Instruction::SDiv:  INTEGER_VECTOR_FUNCTION(sdiv) break;
+    case Instruction::URem:  INTEGER_VECTOR_FUNCTION(urem) break;
+    case Instruction::SRem:  INTEGER_VECTOR_FUNCTION(srem) break;
+    case Instruction::And:   INTEGER_VECTOR_OPERATION(&) break;
+    case Instruction::Or:    INTEGER_VECTOR_OPERATION(|) break;
+    case Instruction::Xor:   INTEGER_VECTOR_OPERATION(^) break;
+    case Instruction::FAdd:  FLOAT_VECTOR_OP(+) break;
+    case Instruction::FSub:  FLOAT_VECTOR_OP(-) break;
+    case Instruction::FMul:  FLOAT_VECTOR_OP(*) break;
+    case Instruction::FDiv:  FLOAT_VECTOR_OP(/) break;
+    case Instruction::FRem:
+      if (dyn_cast<VectorType>(Ty)->getElementType()->isFloatTy())
+        for (unsigned i = 0; i < R.AggregateVal.size(); ++i)
+          R.AggregateVal[i].FloatVal = 
+          fmod(Src1.AggregateVal[i].FloatVal, Src2.AggregateVal[i].FloatVal);
+      else {
+        if (dyn_cast<VectorType>(Ty)->getElementType()->isDoubleTy())
+          for (unsigned i = 0; i < R.AggregateVal.size(); ++i)
+            R.AggregateVal[i].DoubleVal = 
+            fmod(Src1.AggregateVal[i].DoubleVal, Src2.AggregateVal[i].DoubleVal);
+        else {
+          dbgs() << "Unhandled type for Rem instruction: " << *Ty << "\n";
+          llvm_unreachable(0);
+        }
+      }
+      break;
+    }
+  } else {
+    switch (I.getOpcode()) {
+    default:
+      dbgs() << "Don't know how to handle this binary operator!\n-->" << I;
+      llvm_unreachable(0);
+      break;
+    case Instruction::Add:   R.IntVal = Src1.IntVal + Src2.IntVal; break;
+    case Instruction::Sub:   R.IntVal = Src1.IntVal - Src2.IntVal; break;
+    case Instruction::Mul:   R.IntVal = Src1.IntVal * Src2.IntVal; break;
+    case Instruction::FAdd:  executeFAddInst(R, Src1, Src2, Ty); break;
+    case Instruction::FSub:  executeFSubInst(R, Src1, Src2, Ty); break;
+    case Instruction::FMul:  executeFMulInst(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;
+    }
   }
-
   SetValue(&I, R, SF);
 }
 
 static GenericValue executeSelectInst(GenericValue Src1, GenericValue Src2,
                                       GenericValue Src3) {
-  return Src1.BoolVal ? Src2 : Src3;
+  return Src1.IntVal == 0 ? Src3 : Src2;
 }
 
 void Interpreter::visitSelectInst(SelectInst &I) {
@@ -791,9 +808,9 @@ 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.Int32Val);
+  ECStack.clear();
+  runAtExitHandlers();
+  exit(GV.IntVal.zextOrTrunc(32).getZExtValue());
 }
 
 /// Pop the last stack frame off of ECStack and then copy the result
@@ -804,23 +821,24 @@ void Interpreter::exitCalled(GenericValue GV) {
 /// care of switching to the normal destination BB, if we are returning
 /// from an invoke.
 ///
-void Interpreter::popStackAndReturnValueToCaller (const Type *RetTy,
-                                                  GenericValue Result) {
+void Interpreter::popStackAndReturnValueToCaller(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->isIntegral()) {          // Nonvoid return type?
+    if (RetTy && !RetTy->isVoidTy()) {          // Nonvoid return type?
       ExitValue = Result;   // Capture the exit value of the program
     } else {
-      memset(&ExitValue, 0, sizeof(ExitValue));
+      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...
+      // Save result...
+      if (!CallingSF.Caller.getType()->isVoidTy())
         SetValue(I, Result, CallingSF);
       if (InvokeInst *II = dyn_cast<InvokeInst> (I))
         SwitchToNewBasicBlock (II->getNormalDest (), CallingSF);
@@ -831,7 +849,7 @@ void Interpreter::popStackAndReturnValueToCaller (const Type *RetTy,
 
 void Interpreter::visitReturnInst(ReturnInst &I) {
   ExecutionContext &SF = ECStack.back();
-  const Type *RetTy = Type::VoidTy;
+  Type *RetTy = Type::getVoidTy(I.getContext());
   GenericValue Result;
 
   // Save away the return value... (if we are not 'ret void')
@@ -843,27 +861,8 @@ void Interpreter::visitReturnInst(ReturnInst &I) {
   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();
+  report_fatal_error("Program executed an 'unreachable' instruction!");
 }
 
 void Interpreter::visitBranchInst(BranchInst &I) {
@@ -873,7 +872,7 @@ void Interpreter::visitBranchInst(BranchInst &I) {
   Dest = I.getSuccessor(0);          // Uncond branches have a fixed dest...
   if (!I.isUnconditional()) {
     Value *Cond = I.getCondition();
-    if (getOperandValue(Cond, SF).BoolVal == 0) // If false cond...
+    if (getOperandValue(Cond, SF).IntVal == 0) // If false cond...
       Dest = I.getSuccessor(1);
   }
   SwitchToNewBasicBlock(Dest, SF);
@@ -881,22 +880,59 @@ void Interpreter::visitBranchInst(BranchInst &I) {
 
 void Interpreter::visitSwitchInst(SwitchInst &I) {
   ExecutionContext &SF = ECStack.back();
-  GenericValue CondVal = getOperandValue(I.getOperand(0), SF);
-  const Type *ElTy = I.getOperand(0)->getType();
+  Value* Cond = I.getCondition();
+  Type *ElTy = Cond->getType();
+  GenericValue CondVal = getOperandValue(Cond, SF);
 
   // 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).BoolVal) {
-      Dest = cast<BasicBlock>(I.getOperand(i+1));
-      break;
+  for (SwitchInst::CaseIt i = I.case_begin(), e = I.case_end(); i != e; ++i) {
+    IntegersSubset& Case = i.getCaseValueEx();
+    if (Case.isSingleNumber()) {
+      // FIXME: Currently work with ConstantInt based numbers.
+      const ConstantInt *CI = Case.getSingleNumber(0).toConstantInt();
+      GenericValue Val = getOperandValue(const_cast<ConstantInt*>(CI), SF);
+      if (executeICMP_EQ(Val, CondVal, ElTy).IntVal != 0) {
+        Dest = cast<BasicBlock>(i.getCaseSuccessor());
+        break;        
+      }
     }
-
+    if (Case.isSingleNumbersOnly()) {
+      for (unsigned n = 0, en = Case.getNumItems(); n != en; ++n) {
+        // FIXME: Currently work with ConstantInt based numbers.
+        const ConstantInt *CI = Case.getSingleNumber(n).toConstantInt();
+        GenericValue Val = getOperandValue(const_cast<ConstantInt*>(CI), SF);
+        if (executeICMP_EQ(Val, CondVal, ElTy).IntVal != 0) {
+          Dest = cast<BasicBlock>(i.getCaseSuccessor());
+          break;        
+        }
+      }      
+    } else
+      for (unsigned n = 0, en = Case.getNumItems(); n != en; ++n) {
+        IntegersSubset::Range r = Case.getItem(n);
+        // FIXME: Currently work with ConstantInt based numbers.
+        const ConstantInt *LowCI = r.getLow().toConstantInt();
+        const ConstantInt *HighCI = r.getHigh().toConstantInt();
+        GenericValue Low = getOperandValue(const_cast<ConstantInt*>(LowCI), SF);
+        GenericValue High = getOperandValue(const_cast<ConstantInt*>(HighCI), SF);
+        if (executeICMP_ULE(Low, CondVal, ElTy).IntVal != 0 &&
+            executeICMP_ULE(CondVal, High, ElTy).IntVal != 0) {
+          Dest = cast<BasicBlock>(i.getCaseSuccessor());
+          break;        
+        }
+      }
+  }
   if (!Dest) Dest = I.getDefaultDest();   // No cases matched: use default
   SwitchToNewBasicBlock(Dest, SF);
 }
 
+void Interpreter::visitIndirectBrInst(IndirectBrInst &I) {
+  ExecutionContext &SF = ECStack.back();
+  void *Dest = GVTOP(getOperandValue(I.getAddress(), SF));
+  SwitchToNewBasicBlock((BasicBlock*)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.
@@ -939,16 +975,26 @@ void Interpreter::SwitchToNewBasicBlock(BasicBlock *Dest, ExecutionContext &SF){
 //                     Memory Instruction Implementations
 //===----------------------------------------------------------------------===//
 
-void Interpreter::visitAllocationInst(AllocationInst &I) {
+void Interpreter::visitAllocaInst(AllocaInst &I) {
   ExecutionContext &SF = ECStack.back();
 
-  const Type *Ty = I.getType()->getElementType();  // Type to be allocated
+  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).Int32Val;
+  unsigned NumElements = 
+    getOperandValue(I.getOperand(0), SF).IntVal.getZExtValue();
+
+  unsigned TypeSize = (size_t)TD.getTypeAllocSize(Ty);
+
+  // Avoid malloc-ing zero bytes, use max()...
+  unsigned MemToAlloc = std::max(1U, NumElements * TypeSize);
 
   // Allocate enough memory to hold the type...
-  void *Memory = malloc(NumElements * (size_t)TD.getTypeSize(Ty));
+  void *Memory = malloc(MemToAlloc);
+
+  DEBUG(dbgs() << "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!");
@@ -958,57 +1004,51 @@ void Interpreter::visitAllocationInst(AllocationInst &I) {
     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()) &&
+  assert(Ptr->getType()->isPointerTy() &&
          "Cannot getElementOffset of a nonpointer type!");
 
-  PointerTy Total = 0;
+  uint64_t Total = 0;
 
   for (; I != E; ++I) {
-    if (const StructType *STy = dyn_cast<StructType>(*I)) {
+    if (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 += (PointerTy)SLO->MemberOffsets[Index];
+      Total += SLO->getElementOffset(Index);
     } else {
-      const SequentialType *ST = cast<SequentialType>(*I);
+      SequentialType *ST = cast<SequentialType>(*I);
       // Get the index number for the array... which must be long type...
       GenericValue IdxGV = getOperandValue(I.getOperand(), SF);
 
-      uint64_t Idx;
-      switch (I.getOperand()->getType()->getTypeID()) {
-      default: assert(0 && "Illegal getelementptr index for sequential type!");
-      case Type::Int8TyID:  Idx = IdxGV.Int8Val; break;
-      case Type::Int16TyID: Idx = IdxGV.Int16Val; break;
-      case Type::Int32TyID: Idx = IdxGV.Int32Val; break;
-      case Type::Int64TyID: Idx = IdxGV.Int64Val; break;
+      int64_t Idx;
+      unsigned BitWidth = 
+        cast<IntegerType>(I.getOperand()->getType())->getBitWidth();
+      if (BitWidth == 32)
+        Idx = (int64_t)(int32_t)IdxGV.IntVal.getZExtValue();
+      else {
+        assert(BitWidth == 64 && "Invalid index type for getelementptr");
+        Idx = (int64_t)IdxGV.IntVal.getZExtValue();
       }
-      Total += PointerTy(TD.getTypeSize(ST->getElementType())*Idx);
+      Total += TD.getTypeAllocSize(ST->getElementType())*Idx;
     }
   }
 
   GenericValue Result;
-  Result.PointerVal = getOperandValue(Ptr, SF).PointerVal + Total;
+  Result.PointerVal = ((char*)getOperandValue(Ptr, SF).PointerVal) + Total;
+  DEBUG(dbgs() << "GEP Index " << Total << " bytes.\n");
   return Result;
 }
 
 void Interpreter::visitGetElementPtrInst(GetElementPtrInst &I) {
   ExecutionContext &SF = ECStack.back();
-  SetValue(&I, TheEE->executeGEPOperation(I.getPointerOperand(),
+  SetValue(&I, executeGEPOperation(I.getPointerOperand(),
                                    gep_type_begin(I), gep_type_end(I), SF), SF);
 }
 
@@ -1016,8 +1056,11 @@ void Interpreter::visitLoadInst(LoadInst &I) {
   ExecutionContext &SF = ECStack.back();
   GenericValue SRC = getOperandValue(I.getPointerOperand(), SF);
   GenericValue *Ptr = (GenericValue*)GVTOP(SRC);
-  GenericValue Result = LoadValueFromMemory(Ptr, I.getType());
+  GenericValue Result;
+  LoadValueFromMemory(Result, Ptr, I.getType());
   SetValue(&I, Result, SF);
+  if (I.isVolatile() && PrintVolatile)
+    dbgs() << "Volatile load " << I;
 }
 
 void Interpreter::visitStoreInst(StoreInst &I) {
@@ -1026,6 +1069,8 @@ void Interpreter::visitStoreInst(StoreInst &I) {
   GenericValue SRC = getOperandValue(I.getPointerOperand(), SF);
   StoreValueToMemory(Val, (GenericValue *)GVTOP(SRC),
                      I.getOperand(0)->getType());
+  if (I.isVolatile() && PrintVolatile)
+    dbgs() << "Volatile store: " << I;
 }
 
 //===----------------------------------------------------------------------===//
@@ -1036,8 +1081,8 @@ void Interpreter::visitCallSite(CallSite CS) {
   ExecutionContext &SF = ECStack.back();
 
   // Check to see if this is an intrinsic function call...
-  if (Function *F = CS.getCalledFunction())
-   if (F->isExternal ())
+  Function *F = CS.getCalledFunction();
+  if (F && F->isDeclaration())
     switch (F->getIntrinsicID()) {
     case Intrinsic::not_intrinsic:
       break;
@@ -1057,43 +1102,34 @@ void Interpreter::visitCallSite(CallSite CS) {
       // If it is an unknown intrinsic function, use the intrinsic lowering
       // class to transform it into hopefully tasty LLVM code.
       //
-      Instruction *Prev = CS.getInstruction()->getPrev();
+      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 (!Prev) {
+      if (atBegin) {
         SF.CurInst = Parent->begin();
       } else {
-        SF.CurInst = Prev;
+        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) {
+         e = SF.Caller.arg_end(); i != e; ++i, ++pNum) {
     Value *V = *i;
     ArgVals.push_back(getOperandValue(V, SF));
-    // Promote all integral types whose size is < sizeof(int) into ints.  We do
-    // this by zero or sign extending the value as appropriate according to the
-    // source type.
-    const Type *Ty = V->getType();
-    if (Ty->isIntegral() && Ty->getPrimitiveSize() < 4) {
-      if (Ty == Type::Int16Ty)
-        ArgVals.back().Int32Val = ArgVals.back().Int16Val;
-      else if (Ty == Type::Int8Ty)
-        ArgVals.back().Int32Val = ArgVals.back().Int8Val;
-      else if (Ty == Type::BoolTy)
-        ArgVals.back().Int32Val = ArgVals.back().BoolVal;
-      else
-        assert(0 && "Unknown type!");
-    }
   }
 
   // To handle indirect calls, we must get the pointer value from the argument
@@ -1102,246 +1138,257 @@ void Interpreter::visitCallSite(CallSite CS) {
   callFunction((Function*)GVTOP(SRC), ArgVals);
 }
 
-#define IMPLEMENT_SHIFT(OP, TY) \
-   case Type::TY##TyID: Dest.TY##Val = Src1.TY##Val OP Src2.Int8Val; break
-
-#define IMPLEMENT_SIGNLESS_SHIFT(OP, TY, CAST) \
-   case Type::TY##TyID: Dest.TY##Val = ((CAST)Src1.TY##Val) OP Src2.Int8Val; \
-     break
-
-static GenericValue executeShlInst(GenericValue Src1, GenericValue Src2,
-                                   const Type *Ty) {
-  GenericValue Dest;
-  switch (Ty->getTypeID()) {
-    IMPLEMENT_SHIFT(<<, Int8);
-    IMPLEMENT_SHIFT(<<, Int16);
-    IMPLEMENT_SHIFT(<<, Int32);
-    IMPLEMENT_SHIFT(<<, Int64);
-  default:
-    cerr << "Unhandled type for Shl instruction: " << *Ty << "\n";
-  }
-  return Dest;
-}
-
-static GenericValue executeLShrInst(GenericValue Src1, GenericValue Src2,
-                                    const Type *Ty) {
-  GenericValue Dest;
-  switch (Ty->getTypeID()) {
-    IMPLEMENT_SIGNLESS_SHIFT(>>, Int8,  uint8_t);
-    IMPLEMENT_SIGNLESS_SHIFT(>>, Int16, uint16_t);
-    IMPLEMENT_SIGNLESS_SHIFT(>>, Int32, uint32_t);
-    IMPLEMENT_SIGNLESS_SHIFT(>>, Int64, uint64_t);
-  default:
-    cerr << "Unhandled type for LShr instruction: " << *Ty << "\n";
-    abort();
-  }
-  return Dest;
-}
-
-static GenericValue executeAShrInst(GenericValue Src1, GenericValue Src2,
-                                    const Type *Ty) {
-  GenericValue Dest;
-  switch (Ty->getTypeID()) {
-    IMPLEMENT_SIGNLESS_SHIFT(>>, Int8,  int8_t);
-    IMPLEMENT_SIGNLESS_SHIFT(>>, Int16, int16_t);
-    IMPLEMENT_SIGNLESS_SHIFT(>>, Int32, int32_t);
-    IMPLEMENT_SIGNLESS_SHIFT(>>, Int64, int64_t);
-  default:
-    cerr << "Unhandled type for AShr instruction: " << *Ty << "\n";
-    abort();
-  }
-  return Dest;
-}
-
-void Interpreter::visitShl(ShiftInst &I) {
+void Interpreter::visitShl(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 Dest;
-  Dest = executeShlInst (Src1, Src2, Ty);
+  if (Src2.IntVal.getZExtValue() < Src1.IntVal.getBitWidth())
+    Dest.IntVal = Src1.IntVal.shl(Src2.IntVal.getZExtValue());
+  else
+    Dest.IntVal = Src1.IntVal;
+  
   SetValue(&I, Dest, SF);
 }
 
-void Interpreter::visitLShr(ShiftInst &I) {
+void Interpreter::visitLShr(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 Dest;
-  Dest = executeLShrInst (Src1, Src2, Ty);
+  if (Src2.IntVal.getZExtValue() < Src1.IntVal.getBitWidth())
+    Dest.IntVal = Src1.IntVal.lshr(Src2.IntVal.getZExtValue());
+  else
+    Dest.IntVal = Src1.IntVal;
+  
   SetValue(&I, Dest, SF);
 }
 
-void Interpreter::visitAShr(ShiftInst &I) {
+void Interpreter::visitAShr(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 Dest;
-  Dest = executeAShrInst (Src1, Src2, Ty);
+  if (Src2.IntVal.getZExtValue() < Src1.IntVal.getBitWidth())
+    Dest.IntVal = Src1.IntVal.ashr(Src2.IntVal.getZExtValue());
+  else
+    Dest.IntVal = Src1.IntVal;
+  
   SetValue(&I, Dest, SF);
 }
 
-#define IMPLEMENT_CAST_START \
-  switch (DstTy->getTypeID()) {
-
-#define IMPLEMENT_CAST(STY, DTY, CAST) \
-     case Type::STY##TyID: Dest.DTY##Val = (CAST(Src.STY##Val)); break;
-
-#define IMPLEMENT_CAST_CASE(DTY, CAST)          \
-  case Type::DTY##TyID:                         \
-    switch (SrcTy->getTypeID()) {               \
-      IMPLEMENT_CAST(Bool,   DTY, CAST);        \
-      IMPLEMENT_CAST(Int8,   DTY, CAST);        \
-      IMPLEMENT_CAST(Int16,  DTY, CAST);        \
-      IMPLEMENT_CAST(Int32,  DTY, CAST);        \
-      IMPLEMENT_CAST(Int64,  DTY, CAST);        \
-      IMPLEMENT_CAST(Pointer,DTY, CAST);        \
-      IMPLEMENT_CAST(Float,  DTY, CAST);        \
-      IMPLEMENT_CAST(Double, DTY, CAST);        \
-    default:                                    \
-      cerr << "Unhandled cast: "                \
-        << *SrcTy << " to " << *DstTy << "\n";  \
-      abort();                                  \
-    }                                           \
-    break
-
-#define IMPLEMENT_CAST_END                      \
-  default: cerr                                 \
-      << "Unhandled dest type for cast instruction: "  \
-      << *DstTy << "\n";                        \
-    abort();                                    \
-  }
+GenericValue Interpreter::executeTruncInst(Value *SrcVal, Type *DstTy,
+                                           ExecutionContext &SF) {
+  GenericValue Dest, Src = getOperandValue(SrcVal, SF);
+  IntegerType *DITy = cast<IntegerType>(DstTy);
+  unsigned DBitWidth = DITy->getBitWidth();
+  Dest.IntVal = Src.IntVal.trunc(DBitWidth);
+  return Dest;
+}
 
-GenericValue Interpreter::executeCastOperation(Instruction::CastOps opcode,
-                                               Value *SrcVal, const Type *DstTy,
-                                               ExecutionContext &SF) {
-  const Type *SrcTy = SrcVal->getType();
+GenericValue Interpreter::executeSExtInst(Value *SrcVal, Type *DstTy,
+                                          ExecutionContext &SF) {
   GenericValue Dest, Src = getOperandValue(SrcVal, SF);
+  IntegerType *DITy = cast<IntegerType>(DstTy);
+  unsigned DBitWidth = DITy->getBitWidth();
+  Dest.IntVal = Src.IntVal.sext(DBitWidth);
+  return Dest;
+}
 
-  if (opcode == Instruction::Trunc && DstTy->getTypeID() == Type::BoolTyID) {
-    // For truncations to bool, we must clear the high order bits of the source
-    switch (SrcTy->getTypeID()) {
-      case Type::BoolTyID:  Src.BoolVal  &= 1; break;
-      case Type::Int8TyID:  Src.Int8Val  &= 1; break;
-      case Type::Int16TyID: Src.Int16Val &= 1; break;
-      case Type::Int32TyID: Src.Int32Val &= 1; break;
-      case Type::Int64TyID: Src.Int64Val &= 1; break;
-      default:
-        assert(0 && "Can't trunc a non-integer!");
-        break;
-    }
-  } else if (opcode == Instruction::SExt && 
-             SrcTy->getTypeID() == Type::BoolTyID) {
-    // For sign extension from bool, we must extend the source bits.
-    SrcTy = Type::Int64Ty;
-    Src.Int64Val = 0 - Src.BoolVal;
-  }
+GenericValue Interpreter::executeZExtInst(Value *SrcVal, Type *DstTy,
+                                          ExecutionContext &SF) {
+  GenericValue Dest, Src = getOperandValue(SrcVal, SF);
+  IntegerType *DITy = cast<IntegerType>(DstTy);
+  unsigned DBitWidth = DITy->getBitWidth();
+  Dest.IntVal = Src.IntVal.zext(DBitWidth);
+  return Dest;
+}
 
-  switch (opcode) {
-    case Instruction::Trunc:     // src integer, dest integral (can't be long)
-      IMPLEMENT_CAST_START
-      IMPLEMENT_CAST_CASE(Bool , (bool));
-      IMPLEMENT_CAST_CASE(Int8 , (uint8_t));
-      IMPLEMENT_CAST_CASE(Int16, (uint16_t));
-      IMPLEMENT_CAST_CASE(Int32, (uint32_t));
-      IMPLEMENT_CAST_CASE(Int64, (uint64_t));
-      IMPLEMENT_CAST_END
-      break;
-    case Instruction::ZExt:      // src integral (can't be long), dest integer
-      IMPLEMENT_CAST_START
-      IMPLEMENT_CAST_CASE(Int8 , (uint8_t));
-      IMPLEMENT_CAST_CASE(Int16, (uint16_t));
-      IMPLEMENT_CAST_CASE(Int32, (uint32_t));
-      IMPLEMENT_CAST_CASE(Int64, (uint64_t));
-      IMPLEMENT_CAST_END
-      break;
-    case Instruction::SExt:      // src integral (can't be long), dest integer
-      IMPLEMENT_CAST_START
-      IMPLEMENT_CAST_CASE(Int8 , (uint8_t)(int8_t));
-      IMPLEMENT_CAST_CASE(Int16, (uint16_t)(int16_t));
-      IMPLEMENT_CAST_CASE(Int32, (uint32_t)(int32_t));
-      IMPLEMENT_CAST_CASE(Int64, (uint64_t)(int64_t));
-      IMPLEMENT_CAST_END
-      break;
-    case Instruction::FPTrunc:   // src double, dest float
-      IMPLEMENT_CAST_START
-      IMPLEMENT_CAST_CASE(Float  , (float));
-      IMPLEMENT_CAST_END
-      break;
-    case Instruction::FPExt:     // src float, dest double
-      IMPLEMENT_CAST_START
-      IMPLEMENT_CAST_CASE(Double , (double));
-      IMPLEMENT_CAST_END
-      break;
-    case Instruction::UIToFP:    // src integral, dest floating
-      IMPLEMENT_CAST_START
-      IMPLEMENT_CAST_CASE(Float  , (float)(uint64_t));
-      IMPLEMENT_CAST_CASE(Double , (double)(uint64_t));
-      IMPLEMENT_CAST_END
-      break;
-    case Instruction::SIToFP:    // src integeral, dest floating
-      IMPLEMENT_CAST_START
-      IMPLEMENT_CAST_CASE(Float  , (float)(int64_t));
-      IMPLEMENT_CAST_CASE(Double , (double)(int64_t));
-      IMPLEMENT_CAST_END
-      break;
-    case Instruction::FPToUI:    // src floating, dest integral
-      IMPLEMENT_CAST_START
-      IMPLEMENT_CAST_CASE(Bool , (bool));
-      IMPLEMENT_CAST_CASE(Int8 , (uint8_t));
-      IMPLEMENT_CAST_CASE(Int16, (uint16_t));
-      IMPLEMENT_CAST_CASE(Int32, (uint32_t ));
-      IMPLEMENT_CAST_CASE(Int64, (uint64_t));
-      IMPLEMENT_CAST_END
-      break;
-    case Instruction::FPToSI:    // src floating, dest integral
-      IMPLEMENT_CAST_START
-      IMPLEMENT_CAST_CASE(Bool , (bool));
-      IMPLEMENT_CAST_CASE(Int8 , (uint8_t) (int8_t));
-      IMPLEMENT_CAST_CASE(Int16, (uint16_t)(int16_t));
-      IMPLEMENT_CAST_CASE(Int32, (uint32_t)(int32_t));
-      IMPLEMENT_CAST_CASE(Int64, (uint64_t)(int64_t));
-      IMPLEMENT_CAST_END
-      break;
-    case Instruction::PtrToInt:  // src pointer,  dest integral
-      IMPLEMENT_CAST_START
-      IMPLEMENT_CAST_CASE(Bool , (bool));
-      IMPLEMENT_CAST_CASE(Int8 , (uint8_t));
-      IMPLEMENT_CAST_CASE(Int16, (uint16_t));
-      IMPLEMENT_CAST_CASE(Int32, (uint32_t));
-      IMPLEMENT_CAST_CASE(Int64, (uint64_t));
-      IMPLEMENT_CAST_END
-      break;
-    case Instruction::IntToPtr:  // src integral, dest pointer
-      IMPLEMENT_CAST_START
-      IMPLEMENT_CAST_CASE(Pointer, (PointerTy));
-      IMPLEMENT_CAST_END
-      break;
-    case Instruction::BitCast:   // src any, dest any (same size)
-      IMPLEMENT_CAST_START
-      IMPLEMENT_CAST_CASE(Bool   , (bool));
-      IMPLEMENT_CAST_CASE(Int8   , (uint8_t));
-      IMPLEMENT_CAST_CASE(Int16  , (uint16_t));
-      IMPLEMENT_CAST_CASE(Int32  , (uint32_t));
-      IMPLEMENT_CAST_CASE(Int64  , (uint64_t));
-      IMPLEMENT_CAST_CASE(Pointer, (PointerTy));
-      IMPLEMENT_CAST_CASE(Float  , (float));
-      IMPLEMENT_CAST_CASE(Double , (double));
-      IMPLEMENT_CAST_END
-      break;
-    default:
-      cerr << "Invalid cast opcode for cast instruction: " << opcode << "\n";
-      abort();
-  }
+GenericValue Interpreter::executeFPTruncInst(Value *SrcVal, Type *DstTy,
+                                             ExecutionContext &SF) {
+  GenericValue Dest, Src = getOperandValue(SrcVal, SF);
+  assert(SrcVal->getType()->isDoubleTy() && DstTy->isFloatTy() &&
+         "Invalid FPTrunc instruction");
+  Dest.FloatVal = (float) Src.DoubleVal;
+  return Dest;
+}
+
+GenericValue Interpreter::executeFPExtInst(Value *SrcVal, Type *DstTy,
+                                           ExecutionContext &SF) {
+  GenericValue Dest, Src = getOperandValue(SrcVal, SF);
+  assert(SrcVal->getType()->isFloatTy() && DstTy->isDoubleTy() &&
+         "Invalid FPTrunc instruction");
+  Dest.DoubleVal = (double) Src.FloatVal;
+  return Dest;
+}
+
+GenericValue Interpreter::executeFPToUIInst(Value *SrcVal, Type *DstTy,
+                                            ExecutionContext &SF) {
+  Type *SrcTy = SrcVal->getType();
+  uint32_t DBitWidth = cast<IntegerType>(DstTy)->getBitWidth();
+  GenericValue Dest, Src = getOperandValue(SrcVal, SF);
+  assert(SrcTy->isFloatingPointTy() && "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, Type *DstTy,
+                                            ExecutionContext &SF) {
+  Type *SrcTy = SrcVal->getType();
+  uint32_t DBitWidth = cast<IntegerType>(DstTy)->getBitWidth();
+  GenericValue Dest, Src = getOperandValue(SrcVal, SF);
+  assert(SrcTy->isFloatingPointTy() && "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, Type *DstTy,
+                                            ExecutionContext &SF) {
+  GenericValue Dest, Src = getOperandValue(SrcVal, SF);
+  assert(DstTy->isFloatingPointTy() && "Invalid UIToFP instruction");
+
+  if (DstTy->getTypeID() == Type::FloatTyID)
+    Dest.FloatVal = APIntOps::RoundAPIntToFloat(Src.IntVal);
+  else
+    Dest.DoubleVal = APIntOps::RoundAPIntToDouble(Src.IntVal);
   return Dest;
 }
 
-void Interpreter::visitCastInst(CastInst &I) {
+GenericValue Interpreter::executeSIToFPInst(Value *SrcVal, Type *DstTy,
+                                            ExecutionContext &SF) {
+  GenericValue Dest, Src = getOperandValue(SrcVal, SF);
+  assert(DstTy->isFloatingPointTy() && "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, Type *DstTy,
+                                              ExecutionContext &SF) {
+  uint32_t DBitWidth = cast<IntegerType>(DstTy)->getBitWidth();
+  GenericValue Dest, Src = getOperandValue(SrcVal, SF);
+  assert(SrcVal->getType()->isPointerTy() && "Invalid PtrToInt instruction");
+
+  Dest.IntVal = APInt(DBitWidth, (intptr_t) Src.PointerVal);
+  return Dest;
+}
+
+GenericValue Interpreter::executeIntToPtrInst(Value *SrcVal, Type *DstTy,
+                                              ExecutionContext &SF) {
+  GenericValue Dest, Src = getOperandValue(SrcVal, SF);
+  assert(DstTy->isPointerTy() && "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, Type *DstTy,
+                                             ExecutionContext &SF) {
+  
+  Type *SrcTy = SrcVal->getType();
+  GenericValue Dest, Src = getOperandValue(SrcVal, SF);
+  if (DstTy->isPointerTy()) {
+    assert(SrcTy->isPointerTy() && "Invalid BitCast");
+    Dest.PointerVal = Src.PointerVal;
+  } else if (DstTy->isIntegerTy()) {
+    if (SrcTy->isFloatTy()) {
+      Dest.IntVal = APInt::floatToBits(Src.FloatVal);
+    } else if (SrcTy->isDoubleTy()) {
+      Dest.IntVal = APInt::doubleToBits(Src.DoubleVal);
+    } else if (SrcTy->isIntegerTy()) {
+      Dest.IntVal = Src.IntVal;
+    } else 
+      llvm_unreachable("Invalid BitCast");
+  } else if (DstTy->isFloatTy()) {
+    if (SrcTy->isIntegerTy())
+      Dest.FloatVal = Src.IntVal.bitsToFloat();
+    else
+      Dest.FloatVal = Src.FloatVal;
+  } else if (DstTy->isDoubleTy()) {
+    if (SrcTy->isIntegerTy())
+      Dest.DoubleVal = Src.IntVal.bitsToDouble();
+    else
+      Dest.DoubleVal = Src.DoubleVal;
+  } else
+    llvm_unreachable("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, executeCastOperation(I.getOpcode(), I.getOperand(0), 
-                                    I.getType(), SF), SF);
+  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) \
@@ -1355,20 +1402,18 @@ void Interpreter::visitVAArgInst(VAArgInst &I) {
   GenericValue VAList = getOperandValue(I.getOperand(0), SF);
   GenericValue Dest;
   GenericValue Src = ECStack[VAList.UIntPairVal.first]
-   .VarArgs[VAList.UIntPairVal.second];
-  const Type *Ty = I.getType();
+                      .VarArgs[VAList.UIntPairVal.second];
+  Type *Ty = I.getType();
   switch (Ty->getTypeID()) {
-    IMPLEMENT_VAARG(Int8);
-    IMPLEMENT_VAARG(Int16);
-    IMPLEMENT_VAARG(Int32);
-    IMPLEMENT_VAARG(Int64);
-    IMPLEMENT_VAARG(Pointer);
-    IMPLEMENT_VAARG(Float);
-    IMPLEMENT_VAARG(Double);
-    IMPLEMENT_VAARG(Bool);
+  case Type::IntegerTyID:
+    Dest.IntVal = Src.IntVal;
+    break;
+  IMPLEMENT_VAARG(Pointer);
+  IMPLEMENT_VAARG(Float);
+  IMPLEMENT_VAARG(Double);
   default:
-    cerr << "Unhandled dest type for vaarg instruction: " << *Ty << "\n";
-    abort();
+    dbgs() << "Unhandled dest type for vaarg instruction: " << *Ty << "\n";
+    llvm_unreachable(0);
   }
 
   // Set the Value of this Instruction.
@@ -1378,6 +1423,133 @@ void Interpreter::visitVAArgInst(VAArgInst &I) {
   ++VAList.UIntPairVal.second;
 }
 
+void Interpreter::visitExtractElementInst(ExtractElementInst &I) {
+  ExecutionContext &SF = ECStack.back();
+  GenericValue Src1 = getOperandValue(I.getOperand(0), SF);
+  GenericValue Src2 = getOperandValue(I.getOperand(1), SF);
+  GenericValue Dest;
+
+  Type *Ty = I.getType();
+  const unsigned indx = unsigned(Src2.IntVal.getZExtValue());
+
+  if(Src1.AggregateVal.size() > indx) {
+    switch (Ty->getTypeID()) {
+    default:
+      dbgs() << "Unhandled destination type for extractelement instruction: "
+      << *Ty << "\n";
+      llvm_unreachable(0);
+      break;
+    case Type::IntegerTyID:
+      Dest.IntVal = Src1.AggregateVal[indx].IntVal;
+      break;
+    case Type::FloatTyID:
+      Dest.FloatVal = Src1.AggregateVal[indx].FloatVal;
+      break;
+    case Type::DoubleTyID:
+      Dest.DoubleVal = Src1.AggregateVal[indx].DoubleVal;
+      break;
+    }
+  } else {
+    dbgs() << "Invalid index in extractelement instruction\n";
+  }
+
+  SetValue(&I, Dest, SF);
+}
+
+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;
+  Type * Ty = CE->getOperand(0)->getType();
+  switch (CE->getOpcode()) {
+  case Instruction::Add:  Dest.IntVal = Op0.IntVal + Op1.IntVal; break;
+  case Instruction::Sub:  Dest.IntVal = Op0.IntVal - Op1.IntVal; break;
+  case Instruction::Mul:  Dest.IntVal = Op0.IntVal * Op1.IntVal; break;
+  case Instruction::FAdd: executeFAddInst(Dest, Op0, Op1, Ty); break;
+  case Instruction::FSub: executeFSubInst(Dest, Op0, Op1, Ty); break;
+  case Instruction::FMul: executeFMulInst(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 & Op1.IntVal; break;
+  case Instruction::Or:   Dest.IntVal = Op0.IntVal | Op1.IntVal; break;
+  case Instruction::Xor:  Dest.IntVal = Op0.IntVal ^ 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:
+    dbgs() << "Unhandled ConstantExpr: " << *CE << "\n";
+    llvm_unreachable("Unhandled ConstantExpr");
+  }
+  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
 //===----------------------------------------------------------------------===//
@@ -1396,7 +1568,7 @@ void Interpreter::callFunction(Function *F,
   StackFrame.CurFunction = F;
 
   // Special handling for external functions.
-  if (F->isExternal()) {
+  if (F->isDeclaration()) {
     GenericValue Result = callExternalFunction (F, ArgVals);
     // Simulate a 'ret' instruction of the appropriate type.
     popStackAndReturnValueToCaller (F->getReturnType (), Result);
@@ -1414,13 +1586,15 @@ void Interpreter::callFunction(Function *F,
 
   // Handle non-varargs arguments...
   unsigned i = 0;
-  for (Function::arg_iterator AI = F->arg_begin(), E = F->arg_end(); AI != E; ++AI, ++i)
+  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());
 }
 
+
 void Interpreter::run() {
   while (!ECStack.empty()) {
     // Interpret a single instruction & increment the "PC".
@@ -1430,7 +1604,29 @@ void Interpreter::run() {
     // Track the number of dynamic instructions executed.
     ++NumDynamicInsts;
 
-    DOUT << "About to interpret: " << I;
+    DEBUG(dbgs() << "About to interpret: " << I);
     visit(I);   // Dispatch to one of the visit* methods...
+#if 0
+    // This is not safe, as visiting the instruction could lower it and free I.
+DEBUG(
+    if (!isa<CallInst>(I) && !isa<InvokeInst>(I) && 
+        I.getType() != Type::VoidTy) {
+      dbgs() << "  --> ";
+      const GenericValue &Val = SF.Values[&I];
+      switch (I.getType()->getTypeID()) {
+      default: llvm_unreachable("Invalid GenericValue Type");
+      case Type::VoidTyID:    dbgs() << "void"; break;
+      case Type::FloatTyID:   dbgs() << "float " << Val.FloatVal; break;
+      case Type::DoubleTyID:  dbgs() << "double " << Val.DoubleVal; break;
+      case Type::PointerTyID: dbgs() << "void* " << intptr_t(Val.PointerVal);
+        break;
+      case Type::IntegerTyID: 
+        dbgs() << "i" << Val.IntVal.getBitWidth() << " "
+               << Val.IntVal.toStringUnsigned(10)
+               << " (0x" << Val.IntVal.toStringUnsigned(16) << ")\n";
+        break;
+      }
+    });
+#endif
   }
 }