Add support for compiling varargs functions.
[oota-llvm.git] / lib / Target / SparcV9 / SparcV9InstrSelection.cpp
index 4e8259f6d811001ea6282ee4e3665005fb3e3358..709c338fd356c6520a594d9df3d69e2f9df65757 100644 (file)
-// $Id$
-//***************************************************************************
-// File:
-//     SparcInstrSelection.cpp
-// 
-// Purpose:
-//      BURS instruction selection for SPARC V9 architecture.      
-//     
-// History:
-//     7/02/01  -  Vikram Adve  -  Created
-//**************************************************************************/
+//===-- SparcInstrSelection.cpp -------------------------------------------===//
+//
+//  BURS instruction selection for SPARC V9 architecture.      
+//
+//===----------------------------------------------------------------------===//
 
 #include "SparcInternals.h"
 #include "SparcInstrSelectionSupport.h"
 #include "SparcRegClassInfo.h"
 #include "llvm/CodeGen/InstrSelectionSupport.h"
-#include "llvm/CodeGen/MachineInstr.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineInstrAnnot.h"
 #include "llvm/CodeGen/InstrForest.h"
 #include "llvm/CodeGen/InstrSelection.h"
-#include "llvm/CodeGen/MachineCodeForMethod.h"
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineFunctionInfo.h"
 #include "llvm/CodeGen/MachineCodeForInstruction.h"
 #include "llvm/DerivedTypes.h"
 #include "llvm/iTerminators.h"
 #include "llvm/iMemory.h"
 #include "llvm/iOther.h"
-#include "llvm/BasicBlock.h"
 #include "llvm/Function.h"
 #include "llvm/Constants.h"
+#include "llvm/ConstantHandling.h"
+#include "llvm/Intrinsics.h"
 #include "Support/MathExtras.h"
 #include <math.h>
-using std::vector;
 
-//************************* Forward Declarations ***************************/
+static inline void Add3OperandInstr(unsigned Opcode, InstructionNode* Node,
+                                    std::vector<MachineInstr*>& mvec) {
+  mvec.push_back(BuildMI(Opcode, 3).addReg(Node->leftChild()->getValue())
+                                   .addReg(Node->rightChild()->getValue())
+                                   .addRegDef(Node->getValue()));
+}
+
+
+
+//---------------------------------------------------------------------------
+// Function: GetMemInstArgs
+// 
+// Purpose:
+//   Get the pointer value and the index vector for a memory operation
+//   (GetElementPtr, Load, or Store).  If all indices of the given memory
+//   operation are constant, fold in constant indices in a chain of
+//   preceding GetElementPtr instructions (if any), and return the
+//   pointer value of the first instruction in the chain.
+//   All folded instructions are marked so no code is generated for them.
+//
+// Return values:
+//   Returns the pointer Value to use.
+//   Returns the resulting IndexVector in idxVec.
+//   Returns true/false in allConstantIndices if all indices are/aren't const.
+//---------------------------------------------------------------------------
+
+
+//---------------------------------------------------------------------------
+// Function: FoldGetElemChain
+// 
+// Purpose:
+//   Fold a chain of GetElementPtr instructions containing only
+//   constant offsets into an equivalent (Pointer, IndexVector) pair.
+//   Returns the pointer Value, and stores the resulting IndexVector
+//   in argument chainIdxVec. This is a helper function for
+//   FoldConstantIndices that does the actual folding. 
+//---------------------------------------------------------------------------
+
+
+// Check for a constant 0.
+inline bool
+IsZero(Value* idx)
+{
+  return (idx == ConstantSInt::getNullValue(idx->getType()));
+}
+
+static Value*
+FoldGetElemChain(InstrTreeNode* ptrNode, std::vector<Value*>& chainIdxVec,
+                 bool lastInstHasLeadingNonZero)
+{
+  InstructionNode* gepNode = dyn_cast<InstructionNode>(ptrNode);
+  GetElementPtrInst* gepInst =
+    dyn_cast_or_null<GetElementPtrInst>(gepNode ? gepNode->getInstruction() :0);
+
+  // ptr value is not computed in this tree or ptr value does not come from GEP
+  // instruction
+  if (gepInst == NULL)
+    return NULL;
+
+  // Return NULL if we don't fold any instructions in.
+  Value* ptrVal = NULL;
+
+  // Now chase the chain of getElementInstr instructions, if any.
+  // Check for any non-constant indices and stop there.
+  // Also, stop if the first index of child is a non-zero array index
+  // and the last index of the current node is a non-array index:
+  // in that case, a non-array declared type is being accessed as an array
+  // which is not type-safe, but could be legal.
+  // 
+  InstructionNode* ptrChild = gepNode;
+  while (ptrChild && (ptrChild->getOpLabel() == Instruction::GetElementPtr ||
+                      ptrChild->getOpLabel() == GetElemPtrIdx))
+  {
+    // Child is a GetElemPtr instruction
+    gepInst = cast<GetElementPtrInst>(ptrChild->getValue());
+    User::op_iterator OI, firstIdx = gepInst->idx_begin();
+    User::op_iterator lastIdx = gepInst->idx_end();
+    bool allConstantOffsets = true;
+
+    // The first index of every GEP must be an array index.
+    assert((*firstIdx)->getType() == Type::LongTy &&
+           "INTERNAL ERROR: Structure index for a pointer type!");
+
+    // If the last instruction had a leading non-zero index, check if the
+    // current one references a sequential (i.e., indexable) type.
+    // If not, the code is not type-safe and we would create an illegal GEP
+    // by folding them, so don't fold any more instructions.
+    // 
+    if (lastInstHasLeadingNonZero)
+      if (! isa<SequentialType>(gepInst->getType()->getElementType()))
+        break;   // cannot fold in any preceding getElementPtr instrs.
+
+    // Check that all offsets are constant for this instruction
+    for (OI = firstIdx; allConstantOffsets && OI != lastIdx; ++OI)
+      allConstantOffsets = isa<ConstantInt>(*OI);
+
+    if (allConstantOffsets) {
+      // Get pointer value out of ptrChild.
+      ptrVal = gepInst->getPointerOperand();
+
+      // Remember if it has leading zero index: it will be discarded later.
+      lastInstHasLeadingNonZero = ! IsZero(*firstIdx);
+
+      // Insert its index vector at the start, skipping any leading [0]
+      chainIdxVec.insert(chainIdxVec.begin(),
+                         firstIdx + !lastInstHasLeadingNonZero, lastIdx);
+
+      // Mark the folded node so no code is generated for it.
+      ((InstructionNode*) ptrChild)->markFoldedIntoParent();
+
+      // Get the previous GEP instruction and continue trying to fold
+      ptrChild = dyn_cast<InstructionNode>(ptrChild->leftChild());
+    } else // cannot fold this getElementPtr instr. or any preceding ones
+      break;
+  }
 
+  // If the first getElementPtr instruction had a leading [0], add it back.
+  // Note that this instruction is the *last* one successfully folded above.
+  if (ptrVal && ! lastInstHasLeadingNonZero) 
+    chainIdxVec.insert(chainIdxVec.begin(), ConstantSInt::get(Type::LongTy,0));
 
-static void SetMemOperands_Internal     (vector<MachineInstr*>& mvec,
-                                         vector<MachineInstr*>::iterator mvecI,
-                                         const InstructionNode* vmInstrNode,
-                                         Value* ptrVal,
-                                         std::vector<Value*>& idxVec,
-                                         bool allConstantIndices,
-                                         const TargetMachine& target);
+  return ptrVal;
+}
+
+
+//---------------------------------------------------------------------------
+// Function: GetGEPInstArgs
+// 
+// Purpose:
+//   Helper function for GetMemInstArgs that handles the final getElementPtr
+//   instruction used by (or same as) the memory operation.
+//   Extracts the indices of the current instruction and tries to fold in
+//   preceding ones if all indices of the current one are constant.
+//---------------------------------------------------------------------------
+
+static Value *
+GetGEPInstArgs(InstructionNode* gepNode,
+               std::vector<Value*>& idxVec,
+               bool& allConstantIndices)
+{
+  allConstantIndices = true;
+  GetElementPtrInst* gepI = cast<GetElementPtrInst>(gepNode->getInstruction());
+
+  // Default pointer is the one from the current instruction.
+  Value* ptrVal = gepI->getPointerOperand();
+  InstrTreeNode* ptrChild = gepNode->leftChild(); 
+
+  // Extract the index vector of the GEP instructin.
+  // If all indices are constant and first index is zero, try to fold
+  // in preceding GEPs with all constant indices.
+  for (User::op_iterator OI=gepI->idx_begin(),  OE=gepI->idx_end();
+       allConstantIndices && OI != OE; ++OI)
+    if (! isa<Constant>(*OI))
+      allConstantIndices = false;     // note: this also terminates loop!
+
+  // If we have only constant indices, fold chains of constant indices
+  // in this and any preceding GetElemPtr instructions.
+  bool foldedGEPs = false;
+  bool leadingNonZeroIdx = gepI && ! IsZero(*gepI->idx_begin());
+  if (allConstantIndices)
+    if (Value* newPtr = FoldGetElemChain(ptrChild, idxVec, leadingNonZeroIdx)) {
+      ptrVal = newPtr;
+      foldedGEPs = true;
+    }
+
+  // Append the index vector of the current instruction.
+  // Skip the leading [0] index if preceding GEPs were folded into this.
+  idxVec.insert(idxVec.end(),
+                gepI->idx_begin() + (foldedGEPs && !leadingNonZeroIdx),
+                gepI->idx_end());
+
+  return ptrVal;
+}
+
+//---------------------------------------------------------------------------
+// Function: GetMemInstArgs
+// 
+// Purpose:
+//   Get the pointer value and the index vector for a memory operation
+//   (GetElementPtr, Load, or Store).  If all indices of the given memory
+//   operation are constant, fold in constant indices in a chain of
+//   preceding GetElementPtr instructions (if any), and return the
+//   pointer value of the first instruction in the chain.
+//   All folded instructions are marked so no code is generated for them.
+//
+// Return values:
+//   Returns the pointer Value to use.
+//   Returns the resulting IndexVector in idxVec.
+//   Returns true/false in allConstantIndices if all indices are/aren't const.
+//---------------------------------------------------------------------------
+
+static Value*
+GetMemInstArgs(InstructionNode* memInstrNode,
+               std::vector<Value*>& idxVec,
+               bool& allConstantIndices)
+{
+  allConstantIndices = false;
+  Instruction* memInst = memInstrNode->getInstruction();
+  assert(idxVec.size() == 0 && "Need empty vector to return indices");
+
+  // If there is a GetElemPtr instruction to fold in to this instr,
+  // it must be in the left child for Load and GetElemPtr, and in the
+  // right child for Store instructions.
+  InstrTreeNode* ptrChild = (memInst->getOpcode() == Instruction::Store
+                             ? memInstrNode->rightChild()
+                             : memInstrNode->leftChild()); 
+  
+  // Default pointer is the one from the current instruction.
+  Value* ptrVal = ptrChild->getValue(); 
+
+  // Find the "last" GetElemPtr instruction: this one or the immediate child.
+  // There will be none if this is a load or a store from a scalar pointer.
+  InstructionNode* gepNode = NULL;
+  if (isa<GetElementPtrInst>(memInst))
+    gepNode = memInstrNode;
+  else if (isa<InstructionNode>(ptrChild) && isa<GetElementPtrInst>(ptrVal)) {
+    // Child of load/store is a GEP and memInst is its only use.
+    // Use its indices and mark it as folded.
+    gepNode = cast<InstructionNode>(ptrChild);
+    gepNode->markFoldedIntoParent();
+  }
+
+  // If there are no indices, return the current pointer.
+  // Else extract the pointer from the GEP and fold the indices.
+  return gepNode ? GetGEPInstArgs(gepNode, idxVec, allConstantIndices)
+                 : ptrVal;
+}
 
 
 //************************ Internal Functions ******************************/
@@ -55,18 +267,18 @@ ChooseBprInstruction(const InstructionNode* instrNode)
     ((InstructionNode*) instrNode->leftChild())->getInstruction();
   
   switch(setCCInstr->getOpcode())
-    {
-    case Instruction::SetEQ: opCode = BRZ;   break;
-    case Instruction::SetNE: opCode = BRNZ;  break;
-    case Instruction::SetLE: opCode = BRLEZ; break;
-    case Instruction::SetGE: opCode = BRGEZ; break;
-    case Instruction::SetLT: opCode = BRLZ;  break;
-    case Instruction::SetGT: opCode = BRGZ;  break;
-    default:
-      assert(0 && "Unrecognized VM instruction!");
-      opCode = INVALID_OPCODE;
-      break; 
-    }
+  {
+  case Instruction::SetEQ: opCode = V9::BRZ;   break;
+  case Instruction::SetNE: opCode = V9::BRNZ;  break;
+  case Instruction::SetLE: opCode = V9::BRLEZ; break;
+  case Instruction::SetGE: opCode = V9::BRGEZ; break;
+  case Instruction::SetLT: opCode = V9::BRLZ;  break;
+  case Instruction::SetGT: opCode = V9::BRGZ;  break;
+  default:
+    assert(0 && "Unrecognized VM instruction!");
+    opCode = V9::INVALID_OPCODE;
+    break; 
+  }
   
   return opCode;
 }
@@ -76,40 +288,37 @@ static inline MachineOpCode
 ChooseBpccInstruction(const InstructionNode* instrNode,
                       const BinaryOperator* setCCInstr)
 {
-  MachineOpCode opCode = INVALID_OPCODE;
+  MachineOpCode opCode = V9::INVALID_OPCODE;
   
   bool isSigned = setCCInstr->getOperand(0)->getType()->isSigned();
   
-  if (isSigned)
+  if (isSigned) {
+    switch(setCCInstr->getOpcode())
     {
-      switch(setCCInstr->getOpcode())
-        {
-        case Instruction::SetEQ: opCode = BE;  break;
-        case Instruction::SetNE: opCode = BNE; break;
-        case Instruction::SetLE: opCode = BLE; break;
-        case Instruction::SetGE: opCode = BGE; break;
-        case Instruction::SetLT: opCode = BL;  break;
-        case Instruction::SetGT: opCode = BG;  break;
-        default:
-          assert(0 && "Unrecognized VM instruction!");
-          break; 
-        }
+    case Instruction::SetEQ: opCode = V9::BE;  break;
+    case Instruction::SetNE: opCode = V9::BNE; break;
+    case Instruction::SetLE: opCode = V9::BLE; break;
+    case Instruction::SetGE: opCode = V9::BGE; break;
+    case Instruction::SetLT: opCode = V9::BL;  break;
+    case Instruction::SetGT: opCode = V9::BG;  break;
+    default:
+      assert(0 && "Unrecognized VM instruction!");
+      break; 
     }
-  else
+  } else {
+    switch(setCCInstr->getOpcode())
     {
-      switch(setCCInstr->getOpcode())
-        {
-        case Instruction::SetEQ: opCode = BE;   break;
-        case Instruction::SetNE: opCode = BNE;  break;
-        case Instruction::SetLE: opCode = BLEU; break;
-        case Instruction::SetGE: opCode = BCC;  break;
-        case Instruction::SetLT: opCode = BCS;  break;
-        case Instruction::SetGT: opCode = BGU;  break;
-        default:
-          assert(0 && "Unrecognized VM instruction!");
-          break; 
-        }
+    case Instruction::SetEQ: opCode = V9::BE;   break;
+    case Instruction::SetNE: opCode = V9::BNE;  break;
+    case Instruction::SetLE: opCode = V9::BLEU; break;
+    case Instruction::SetGE: opCode = V9::BCC;  break;
+    case Instruction::SetLT: opCode = V9::BCS;  break;
+    case Instruction::SetGT: opCode = V9::BGU;  break;
+    default:
+      assert(0 && "Unrecognized VM instruction!");
+      break; 
     }
+  }
   
   return opCode;
 }
@@ -118,20 +327,20 @@ static inline MachineOpCode
 ChooseBFpccInstruction(const InstructionNode* instrNode,
                        const BinaryOperator* setCCInstr)
 {
-  MachineOpCode opCode = INVALID_OPCODE;
+  MachineOpCode opCode = V9::INVALID_OPCODE;
   
   switch(setCCInstr->getOpcode())
-    {
-    case Instruction::SetEQ: opCode = FBE;  break;
-    case Instruction::SetNE: opCode = FBNE; break;
-    case Instruction::SetLE: opCode = FBLE; break;
-    case Instruction::SetGE: opCode = FBGE; break;
-    case Instruction::SetLT: opCode = FBL;  break;
-    case Instruction::SetGT: opCode = FBG;  break;
-    default:
-      assert(0 && "Unrecognized VM instruction!");
-      break; 
-    }
+  {
+  case Instruction::SetEQ: opCode = V9::FBE;  break;
+  case Instruction::SetNE: opCode = V9::FBNE; break;
+  case Instruction::SetLE: opCode = V9::FBLE; break;
+  case Instruction::SetGE: opCode = V9::FBGE; break;
+  case Instruction::SetLT: opCode = V9::FBL;  break;
+  case Instruction::SetGT: opCode = V9::FBG;  break;
+  default:
+    assert(0 && "Unrecognized VM instruction!");
+    break; 
+  }
   
   return opCode;
 }
@@ -148,17 +357,16 @@ ChooseBFpccInstruction(const InstructionNode* instrNode,
 static TmpInstruction*
 GetTmpForCC(Value* boolVal, const Function *F, const Type* ccType)
 {
-  typedef std::hash_map<const Value*, TmpInstruction*> BoolTmpCache;
+  typedef hash_map<const Value*, TmpInstruction*> BoolTmpCache;
   static BoolTmpCache boolToTmpCache;     // Map boolVal -> TmpInstruction*
   static const Function *lastFunction = 0;// Use to flush cache between funcs
   
   assert(boolVal->getType() == Type::BoolTy && "Weird but ok! Delete assert");
   
-  if (lastFunction != F)
-    {
-      lastFunction = F;
-      boolToTmpCache.clear();
-    }
+  if (lastFunction != F) {
+    lastFunction = F;
+    boolToTmpCache.clear();
+  }
   
   // Look for tmpI and create a new one otherwise.  The new value is
   // directly written to map using the ref returned by operator[].
@@ -175,7 +383,8 @@ ChooseBccInstruction(const InstructionNode* instrNode,
                      bool& isFPBranch)
 {
   InstructionNode* setCCNode = (InstructionNode*) instrNode->leftChild();
-  BinaryOperator* setCCInstr = (BinaryOperator*) setCCNode->getInstruction();
+  assert(setCCNode->getOpLabel() == SetCCOp);
+  BinaryOperator* setCCInstr =cast<BinaryOperator>(setCCNode->getInstruction());
   const Type* setCCType = setCCInstr->getOperand(0)->getType();
   
   isFPBranch = setCCType->isFloatingPoint(); // Return value: don't delete!
@@ -190,20 +399,20 @@ ChooseBccInstruction(const InstructionNode* instrNode,
 static inline MachineOpCode 
 ChooseMovFpccInstruction(const InstructionNode* instrNode)
 {
-  MachineOpCode opCode = INVALID_OPCODE;
+  MachineOpCode opCode = V9::INVALID_OPCODE;
   
   switch(instrNode->getInstruction()->getOpcode())
-    {
-    case Instruction::SetEQ: opCode = MOVFE;  break;
-    case Instruction::SetNE: opCode = MOVFNE; break;
-    case Instruction::SetLE: opCode = MOVFLE; break;
-    case Instruction::SetGE: opCode = MOVFGE; break;
-    case Instruction::SetLT: opCode = MOVFL;  break;
-    case Instruction::SetGT: opCode = MOVFG;  break;
-    default:
-      assert(0 && "Unrecognized VM instruction!");
-      break; 
-    }
+  {
+  case Instruction::SetEQ: opCode = V9::MOVFE;  break;
+  case Instruction::SetNE: opCode = V9::MOVFNE; break;
+  case Instruction::SetLE: opCode = V9::MOVFLE; break;
+  case Instruction::SetGE: opCode = V9::MOVFGE; break;
+  case Instruction::SetLT: opCode = V9::MOVFL;  break;
+  case Instruction::SetGT: opCode = V9::MOVFG;  break;
+  default:
+    assert(0 && "Unrecognized VM instruction!");
+    break; 
+  }
   
   return opCode;
 }
@@ -222,20 +431,20 @@ ChooseMovpccAfterSub(const InstructionNode* instrNode,
                      bool& mustClearReg,
                      int& valueToMove)
 {
-  MachineOpCode opCode = INVALID_OPCODE;
+  MachineOpCode opCode = V9::INVALID_OPCODE;
   mustClearReg = true;
   valueToMove = 1;
   
   switch(instrNode->getInstruction()->getOpcode())
-    {
-    case Instruction::SetEQ: opCode = MOVE;  break;
-    case Instruction::SetLE: opCode = MOVLE; break;
-    case Instruction::SetGE: opCode = MOVGE; break;
-    case Instruction::SetLT: opCode = MOVL;  break;
-    case Instruction::SetGT: opCode = MOVG;  break;
-    case Instruction::SetNE: assert(0 && "No move required!"); break;
-    default:                assert(0 && "Unrecognized VM instr!"); break; 
-    }
+  {
+  case Instruction::SetEQ: opCode = V9::MOVE;  break;
+  case Instruction::SetLE: opCode = V9::MOVLE; break;
+  case Instruction::SetGE: opCode = V9::MOVGE; break;
+  case Instruction::SetLT: opCode = V9::MOVL;  break;
+  case Instruction::SetGT: opCode = V9::MOVG;  break;
+  case Instruction::SetNE: assert(0 && "No move required!"); break;
+  default:                  assert(0 && "Unrecognized VM instr!"); break; 
+  }
   
   return opCode;
 }
@@ -243,93 +452,127 @@ ChooseMovpccAfterSub(const InstructionNode* instrNode,
 static inline MachineOpCode
 ChooseConvertToFloatInstr(OpLabel vopCode, const Type* opType)
 {
-  MachineOpCode opCode = INVALID_OPCODE;
+  MachineOpCode opCode = V9::INVALID_OPCODE;
   
   switch(vopCode)
-    {
-    case ToFloatTy: 
-      if (opType == Type::SByteTy || opType == Type::ShortTy || opType == Type::IntTy)
-        opCode = FITOS;
-      else if (opType == Type::LongTy)
-        opCode = FXTOS;
-      else if (opType == Type::DoubleTy)
-        opCode = FDTOS;
-      else if (opType == Type::FloatTy)
-        ;
-      else
-        assert(0 && "Cannot convert this type to FLOAT on SPARC");
-      break;
+  {
+  case ToFloatTy: 
+    if (opType == Type::SByteTy || opType == Type::ShortTy ||
+        opType == Type::IntTy)
+      opCode = V9::FITOS;
+    else if (opType == Type::LongTy)
+      opCode = V9::FXTOS;
+    else if (opType == Type::DoubleTy)
+      opCode = V9::FDTOS;
+    else if (opType == Type::FloatTy)
+      ;
+    else
+      assert(0 && "Cannot convert this type to FLOAT on SPARC");
+    break;
       
-    case ToDoubleTy: 
-      // This is usually used in conjunction with CreateCodeToCopyIntToFloat().
-      // Both functions should treat the integer as a 32-bit value for types
-      // of 4 bytes or less, and as a 64-bit value otherwise.
-      if (opType == Type::SByteTy || opType == Type::UByteTy ||
-          opType == Type::ShortTy || opType == Type::UShortTy ||
-          opType == Type::IntTy   || opType == Type::UIntTy)
-        opCode = FITOD;
-      else if (opType == Type::LongTy || opType == Type::ULongTy)
-        opCode = FXTOD;
-      else if (opType == Type::FloatTy)
-        opCode = FSTOD;
-      else if (opType == Type::DoubleTy)
-        ;
-      else
-        assert(0 && "Cannot convert this type to DOUBLE on SPARC");
-      break;
+  case ToDoubleTy: 
+    // This is usually used in conjunction with CreateCodeToCopyIntToFloat().
+    // Both functions should treat the integer as a 32-bit value for types
+    // of 4 bytes or less, and as a 64-bit value otherwise.
+    if (opType == Type::SByteTy || opType == Type::UByteTy ||
+        opType == Type::ShortTy || opType == Type::UShortTy ||
+        opType == Type::IntTy   || opType == Type::UIntTy)
+      opCode = V9::FITOD;
+    else if (opType == Type::LongTy || opType == Type::ULongTy)
+      opCode = V9::FXTOD;
+    else if (opType == Type::FloatTy)
+      opCode = V9::FSTOD;
+    else if (opType == Type::DoubleTy)
+      ;
+    else
+      assert(0 && "Cannot convert this type to DOUBLE on SPARC");
+    break;
       
-    default:
-      break;
-    }
+  default:
+    break;
+  }
   
   return opCode;
 }
 
 static inline MachineOpCode 
-ChooseConvertToIntInstr(OpLabel vopCode, const Type* opType)
+ChooseConvertFPToIntInstr(Type::PrimitiveID tid, const Type* opType)
 {
-  MachineOpCode opCode = INVALID_OPCODE;;
-  
-  if (vopCode == ToSByteTy || vopCode == ToShortTy || vopCode == ToIntTy)
-    {
-      switch (opType->getPrimitiveID())
-        {
-        case Type::FloatTyID:   opCode = FSTOI; break;
-        case Type::DoubleTyID:  opCode = FDTOI; break;
-        default:
-          assert(0 && "Non-numeric non-bool type cannot be converted to Int");
-          break;
-        }
-    }
-  else if (vopCode == ToLongTy)
-    {
-      switch (opType->getPrimitiveID())
-        {
-        case Type::FloatTyID:   opCode = FSTOX; break;
-        case Type::DoubleTyID:  opCode = FDTOX; break;
-        default:
-          assert(0 && "Non-numeric non-bool type cannot be converted to Long");
-          break;
-        }
-    }
-  else
-      assert(0 && "Should not get here, Mo!");
-  
+  MachineOpCode opCode = V9::INVALID_OPCODE;;
+
+  assert((opType == Type::FloatTy || opType == Type::DoubleTy)
+         && "This function should only be called for FLOAT or DOUBLE");
+
+  if (tid == Type::UIntTyID) {
+    assert(tid != Type::UIntTyID && "FP-to-uint conversions must be expanded"
+           " into FP->long->uint for SPARC v9:  SO RUN PRESELECTION PASS!");
+  } else if (tid == Type::SByteTyID || tid == Type::ShortTyID || 
+             tid == Type::IntTyID   || tid == Type::UByteTyID ||
+             tid == Type::UShortTyID) {
+    opCode = (opType == Type::FloatTy)? V9::FSTOI : V9::FDTOI;
+  } else if (tid == Type::LongTyID || tid == Type::ULongTyID) {
+      opCode = (opType == Type::FloatTy)? V9::FSTOX : V9::FDTOX;
+  } else
+    assert(0 && "Should not get here, Mo!");
+
   return opCode;
 }
 
 MachineInstr*
-CreateConvertToIntInstr(OpLabel vopCode, Value* srcVal, Value* destVal)
+CreateConvertFPToIntInstr(Type::PrimitiveID destTID,
+                          Value* srcVal, Value* destVal)
 {
-  MachineOpCode opCode = ChooseConvertToIntInstr(vopCode, srcVal->getType());
-  assert(opCode != INVALID_OPCODE && "Expected to need conversion!");
-  
-  MachineInstr* M = new MachineInstr(opCode);
-  M->SetMachineOperandVal(0, MachineOperand::MO_VirtualRegister, srcVal);
-  M->SetMachineOperandVal(1, MachineOperand::MO_VirtualRegister, destVal);
-  return M;
+  MachineOpCode opCode = ChooseConvertFPToIntInstr(destTID, srcVal->getType());
+  assert(opCode != V9::INVALID_OPCODE && "Expected to need conversion!");
+  return BuildMI(opCode, 2).addReg(srcVal).addRegDef(destVal);
+}
+
+// CreateCodeToConvertFloatToInt: Convert FP value to signed or unsigned integer
+// The FP value must be converted to the dest type in an FP register,
+// and the result is then copied from FP to int register via memory.
+//
+// Since fdtoi converts to signed integers, any FP value V between MAXINT+1
+// and MAXUNSIGNED (i.e., 2^31 <= V <= 2^32-1) would be converted incorrectly
+// *only* when converting to an unsigned.  (Unsigned byte, short or long
+// don't have this problem.)
+// For unsigned int, we therefore have to generate the code sequence:
+// 
+//      if (V > (float) MAXINT) {
+//        unsigned result = (unsigned) (V  - (float) MAXINT);
+//        result = result + (unsigned) MAXINT;
+//      }
+//      else
+//        result = (unsigned) V;
+// 
+static void
+CreateCodeToConvertFloatToInt(const TargetMachine& target,
+                              Value* opVal,
+                              Instruction* destI,
+                              std::vector<MachineInstr*>& mvec,
+                              MachineCodeForInstruction& mcfi)
+{
+  // Create a temporary to represent the FP register into which the
+  // int value will placed after conversion.  The type of this temporary
+  // depends on the type of FP register to use: single-prec for a 32-bit
+  // int or smaller; double-prec for a 64-bit int.
+  // 
+  size_t destSize = target.getTargetData().getTypeSize(destI->getType());
+  const Type* destTypeToUse = (destSize > 4)? Type::DoubleTy : Type::FloatTy;
+  TmpInstruction* destForCast = new TmpInstruction(destTypeToUse, opVal);
+  mcfi.addTemp(destForCast);
+
+  // Create the fp-to-int conversion code
+  MachineInstr* M =CreateConvertFPToIntInstr(destI->getType()->getPrimitiveID(),
+                                             opVal, destForCast);
+  mvec.push_back(M);
+
+  // Create the fpreg-to-intreg copy code
+  target.getInstrInfo().
+    CreateCodeToCopyFloatToInt(target, destI->getParent()->getParent(),
+                               destForCast, destI, mvec, mcfi);
 }
 
+
 static inline MachineOpCode 
 ChooseAddInstruction(const InstructionNode* instrNode)
 {
@@ -341,13 +584,9 @@ static inline MachineInstr*
 CreateMovFloatInstruction(const InstructionNode* instrNode,
                           const Type* resultType)
 {
-  MachineInstr* minstr = new MachineInstr((resultType == Type::FloatTy)
-                                          ? FMOVS : FMOVD);
-  minstr->SetMachineOperandVal(0, MachineOperand::MO_VirtualRegister,
-                               instrNode->leftChild()->getValue());
-  minstr->SetMachineOperandVal(1, MachineOperand::MO_VirtualRegister,
-                               instrNode->getValue());
-  return minstr;
+  return BuildMI((resultType == Type::FloatTy) ? V9::FMOVS : V9::FMOVD, 2)
+                   .addReg(instrNode->leftChild()->getValue())
+                   .addRegDef(instrNode->getValue());
 }
 
 static inline MachineInstr* 
@@ -363,11 +602,11 @@ CreateAddConstInstruction(const InstructionNode* instrNode)
   //    instead of an FADD (1 vs 3 cycles).  There is no integer MOV.
   // 
   if (ConstantFP *FPC = dyn_cast<ConstantFP>(constOp)) {
-      double dval = FPC->getValue();
-      if (dval == 0.0)
-        minstr = CreateMovFloatInstruction(instrNode,
-                                   instrNode->getInstruction()->getType());
-    }
+    double dval = FPC->getValue();
+    if (dval == 0.0)
+      minstr = CreateMovFloatInstruction(instrNode,
+                                        instrNode->getInstruction()->getType());
+  }
   
   return minstr;
 }
@@ -376,20 +615,19 @@ CreateAddConstInstruction(const InstructionNode* instrNode)
 static inline MachineOpCode 
 ChooseSubInstructionByType(const Type* resultType)
 {
-  MachineOpCode opCode = INVALID_OPCODE;
+  MachineOpCode opCode = V9::INVALID_OPCODE;
   
-  if (resultType->isIntegral() || isa<PointerType>(resultType))
+  if (resultType->isInteger() || isa<PointerType>(resultType)) {
+      opCode = V9::SUB;
+  } else {
+    switch(resultType->getPrimitiveID())
     {
-      opCode = SUB;
+    case Type::FloatTyID:  opCode = V9::FSUBS; break;
+    case Type::DoubleTyID: opCode = V9::FSUBD; break;
+    default: assert(0 && "Invalid type for SUB instruction"); break; 
     }
-  else
-    switch(resultType->getPrimitiveID())
-      {
-      case Type::FloatTyID:  opCode = FSUBS; break;
-      case Type::DoubleTyID: opCode = FSUBD; break;
-      default: assert(0 && "Invalid type for SUB instruction"); break; 
-      }
-  
+  }
+
   return opCode;
 }
 
@@ -420,12 +658,12 @@ CreateSubConstInstruction(const InstructionNode* instrNode)
 static inline MachineOpCode 
 ChooseFcmpInstruction(const InstructionNode* instrNode)
 {
-  MachineOpCode opCode = INVALID_OPCODE;
+  MachineOpCode opCode = V9::INVALID_OPCODE;
   
   Value* operand = ((InstrTreeNode*) instrNode->leftChild())->getValue();
   switch(operand->getType()->getPrimitiveID()) {
-  case Type::FloatTyID:  opCode = FCMPS; break;
-  case Type::DoubleTyID: opCode = FCMPD; break;
+  case Type::FloatTyID:  opCode = V9::FCMPS; break;
+  case Type::DoubleTyID: opCode = V9::FCMPD; break;
   default: assert(0 && "Invalid type for FCMP instruction"); break; 
   }
   
@@ -454,15 +692,15 @@ BothFloatToDouble(const InstructionNode* instrNode)
 static inline MachineOpCode 
 ChooseMulInstructionByType(const Type* resultType)
 {
-  MachineOpCode opCode = INVALID_OPCODE;
+  MachineOpCode opCode = V9::INVALID_OPCODE;
   
-  if (resultType->isIntegral())
-    opCode = MULX;
+  if (resultType->isInteger())
+    opCode = V9::MULX;
   else
     switch(resultType->getPrimitiveID())
       {
-      case Type::FloatTyID:  opCode = FMULS; break;
-      case Type::DoubleTyID: opCode = FMULD; break;
+      case Type::FloatTyID:  opCode = V9::FMULS; break;
+      case Type::DoubleTyID: opCode = V9::FMULD; break;
       default: assert(0 && "Invalid type for MUL instruction"); break; 
       }
   
@@ -475,11 +713,8 @@ static inline MachineInstr*
 CreateIntNegInstruction(const TargetMachine& target,
                         Value* vreg)
 {
-  MachineInstr* minstr = new MachineInstr(SUB);
-  minstr->SetMachineOperandReg(0, target.getRegInfo().getZeroRegNum());
-  minstr->SetMachineOperandVal(1, MachineOperand::MO_VirtualRegister, vreg);
-  minstr->SetMachineOperandVal(2, MachineOperand::MO_VirtualRegister, vreg);
-  return minstr;
+  return BuildMI(V9::SUB, 3).addMReg(target.getRegInfo().getZeroRegNum())
+    .addReg(vreg).addRegDef(vreg);
 }
 
 
@@ -497,9 +732,9 @@ CreateShiftInstructions(const TargetMachine& target,
                         MachineOpCode shiftOpCode,
                         Value* argVal1,
                         Value* optArgVal2, /* Use optArgVal2 if not NULL */
-                        unsigned int optShiftNum, /* else use optShiftNum */
+                        unsigned optShiftNum, /* else use optShiftNum */
                         Instruction* destVal,
-                        vector<MachineInstr*>& mvec,
+                        std::vector<MachineInstr*>& mvec,
                         MachineCodeForInstruction& mcfi)
 {
   assert((optArgVal2 != NULL || optShiftNum <= 64) &&
@@ -511,26 +746,26 @@ CreateShiftInstructions(const TargetMachine& target,
   // of dest, so we need to put the result of the SLL into a temporary.
   // 
   Value* shiftDest = destVal;
-  const Type* opType = argVal1->getType();
-  unsigned opSize = target.DataLayout.getTypeSize(argVal1->getType());
-  if ((shiftOpCode == SLL || shiftOpCode == SLLX)
-      && opSize < target.DataLayout.getIntegerRegize())
+  unsigned opSize = target.getTargetData().getTypeSize(argVal1->getType());
+  if ((shiftOpCode == V9::SLL || shiftOpCode == V9::SLLX) && opSize < 8)
     { // put SLL result into a temporary
       shiftDest = new TmpInstruction(argVal1, optArgVal2, "sllTmp");
       mcfi.addTemp(shiftDest);
     }
   
   MachineInstr* M = (optArgVal2 != NULL)
-    ? Create3OperandInstr(shiftOpCode, argVal1, optArgVal2, shiftDest)
-    : Create3OperandInstr_UImmed(shiftOpCode, argVal1, optShiftNum, shiftDest);
+    ? BuildMI(shiftOpCode, 3).addReg(argVal1).addReg(optArgVal2)
+                             .addReg(shiftDest, MOTy::Def)
+    : BuildMI(shiftOpCode, 3).addReg(argVal1).addZImm(optShiftNum)
+                             .addReg(shiftDest, MOTy::Def);
   mvec.push_back(M);
   
   if (shiftDest != destVal)
     { // extend the sign-bit of the result into all upper bits of dest
       assert(8*opSize <= 32 && "Unexpected type size > 4 and < IntRegSize?");
       target.getInstrInfo().
-        CreateSignExtensionInstructions(target, F, shiftDest, 8*opSize,
-                                        destVal, mvec, mcfi);
+        CreateSignExtensionInstructions(target, F, shiftDest, destVal,
+                                        8*opSize, mvec, mcfi);
     }
 }
 
@@ -539,15 +774,15 @@ CreateShiftInstructions(const TargetMachine& target,
 // create a cheaper instruction.
 // This returns the approximate cost of the instructions generated,
 // which is used to pick the cheapest when both operands are constant.
-static inline unsigned int
+static inline unsigned
 CreateMulConstInstruction(const TargetMachine &target, Function* F,
                           Value* lval, Value* rval, Instruction* destVal,
-                          vector<MachineInstr*>& mvec,
+                          std::vector<MachineInstr*>& mvec,
                           MachineCodeForInstruction& mcfi)
 {
   /* Use max. multiply cost, viz., cost of MULX */
-  unsigned int cost = target.getInstrInfo().minLatency(MULX);
-  unsigned int firstNewInstr = mvec.size();
+  unsigned cost = target.getInstrInfo().minLatency(V9::MULX);
+  unsigned firstNewInstr = mvec.size();
   
   Value* constOp = rval;
   if (! isa<Constant>(constOp))
@@ -559,70 +794,57 @@ CreateMulConstInstruction(const TargetMachine &target, Function* F,
   // 
   const Type* resultType = destVal->getType();
   
-  if (resultType->isIntegral() || isa<PointerType>(resultType))
-    {
-      bool isValidConst;
-      int64_t C = GetConstantValueAsSignedInt(constOp, isValidConst);
-      if (isValidConst)
-        {
-          unsigned pow;
-          bool needNeg = false;
-          if (C < 0)
-            {
-              needNeg = true;
-              C = -C;
-            }
+  if (resultType->isInteger() || isa<PointerType>(resultType)) {
+    bool isValidConst;
+    int64_t C = GetConstantValueAsSignedInt(constOp, isValidConst);
+    if (isValidConst) {
+      unsigned pow;
+      bool needNeg = false;
+      if (C < 0) {
+        needNeg = true;
+        C = -C;
+      }
           
-          if (C == 0 || C == 1)
-            {
-              cost = target.getInstrInfo().minLatency(ADD);
-              MachineInstr* M = (C == 0)
-                ? Create3OperandInstr_Reg(ADD,
-                                          target.getRegInfo().getZeroRegNum(),
-                                          target.getRegInfo().getZeroRegNum(),
-                                          destVal)
-                : Create3OperandInstr_Reg(ADD, lval,
-                                          target.getRegInfo().getZeroRegNum(),
-                                          destVal);
-              mvec.push_back(M);
-            }
-          else if (isPowerOf2(C, pow))
-            {
-              unsigned int opSize = target.DataLayout.getTypeSize(resultType);
-              MachineOpCode opCode = (opSize <= 32)? SLL : SLLX;
-              CreateShiftInstructions(target, F, opCode, lval, NULL, pow,
-                                      destVal, mvec, mcfi); 
-            }
+      if (C == 0 || C == 1) {
+        cost = target.getInstrInfo().minLatency(V9::ADD);
+        unsigned Zero = target.getRegInfo().getZeroRegNum();
+        MachineInstr* M;
+        if (C == 0)
+          M = BuildMI(V9::ADD,3).addMReg(Zero).addMReg(Zero).addRegDef(destVal);
+        else
+          M = BuildMI(V9::ADD,3).addReg(lval).addMReg(Zero).addRegDef(destVal);
+        mvec.push_back(M);
+      }
+      else if (isPowerOf2(C, pow)) {
+        unsigned opSize = target.getTargetData().getTypeSize(resultType);
+        MachineOpCode opCode = (opSize <= 32)? V9::SLL : V9::SLLX;
+        CreateShiftInstructions(target, F, opCode, lval, NULL, pow,
+                                destVal, mvec, mcfi);
+      }
           
-          if (mvec.size() > 0 && needNeg)
-            { // insert <reg = SUB 0, reg> after the instr to flip the sign
-              MachineInstr* M = CreateIntNegInstruction(target, destVal);
-              mvec.push_back(M);
-            }
-        }
+      if (mvec.size() > 0 && needNeg)
+      { // insert <reg = SUB 0, reg> after the instr to flip the sign
+        MachineInstr* M = CreateIntNegInstruction(target, destVal);
+        mvec.push_back(M);
+      }
     }
-  else
-    {
-      if (ConstantFP *FPC = dyn_cast<ConstantFP>(constOp))
-        {
-          double dval = FPC->getValue();
-          if (fabs(dval) == 1)
-            {
-              MachineOpCode opCode =  (dval < 0)
-                ? (resultType == Type::FloatTy? FNEGS : FNEGD)
-                : (resultType == Type::FloatTy? FMOVS : FMOVD);
-              MachineInstr* M = Create2OperandInstr(opCode, lval, destVal);
-              mvec.push_back(M);
-            } 
-        }
+  } else {
+    if (ConstantFP *FPC = dyn_cast<ConstantFP>(constOp)) {
+      double dval = FPC->getValue();
+      if (fabs(dval) == 1) {
+        MachineOpCode opCode =  (dval < 0)
+          ? (resultType == Type::FloatTy? V9::FNEGS : V9::FNEGD)
+          : (resultType == Type::FloatTy? V9::FMOVS : V9::FMOVD);
+        mvec.push_back(BuildMI(opCode,2).addReg(lval).addRegDef(destVal));
+      } 
     }
+  }
   
-  if (firstNewInstr < mvec.size())
-    {
-      cost = 0;
-      for (unsigned int i=firstNewInstr; i < mvec.size(); ++i)
-        cost += target.getInstrInfo().minLatency(mvec[i]->getOpCode());
-    }
+  if (firstNewInstr < mvec.size()) {
+    cost = 0;
+    for (unsigned i=firstNewInstr; i < mvec.size(); ++i)
+      cost += target.getInstrInfo().minLatency(mvec[i]->getOpCode());
+  }
   
   return cost;
 }
@@ -636,23 +858,15 @@ CreateCheapestMulConstInstruction(const TargetMachine &target,
                                   Function* F,
                                   Value* lval, Value* rval,
                                   Instruction* destVal,
-                                  vector<MachineInstr*>& mvec,
+                                  std::vector<MachineInstr*>& mvec,
                                   MachineCodeForInstruction& mcfi)
 {
   Value* constOp;
   if (isa<Constant>(lval) && isa<Constant>(rval))
-    { // both operands are constant: try both orders!
-      vector<MachineInstr*> mvec1, mvec2;
-      unsigned int lcost = CreateMulConstInstruction(target, F, lval, rval,
-                                                     destVal, mvec1, mcfi);
-      unsigned int rcost = CreateMulConstInstruction(target, F, rval, lval,
-                                                     destVal, mvec2, mcfi);
-      vector<MachineInstr*>& mincostMvec =  (lcost <= rcost)? mvec1 : mvec2;
-      vector<MachineInstr*>& maxcostMvec =  (lcost <= rcost)? mvec2 : mvec1;
-      mvec.insert(mvec.end(), mincostMvec.begin(), mincostMvec.end()); 
-
-      for (unsigned int i=0; i < maxcostMvec.size(); ++i)
-        delete maxcostMvec[i];
+    { // both operands are constant: evaluate and "set" in dest
+      Constant* P = ConstantFoldBinaryInstruction(Instruction::Mul,
+                                  cast<Constant>(lval), cast<Constant>(rval));
+      target.getInstrInfo().CreateCodeToLoadConst(target,F,P,destVal,mvec,mcfi);
     }
   else if (isa<Constant>(rval))         // rval is constant, but not lval
     CreateMulConstInstruction(target, F, lval, rval, destVal, mvec, mcfi);
@@ -667,25 +881,22 @@ CreateCheapestMulConstInstruction(const TargetMachine &target,
 static inline void
 CreateMulInstruction(const TargetMachine &target, Function* F,
                      Value* lval, Value* rval, Instruction* destVal,
-                     vector<MachineInstr*>& mvec,
+                     std::vector<MachineInstr*>& mvec,
                      MachineCodeForInstruction& mcfi,
                      MachineOpCode forceMulOp = INVALID_MACHINE_OPCODE)
 {
-  unsigned int L = mvec.size();
+  unsigned L = mvec.size();
   CreateCheapestMulConstInstruction(target,F, lval, rval, destVal, mvec, mcfi);
-  if (mvec.size() == L)
-    { // no instructions were added so create MUL reg, reg, reg.
-      // Use FSMULD if both operands are actually floats cast to doubles.
-      // Otherwise, use the default opcode for the appropriate type.
-      MachineOpCode mulOp = ((forceMulOp != INVALID_MACHINE_OPCODE)
-                             ? forceMulOp 
-                             : ChooseMulInstructionByType(destVal->getType()));
-      MachineInstr* M = new MachineInstr(mulOp);
-      M->SetMachineOperandVal(0, MachineOperand::MO_VirtualRegister, lval);
-      M->SetMachineOperandVal(1, MachineOperand::MO_VirtualRegister, rval);
-      M->SetMachineOperandVal(2, MachineOperand::MO_VirtualRegister, destVal);
-      mvec.push_back(M);
-    }
+  if (mvec.size() == L) {
+    // no instructions were added so create MUL reg, reg, reg.
+    // Use FSMULD if both operands are actually floats cast to doubles.
+    // Otherwise, use the default opcode for the appropriate type.
+    MachineOpCode mulOp = ((forceMulOp != INVALID_MACHINE_OPCODE)
+                           ? forceMulOp 
+                           : ChooseMulInstructionByType(destVal->getType()));
+    mvec.push_back(BuildMI(mulOp, 3).addReg(lval).addReg(rval)
+                   .addRegDef(destVal));
+  }
 }
 
 
@@ -697,17 +908,17 @@ static inline MachineOpCode
 ChooseDivInstruction(TargetMachine &target,
                      const InstructionNode* instrNode)
 {
-  MachineOpCode opCode = INVALID_OPCODE;
+  MachineOpCode opCode = V9::INVALID_OPCODE;
   
   const Type* resultType = instrNode->getInstruction()->getType();
   
-  if (resultType->isIntegral())
-    opCode = resultType->isSigned()? SDIVX : UDIVX;
+  if (resultType->isInteger())
+    opCode = resultType->isSigned()? V9::SDIVX : V9::UDIVX;
   else
     switch(resultType->getPrimitiveID())
       {
-      case Type::FloatTyID:  opCode = FDIVS; break;
-      case Type::DoubleTyID: opCode = FDIVD; break;
+      case Type::FloatTyID:  opCode = V9::FDIVS; break;
+      case Type::DoubleTyID: opCode = V9::FDIVD; break;
       default: assert(0 && "Invalid type for DIV instruction"); break; 
       }
   
@@ -715,199 +926,182 @@ ChooseDivInstruction(TargetMachine &target,
 }
 
 
-// Return NULL if we cannot exploit constant to create a cheaper instruction
+// Return if we cannot exploit constant to create a cheaper instruction
 static inline void
 CreateDivConstInstruction(TargetMachine &target,
                           const InstructionNode* instrNode,
-                          vector<MachineInstr*>& mvec)
+                          std::vector<MachineInstr*>& mvec)
 {
-  MachineInstr* minstr1 = NULL;
-  MachineInstr* minstr2 = NULL;
-  
+  Value* LHS  = instrNode->leftChild()->getValue();
   Value* constOp = ((InstrTreeNode*) instrNode->rightChild())->getValue();
-  if (! isa<Constant>(constOp))
+  if (!isa<Constant>(constOp))
     return;
+
+  Value* DestVal = instrNode->getValue();
+  unsigned ZeroReg = target.getRegInfo().getZeroRegNum();
   
   // Cases worth optimizing are:
   // (1) Divide by 1 for any type: replace with copy (ADD or FMOV)
   // (2) Divide by 2^x for integer types: replace with SR[L or A]{X}
   // 
   const Type* resultType = instrNode->getInstruction()->getType();
-  
-  if (resultType->isIntegral())
-    {
-      unsigned pow;
-      bool isValidConst;
-      int64_t C = GetConstantValueAsSignedInt(constOp, isValidConst);
-      if (isValidConst)
-        {
-          bool needNeg = false;
-          if (C < 0)
-            {
-              needNeg = true;
-              C = -C;
-            }
+  if (resultType->isInteger())
+  {
+    unsigned pow;
+    bool isValidConst;
+    int64_t C = GetConstantValueAsSignedInt(constOp, isValidConst);
+    if (isValidConst) {
+      bool needNeg = false;
+      if (C < 0) {
+        needNeg = true;
+        C = -C;
+      }
           
-          if (C == 1)
-            {
-              minstr1 = new MachineInstr(ADD);
-              minstr1->SetMachineOperandVal(0,
-                                           MachineOperand::MO_VirtualRegister,
-                                           instrNode->leftChild()->getValue());
-              minstr1->SetMachineOperandReg(1,
-                                        target.getRegInfo().getZeroRegNum());
-            }
-          else if (isPowerOf2(C, pow))
-            {
-              MachineOpCode opCode= ((resultType->isSigned())
-                                     ? (resultType==Type::LongTy)? SRAX : SRA
-                                     : (resultType==Type::LongTy)? SRLX : SRL);
-              minstr1 = new MachineInstr(opCode);
-              minstr1->SetMachineOperandVal(0,
-                                           MachineOperand::MO_VirtualRegister,
-                                           instrNode->leftChild()->getValue());
-              minstr1->SetMachineOperandConst(1,
-                                          MachineOperand::MO_UnextendedImmed,
-                                          pow);
-            }
+      if (C == 1) {
+        mvec.push_back(BuildMI(V9::ADD, 3).addReg(LHS).addMReg(ZeroReg)
+                       .addRegDef(DestVal));
+      } else if (isPowerOf2(C, pow)) {
+        unsigned opCode= ((resultType->isSigned())
+                          ? (resultType==Type::LongTy) ? V9::SRAX : V9::SRA
+                          : (resultType==Type::LongTy) ? V9::SRLX : V9::SRL);
+        mvec.push_back(BuildMI(opCode, 3).addReg(LHS).addZImm(pow)
+                       .addRegDef(DestVal));
+      }
           
-          if (minstr1 && needNeg)
-            { // insert <reg = SUB 0, reg> after the instr to flip the sign
-              minstr2 = CreateIntNegInstruction(target,
-                                                   instrNode->getValue());
-            }
-        }
+      if (needNeg && (C == 1 || isPowerOf2(C, pow))) {
+        // insert <reg = SUB 0, reg> after the instr to flip the sign
+        mvec.push_back(CreateIntNegInstruction(target, DestVal));
+      }
     }
-  else
-    {
-      if (ConstantFP *FPC = dyn_cast<ConstantFP>(constOp))
-        {
-          double dval = FPC->getValue();
-          if (fabs(dval) == 1)
-            {
-              bool needNeg = (dval < 0);
-              
-              MachineOpCode opCode = needNeg
-                ? (resultType == Type::FloatTy? FNEGS : FNEGD)
-                : (resultType == Type::FloatTy? FMOVS : FMOVD);
+  } else {
+    if (ConstantFP *FPC = dyn_cast<ConstantFP>(constOp)) {
+      double dval = FPC->getValue();
+      if (fabs(dval) == 1) {
+        unsigned opCode = 
+          (dval < 0) ? (resultType == Type::FloatTy? V9::FNEGS : V9::FNEGD)
+          : (resultType == Type::FloatTy? V9::FMOVS : V9::FMOVD);
               
-              minstr1 = new MachineInstr(opCode);
-              minstr1->SetMachineOperandVal(0,
-                                           MachineOperand::MO_VirtualRegister,
-                                           instrNode->leftChild()->getValue());
-            } 
-        }
+        mvec.push_back(BuildMI(opCode, 2).addReg(LHS).addRegDef(DestVal));
+      } 
     }
-  
-  if (minstr1 != NULL)
-    minstr1->SetMachineOperandVal(2, MachineOperand::MO_VirtualRegister,
-                                 instrNode->getValue());   
-  
-  if (minstr1)
-    mvec.push_back(minstr1);
-  if (minstr2)
-    mvec.push_back(minstr2);
+  }
 }
 
 
 static void
 CreateCodeForVariableSizeAlloca(const TargetMachine& target,
                                 Instruction* result,
-                                unsigned int tsize,
+                                unsigned tsize,
                                 Value* numElementsVal,
-                                vector<MachineInstr*>& getMvec)
+                                std::vector<MachineInstr*>& getMvec)
 {
+  Value* totalSizeVal;
   MachineInstr* M;
-  
-  // Create a Value to hold the (constant) element size
-  Value* tsizeVal = ConstantSInt::get(Type::IntTy, tsize);
+  MachineCodeForInstruction& mcfi = MachineCodeForInstruction::get(result);
+  Function *F = result->getParent()->getParent();
+
+  // Enforce the alignment constraints on the stack pointer at
+  // compile time if the total size is a known constant.
+  if (isa<Constant>(numElementsVal))
+    {
+      bool isValid;
+      int64_t numElem = GetConstantValueAsSignedInt(numElementsVal, isValid);
+      assert(isValid && "Unexpectedly large array dimension in alloca!");
+      int64_t total = numElem * tsize;
+      if (int extra= total % target.getFrameInfo().getStackFrameSizeAlignment())
+        total += target.getFrameInfo().getStackFrameSizeAlignment() - extra;
+      totalSizeVal = ConstantSInt::get(Type::IntTy, total);
+    }
+  else
+    {
+      // The size is not a constant.  Generate code to compute it and
+      // code to pad the size for stack alignment.
+      // Create a Value to hold the (constant) element size
+      Value* tsizeVal = ConstantSInt::get(Type::IntTy, tsize);
+
+      // Create temporary values to hold the result of MUL, SLL, SRL
+      // THIS CASE IS INCOMPLETE AND WILL BE FIXED SHORTLY.
+      TmpInstruction* tmpProd = new TmpInstruction(numElementsVal, tsizeVal);
+      TmpInstruction* tmpSLL  = new TmpInstruction(numElementsVal, tmpProd);
+      TmpInstruction* tmpSRL  = new TmpInstruction(numElementsVal, tmpSLL);
+      mcfi.addTemp(tmpProd);
+      mcfi.addTemp(tmpSLL);
+      mcfi.addTemp(tmpSRL);
+
+      // Instruction 1: mul numElements, typeSize -> tmpProd
+      // This will optimize the MUL as far as possible.
+      CreateMulInstruction(target, F, numElementsVal, tsizeVal, tmpProd,getMvec,
+                           mcfi, INVALID_MACHINE_OPCODE);
+
+      assert(0 && "Need to insert padding instructions here!");
+
+      totalSizeVal = tmpProd;
+    }
 
   // Get the constant offset from SP for dynamically allocated storage
   // and create a temporary Value to hold it.
-  assert(result && result->getParent() && "Result value is not part of a fn?");
-  Function *F = result->getParent()->getParent();
-  MachineCodeForMethod& mcInfo = MachineCodeForMethod::get(F);
+  MachineFunction& mcInfo = MachineFunction::get(F);
   bool growUp;
   ConstantSInt* dynamicAreaOffset =
     ConstantSInt::get(Type::IntTy,
-                      target.getFrameInfo().getDynamicAreaOffset(mcInfo,growUp));
+                     target.getFrameInfo().getDynamicAreaOffset(mcInfo,growUp));
   assert(! growUp && "Has SPARC v9 stack frame convention changed?");
 
-  // Create a temporary value to hold the result of MUL
-  TmpInstruction* tmpProd = new TmpInstruction(numElementsVal, tsizeVal);
-  MachineCodeForInstruction::get(result).addTemp(tmpProd);
-  
-  // Instruction 1: mul numElements, typeSize -> tmpProd
-  M = new MachineInstr(MULX);
-  M->SetMachineOperandVal(0, MachineOperand::MO_VirtualRegister, numElementsVal);
-  M->SetMachineOperandVal(1, MachineOperand::MO_VirtualRegister, tsizeVal);
-  M->SetMachineOperandVal(2, MachineOperand::MO_VirtualRegister, tmpProd);
-  getMvec.push_back(M);
-        
-  // Instruction 2: sub %sp, tmpProd -> %sp
-  M = new MachineInstr(SUB);
-  M->SetMachineOperandReg(0, target.getRegInfo().getStackPointer());
-  M->SetMachineOperandVal(1, MachineOperand::MO_VirtualRegister, tmpProd);
-  M->SetMachineOperandReg(2, target.getRegInfo().getStackPointer());
-  getMvec.push_back(M);
-  
+  unsigned SPReg = target.getRegInfo().getStackPointer();
+
+  // Instruction 2: sub %sp, totalSizeVal -> %sp
+  getMvec.push_back(BuildMI(V9::SUB, 3).addMReg(SPReg).addReg(totalSizeVal)
+                    .addMReg(SPReg,MOTy::Def));
+
   // Instruction 3: add %sp, frameSizeBelowDynamicArea -> result
-  M = new MachineInstr(ADD);
-  M->SetMachineOperandReg(0, target.getRegInfo().getStackPointer());
-  M->SetMachineOperandVal(1, MachineOperand::MO_VirtualRegister, dynamicAreaOffset);
-  M->SetMachineOperandVal(2, MachineOperand::MO_VirtualRegister, result);
-  getMvec.push_back(M);
+  getMvec.push_back(BuildMI(V9::ADD, 3).addMReg(SPReg).addReg(dynamicAreaOffset)
+                    .addRegDef(result));
 }        
 
 
 static void
 CreateCodeForFixedSizeAlloca(const TargetMachine& target,
                              Instruction* result,
-                             unsigned int tsize,
-                             unsigned int numElements,
-                             vector<MachineInstr*>& getMvec)
+                             unsigned tsize,
+                             unsigned numElements,
+                             std::vector<MachineInstr*>& getMvec)
 {
+  assert(tsize > 0 && "Illegal (zero) type size for alloca");
   assert(result && result->getParent() &&
          "Result value is not part of a function?");
   Function *F = result->getParent()->getParent();
-  MachineCodeForMethod &mcInfo = MachineCodeForMethod::get(F);
+  MachineFunction &mcInfo = MachineFunction::get(F);
 
   // Check if the offset would small enough to use as an immediate in
   // load/stores (check LDX because all load/stores have the same-size immediate
   // field).  If not, put the variable in the dynamically sized area of the
   // frame.
-  unsigned int paddedSizeIgnored;
-  int offsetFromFP = mcInfo.computeOffsetforLocalVar(target, result,
+  unsigned paddedSizeIgnored;
+  int offsetFromFP = mcInfo.getInfo()->computeOffsetforLocalVar(result,
                                                      paddedSizeIgnored,
                                                      tsize * numElements);
-  if (! target.getInstrInfo().constantFitsInImmedField(LDX, offsetFromFP))
-    {
-      CreateCodeForVariableSizeAlloca(target, result, tsize, 
-                                      ConstantSInt::get(Type::IntTy,numElements),
-                                      getMvec);
-      return;
-    }
+  if (! target.getInstrInfo().constantFitsInImmedField(V9::LDX, offsetFromFP)) {
+    CreateCodeForVariableSizeAlloca(target, result, tsize, 
+                                   ConstantSInt::get(Type::IntTy,numElements),
+                                   getMvec);
+    return;
+  }
   
   // else offset fits in immediate field so go ahead and allocate it.
-  offsetFromFP = mcInfo.allocateLocalVar(target, result, tsize * numElements);
+  offsetFromFP = mcInfo.getInfo()->allocateLocalVar(result, tsize *numElements);
   
   // Create a temporary Value to hold the constant offset.
   // This is needed because it may not fit in the immediate field.
   ConstantSInt* offsetVal = ConstantSInt::get(Type::IntTy, offsetFromFP);
   
   // Instruction 1: add %fp, offsetFromFP -> result
-  MachineInstr* M = new MachineInstr(ADD);
-  M->SetMachineOperandReg(0, target.getRegInfo().getFramePointer());
-  M->SetMachineOperandVal(1, MachineOperand::MO_VirtualRegister, offsetVal); 
-  M->SetMachineOperandVal(2, MachineOperand::MO_VirtualRegister, result);
-  
-  getMvec.push_back(M);
+  unsigned FPReg = target.getRegInfo().getFramePointer();
+  getMvec.push_back(BuildMI(V9::ADD, 3).addMReg(FPReg).addReg(offsetVal)
+                    .addRegDef(result));
 }
 
 
-
-
-
 //------------------------------------------------------------------------ 
 // Function SetOperandsForMemInstr
 //
@@ -925,89 +1119,28 @@ CreateCodeForFixedSizeAlloca(const TargetMachine& target,
 //------------------------------------------------------------------------ 
 
 static void
-SetOperandsForMemInstr(vector<MachineInstr*>& mvec,
-                       vector<MachineInstr*>::iterator mvecI,
-                       const InstructionNode* vmInstrNode,
+SetOperandsForMemInstr(unsigned Opcode,
+                       std::vector<MachineInstr*>& mvec,
+                       InstructionNode* vmInstrNode,
                        const TargetMachine& target)
 {
-  MemAccessInst* memInst = (MemAccessInst*) vmInstrNode->getInstruction();
-  
-  // Variables to hold the index vector and ptr value.
-  // The major work here is to extract these for all 3 instruction types
-  // and to try to fold chains of constant indices into a single offset.
-  // After that, we call SetMemOperands_Internal(), which creates the
-  // appropriate operands for the machine instruction.
-  vector<Value*> idxVec;
-  bool allConstantIndices = true;
-  Value* ptrVal = memInst->getPointerOperand();
-  
-  // If there is a GetElemPtr instruction to fold in to this instr,
-  // it must be in the left child for Load and GetElemPtr, and in the
-  // right child for Store instructions.
-  InstrTreeNode* ptrChild = (vmInstrNode->getOpLabel() == Instruction::Store
-                             ? vmInstrNode->rightChild()
-                             : vmInstrNode->leftChild()); 
-  
-  // Check if all indices are constant for this instruction
-  for (MemAccessInst::op_iterator OI=memInst->idx_begin();
-       OI != memInst->idx_end(); ++OI)
-    if (! isa<ConstantUInt>(*OI))
-      {
-        allConstantIndices = false; 
-        break;
-      }
-  
-  // If we have only constant indices, fold chains of constant indices
-  // in this and any preceding GetElemPtr instructions.
-  if (allConstantIndices &&
-      ptrChild->getOpLabel() == Instruction::GetElementPtr ||
-      ptrChild->getOpLabel() == GetElemPtrIdx)
-    {
-      Value* newPtr = FoldGetElemChain((InstructionNode*) ptrChild, idxVec);
-      if (newPtr)
-        ptrVal = newPtr;
-    }
-  
-  // Append the index vector of the current instruction, if any.
-  // Discard any leading [0] index.
-  if (memInst->idx_begin() != memInst->idx_end())
-    {
-      const ConstantUInt* CV = dyn_cast<ConstantUInt>(memInst->idx_begin()->get());
-      unsigned zeroOrIOne = (CV && CV->getType() == Type::UIntTy &&
-                             (CV->getValue() == 0))? 1 : 0;
-      idxVec.insert(idxVec.end(),
-                    memInst->idx_begin()+zeroOrIOne, memInst->idx_end());
-    }
-  
-  // Now create the appropriate operands for the machine instruction
-  SetMemOperands_Internal(mvec, mvecI, vmInstrNode,
-                          ptrVal, idxVec, allConstantIndices, target);
-}
-
-
-// Generate the correct operands (and additional instructions if needed)
-// for the given pointer and given index vector.
-//
-static void
-SetMemOperands_Internal(vector<MachineInstr*>& mvec,
-                        vector<MachineInstr*>::iterator mvecI,
-                        const InstructionNode* vmInstrNode,
-                        Value* ptrVal,
-                        vector<Value*>& idxVec,
-                        bool allConstantIndices,
-                        const TargetMachine& target)
-{
-  MemAccessInst* memInst = (MemAccessInst*) vmInstrNode->getInstruction();
-  
-  // Initialize so we default to storing the offset in a register.
+  Instruction* memInst = vmInstrNode->getInstruction();
+  // Index vector, ptr value, and flag if all indices are const.
+  std::vector<Value*> idxVec;
+  bool allConstantIndices;
+  Value* ptrVal = GetMemInstArgs(vmInstrNode, idxVec, allConstantIndices);
+
+  // Now create the appropriate operands for the machine instruction.
+  // First, initialize so we default to storing the offset in a register.
   int64_t smallConstOffset = 0;
   Value* valueForRegOffset = NULL;
-  MachineOperand::MachineOperandType offsetOpType =MachineOperand::MO_VirtualRegister;
+  MachineOperand::MachineOperandType offsetOpType =
+    MachineOperand::MO_VirtualRegister;
 
   // Check if there is an index vector and if so, compute the
   // right offset for structures and for arrays 
   // 
-  if (idxVec.size() > 0)
+  if (!idxVec.empty())
     {
       const PointerType* ptrType = cast<PointerType>(ptrVal->getType());
       
@@ -1016,63 +1149,49 @@ SetMemOperands_Internal(vector<MachineInstr*>& mvec,
         {
           // Compute the offset value using the index vector. Create a
           // virtual reg. for it since it may not fit in the immed field.
-          uint64_t offset = target.DataLayout.getIndexedOffset(ptrType,idxVec);
+          uint64_t offset = target.getTargetData().getIndexedOffset(ptrType,idxVec);
           valueForRegOffset = ConstantSInt::get(Type::LongTy, offset);
         }
       else
         {
           // There is at least one non-constant offset.  Therefore, this must
-          // be an array ref, and must have been lowered to a single offset.
-          assert((memInst->getNumOperands()
-                  == (unsigned) 1 + memInst->getFirstIndexOperandNumber())
+          // be an array ref, and must have been lowered to a single non-zero
+          // offset.  (An extra leading zero offset, if any, can be ignored.)
+          // Generate code sequence to compute address from index.
+          // 
+          bool firstIdxIsZero = IsZero(idxVec[0]);
+          assert(idxVec.size() == 1U + firstIdxIsZero 
                  && "Array refs must be lowered before Instruction Selection");
-          
-          Value* arrayOffsetVal =  * memInst->idx_begin();
-          
-          // Handle special common case of leading [0] index.
-          ConstantUInt* CV = dyn_cast<ConstantUInt>(idxVec.front());
-          bool firstIndexIsZero = bool(CV && CV->getType() == Type::UIntTy &&
-                                       (CV->getValue() == 0));
-      
-          // If index is 0, the offset value is just 0.  Otherwise, 
-          // generate a MUL instruction to compute address from index.
+
+          Value* idxVal = idxVec[firstIdxIsZero];
+
+          std::vector<MachineInstr*> mulVec;
+          Instruction* addr = new TmpInstruction(Type::ULongTy, memInst);
+          MachineCodeForInstruction::get(memInst).addTemp(addr);
+
+          // Get the array type indexed by idxVal, and compute its element size.
           // The call to getTypeSize() will fail if size is not constant.
+          const Type* vecType = (firstIdxIsZero
+                                 ? GetElementPtrInst::getIndexedType(ptrType,
+                                           std::vector<Value*>(1U, idxVec[0]),
+                                           /*AllowCompositeLeaf*/ true)
+                                 : ptrType);
+          const Type* eltType = cast<SequentialType>(vecType)->getElementType();
+          ConstantUInt* eltSizeVal = ConstantUInt::get(Type::ULongTy,
+                                       target.getTargetData().getTypeSize(eltType));
+
           // CreateMulInstruction() folds constants intelligently enough.
-          // 
-          if (firstIndexIsZero)
-            {
-              offsetOpType = MachineOperand::MO_SignExtendedImmed;
-              smallConstOffset = 0;
-            }
-          else
-            {
-              vector<MachineInstr*> mulVec;
-              Instruction* addr = new TmpInstruction(Type::UIntTy, memInst);
-              MachineCodeForInstruction::get(memInst).addTemp(addr);
-              
-              unsigned int eltSize =
-                target.DataLayout.getTypeSize(ptrType->getElementType());
-              assert(eltSize > 0 && "Invalid or non-const array element size");
-              ConstantUInt* eltVal = ConstantUInt::get(Type::UIntTy, eltSize);
-              
-              CreateMulInstruction(target,
-                                   memInst->getParent()->getParent(),
-                                   arrayOffsetVal, /* lval, not likely const */
-                                   eltVal,         /* rval, likely constant */
-                                   addr,           /* result*/
-                                   mulVec,
-                                   MachineCodeForInstruction::get(memInst),
-                                   INVALID_MACHINE_OPCODE);
-              assert(mulVec.size() > 0 && "No multiply instruction created?");
-              for (vector<MachineInstr*>::const_iterator I = mulVec.begin();
-                   I != mulVec.end(); ++I)
-                {
-                  mvecI = mvec.insert(mvecI, *I);   // ptr to inserted value
-                  ++mvecI;                          // ptr to mem. instr.
-                }
-              
-              valueForRegOffset = addr;
-            }
+          CreateMulInstruction(target, memInst->getParent()->getParent(),
+                               idxVal,         /* lval, not likely to be const*/
+                               eltSizeVal,     /* rval, likely to be constant */
+                               addr,           /* result */
+                               mulVec, MachineCodeForInstruction::get(memInst),
+                               INVALID_MACHINE_OPCODE);
+
+          assert(mulVec.size() > 0 && "No multiply code created?");
+          mvec.insert(mvec.end(), mulVec.begin(), mulVec.end());
+
+          valueForRegOffset = addr;
         }
     }
   else
@@ -1080,40 +1199,30 @@ SetMemOperands_Internal(vector<MachineInstr*>& mvec,
       offsetOpType = MachineOperand::MO_SignExtendedImmed;
       smallConstOffset = 0;
     }
-  
+
   // For STORE:
   //   Operand 0 is value, operand 1 is ptr, operand 2 is offset
   // For LOAD or GET_ELEMENT_PTR,
   //   Operand 0 is ptr, operand 1 is offset, operand 2 is result.
   // 
   unsigned offsetOpNum, ptrOpNum;
-  if (memInst->getOpcode() == Instruction::Store)
-    {
-      (*mvecI)->SetMachineOperandVal(0, MachineOperand::MO_VirtualRegister,
-                                     vmInstrNode->leftChild()->getValue());
-      ptrOpNum = 1;
-      offsetOpNum = 2;
-    }
-  else
-    {
-      ptrOpNum = 0;
-      offsetOpNum = 1;
-      (*mvecI)->SetMachineOperandVal(2, MachineOperand::MO_VirtualRegister,
-                                     memInst);
-    }
-  
-  (*mvecI)->SetMachineOperandVal(ptrOpNum, MachineOperand::MO_VirtualRegister,
-                                 ptrVal);
-  
-  if (offsetOpType == MachineOperand::MO_VirtualRegister)
-    {
-      assert(valueForRegOffset != NULL);
-      (*mvecI)->SetMachineOperandVal(offsetOpNum, offsetOpType,
-                                     valueForRegOffset); 
-    }
-  else
-    (*mvecI)->SetMachineOperandConst(offsetOpNum, offsetOpType,
-                                     smallConstOffset);
+  MachineInstr *MI;
+  if (memInst->getOpcode() == Instruction::Store) {
+    if (offsetOpType == MachineOperand::MO_VirtualRegister)
+      MI = BuildMI(Opcode, 3).addReg(vmInstrNode->leftChild()->getValue())
+                             .addReg(ptrVal).addReg(valueForRegOffset);
+    else
+      MI = BuildMI(Opcode, 3).addReg(vmInstrNode->leftChild()->getValue())
+                             .addReg(ptrVal).addSImm(smallConstOffset);
+  } else {
+    if (offsetOpType == MachineOperand::MO_VirtualRegister)
+      MI = BuildMI(Opcode, 3).addReg(ptrVal).addReg(valueForRegOffset)
+                             .addRegDef(memInst);
+    else
+      MI = BuildMI(Opcode, 3).addReg(ptrVal).addSImm(smallConstOffset)
+                             .addRegDef(memInst);
+  }
+  mvec.push_back(MI);
 }
 
 
@@ -1163,7 +1272,7 @@ ForwardOperand(InstructionNode* treeNode,
           for (unsigned i=0, numOps=minstr->getNumOperands(); i < numOps; ++i)
             {
               const MachineOperand& mop = minstr->getOperand(i);
-              if (mop.getOperandType() == MachineOperand::MO_VirtualRegister &&
+              if (mop.getType() == MachineOperand::MO_VirtualRegister &&
                   mop.getVRegValue() == unusedOp)
                 minstr->SetMachineOperandVal(i,
                                 MachineOperand::MO_VirtualRegister, fwdOp);
@@ -1172,7 +1281,8 @@ ForwardOperand(InstructionNode* treeNode,
           for (unsigned i=0,numOps=minstr->getNumImplicitRefs(); i<numOps; ++i)
             if (minstr->getImplicitRef(i) == unusedOp)
               minstr->setImplicitRef(i, fwdOp,
-                                     minstr->implicitRefIsDefined(i));
+                                     minstr->implicitRefIsDefined(i),
+                                     minstr->implicitRefIsDefinedAndUsed(i));
         }
     }
 }
@@ -1189,6 +1299,46 @@ AllUsesAreBranches(const Instruction* setccI)
   return true;
 }
 
+// Generate code for any intrinsic that needs a special code sequence
+// instead of a regular call.  If not that kind of intrinsic, do nothing.
+// Returns true if code was generated, otherwise false.
+// 
+bool CodeGenIntrinsic(LLVMIntrinsic::ID iid, CallInst &callInstr,
+                      TargetMachine &target,
+                      std::vector<MachineInstr*>& mvec)
+{
+  switch (iid) {
+  case LLVMIntrinsic::va_start: {
+    // Get the address of the first vararg value on stack and copy it to
+    // the argument of va_start(va_list* ap).
+    bool ignore;
+    Function* func = cast<Function>(callInstr.getParent()->getParent());
+    int numFixedArgs   = func->getFunctionType()->getNumParams();
+    int fpReg          = target.getFrameInfo().getIncomingArgBaseRegNum();
+    int argSize        = target.getFrameInfo().getSizeOfEachArgOnStack();
+    int firstVarArgOff = numFixedArgs * argSize + target.getFrameInfo().
+      getFirstIncomingArgOffset(MachineFunction::get(func), ignore);
+    mvec.push_back(BuildMI(V9::ADD, 3).addMReg(fpReg).addSImm(firstVarArgOff).
+                   addReg(callInstr.getOperand(1)));
+    return true;
+  }
+
+  case LLVMIntrinsic::va_end:
+    return true;                        // no-op on Sparc
+
+  case LLVMIntrinsic::va_copy:
+    // Simple copy of current va_list (arg2) to new va_list (arg1)
+    mvec.push_back(BuildMI(V9::OR, 3).
+                   addMReg(target.getRegInfo().getZeroRegNum()).
+                   addReg(callInstr.getOperand(2)).
+                   addReg(callInstr.getOperand(1)));
+    return true;
+
+  default:
+    return false;
+  }
+}
+
 //******************* Externally Visible Functions *************************/
 
 //------------------------------------------------------------------------ 
@@ -1204,7 +1354,6 @@ ThisIsAChainRule(int eruleno)
   switch(eruleno)
     {
     case 111:  // stmt:  reg
-    case 113:  // stmt:  bool
     case 123:
     case 124:
     case 125:
@@ -1223,9 +1372,10 @@ ThisIsAChainRule(int eruleno)
     case 242:
     case 243:
     case 244:
+    case 245:
     case 321:
       return true; break;
-      
+
     default:
       return false; break;
     }
@@ -1245,14 +1395,15 @@ GetInstructionsByRule(InstructionNode* subtreeRoot,
                       int ruleForNode,
                       short* nts,
                       TargetMachine &target,
-                      vector<MachineInstr*>& mvec)
+                      std::vector<MachineInstr*>& mvec)
 {
   bool checkCast = false;              // initialize here to use fall-through
+  bool maskUnsignedResult = false;
   int nextRule;
   int forwardOperandNum = -1;
-  unsigned int allocaSize = 0;
+  unsigned allocaSize = 0;
   MachineInstr* M, *M2;
-  unsigned int L;
+  unsigned L;
 
   mvec.clear(); 
   
@@ -1294,41 +1445,34 @@ GetInstructionsByRule(InstructionNode* subtreeRoot,
         Instruction* returnReg = new TmpInstruction(returnInstr);
         MachineCodeForInstruction::get(returnInstr).addTemp(returnReg);
         
-        M = new MachineInstr(JMPLRET);
-        M->SetMachineOperandReg(0, MachineOperand::MO_VirtualRegister,
-                                      returnReg);
-        M->SetMachineOperandConst(1,MachineOperand::MO_SignExtendedImmed,
-                                   (int64_t)8);
-        M->SetMachineOperandReg(2, target.getRegInfo().getZeroRegNum());
+        M = BuildMI(V9::JMPLRET, 3).addReg(returnReg).addSImm(8)
+          .addMReg(target.getRegInfo().getZeroRegNum(), MOTy::Def);
         
         if (returnInstr->getReturnValue() != NULL)
           M->addImplicitRef(returnInstr->getReturnValue());
         
         mvec.push_back(M);
-        mvec.push_back(new MachineInstr(NOP));
+        mvec.push_back(BuildMI(V9::NOP, 0));
         
         break;
       }  
         
       case 3:  // stmt:   Store(reg,reg)
       case 4:  // stmt:   Store(reg,ptrreg)
-        mvec.push_back(new MachineInstr(
-                         ChooseStoreInstruction(
-                            subtreeRoot->leftChild()->getValue()->getType())));
-        SetOperandsForMemInstr(mvec, mvec.end()-1, subtreeRoot, target);
+        SetOperandsForMemInstr(ChooseStoreInstruction(
+                        subtreeRoot->leftChild()->getValue()->getType()),
+                               mvec, subtreeRoot, target);
         break;
 
       case 5:  // stmt:   BrUncond
-        M = new MachineInstr(BA);
-        M->SetMachineOperandVal(0, MachineOperand::MO_CCRegister,
-                                      (Value*)NULL);
-        M->SetMachineOperandVal(1, MachineOperand::MO_PCRelativeDisp,
-             cast<BranchInst>(subtreeRoot->getInstruction())->getSuccessor(0));
-        mvec.push_back(M);
+        {
+          BranchInst *BI = cast<BranchInst>(subtreeRoot->getInstruction());
+          mvec.push_back(BuildMI(V9::BA, 1).addPCDisp(BI->getSuccessor(0)));
         
-        // delay slot
-        mvec.push_back(new MachineInstr(NOP));
-        break;
+          // delay slot
+          mvec.push_back(BuildMI(V9::NOP, 0));
+          break;
+        }
 
       case 206:        // stmt:   BrCond(setCCconst)
       { // setCCconst => boolean was computed with `%b = setCC type reg1 const'
@@ -1342,7 +1486,7 @@ GetInstructionsByRule(InstructionNode* subtreeRoot,
         Constant *constVal = cast<Constant>(constNode->getValue());
         bool isValidConst;
         
-        if ((constVal->getType()->isIntegral()
+        if ((constVal->getType()->isInteger()
              || isa<PointerType>(constVal->getType()))
             && GetConstantValueAsSignedInt(constVal, isValidConst) == 0
             && isValidConst)
@@ -1357,66 +1501,49 @@ GetInstructionsByRule(InstructionNode* subtreeRoot,
             
             BranchInst* brInst=cast<BranchInst>(subtreeRoot->getInstruction());
             
-            M = new MachineInstr(ChooseBprInstruction(subtreeRoot));
-            M->SetMachineOperandVal(0, MachineOperand::MO_VirtualRegister,
-                                    setCCNode->leftChild()->getValue());
-            M->SetMachineOperandVal(1, MachineOperand::MO_PCRelativeDisp,
-                                    brInst->getSuccessor(0));
+            M = BuildMI(ChooseBprInstruction(subtreeRoot), 2)
+                                .addReg(setCCNode->leftChild()->getValue())
+                                .addPCDisp(brInst->getSuccessor(0));
             mvec.push_back(M);
             
             // delay slot
-            mvec.push_back(new MachineInstr(NOP));
+            mvec.push_back(BuildMI(V9::NOP, 0));
 
             // false branch
-            M = new MachineInstr(BA);
-            M->SetMachineOperandVal(0, MachineOperand::MO_CCRegister,
-                                    (Value*) NULL);
-            M->SetMachineOperandVal(1, MachineOperand::MO_PCRelativeDisp,
-                                    brInst->getSuccessor(1));
-            mvec.push_back(M);
+            mvec.push_back(BuildMI(V9::BA, 1)
+                           .addPCDisp(brInst->getSuccessor(1)));
             
             // delay slot
-            mvec.push_back(new MachineInstr(NOP));
-            
+            mvec.push_back(BuildMI(V9::NOP, 0));
             break;
           }
         // ELSE FALL THROUGH
       }
 
-      case 6:  // stmt:   BrCond(bool)
-      { // bool => boolean was computed with some boolean operator
-        // (SetCC, Not, ...).  We need to check whether the type was a FP,
-        // signed int or unsigned int, and check the branching condition in
-        // order to choose the branch to use.
+      case 6:  // stmt:   BrCond(setCC)
+      { // bool => boolean was computed with SetCC.
+        // The branch to use depends on whether it is FP, signed, or unsigned.
         // If it is an integer CC, we also need to find the unique
         // TmpInstruction representing that CC.
         // 
         BranchInst* brInst = cast<BranchInst>(subtreeRoot->getInstruction());
         bool isFPBranch;
-        M = new MachineInstr(ChooseBccInstruction(subtreeRoot, isFPBranch));
-        
+        unsigned Opcode = ChooseBccInstruction(subtreeRoot, isFPBranch);
         Value* ccValue = GetTmpForCC(subtreeRoot->leftChild()->getValue(),
                                      brInst->getParent()->getParent(),
                                      isFPBranch? Type::FloatTy : Type::IntTy);
-        
-        M->SetMachineOperandVal(0, MachineOperand::MO_CCRegister, ccValue);
-        M->SetMachineOperandVal(1, MachineOperand::MO_PCRelativeDisp,
-                                   brInst->getSuccessor(0));
+        M = BuildMI(Opcode, 2).addCCReg(ccValue)
+                              .addPCDisp(brInst->getSuccessor(0));
         mvec.push_back(M);
-        
+
         // delay slot
-        mvec.push_back(new MachineInstr(NOP));
-        
+        mvec.push_back(BuildMI(V9::NOP, 0));
+
         // false branch
-        M = new MachineInstr(BA);
-        M->SetMachineOperandVal(0, MachineOperand::MO_CCRegister,
-                                   (Value*) NULL);
-        M->SetMachineOperandVal(1, MachineOperand::MO_PCRelativeDisp,
-                                   brInst->getSuccessor(1));
-        mvec.push_back(M);
-        
+        mvec.push_back(BuildMI(V9::BA, 1).addPCDisp(brInst->getSuccessor(1)));
+
         // delay slot
-        mvec.push_back(new MachineInstr(NOP));
+        mvec.push_back(BuildMI(V9::NOP, 0));
         break;
       }
         
@@ -1427,15 +1554,12 @@ GetInstructionsByRule(InstructionNode* subtreeRoot,
           cast<Constant>(subtreeRoot->leftChild()->getValue());
         unsigned dest = cast<ConstantBool>(constVal)->getValue()? 0 : 1;
         
-        M = new MachineInstr(BA);
-        M->SetMachineOperandVal(0, MachineOperand::MO_CCRegister,
-                                (Value*) NULL);
-        M->SetMachineOperandVal(1, MachineOperand::MO_PCRelativeDisp,
+        M = BuildMI(V9::BA, 1).addPCDisp(
           cast<BranchInst>(subtreeRoot->getInstruction())->getSuccessor(dest));
         mvec.push_back(M);
         
         // delay slot
-        mvec.push_back(new MachineInstr(NOP));
+        mvec.push_back(BuildMI(V9::NOP, 0));
         break;
       }
         
@@ -1443,26 +1567,19 @@ GetInstructionsByRule(InstructionNode* subtreeRoot,
       { // boolreg   => boolean is stored in an existing register.
         // Just use the branch-on-integer-register instruction!
         // 
-        M = new MachineInstr(BRNZ);
-        M->SetMachineOperandVal(0, MachineOperand::MO_VirtualRegister,
-                                      subtreeRoot->leftChild()->getValue());
-        M->SetMachineOperandVal(1, MachineOperand::MO_PCRelativeDisp,
-              cast<BranchInst>(subtreeRoot->getInstruction())->getSuccessor(0));
+        BranchInst *BI = cast<BranchInst>(subtreeRoot->getInstruction());
+        M = BuildMI(V9::BRNZ, 2).addReg(subtreeRoot->leftChild()->getValue())
+          .addPCDisp(BI->getSuccessor(0));
         mvec.push_back(M);
 
         // delay slot
-        mvec.push_back(new MachineInstr(NOP));
+        mvec.push_back(BuildMI(V9::NOP, 0));
 
         // false branch
-        M = new MachineInstr(BA);
-        M->SetMachineOperandVal(0, MachineOperand::MO_CCRegister,
-                                (Value*) NULL);
-        M->SetMachineOperandVal(1, MachineOperand::MO_PCRelativeDisp,
-              cast<BranchInst>(subtreeRoot->getInstruction())->getSuccessor(1));
-        mvec.push_back(M);
+        mvec.push_back(BuildMI(V9::BA, 1).addPCDisp(BI->getSuccessor(1)));
         
         // delay slot
-        mvec.push_back(new MachineInstr(NOP));
+        mvec.push_back(BuildMI(V9::NOP, 0));
         break;
       }  
       
@@ -1474,175 +1591,145 @@ GetInstructionsByRule(InstructionNode* subtreeRoot,
         assert(0 && "VRegList should never be the topmost non-chain rule");
         break;
 
-      case 21: // bool:  Not(bool):    Both these are implemented as:
-      case 421:        // reg:   BNot(reg) :        reg = reg XOR-NOT 0
-        M = new MachineInstr(XNOR);
-        M->SetMachineOperandVal(0, MachineOperand::MO_VirtualRegister,
-                                subtreeRoot->leftChild()->getValue());
-        M->SetMachineOperandReg(1, target.getRegInfo().getZeroRegNum());
-        M->SetMachineOperandVal(2, MachineOperand::MO_VirtualRegister,
-                                subtreeRoot->getValue());
-        mvec.push_back(M);
+      case 21: // bool:  Not(bool,reg): Both these are implemented as:
+      case 421:        // reg:   BNot(reg,reg):        reg = reg XOR-NOT 0
+      { // First find the unary operand. It may be left or right, usually right.
+        Value* notArg = BinaryOperator::getNotArgument(
+                           cast<BinaryOperator>(subtreeRoot->getInstruction()));
+        unsigned ZeroReg = target.getRegInfo().getZeroRegNum();
+        mvec.push_back(BuildMI(V9::XNOR, 3).addReg(notArg).addMReg(ZeroReg)
+                                       .addRegDef(subtreeRoot->getValue()));
         break;
+      }
 
-      case 322:        // reg:   ToBoolTy(bool):
       case 22: // reg:   ToBoolTy(reg):
       {
         const Type* opType = subtreeRoot->leftChild()->getValue()->getType();
-        assert(opType->isIntegral() || isa<PointerType>(opType)
-               || opType == Type::BoolTy);
+        assert(opType->isIntegral() || isa<PointerType>(opType));
         forwardOperandNum = 0;          // forward first operand to user
         break;
       }
       
       case 23: // reg:   ToUByteTy(reg)
+      case 24: // reg:   ToSByteTy(reg)
       case 25: // reg:   ToUShortTy(reg)
+      case 26: // reg:   ToShortTy(reg)
       case 27: // reg:   ToUIntTy(reg)
-      case 29: // reg:   ToULongTy(reg)
+      case 28: // reg:   ToIntTy(reg)
       {
+        //======================================================================
+        // Rules for integer conversions:
+        // 
+        //--------
+        // From ISO 1998 C++ Standard, Sec. 4.7:
+        //
+        // 2. If the destination type is unsigned, the resulting value is
+        // the least unsigned integer congruent to the source integer
+        // (modulo 2n where n is the number of bits used to represent the
+        // unsigned type). [Note: In a two s complement representation,
+        // this conversion is conceptual and there is no change in the
+        // bit pattern (if there is no truncation). ]
+        // 
+        // 3. If the destination type is signed, the value is unchanged if
+        // it can be represented in the destination type (and bitfield width);
+        // otherwise, the value is implementation-defined.
+        //--------
+        // 
+        // Since we assume 2s complement representations, this implies:
+        // 
+        // -- if operand is smaller than destination, zero-extend or sign-extend
+        //    according to the signedness of the *operand*: source decides.
+        //    ==> we have to do nothing here!
+        // 
+        // -- if operand is same size as or larger than destination, and the
+        //    destination is *unsigned*, zero-extend the operand: dest. decides
+        // 
+        // -- if operand is same size as or larger than destination, and the
+        //    destination is *signed*, the choice is implementation defined:
+        //    we sign-extend the operand: i.e., again dest. decides.
+        //    Note: this matches both Sun's cc and gcc3.2.
+        //======================================================================
+
         Instruction* destI =  subtreeRoot->getInstruction();
         Value* opVal = subtreeRoot->leftChild()->getValue();
-        const Type* opType = subtreeRoot->leftChild()->getValue()->getType();
-        assert(opType->isIntegral() ||
-               isa<PointerType>(opType) ||
-               opType == Type::BoolTy && "Cast is illegal for other types");
-        
-        unsigned opSize = target.DataLayout.getTypeSize(opType);
-        unsigned destSize = target.DataLayout.getTypeSize(destI->getType());
-        
-        if (opSize > destSize ||
-            (opType->isSigned()
-             && destSize < target.DataLayout.getIntegerRegize()))
-          { // operand is larger than dest,
-            //    OR both are equal but smaller than the full register size
-            //       AND operand is signed, so it may have extra sign bits:
-            // mask high bits using AND
-            // 
-            M = Create3OperandInstr(AND, opVal,
-                                    ConstantUInt::get(Type::ULongTy,
-                                          ((uint64_t) 1 << 8*destSize) - 1),
-                                    destI);
-            mvec.push_back(M);
+        const Type* opType = opVal->getType();
+        if (opType->isIntegral() || isa<PointerType>(opType))
+          {
+            unsigned opSize = target.getTargetData().getTypeSize(opType);
+            unsigned destSize = target.getTargetData().getTypeSize(destI->getType());
+            if (opSize >= destSize)
+              { // Operand is same size as or larger than dest:
+                // zero- or sign-extend, according to the signeddness of
+                // the destination (see above).
+                if (destI->getType()->isSigned())
+                  target.getInstrInfo().CreateSignExtensionInstructions(target,
+                    destI->getParent()->getParent(), opVal, destI, 8*destSize,
+                    mvec, MachineCodeForInstruction::get(destI));
+                else
+                  target.getInstrInfo().CreateZeroExtensionInstructions(target,
+                    destI->getParent()->getParent(), opVal, destI, 8*destSize,
+                    mvec, MachineCodeForInstruction::get(destI));
+              }
+            else
+              forwardOperandNum = 0;          // forward first operand to user
+          }
+        else if (opType->isFloatingPoint())
+          {
+            CreateCodeToConvertFloatToInt(target, opVal, destI, mvec,
+                                         MachineCodeForInstruction::get(destI));
+            if (destI->getType()->isUnsigned())
+              maskUnsignedResult = true; // not handled by fp->int code
           }
         else
-          forwardOperandNum = 0;          // forward first operand to user
-        
+          assert(0 && "Unrecognized operand type for convert-to-unsigned");
+
         break;
       }
-      
-      case 24: // reg:   ToSByteTy(reg)
-      case 26: // reg:   ToShortTy(reg)
-      case 28: // reg:   ToIntTy(reg)
+
+      case 29: // reg:   ToULongTy(reg)
       case 30: // reg:   ToLongTy(reg)
       {
-        unsigned int oldMvecSize = mvec.size(); // to check if it grew
-        Instruction* destI =  subtreeRoot->getInstruction();
         Value* opVal = subtreeRoot->leftChild()->getValue();
-        MachineCodeForInstruction& mcfi =MachineCodeForInstruction::get(destI);
-        
         const Type* opType = opVal->getType();
-        if (opType->isIntegral()
-            || isa<PointerType>(opType)
-            || opType == Type::BoolTy)
+        if (opType->isIntegral() || isa<PointerType>(opType))
+          forwardOperandNum = 0;          // forward first operand to user
+        else if (opType->isFloatingPoint())
           {
-            // These operand types have the same format as the destination,
-            // but may have different size: add sign bits or mask as needed.
-            // 
-            const Type* destType = destI->getType();
-            unsigned opSize = target.DataLayout.getTypeSize(opType);
-            unsigned destSize = target.DataLayout.getTypeSize(destType);
-            if (opSize <= destSize && !opType->isSigned())
-              { // operand is smaller than or same size as dest:
-                // -- if operand is signed (checked above), nothing to do
-                // -- if operand is unsigned, sign-extend the value:
-                // 
-                target.getInstrInfo().CreateSignExtensionInstructions(target, destI->getParent()->getParent(), opVal, 8*opSize, destI, mvec, mcfi);
-              }
-            else if (opSize > destSize)
-              { // operand is larger than dest: mask high bits using AND
-                // and then sign-extend using SRA by 0!
-                // 
-                TmpInstruction *tmpI = new TmpInstruction(destType, opVal,
-                                                          destI, "maskHi");
-                mcfi.addTemp(tmpI);
-                M = Create3OperandInstr(AND, opVal,
-                                        ConstantUInt::get(Type::UIntTy,
-                                              ((uint64_t) 1 << 8*destSize)-1),
-                                        tmpI);
-                mvec.push_back(M);
-                
-                target.getInstrInfo().CreateSignExtensionInstructions(target, destI->getParent()->getParent(), tmpI, 8*destSize, destI, mvec, mcfi);
-              }
+            Instruction* destI =  subtreeRoot->getInstruction();
+            CreateCodeToConvertFloatToInt(target, opVal, destI, mvec,
+                                         MachineCodeForInstruction::get(destI));
           }
         else
-          {
-            // If the source operand is an FP type, the int result must be
-            // copied from float to int register via memory!
-            Value* leftVal = subtreeRoot->leftChild()->getValue();
-            Value* destForCast;
-            vector<MachineInstr*> minstrVec;
-            
-            if (opType->isFloatingPoint())
-              {
-                // Create a temporary to represent the INT register
-                // into which the FP value will be copied via memory.
-                // The type of this temporary will determine the FP
-                // register used: single-prec for a 32-bit int or smaller,
-                // double-prec for a 64-bit int.
-                // 
-                const Type* destTypeToUse =
-                  (destI->getType() == Type::LongTy)? Type::DoubleTy
-                                                   : Type::FloatTy;
-                destForCast = new TmpInstruction(destTypeToUse, leftVal);
-                MachineCodeForInstruction &destMCFI = 
-                  MachineCodeForInstruction::get(destI);
-                destMCFI.addTemp(destForCast);
-                
-                target.getInstrInfo().
-                  CreateCodeToCopyFloatToInt(target,
-                                             destI->getParent()->getParent(),
-                                             (TmpInstruction*) destForCast,
-                                             destI, minstrVec, destMCFI);
-              }
-            else
-              destForCast = leftVal;
-            
-            M = CreateConvertToIntInstr(subtreeRoot->getOpLabel(),
-                                        leftVal, destForCast);
-            mvec.push_back(M);
-            
-            // Append the copy code, if any, after the conversion instr.
-            mvec.insert(mvec.end(), minstrVec.begin(), minstrVec.end());
-          }
-        
-        if (oldMvecSize == mvec.size()) // no instruction was generated
-          forwardOperandNum = 0;  // forward first operand to user
-        
+          assert(0 && "Unrecognized operand type for convert-to-signed");
         break;
-      }  
+      }
       
       case  31:        // reg:   ToFloatTy(reg):
       case  32:        // reg:   ToDoubleTy(reg):
       case 232:        // reg:   ToDoubleTy(Constant):
-        
+      
         // If this instruction has a parent (a user) in the tree 
         // and the user is translated as an FsMULd instruction,
         // then the cast is unnecessary.  So check that first.
         // In the future, we'll want to do the same for the FdMULq instruction,
         // so do the check here instead of only for ToFloatTy(reg).
         // 
-        if (subtreeRoot->parent() != NULL &&
-            MachineCodeForInstruction::get(((InstructionNode*)subtreeRoot->parent())->getInstruction())[0]->getOpCode() == FSMULD)
+        if (subtreeRoot->parent() != NULL)
           {
-            forwardOperandNum = 0;          // forward first operand to user
+            const MachineCodeForInstruction& mcfi =
+              MachineCodeForInstruction::get(
+                cast<InstructionNode>(subtreeRoot->parent())->getInstruction());
+            if (mcfi.size() == 0 || mcfi.front()->getOpCode() == V9::FSMULD)
+              forwardOperandNum = 0;    // forward first operand to user
           }
-        else
+
+        if (forwardOperandNum != 0)     // we do need the cast
           {
             Value* leftVal = subtreeRoot->leftChild()->getValue();
             const Type* opType = leftVal->getType();
             MachineOpCode opCode=ChooseConvertToFloatInstr(
                                        subtreeRoot->getOpLabel(), opType);
-            if (opCode == INVALID_OPCODE)      // no conversion needed
+            if (opCode == V9::INVALID_OPCODE)  // no conversion needed
               {
                 forwardOperandNum = 0;      // forward first operand to user
               }
@@ -1661,28 +1748,24 @@ GetInstructionsByRule(InstructionNode* subtreeRoot,
                     // register used: single-prec for a 32-bit int or smaller,
                     // double-prec for a 64-bit int.
                     // 
-                    const Type* srcTypeToUse =
-                      (leftVal->getType() == Type::LongTy)? Type::DoubleTy
-                                                          : Type::FloatTy;
-                    
-                    srcForCast = new TmpInstruction(srcTypeToUse, dest);
+                    uint64_t srcSize =
+                      target.getTargetData().getTypeSize(leftVal->getType());
+                    Type* tmpTypeToUse =
+                      (srcSize <= 4)? Type::FloatTy : Type::DoubleTy;
+                    srcForCast = new TmpInstruction(tmpTypeToUse, dest);
                     MachineCodeForInstruction &destMCFI = 
                       MachineCodeForInstruction::get(dest);
                     destMCFI.addTemp(srcForCast);
-                    
+
                     target.getInstrInfo().CreateCodeToCopyIntToFloat(target,
                          dest->getParent()->getParent(),
-                         leftVal, (TmpInstruction*) srcForCast,
+                         leftVal, cast<Instruction>(srcForCast),
                          mvec, destMCFI);
                   }
                 else
                   srcForCast = leftVal;
-                
-                M = new MachineInstr(opCode);
-                M->SetMachineOperandVal(0, MachineOperand::MO_VirtualRegister,
-                                           srcForCast);
-                M->SetMachineOperandVal(1, MachineOperand::MO_VirtualRegister,
-                                           dest);
+
+                M = BuildMI(opCode, 2).addReg(srcForCast).addRegDef(dest);
                 mvec.push_back(M);
               }
           }
@@ -1694,6 +1777,7 @@ GetInstructionsByRule(InstructionNode* subtreeRoot,
         break;
 
       case 233:        // reg:   Add(reg, Constant)
+        maskUnsignedResult = true;
         M = CreateAddConstInstruction(subtreeRoot);
         if (M != NULL)
           {
@@ -1703,11 +1787,12 @@ GetInstructionsByRule(InstructionNode* subtreeRoot,
         // ELSE FALL THROUGH
         
       case 33: // reg:   Add(reg, reg)
-        mvec.push_back(new MachineInstr(ChooseAddInstruction(subtreeRoot)));
-        Set3OperandsFromInstr(mvec.back(), subtreeRoot, target);
+        maskUnsignedResult = true;
+        Add3OperandInstr(ChooseAddInstruction(subtreeRoot), subtreeRoot, mvec);
         break;
 
       case 234:        // reg:   Sub(reg, Constant)
+        maskUnsignedResult = true;
         M = CreateSubConstInstruction(subtreeRoot);
         if (M != NULL)
           {
@@ -1717,9 +1802,10 @@ GetInstructionsByRule(InstructionNode* subtreeRoot,
         // ELSE FALL THROUGH
         
       case 34: // reg:   Sub(reg, reg)
-        mvec.push_back(new MachineInstr(ChooseSubInstructionByType(
-                                   subtreeRoot->getInstruction()->getType())));
-        Set3OperandsFromInstr(mvec.back(), subtreeRoot, target);
+        maskUnsignedResult = true;
+        Add3OperandInstr(ChooseSubInstructionByType(
+                                   subtreeRoot->getInstruction()->getType()),
+                         subtreeRoot, mvec);
         break;
 
       case 135:        // reg:   Mul(todouble, todouble)
@@ -1728,8 +1814,9 @@ GetInstructionsByRule(InstructionNode* subtreeRoot,
 
       case 35: // reg:   Mul(reg, reg)
       {
+        maskUnsignedResult = true;
         MachineOpCode forceOp = ((checkCast && BothFloatToDouble(subtreeRoot))
-                                 ? FSMULD
+                                 ? V9::FSMULD
                                  : INVALID_MACHINE_OPCODE);
         Instruction* mulInstr = subtreeRoot->getInstruction();
         CreateMulInstruction(target, mulInstr->getParent()->getParent(),
@@ -1745,8 +1832,9 @@ GetInstructionsByRule(InstructionNode* subtreeRoot,
 
       case 235:        // reg:   Mul(reg, Constant)
       {
+        maskUnsignedResult = true;
         MachineOpCode forceOp = ((checkCast && BothFloatToDouble(subtreeRoot))
-                                 ? FSMULD
+                                 ? V9::FSMULD
                                  : INVALID_MACHINE_OPCODE);
         Instruction* mulInstr = subtreeRoot->getInstruction();
         CreateMulInstruction(target, mulInstr->getParent()->getParent(),
@@ -1758,6 +1846,7 @@ GetInstructionsByRule(InstructionNode* subtreeRoot,
         break;
       }
       case 236:        // reg:   Div(reg, Constant)
+        maskUnsignedResult = true;
         L = mvec.size();
         CreateDivConstInstruction(target, subtreeRoot, mvec);
         if (mvec.size() > L)
@@ -1765,13 +1854,15 @@ GetInstructionsByRule(InstructionNode* subtreeRoot,
         // ELSE FALL THROUGH
       
       case 36: // reg:   Div(reg, reg)
-        mvec.push_back(new MachineInstr(ChooseDivInstruction(target, subtreeRoot)));
-        Set3OperandsFromInstr(mvec.back(), subtreeRoot, target);
+        maskUnsignedResult = true;
+        Add3OperandInstr(ChooseDivInstruction(target, subtreeRoot),
+                         subtreeRoot, mvec);
         break;
 
       case  37:        // reg:   Rem(reg, reg)
       case 237:        // reg:   Rem(reg, Constant)
       {
+        maskUnsignedResult = true;
         Instruction* remInstr = subtreeRoot->getInstruction();
         
         TmpInstruction* quot = new TmpInstruction(
@@ -1782,25 +1873,24 @@ GetInstructionsByRule(InstructionNode* subtreeRoot,
                                         subtreeRoot->rightChild()->getValue());
         MachineCodeForInstruction::get(remInstr).addTemp(quot).addTemp(prod); 
         
-        M = new MachineInstr(ChooseDivInstruction(target, subtreeRoot));
-        Set3OperandsFromInstr(M, subtreeRoot, target);
-        M->SetMachineOperandVal(2, MachineOperand::MO_VirtualRegister,quot);
+        M = BuildMI(ChooseDivInstruction(target, subtreeRoot), 3)
+                             .addReg(subtreeRoot->leftChild()->getValue())
+                             .addReg(subtreeRoot->rightChild()->getValue())
+                             .addRegDef(quot);
         mvec.push_back(M);
         
-        M = new MachineInstr(ChooseMulInstructionByType(
-                                   subtreeRoot->getInstruction()->getType()));
-        M->SetMachineOperandVal(0, MachineOperand::MO_VirtualRegister,quot);
-        M->SetMachineOperandVal(1, MachineOperand::MO_VirtualRegister,
-                                      subtreeRoot->rightChild()->getValue());
-        M->SetMachineOperandVal(2, MachineOperand::MO_VirtualRegister,prod);
+        unsigned MulOpcode =
+          ChooseMulInstructionByType(subtreeRoot->getInstruction()->getType());
+        Value *MulRHS = subtreeRoot->rightChild()->getValue();
+        M = BuildMI(MulOpcode, 3).addReg(quot).addReg(MulRHS).addReg(prod,
+                                                                     MOTy::Def);
         mvec.push_back(M);
         
-        M = new MachineInstr(ChooseSubInstructionByType(
-                                   subtreeRoot->getInstruction()->getType()));
-        Set3OperandsFromInstr(M, subtreeRoot, target);
-        M->SetMachineOperandVal(1, MachineOperand::MO_VirtualRegister,prod);
+        unsigned Opcode = ChooseSubInstructionByType(
+                                   subtreeRoot->getInstruction()->getType());
+        M = BuildMI(Opcode, 3).addReg(subtreeRoot->leftChild()->getValue())
+                              .addReg(prod).addRegDef(subtreeRoot->getValue());
         mvec.push_back(M);
-        
         break;
       }
       
@@ -1808,43 +1898,67 @@ GetInstructionsByRule(InstructionNode* subtreeRoot,
       case 238:        // bool:   And(bool, boolconst)
       case 338:        // reg :   BAnd(reg, reg)
       case 538:        // reg :   BAnd(reg, Constant)
-        mvec.push_back(new MachineInstr(AND));
-        Set3OperandsFromInstr(mvec.back(), subtreeRoot, target);
+        Add3OperandInstr(V9::AND, subtreeRoot, mvec);
         break;
 
       case 138:        // bool:   And(bool, not)
-      case 438:        // bool:   BAnd(bool, not)
-        mvec.push_back(new MachineInstr(ANDN));
-        Set3OperandsFromInstr(mvec.back(), subtreeRoot, target);
+      case 438:        // bool:   BAnd(bool, bnot)
+      { // Use the argument of NOT as the second argument!
+        // Mark the NOT node so that no code is generated for it.
+        InstructionNode* notNode = (InstructionNode*) subtreeRoot->rightChild();
+        Value* notArg = BinaryOperator::getNotArgument(
+                           cast<BinaryOperator>(notNode->getInstruction()));
+        notNode->markFoldedIntoParent();
+        Value *LHS = subtreeRoot->leftChild()->getValue();
+        Value *Dest = subtreeRoot->getValue();
+        mvec.push_back(BuildMI(V9::ANDN, 3).addReg(LHS).addReg(notArg)
+                                       .addReg(Dest, MOTy::Def));
         break;
+      }
 
       case  39:        // bool:   Or(bool, bool)
       case 239:        // bool:   Or(bool, boolconst)
       case 339:        // reg :   BOr(reg, reg)
       case 539:        // reg :   BOr(reg, Constant)
-        mvec.push_back(new MachineInstr(OR));
-        Set3OperandsFromInstr(mvec.back(), subtreeRoot, target);
+        Add3OperandInstr(V9::OR, subtreeRoot, mvec);
         break;
 
       case 139:        // bool:   Or(bool, not)
-      case 439:        // bool:   BOr(bool, not)
-        mvec.push_back(new MachineInstr(ORN));
-        Set3OperandsFromInstr(mvec.back(), subtreeRoot, target);
+      case 439:        // bool:   BOr(bool, bnot)
+      { // Use the argument of NOT as the second argument!
+        // Mark the NOT node so that no code is generated for it.
+        InstructionNode* notNode = (InstructionNode*) subtreeRoot->rightChild();
+        Value* notArg = BinaryOperator::getNotArgument(
+                           cast<BinaryOperator>(notNode->getInstruction()));
+        notNode->markFoldedIntoParent();
+        Value *LHS = subtreeRoot->leftChild()->getValue();
+        Value *Dest = subtreeRoot->getValue();
+        mvec.push_back(BuildMI(V9::ORN, 3).addReg(LHS).addReg(notArg)
+                       .addReg(Dest, MOTy::Def));
         break;
+      }
 
       case  40:        // bool:   Xor(bool, bool)
       case 240:        // bool:   Xor(bool, boolconst)
       case 340:        // reg :   BXor(reg, reg)
       case 540:        // reg :   BXor(reg, Constant)
-        mvec.push_back(new MachineInstr(XOR));
-        Set3OperandsFromInstr(mvec.back(), subtreeRoot, target);
+        Add3OperandInstr(V9::XOR, subtreeRoot, mvec);
         break;
 
       case 140:        // bool:   Xor(bool, not)
-      case 440:        // bool:   BXor(bool, not)
-        mvec.push_back(new MachineInstr(XNOR));
-        Set3OperandsFromInstr(mvec.back(), subtreeRoot, target);
+      case 440:        // bool:   BXor(bool, bnot)
+      { // Use the argument of NOT as the second argument!
+        // Mark the NOT node so that no code is generated for it.
+        InstructionNode* notNode = (InstructionNode*) subtreeRoot->rightChild();
+        Value* notArg = BinaryOperator::getNotArgument(
+                           cast<BinaryOperator>(notNode->getInstruction()));
+        notNode->markFoldedIntoParent();
+        Value *LHS = subtreeRoot->leftChild()->getValue();
+        Value *Dest = subtreeRoot->getValue();
+        mvec.push_back(BuildMI(V9::XNOR, 3).addReg(LHS).addReg(notArg)
+                       .addReg(Dest, MOTy::Def));
         break;
+      }
 
       case 41: // boolconst:   SetCC(reg, Constant)
         // 
@@ -1858,7 +1972,8 @@ GetInstructionsByRule(InstructionNode* subtreeRoot,
         // a result register, and setting a condition code.
         // 
         // If the boolean result of the SetCC is used by anything other
-        // than a branch instruction, the boolean must be
+        // than a branch instruction, or if it is used outside the current
+        // basic block, the boolean must be
         // computed and stored in the result register.  Otherwise, discard
         // the difference (by using %g0) and keep only the condition code.
         // 
@@ -1870,7 +1985,8 @@ GetInstructionsByRule(InstructionNode* subtreeRoot,
         InstructionNode* parentNode = (InstructionNode*) subtreeRoot->parent();
         Instruction* setCCInstr = subtreeRoot->getInstruction();
         
-        bool keepBoolVal = ! AllUsesAreBranches(setCCInstr);
+        bool keepBoolVal = parentNode == NULL ||
+                           ! AllUsesAreBranches(setCCInstr);
         bool subValIsBoolVal = setCCInstr->getOpcode() == Instruction::SetNE;
         bool keepSubVal = keepBoolVal && subValIsBoolVal;
         bool computeBoolVal = keepBoolVal && ! subValIsBoolVal;
@@ -1878,7 +1994,7 @@ GetInstructionsByRule(InstructionNode* subtreeRoot,
         bool mustClearReg;
         int valueToMove;
         MachineOpCode movOpCode = 0;
-
+        
         // Mark the 4th operand as being a CC register, and as a def
         // A TmpInstruction is created to represent the CC "result".
         // Unlike other instances of TmpInstruction, this one is used
@@ -1905,10 +2021,19 @@ GetInstructionsByRule(InstructionNode* subtreeRoot,
             // a separate instruction to compute the bool result, so discard
             // result of SUBcc instruction anyway.
             // 
-            M = new MachineInstr(SUBcc);
-            Set3OperandsFromInstr(M, subtreeRoot, target, ! keepSubVal);
-            M->SetMachineOperandVal(3, MachineOperand::MO_CCRegister,
-                                    tmpForCC, /*def*/true);
+            if (keepSubVal) {
+              M = BuildMI(V9::SUBcc, 4)
+                .addReg(subtreeRoot->leftChild()->getValue())
+                .addReg(subtreeRoot->rightChild()->getValue())
+                .addRegDef(subtreeRoot->getValue())
+                .addCCReg(tmpForCC, MOTy::Def);
+            } else {
+              M = BuildMI(V9::SUBcc, 4)
+                .addReg(subtreeRoot->leftChild()->getValue())
+                .addReg(subtreeRoot->rightChild()->getValue())
+                .addMReg(target.getRegInfo().getZeroRegNum(), MOTy::Def)
+                .addCCReg(tmpForCC, MOTy::Def);
+            }
             mvec.push_back(M);
             
             if (computeBoolVal)
@@ -1920,13 +2045,10 @@ GetInstructionsByRule(InstructionNode* subtreeRoot,
         else
           {
             // FP condition: dest of FCMP should be some FCCn register
-            M = new MachineInstr(ChooseFcmpInstruction(subtreeRoot));
-            M->SetMachineOperandVal(0, MachineOperand::MO_CCRegister,
-                                          tmpForCC);
-            M->SetMachineOperandVal(1,MachineOperand::MO_VirtualRegister,
-                                         subtreeRoot->leftChild()->getValue());
-            M->SetMachineOperandVal(2,MachineOperand::MO_VirtualRegister,
-                                        subtreeRoot->rightChild()->getValue());
+            M = BuildMI(ChooseFcmpInstruction(subtreeRoot), 3)
+                         .addCCReg(tmpForCC, MOTy::Def)
+                         .addReg(subtreeRoot->leftChild()->getValue())
+                         .addRegDef(subtreeRoot->rightChild()->getValue());
             mvec.push_back(M);
             
             if (computeBoolVal)
@@ -1941,59 +2063,45 @@ GetInstructionsByRule(InstructionNode* subtreeRoot,
           {
             if (mustClearReg)
               {// Unconditionally set register to 0
-                M = new MachineInstr(SETHI);
-                M->SetMachineOperandConst(0,MachineOperand::MO_UnextendedImmed,
-                                          (int64_t)0);
-                M->SetMachineOperandVal(1, MachineOperand::MO_VirtualRegister,
-                                        setCCInstr);
+                M = BuildMI(V9::SETHI, 2).addZImm(0).addRegDef(setCCInstr);
                 mvec.push_back(M);
               }
             
             // Now conditionally move `valueToMove' (0 or 1) into the register
-            M = new MachineInstr(movOpCode);
-            M->SetMachineOperandVal(0, MachineOperand::MO_CCRegister,
-                                    tmpForCC);
-            M->SetMachineOperandConst(1, MachineOperand::MO_UnextendedImmed,
-                                      valueToMove);
-            M->SetMachineOperandVal(2, MachineOperand::MO_VirtualRegister,
-                                    setCCInstr);
+            // Mark the register as a use (as well as a def) because the old
+            // value should be retained if the condition is false.
+            M = BuildMI(movOpCode, 3).addCCReg(tmpForCC).addZImm(valueToMove)
+                                     .addReg(setCCInstr, MOTy::UseAndDef);
             mvec.push_back(M);
           }
         break;
       }    
 
-      case 43: // boolreg: VReg
-      case 44: // boolreg: Constant
-        break;
-
       case 51: // reg:   Load(reg)
       case 52: // reg:   Load(ptrreg)
-      case 53: // reg:   LoadIdx(reg,reg)
-      case 54: // reg:   LoadIdx(ptrreg,reg)
-        mvec.push_back(new MachineInstr(ChooseLoadInstruction(
-                                     subtreeRoot->getValue()->getType())));
-        SetOperandsForMemInstr(mvec, mvec.end()-1, subtreeRoot, target);
+        SetOperandsForMemInstr(ChooseLoadInstruction(
+                                   subtreeRoot->getValue()->getType()),
+                               mvec, subtreeRoot, target);
         break;
 
       case 55: // reg:   GetElemPtr(reg)
       case 56: // reg:   GetElemPtrIdx(reg,reg)
         // If the GetElemPtr was folded into the user (parent), it will be
         // caught above.  For other cases, we have to compute the address.
-        mvec.push_back(new MachineInstr(ADD));
-        SetOperandsForMemInstr(mvec, mvec.end()-1, subtreeRoot, target);
+        SetOperandsForMemInstr(V9::ADD, mvec, subtreeRoot, target);
         break;
-        
+
       case 57: // reg:  Alloca: Implement as 1 instruction:
       {         //         add %fp, offsetFromFP -> result
         AllocationInst* instr =
           cast<AllocationInst>(subtreeRoot->getInstruction());
-        unsigned int tsize =
-          target.findOptimalStorageSize(instr->getAllocatedType());
+        unsigned tsize =
+          target.getTargetData().getTypeSize(instr->getAllocatedType());
         assert(tsize != 0);
         CreateCodeForFixedSizeAlloca(target, instr, tsize, 1, mvec);
         break;
       }
-      
+
       case 58: // reg:   Alloca(reg): Implement as 3 instructions:
                 //     mul num, typeSz -> tmp
                 //     sub %sp, tmp    -> %sp
@@ -2003,14 +2111,14 @@ GetInstructionsByRule(InstructionNode* subtreeRoot,
         const Type* eltType = instr->getAllocatedType();
         
         // If #elements is constant, use simpler code for fixed-size allocas
-        int tsize = (int) target.findOptimalStorageSize(eltType);
+        int tsize = (int) target.getTargetData().getTypeSize(eltType);
         Value* numElementsVal = NULL;
         bool isArray = instr->isArrayAllocation();
         
         if (!isArray ||
             isa<Constant>(numElementsVal = instr->getArraySize()))
           { // total size is constant: generate code for fixed-size alloca
-            unsigned int numElements = isArray? 
+            unsigned numElements = isArray? 
               cast<ConstantUInt>(numElementsVal)->getValue() : 1;
             CreateCodeForFixedSizeAlloca(target, instr, tsize,
                                          numElements, mvec);
@@ -2020,13 +2128,13 @@ GetInstructionsByRule(InstructionNode* subtreeRoot,
                                           numElementsVal, mvec);
         break;
       }
-      
+
       case 61: // reg:   Call
-      {         // Generate a direct (CALL) or indirect (JMPL). depending
-                // Mark the return-address register and the indirection
-                // register (if any) as hidden virtual registers.
-                // Also, mark the operands of the Call and return value (if
-                // any) as implicit operands of the CALL machine instruction.
+      {         // Generate a direct (CALL) or indirect (JMPL) call.
+                // Mark the return-address register, the indirection
+                // register (for indirect calls), the operands of the Call,
+                // and the return value (if any) as implicit operands
+                // of the machine instruction.
                 // 
                 // If this is a varargs function, floating point arguments
                 // have to passed in integer registers so insert
@@ -2034,107 +2142,112 @@ GetInstructionsByRule(InstructionNode* subtreeRoot,
                 // 
         CallInst *callInstr = cast<CallInst>(subtreeRoot->getInstruction());
         Value *callee = callInstr->getCalledValue();
-        
-        // Create hidden virtual register for return address, with type void*. 
-        TmpInstruction* retAddrReg =
-          new TmpInstruction(PointerType::get(Type::VoidTy), callInstr);
-        MachineCodeForInstruction::get(callInstr).addTemp(retAddrReg);
-        
-        // Generate the machine instruction and its operands.
-        // Use CALL for direct function calls; this optimistically assumes
-        // the PC-relative address fits in the CALL address field (22 bits).
-        // Use JMPL for indirect calls.
+        Function* calledFunc = dyn_cast<Function>(callee);
+
+        // Check if this is an intrinsic function that needs a special code
+        // sequence (e.g., va_start).  Indirect calls cannot be special.
         // 
-        if (isa<Function>(callee))
-          { // direct function call
-            M = new MachineInstr(CALL);
-            M->SetMachineOperandVal(0, MachineOperand::MO_PCRelativeDisp,
-                                    callee);
-          } 
-        else
-          { // indirect function call
-            M = new MachineInstr(JMPLCALL);
-            M->SetMachineOperandVal(0, MachineOperand::MO_VirtualRegister,
-                                    callee);
-            M->SetMachineOperandConst(1, MachineOperand::MO_SignExtendedImmed,
-                                      (int64_t) 0);
-            M->SetMachineOperandVal(2, MachineOperand::MO_VirtualRegister,
-                                    retAddrReg);
-          }
-        
-        mvec.push_back(M);
+        bool specialIntrinsic = false;
+        LLVMIntrinsic::ID iid;
+        if (calledFunc && (iid=(LLVMIntrinsic::ID)calledFunc->getIntrinsicID()))
+          specialIntrinsic = CodeGenIntrinsic(iid, *callInstr, target, mvec);
 
-        const FunctionType* funcType =
-          cast<FunctionType>(cast<PointerType>(callee->getType())
-                             ->getElementType());
-        bool isVarArgs = funcType->isVarArg();
-        bool noPrototype = isVarArgs && funcType->getNumParams() == 0;
-        
-        // Use an annotation to pass information about call arguments
-        // to the register allocator.
-        CallArgsDescriptor* argDesc = new CallArgsDescriptor(callInstr,
-                                         retAddrReg, isVarArgs, noPrototype);
-        M->addAnnotation(argDesc);
-        
-        assert(callInstr->getOperand(0) == callee
-               && "This is assumed in the loop below!");
-        
-        for (unsigned i=1, N=callInstr->getNumOperands(); i < N; ++i)
+        // If not, generate the normal call sequence for the function.
+        // This can also handle any intrinsics that are just function calls.
+        // 
+        if (! specialIntrinsic)
           {
-            Value* argVal = callInstr->getOperand(i);
-            Instruction* intArgReg = NULL;
+            // Create hidden virtual register for return address with type void*
+            TmpInstruction* retAddrReg =
+              new TmpInstruction(PointerType::get(Type::VoidTy), callInstr);
+            MachineCodeForInstruction::get(callInstr).addTemp(retAddrReg);
+
+            // Generate the machine instruction and its operands.
+            // Use CALL for direct function calls; this optimistically assumes
+            // the PC-relative address fits in the CALL address field (22 bits).
+            // Use JMPL for indirect calls.
+            // 
+            if (calledFunc)             // direct function call
+              M = BuildMI(V9::CALL, 1).addPCDisp(callee);
+            else                        // indirect function call
+              M = BuildMI(V9::JMPLCALL, 3).addReg(callee).addSImm((int64_t)0)
+                .addRegDef(retAddrReg);
+            mvec.push_back(M);
+
+            const FunctionType* funcType =
+              cast<FunctionType>(cast<PointerType>(callee->getType())
+                                 ->getElementType());
+            bool isVarArgs = funcType->isVarArg();
+            bool noPrototype = isVarArgs && funcType->getNumParams() == 0;
+        
+            // Use a descriptor to pass information about call arguments
+            // to the register allocator.  This descriptor will be "owned"
+            // and freed automatically when the MachineCodeForInstruction
+            // object for the callInstr goes away.
+            CallArgsDescriptor* argDesc = new CallArgsDescriptor(callInstr,
+                                             retAddrReg, isVarArgs,noPrototype);
+            
+            assert(callInstr->getOperand(0) == callee
+                   && "This is assumed in the loop below!");
             
-            // Check for FP arguments to varargs functions.
-            // Any such argument in the first $K$ args must be passed in an
-            // integer register, where K = #integer argument registers.
-            if (isVarArgs && argVal->getType()->isFloatingPoint())
+            for (unsigned i=1, N=callInstr->getNumOperands(); i < N; ++i)
               {
-                // If it is a function with no prototype, pass value
-                // as an FP value as well as a varargs value
-                if (noPrototype)
-                  argDesc->getArgInfo(i-1).setUseFPArgReg();
-                
-                // If this arg. is in the first $K$ regs, add a copy
-                // float-to-int instruction to pass the value as an integer.
-                if (i < target.getRegInfo().GetNumOfIntArgRegs())
+                Value* argVal = callInstr->getOperand(i);
+                Instruction* intArgReg = NULL;
+            
+                // Check for FP arguments to varargs functions.
+                // Any such argument in the first $K$ args must be passed in an
+                // integer register, where K = #integer argument registers.
+                if (isVarArgs && argVal->getType()->isFloatingPoint())
                   {
-                    MachineCodeForInstruction &destMCFI = 
-                      MachineCodeForInstruction::get(callInstr);   
-                    intArgReg = new TmpInstruction(Type::IntTy, argVal);
-                    destMCFI.addTemp(intArgReg);
+                    // If it is a function with no prototype, pass value
+                    // as an FP value as well as a varargs value
+                    if (noPrototype)
+                      argDesc->getArgInfo(i-1).setUseFPArgReg();
+                
+                    // If this arg. is in the first $K$ regs, add a copy
+                    // float-to-int instruction to pass the value as an integer.
+                    if (i <= target.getRegInfo().getNumOfIntArgRegs())
+                      {
+                        MachineCodeForInstruction &destMCFI = 
+                          MachineCodeForInstruction::get(callInstr);   
+                        intArgReg = new TmpInstruction(Type::IntTy, argVal);
+                        destMCFI.addTemp(intArgReg);
                     
-                    vector<MachineInstr*> copyMvec;
-                    target.getInstrInfo().CreateCodeToCopyFloatToInt(target,
-                                           callInstr->getParent()->getParent(),
-                                           argVal, (TmpInstruction*) intArgReg,
-                                           copyMvec, destMCFI);
-                    mvec.insert(mvec.begin(),copyMvec.begin(),copyMvec.end());
+                        std::vector<MachineInstr*> copyMvec;
+                        target.getInstrInfo().CreateCodeToCopyFloatToInt(target,
+                                         callInstr->getParent()->getParent(),
+                                         argVal, (TmpInstruction*) intArgReg,
+                                         copyMvec, destMCFI);
+                        mvec.insert(mvec.begin(),copyMvec.begin(),copyMvec.end());
                     
-                    argDesc->getArgInfo(i-1).setUseIntArgReg();
-                    argDesc->getArgInfo(i-1).setArgCopy(intArgReg);
+                        argDesc->getArgInfo(i-1).setUseIntArgReg();
+                        argDesc->getArgInfo(i-1).setArgCopy(intArgReg);
+                      }
+                    else
+                      // Cannot fit in first $K$ regs so pass arg on stack
+                      argDesc->getArgInfo(i-1).setUseStackSlot();
                   }
-                else
-                  // Cannot fit in first $K$ regs so pass the arg on the stack
-                  argDesc->getArgInfo(i-1).setUseStackSlot();
-              }
             
-            if (intArgReg)
-              mvec.back()->addImplicitRef(intArgReg);
+                if (intArgReg)
+                  mvec.back()->addImplicitRef(intArgReg);
             
-            mvec.back()->addImplicitRef(argVal);
-          }
+                mvec.back()->addImplicitRef(argVal);
+              }
         
-        // Add the return value as an implicit ref.  The call operands
-        // were added above.
-        if (callInstr->getType() != Type::VoidTy)
-          mvec.back()->addImplicitRef(callInstr, /*isDef*/ true);
+            // Add the return value as an implicit ref.  The call operands
+            // were added above.
+            if (callInstr->getType() != Type::VoidTy)
+              mvec.back()->addImplicitRef(callInstr, /*isDef*/ true);
         
-        // For the CALL instruction, the ret. addr. reg. is also implicit
-        if (isa<Function>(callee))
-          mvec.back()->addImplicitRef(retAddrReg, /*isDef*/ true);
+            // For the CALL instruction, the ret. addr. reg. is also implicit
+            if (isa<Function>(callee))
+              mvec.back()->addImplicitRef(retAddrReg, /*isDef*/ true);
         
-        // delay slot
-        mvec.push_back(new MachineInstr(NOP));
+            // delay slot
+            mvec.push_back(BuildMI(V9::NOP, 0));
+          }
+
         break;
       }
       
@@ -2145,12 +2258,11 @@ GetInstructionsByRule(InstructionNode* subtreeRoot,
         Instruction* shlInstr = subtreeRoot->getInstruction();
         
         const Type* opType = argVal1->getType();
-        assert(opType->isIntegral()
-               || opType == Type::BoolTy
-               || isa<PointerType>(opType)&&"Shl unsupported for other types");
+        assert((opType->isInteger() || isa<PointerType>(opType)) &&
+               "Shl unsupported for other types");
         
         CreateShiftInstructions(target, shlInstr->getParent()->getParent(),
-                                (opType == Type::LongTy)? SLLX : SLL,
+                                (opType == Type::LongTy)? V9::SLLX : V9::SLL,
                                 argVal1, argVal2, 0, shlInstr, mvec,
                                 MachineCodeForInstruction::get(shlInstr));
         break;
@@ -2158,33 +2270,30 @@ GetInstructionsByRule(InstructionNode* subtreeRoot,
       
       case 63: // reg:   Shr(reg, reg)
       { const Type* opType = subtreeRoot->leftChild()->getValue()->getType();
-        assert(opType->isIntegral()
-               || isa<PointerType>(opType)&&"Shr unsupported for other types");
-        mvec.push_back(new MachineInstr((opType->isSigned()
-                                   ? ((opType == Type::LongTy)? SRAX : SRA)
-                                   : ((opType == Type::LongTy)? SRLX : SRL))));
-        Set3OperandsFromInstr(mvec.back(), subtreeRoot, target);
+        assert((opType->isInteger() || isa<PointerType>(opType)) &&
+               "Shr unsupported for other types");
+        Add3OperandInstr(opType->isSigned()
+                               ? (opType == Type::LongTy ? V9::SRAX : V9::SRA)
+                               : (opType == Type::LongTy ? V9::SRLX : V9::SRL),
+                         subtreeRoot, mvec);
         break;
       }
       
       case 64: // reg:   Phi(reg,reg)
         break;                          // don't forward the value
 
-#undef NEED_PHI_MACHINE_INSTRS
-#ifdef NEED_PHI_MACHINE_INSTRS
-      {                // This instruction has variable #operands, so resultPos is 0.
-        Instruction* phi = subtreeRoot->getInstruction();
-        M = new MachineInstr(PHI, 1 + phi->getNumOperands());
-        M->SetMachineOperandVal(0, MachineOperand::MO_VirtualRegister,
-                                      subtreeRoot->getValue());
-        for (unsigned i=0, N=phi->getNumOperands(); i < N; i++)
-          M->SetMachineOperandVal(i+1, MachineOperand::MO_VirtualRegister,
-                                  phi->getOperand(i));
-        mvec.push_back(M);
+      case 65: // reg:   VaArg(reg)
+      {
+        // Use value initialized by va_start as pointer to args on the stack.
+        // Load argument via current pointer value, then increment pointer.
+        int argSize = target.getFrameInfo().getSizeOfEachArgOnStack();
+        Instruction* vaArgI = subtreeRoot->getInstruction();
+        mvec.push_back(BuildMI(V9::LDX, 3).addReg(vaArgI->getOperand(0)).
+                       addSImm(0).addRegDef(vaArgI));
+        mvec.push_back(BuildMI(V9::ADD, 3).addReg(vaArgI->getOperand(0)).
+                       addSImm(argSize).addRegDef(vaArgI->getOperand(0)));
         break;
-      }  
-#endif // NEED_PHI_MACHINE_INSTRS
-      
+      }
       
       case 71: // reg:     VReg
       case 72: // reg:     Constant
@@ -2195,7 +2304,7 @@ GetInstructionsByRule(InstructionNode* subtreeRoot,
         break;
       }
     }
-  
+
   if (forwardOperandNum >= 0)
     { // We did not generate a machine instruction but need to use operand.
       // If user is in the same tree, replace Value in its machine operand.
@@ -2205,7 +2314,7 @@ GetInstructionsByRule(InstructionNode* subtreeRoot,
         ForwardOperand(subtreeRoot, subtreeRoot->parent(), forwardOperandNum);
       else
         {
-          vector<MachineInstr*> minstrVec;
+          std::vector<MachineInstr*> minstrVec;
           Instruction* instr = subtreeRoot->getInstruction();
           target.getInstrInfo().
             CreateCopyInstructionsByType(target,
@@ -2217,4 +2326,33 @@ GetInstructionsByRule(InstructionNode* subtreeRoot,
           mvec.insert(mvec.end(), minstrVec.begin(), minstrVec.end());
         }
     }
+
+  if (maskUnsignedResult)
+    { // If result is unsigned and smaller than int reg size,
+      // we need to clear high bits of result value.
+      assert(forwardOperandNum < 0 && "Need mask but no instruction generated");
+      Instruction* dest = subtreeRoot->getInstruction();
+      if (dest->getType()->isUnsigned())
+        {
+          unsigned destSize=target.getTargetData().getTypeSize(dest->getType());
+          if (destSize <= 4)
+            { // Mask high bits.  Use a TmpInstruction to represent the
+              // intermediate result before masking.  Since those instructions
+              // have already been generated, go back and substitute tmpI
+              // for dest in the result position of each one of them.
+              TmpInstruction *tmpI = new TmpInstruction(dest->getType(), dest,
+                                                        NULL, "maskHi");
+              MachineCodeForInstruction::get(dest).addTemp(tmpI);
+
+              for (unsigned i=0, N=mvec.size(); i < N; ++i)
+                mvec[i]->substituteValue(dest, tmpI);
+
+              M = BuildMI(V9::SRL, 3).addReg(tmpI).addZImm(8*(4-destSize))
+                .addReg(dest, MOTy::Def);
+              mvec.push_back(M);
+            }
+          else if (destSize < 8)
+            assert(0 && "Unsupported type size: 32 < size < 64 bits");
+        }
+    }
 }