Sync upstream to r102410.
Re-turn on sdk.

Change-Id: I91a890863989a67243b4d2dfd1ae09b843ebaeaf
diff --git a/lib/CodeGen/SelectionDAG/CMakeLists.txt b/lib/CodeGen/SelectionDAG/CMakeLists.txt
index 80c7d7c..0cfd5e1 100644
--- a/lib/CodeGen/SelectionDAG/CMakeLists.txt
+++ b/lib/CodeGen/SelectionDAG/CMakeLists.txt
@@ -20,6 +20,7 @@
   SelectionDAGISel.cpp
   SelectionDAGPrinter.cpp
   TargetLowering.cpp
+  TargetSelectionDAGInfo.cpp
   )
 
 target_link_libraries (LLVMSelectionDAG LLVMAnalysis LLVMAsmPrinter LLVMCodeGen)
diff --git a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 3be6b43..9192593 100644
--- a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -129,6 +129,14 @@
     bool CombineToPreIndexedLoadStore(SDNode *N);
     bool CombineToPostIndexedLoadStore(SDNode *N);
 
+    void ReplaceLoadWithPromotedLoad(SDNode *Load, SDNode *ExtLoad);
+    SDValue PromoteOperand(SDValue Op, EVT PVT, bool &Replace);
+    SDValue SExtPromoteOperand(SDValue Op, EVT PVT);
+    SDValue ZExtPromoteOperand(SDValue Op, EVT PVT);
+    SDValue PromoteIntBinOp(SDValue Op);
+    SDValue PromoteIntShiftOp(SDValue Op);
+    SDValue PromoteExtend(SDValue Op);
+    bool PromoteLoad(SDValue Op);
 
     /// combine - call the node-specific routine that knows how to fold each
     /// particular type of node. If that doesn't do anything, try the
@@ -254,24 +262,28 @@
     /// looking for a better chain (aliasing node.)
     SDValue FindBetterChain(SDNode *N, SDValue Chain);
 
-    /// getShiftAmountTy - Returns a type large enough to hold any valid
-    /// shift amount - before type legalization these can be huge.
-    EVT getShiftAmountTy() {
-      return LegalTypes ?  TLI.getShiftAmountTy() : TLI.getPointerTy();
-    }
-
-public:
+  public:
     DAGCombiner(SelectionDAG &D, AliasAnalysis &A, CodeGenOpt::Level OL)
-      : DAG(D),
-        TLI(D.getTargetLoweringInfo()),
-        Level(Unrestricted),
-        OptLevel(OL),
-        LegalOperations(false),
-        LegalTypes(false),
-        AA(A) {}
+      : DAG(D), TLI(D.getTargetLoweringInfo()), Level(Unrestricted),
+        OptLevel(OL), LegalOperations(false), LegalTypes(false), AA(A) {}
 
     /// Run - runs the dag combiner on all nodes in the work list
     void Run(CombineLevel AtLevel);
+    
+    SelectionDAG &getDAG() const { return DAG; }
+    
+    /// getShiftAmountTy - Returns a type large enough to hold any valid
+    /// shift amount - before type legalization these can be huge.
+    EVT getShiftAmountTy() {
+      return LegalTypes ? TLI.getShiftAmountTy() : TLI.getPointerTy();
+    }
+    
+    /// isTypeLegal - This method returns true if we are running before type
+    /// legalization or if the specified VT is legal.
+    bool isTypeLegal(const EVT &VT) {
+      if (!LegalTypes) return true;
+      return TLI.isTypeLegal(VT);
+    }
   };
 }
 
@@ -577,9 +589,8 @@
   return SDValue(N, 0);
 }
 
-void
-DAGCombiner::CommitTargetLoweringOpt(const TargetLowering::TargetLoweringOpt &
-                                                                          TLO) {
+void DAGCombiner::
+CommitTargetLoweringOpt(const TargetLowering::TargetLoweringOpt &TLO) {
   // Replace all uses.  If any nodes become isomorphic to other nodes and
   // are deleted, make sure to remove them from our worklist.
   WorkListRemover DeadNodes(*this);
@@ -609,7 +620,7 @@
 /// it can be simplified or if things it uses can be simplified by bit
 /// propagation.  If so, return true.
 bool DAGCombiner::SimplifyDemandedBits(SDValue Op, const APInt &Demanded) {
-  TargetLowering::TargetLoweringOpt TLO(DAG);
+  TargetLowering::TargetLoweringOpt TLO(DAG, LegalTypes, LegalOperations);
   APInt KnownZero, KnownOne;
   if (!TLI.SimplifyDemandedBits(Op, Demanded, KnownZero, KnownOne, TLO))
     return false;
@@ -629,6 +640,260 @@
   return true;
 }
 
+void DAGCombiner::ReplaceLoadWithPromotedLoad(SDNode *Load, SDNode *ExtLoad) {
+  DebugLoc dl = Load->getDebugLoc();
+  EVT VT = Load->getValueType(0);
+  SDValue Trunc = DAG.getNode(ISD::TRUNCATE, dl, VT, SDValue(ExtLoad, 0));
+
+  DEBUG(dbgs() << "\nReplacing.9 ";
+        Load->dump(&DAG);
+        dbgs() << "\nWith: ";
+        Trunc.getNode()->dump(&DAG);
+        dbgs() << '\n');
+  WorkListRemover DeadNodes(*this);
+  DAG.ReplaceAllUsesOfValueWith(SDValue(Load, 0), Trunc, &DeadNodes);
+  DAG.ReplaceAllUsesOfValueWith(SDValue(Load, 1), SDValue(ExtLoad, 1),
+                                &DeadNodes);
+  removeFromWorkList(Load);
+  DAG.DeleteNode(Load);
+}
+
+SDValue DAGCombiner::PromoteOperand(SDValue Op, EVT PVT, bool &Replace) {
+  Replace = false;
+  DebugLoc dl = Op.getDebugLoc();
+  if (LoadSDNode *LD = dyn_cast<LoadSDNode>(Op)) {
+    ISD::LoadExtType ExtType =
+      ISD::isNON_EXTLoad(LD) ? ISD::EXTLOAD : LD->getExtensionType();
+    Replace = true;
+    return DAG.getExtLoad(ExtType, dl, PVT,
+                          LD->getChain(), LD->getBasePtr(),
+                          LD->getSrcValue(), LD->getSrcValueOffset(),
+                          LD->getMemoryVT(), LD->isVolatile(),
+                          LD->isNonTemporal(), LD->getAlignment());
+  }
+
+  unsigned Opc = Op.getOpcode();
+  switch (Opc) {
+  default: break;
+  case ISD::AssertSext:
+    return DAG.getNode(ISD::AssertSext, dl, PVT,
+                       SExtPromoteOperand(Op.getOperand(0), PVT),
+                       Op.getOperand(1));
+  case ISD::AssertZext:
+    return DAG.getNode(ISD::AssertZext, dl, PVT,
+                       ZExtPromoteOperand(Op.getOperand(0), PVT),
+                       Op.getOperand(1));
+  case ISD::Constant: {
+    unsigned ExtOpc =
+      Op.getValueType().isByteSized() ? ISD::SIGN_EXTEND : ISD::ZERO_EXTEND;
+    return DAG.getNode(ExtOpc, dl, PVT, Op);
+  }    
+  }
+
+  if (!TLI.isOperationLegal(ISD::ANY_EXTEND, PVT))
+    return SDValue();
+  return DAG.getNode(ISD::ANY_EXTEND, dl, PVT, Op);
+}
+
+SDValue DAGCombiner::SExtPromoteOperand(SDValue Op, EVT PVT) {
+  if (!TLI.isOperationLegal(ISD::SIGN_EXTEND_INREG, PVT))
+    return SDValue();
+  EVT OldVT = Op.getValueType();
+  DebugLoc dl = Op.getDebugLoc();
+  bool Replace = false;
+  SDValue NewOp = PromoteOperand(Op, PVT, Replace);
+  if (NewOp.getNode() == 0)
+    return SDValue();
+
+  if (Replace)
+    ReplaceLoadWithPromotedLoad(Op.getNode(), NewOp.getNode());
+  return DAG.getNode(ISD::SIGN_EXTEND_INREG, dl, NewOp.getValueType(), NewOp,
+                     DAG.getValueType(OldVT));
+}
+
+SDValue DAGCombiner::ZExtPromoteOperand(SDValue Op, EVT PVT) {
+  EVT OldVT = Op.getValueType();
+  DebugLoc dl = Op.getDebugLoc();
+  bool Replace = false;
+  SDValue NewOp = PromoteOperand(Op, PVT, Replace);
+  if (NewOp.getNode() == 0)
+    return SDValue();
+
+  if (Replace)
+    ReplaceLoadWithPromotedLoad(Op.getNode(), NewOp.getNode());
+  return DAG.getZeroExtendInReg(NewOp, dl, OldVT);
+}
+
+/// PromoteIntBinOp - Promote the specified integer binary operation if the
+/// target indicates it is beneficial. e.g. On x86, it's usually better to
+/// promote i16 operations to i32 since i16 instructions are longer.
+SDValue DAGCombiner::PromoteIntBinOp(SDValue Op) {
+  if (!LegalOperations)
+    return SDValue();
+
+  EVT VT = Op.getValueType();
+  if (VT.isVector() || !VT.isInteger())
+    return SDValue();
+
+  // If operation type is 'undesirable', e.g. i16 on x86, consider
+  // promoting it.
+  unsigned Opc = Op.getOpcode();
+  if (TLI.isTypeDesirableForOp(Opc, VT))
+    return SDValue();
+
+  EVT PVT = VT;
+  // Consult target whether it is a good idea to promote this operation and
+  // what's the right type to promote it to.
+  if (TLI.IsDesirableToPromoteOp(Op, PVT)) {
+    assert(PVT != VT && "Don't know what type to promote to!");
+
+    bool Replace0 = false;
+    SDValue N0 = Op.getOperand(0);
+    SDValue NN0 = PromoteOperand(N0, PVT, Replace0);
+    if (NN0.getNode() == 0)
+      return SDValue();
+
+    bool Replace1 = false;
+    SDValue N1 = Op.getOperand(1);
+    SDValue NN1 = PromoteOperand(N1, PVT, Replace1);
+    if (NN1.getNode() == 0)
+      return SDValue();
+
+    AddToWorkList(NN0.getNode());
+    AddToWorkList(NN1.getNode());
+
+    if (Replace0)
+      ReplaceLoadWithPromotedLoad(N0.getNode(), NN0.getNode());
+    if (Replace1)
+      ReplaceLoadWithPromotedLoad(N1.getNode(), NN1.getNode());
+
+    DebugLoc dl = Op.getDebugLoc();
+    return DAG.getNode(ISD::TRUNCATE, dl, VT,
+                       DAG.getNode(Opc, dl, PVT, NN0, NN1));
+  }
+  return SDValue();
+}
+
+/// PromoteIntShiftOp - Promote the specified integer shift operation if the
+/// target indicates it is beneficial. e.g. On x86, it's usually better to
+/// promote i16 operations to i32 since i16 instructions are longer.
+SDValue DAGCombiner::PromoteIntShiftOp(SDValue Op) {
+  if (!LegalOperations)
+    return SDValue();
+
+  EVT VT = Op.getValueType();
+  if (VT.isVector() || !VT.isInteger())
+    return SDValue();
+
+  // If operation type is 'undesirable', e.g. i16 on x86, consider
+  // promoting it.
+  unsigned Opc = Op.getOpcode();
+  if (TLI.isTypeDesirableForOp(Opc, VT))
+    return SDValue();
+
+  EVT PVT = VT;
+  // Consult target whether it is a good idea to promote this operation and
+  // what's the right type to promote it to.
+  if (TLI.IsDesirableToPromoteOp(Op, PVT)) {
+    assert(PVT != VT && "Don't know what type to promote to!");
+
+    bool Replace = false;
+    SDValue N0 = Op.getOperand(0);
+    if (Opc == ISD::SRA)
+      N0 = SExtPromoteOperand(Op.getOperand(0), PVT);
+    else if (Opc == ISD::SRL)
+      N0 = ZExtPromoteOperand(Op.getOperand(0), PVT);
+    else
+      N0 = PromoteOperand(N0, PVT, Replace);
+    if (N0.getNode() == 0)
+      return SDValue();
+
+    AddToWorkList(N0.getNode());
+    if (Replace)
+      ReplaceLoadWithPromotedLoad(Op.getOperand(0).getNode(), N0.getNode());
+
+    DebugLoc dl = Op.getDebugLoc();
+    return DAG.getNode(ISD::TRUNCATE, dl, VT,
+                       DAG.getNode(Opc, dl, PVT, N0, Op.getOperand(1)));
+  }
+  return SDValue();
+}
+
+SDValue DAGCombiner::PromoteExtend(SDValue Op) {
+  if (!LegalOperations)
+    return SDValue();
+
+  EVT VT = Op.getValueType();
+  if (VT.isVector() || !VT.isInteger())
+    return SDValue();
+
+  // If operation type is 'undesirable', e.g. i16 on x86, consider
+  // promoting it.
+  unsigned Opc = Op.getOpcode();
+  if (TLI.isTypeDesirableForOp(Opc, VT))
+    return SDValue();
+
+  EVT PVT = VT;
+  // Consult target whether it is a good idea to promote this operation and
+  // what's the right type to promote it to.
+  if (TLI.IsDesirableToPromoteOp(Op, PVT)) {
+    assert(PVT != VT && "Don't know what type to promote to!");
+    // fold (aext (aext x)) -> (aext x)
+    // fold (aext (zext x)) -> (zext x)
+    // fold (aext (sext x)) -> (sext x)
+    return DAG.getNode(Op.getOpcode(), Op.getDebugLoc(), VT, Op.getOperand(0));
+  }
+  return SDValue();
+}
+
+bool DAGCombiner::PromoteLoad(SDValue Op) {
+  if (!LegalOperations)
+    return false;
+
+  EVT VT = Op.getValueType();
+  if (VT.isVector() || !VT.isInteger())
+    return false;
+
+  // If operation type is 'undesirable', e.g. i16 on x86, consider
+  // promoting it.
+  unsigned Opc = Op.getOpcode();
+  if (TLI.isTypeDesirableForOp(Opc, VT))
+    return false;
+
+  EVT PVT = VT;
+  // Consult target whether it is a good idea to promote this operation and
+  // what's the right type to promote it to.
+  if (TLI.IsDesirableToPromoteOp(Op, PVT)) {
+    assert(PVT != VT && "Don't know what type to promote to!");
+
+    DebugLoc dl = Op.getDebugLoc();
+    SDNode *N = Op.getNode();
+    LoadSDNode *LD = cast<LoadSDNode>(N);
+    ISD::LoadExtType ExtType =
+      ISD::isNON_EXTLoad(LD) ? ISD::EXTLOAD : LD->getExtensionType();
+    SDValue NewLD = DAG.getExtLoad(ExtType, dl, PVT,
+                                   LD->getChain(), LD->getBasePtr(),
+                                   LD->getSrcValue(), LD->getSrcValueOffset(),
+                                   LD->getMemoryVT(), LD->isVolatile(),
+                                   LD->isNonTemporal(), LD->getAlignment());
+    SDValue Result = DAG.getNode(ISD::TRUNCATE, dl, VT, NewLD);
+
+    DEBUG(dbgs() << "\nPromoting ";
+          N->dump(&DAG);
+          dbgs() << "\nTo: ";
+          Result.getNode()->dump(&DAG);
+          dbgs() << '\n');
+    WorkListRemover DeadNodes(*this);
+    DAG.ReplaceAllUsesOfValueWith(SDValue(N, 0), Result, &DeadNodes);
+    DAG.ReplaceAllUsesOfValueWith(SDValue(N, 1), NewLD.getValue(1), &DeadNodes);
+    removeFromWorkList(N);
+    DAG.DeleteNode(N);
+    return true;
+  }
+  return false;
+}
+
+
 //===----------------------------------------------------------------------===//
 //  Main DAG Combiner implementation
 //===----------------------------------------------------------------------===//
@@ -1108,7 +1373,7 @@
                                        N0.getOperand(0).getOperand(1),
                                        N0.getOperand(1)));
 
-  return SDValue();
+  return PromoteIntBinOp(SDValue(N, 0));
 }
 
 SDValue DAGCombiner::visitADDC(SDNode *N) {
@@ -1246,7 +1511,7 @@
                                  VT);
     }
 
-  return SDValue();
+  return PromoteIntBinOp(SDValue(N, 0));
 }
 
 SDValue DAGCombiner::visitMUL(SDNode *N) {
@@ -1339,7 +1604,7 @@
   if (RMUL.getNode() != 0)
     return RMUL;
 
-  return SDValue();
+  return PromoteIntBinOp(SDValue(N, 0));
 }
 
 SDValue DAGCombiner::visitSDIV(SDNode *N) {
@@ -1720,8 +1985,10 @@
   // into a vsetcc.
   EVT Op0VT = N0.getOperand(0).getValueType();
   if ((N0.getOpcode() == ISD::ZERO_EXTEND ||
-       N0.getOpcode() == ISD::ANY_EXTEND  ||
        N0.getOpcode() == ISD::SIGN_EXTEND ||
+       // Avoid infinite looping with PromoteIntBinOp.
+       (N0.getOpcode() == ISD::ANY_EXTEND &&
+        (!LegalTypes || TLI.isTypeDesirableForOp(N->getOpcode(), Op0VT))) ||
        (N0.getOpcode() == ISD::TRUNCATE && TLI.isTypeLegal(Op0VT))) &&
       !VT.isVector() &&
       Op0VT == N1.getOperand(0).getValueType() &&
@@ -1983,7 +2250,7 @@
     }
   }
 
-  return SDValue();
+  return PromoteIntBinOp(SDValue(N, 0));
 }
 
 SDValue DAGCombiner::visitOR(SDNode *N) {
@@ -2109,7 +2376,7 @@
   if (SDNode *Rot = MatchRotate(N0, N1, N->getDebugLoc()))
     return SDValue(Rot, 0);
 
-  return SDValue();
+  return PromoteIntBinOp(SDValue(N, 0));
 }
 
 /// MatchRotateHalf - Match "(X shl/srl V1) & V2" where V2 may not be present.
@@ -2418,7 +2685,7 @@
       SimplifyDemandedBits(SDValue(N, 0)))
     return SDValue(N, 0);
 
-  return SDValue();
+  return PromoteIntBinOp(SDValue(N, 0));
 }
 
 /// visitShiftByConstant - Handle transforms common to the three shifts, when
@@ -2579,7 +2846,13 @@
                        HiBitsMask);
   }
 
-  return N1C ? visitShiftByConstant(N, N1C->getZExtValue()) : SDValue();
+  if (N1C) {
+    SDValue NewSHL = visitShiftByConstant(N, N1C->getZExtValue());
+    if (NewSHL.getNode())
+      return NewSHL;
+  }
+
+  return PromoteIntShiftOp(SDValue(N, 0));
 }
 
 SDValue DAGCombiner::visitSRA(SDNode *N) {
@@ -2693,7 +2966,13 @@
   if (DAG.SignBitIsZero(N0))
     return DAG.getNode(ISD::SRL, N->getDebugLoc(), VT, N0, N1);
 
-  return N1C ? visitShiftByConstant(N, N1C->getZExtValue()) : SDValue();
+  if (N1C) {
+    SDValue NewSRA = visitShiftByConstant(N, N1C->getZExtValue());
+    if (NewSRA.getNode())
+      return NewSRA;
+  }
+
+  return PromoteIntShiftOp(SDValue(N, 0));
 }
 
 SDValue DAGCombiner::visitSRL(SDNode *N) {
@@ -2731,6 +3010,15 @@
     return DAG.getNode(ISD::SRL, N->getDebugLoc(), VT, N0.getOperand(0),
                        DAG.getConstant(c1 + c2, N1.getValueType()));
   }
+  
+  // fold (srl (shl x, c), c) -> (and x, cst2)
+  if (N1C && N0.getOpcode() == ISD::SHL && N0.getOperand(1) == N1 &&
+      N0.getValueSizeInBits() <= 64) {
+    uint64_t ShAmt = N1C->getZExtValue()+64-N0.getValueSizeInBits();
+    return DAG.getNode(ISD::AND, N->getDebugLoc(), VT, N0.getOperand(0),
+                       DAG.getConstant(~0ULL >> ShAmt, VT));
+  }
+  
 
   // fold (srl (anyextend x), c) -> (anyextend (srl x, c))
   if (N1C && N0.getOpcode() == ISD::ANY_EXTEND) {
@@ -2739,10 +3027,12 @@
     if (N1C->getZExtValue() >= SmallVT.getSizeInBits())
       return DAG.getUNDEF(VT);
 
-    SDValue SmallShift = DAG.getNode(ISD::SRL, N0.getDebugLoc(), SmallVT,
-                                     N0.getOperand(0), N1);
-    AddToWorkList(SmallShift.getNode());
-    return DAG.getNode(ISD::ANY_EXTEND, N->getDebugLoc(), VT, SmallShift);
+    if (!LegalTypes || TLI.isTypeDesirableForOp(ISD::SRL, SmallVT)) {
+      SDValue SmallShift = DAG.getNode(ISD::SRL, N0.getDebugLoc(), SmallVT,
+                                       N0.getOperand(0), N1);
+      AddToWorkList(SmallShift.getNode());
+      return DAG.getNode(ISD::ANY_EXTEND, N->getDebugLoc(), VT, SmallShift);
+    }
   }
 
   // fold (srl (sra X, Y), 31) -> (srl X, 31).  This srl only looks at the sign
@@ -2848,7 +3138,7 @@
     }
   }
 
-  return SDValue();
+  return PromoteIntShiftOp(SDValue(N, 0));
 }
 
 SDValue DAGCombiner::visitCTLZ(SDNode *N) {
@@ -3221,8 +3511,9 @@
     }
     
     // sext(setcc x, y, cc) -> (select_cc x, y, -1, 0, cc)
+    unsigned ElementWidth = VT.getScalarType().getSizeInBits();
     SDValue NegOne =
-      DAG.getConstant(APInt::getAllOnesValue(VT.getSizeInBits()), VT);
+      DAG.getConstant(APInt::getAllOnesValue(ElementWidth), VT);
     SDValue SCC =
       SimplifySelectCC(N->getDebugLoc(), N0.getOperand(0), N0.getOperand(1),
                        NegOne, DAG.getConstant(0, VT),
@@ -3245,7 +3536,7 @@
       DAG.SignBitIsZero(N0))
     return DAG.getNode(ISD::ZERO_EXTEND, N->getDebugLoc(), VT, N0);
 
-  return SDValue();
+  return PromoteExtend(SDValue(N, 0));
 }
 
 SDValue DAGCombiner::visitZERO_EXTEND(SDNode *N) {
@@ -3408,7 +3699,7 @@
                                    N0.getOperand(1)));
   }
 
-  return SDValue();
+  return PromoteExtend(SDValue(N, 0));
 }
 
 SDValue DAGCombiner::visitANY_EXTEND(SDNode *N) {
@@ -3544,7 +3835,7 @@
       return SCC;
   }
 
-  return SDValue();
+  return PromoteExtend(SDValue(N, 0));
 }
 
 /// GetDemandedBits - See if the specified operand can be simplified with the
@@ -3624,7 +3915,7 @@
   // Do not generate loads of non-round integer types since these can
   // be expensive (and would be wrong if the type is not byte sized).
   if (isa<LoadSDNode>(N0) && N0.hasOneUse() && ExtVT.isRound() &&
-      cast<LoadSDNode>(N0)->getMemoryVT().getSizeInBits() > EVTBits &&
+      cast<LoadSDNode>(N0)->getMemoryVT().getSizeInBits() >= EVTBits &&
       // Do not change the width of a volatile load.
       !cast<LoadSDNode>(N0)->isVolatile()) {
     LoadSDNode *LN0 = cast<LoadSDNode>(N0);
@@ -3694,7 +3985,8 @@
   // if x is small enough.
   if (N0.getOpcode() == ISD::SIGN_EXTEND || N0.getOpcode() == ISD::ANY_EXTEND) {
     SDValue N00 = N0.getOperand(0);
-    if (N00.getValueType().getScalarType().getSizeInBits() < EVTBits)
+    if (N00.getValueType().getScalarType().getSizeInBits() <= EVTBits &&
+        (!LegalOperations || TLI.isOperationLegal(ISD::SIGN_EXTEND, VT)))
       return DAG.getNode(ISD::SIGN_EXTEND, N->getDebugLoc(), VT, N00, N1);
   }
 
@@ -3779,7 +4071,8 @@
   if (N0.getOpcode() == ISD::TRUNCATE)
     return DAG.getNode(ISD::TRUNCATE, N->getDebugLoc(), VT, N0.getOperand(0));
   // fold (truncate (ext x)) -> (ext x) or (truncate x) or x
-  if (N0.getOpcode() == ISD::ZERO_EXTEND || N0.getOpcode() == ISD::SIGN_EXTEND||
+  if (N0.getOpcode() == ISD::ZERO_EXTEND ||
+      N0.getOpcode() == ISD::SIGN_EXTEND ||
       N0.getOpcode() == ISD::ANY_EXTEND) {
     if (N0.getOperand(0).getValueType().bitsLT(VT))
       // if the source is smaller than the dest, we still need an extend
@@ -3805,7 +4098,9 @@
 
   // fold (truncate (load x)) -> (smaller load x)
   // fold (truncate (srl (load x), c)) -> (smaller load (x+c/evtbits))
-  return ReduceLoadWidth(N);
+  if (!LegalTypes || TLI.isTypeDesirableForOp(N0.getOpcode(), VT))
+    return ReduceLoadWidth(N);
+  return SDValue();
 }
 
 static SDNode *getBuildPairElt(SDNode *N, unsigned i) {
@@ -3949,7 +4244,7 @@
       VT.isInteger() && !VT.isVector()) {
     unsigned OrigXWidth = N0.getOperand(1).getValueType().getSizeInBits();
     EVT IntXVT = EVT::getIntegerVT(*DAG.getContext(), OrigXWidth);
-    if (TLI.isTypeLegal(IntXVT) || !LegalTypes) {
+    if (isTypeLegal(IntXVT)) {
       SDValue X = DAG.getNode(ISD::BIT_CONVERT, N0.getDebugLoc(),
                               IntXVT, N0.getOperand(1));
       AddToWorkList(X.getNode());
@@ -4075,8 +4370,8 @@
         if (Op.getOpcode() == ISD::UNDEF) continue;
         EltIsUndef = false;
 
-        NewBits |= (APInt(cast<ConstantSDNode>(Op)->getAPIntValue()).
-                    zextOrTrunc(SrcBitSize).zext(DstBitSize));
+        NewBits |= APInt(cast<ConstantSDNode>(Op)->getAPIntValue()).
+                   zextOrTrunc(SrcBitSize).zext(DstBitSize);
       }
 
       if (EltIsUndef)
@@ -4464,7 +4759,7 @@
   ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
 
   // fold (fp_round_inreg c1fp) -> c1fp
-  if (N0CFP && (TLI.isTypeLegal(EVT) || !LegalTypes)) {
+  if (N0CFP && isTypeLegal(EVT)) {
     SDValue Round = DAG.getConstantFP(*N0CFP->getConstantFPValue(), EVT);
     return DAG.getNode(ISD::FP_EXTEND, N->getDebugLoc(), VT, Round);
   }
@@ -4605,7 +4900,7 @@
 
   SDNode *Trunc = 0;
   if (N1.getOpcode() == ISD::TRUNCATE && N1.hasOneUse()) {
-    // Look pass truncate.
+    // Look past truncate.
     Trunc = N1.getNode();
     N1 = N1.getOperand(0);
   }
@@ -4676,7 +4971,7 @@
     if (Op0.getOpcode() == Op1.getOpcode()) {
       // Avoid missing important xor optimizations.
       SDValue Tmp = visitXOR(TheXor);
-      if (Tmp.getNode()) {
+      if (Tmp.getNode() && Tmp.getNode() != TheXor) {
         DEBUG(dbgs() << "\nReplacing.8 ";
               TheXor->dump(&DAG);
               dbgs() << "\nWith: ";
@@ -4700,7 +4995,9 @@
           Equal = true;
         }
 
-      EVT SetCCVT = N1.getValueType();
+      SDValue NodeToReplace = Trunc ? SDValue(Trunc, 0) : N1;
+      
+      EVT SetCCVT = NodeToReplace.getValueType();
       if (LegalTypes)
         SetCCVT = TLI.getSetCCResultType(SetCCVT);
       SDValue SetCC = DAG.getSetCC(TheXor->getDebugLoc(),
@@ -4709,9 +5006,9 @@
                                    Equal ? ISD::SETEQ : ISD::SETNE);
       // Replace the uses of XOR with SETCC
       WorkListRemover DeadNodes(*this);
-      DAG.ReplaceAllUsesOfValueWith(N1, SetCC, &DeadNodes);
-      removeFromWorkList(N1.getNode());
-      DAG.DeleteNode(N1.getNode());
+      DAG.ReplaceAllUsesOfValueWith(NodeToReplace, SetCC, &DeadNodes);
+      removeFromWorkList(NodeToReplace.getNode());
+      DAG.DeleteNode(NodeToReplace.getNode());
       return DAG.getNode(ISD::BRCOND, N->getDebugLoc(),
                          MVT::Other, Chain, SetCC, N2);
     }
@@ -5020,18 +5317,6 @@
   SDValue Chain = LD->getChain();
   SDValue Ptr   = LD->getBasePtr();
 
-  // Try to infer better alignment information than the load already has.
-  if (OptLevel != CodeGenOpt::None && LD->isUnindexed()) {
-    if (unsigned Align = DAG.InferPtrAlignment(Ptr)) {
-      if (Align > LD->getAlignment())
-        return DAG.getExtLoad(LD->getExtensionType(), N->getDebugLoc(),
-                              LD->getValueType(0),
-                              Chain, Ptr, LD->getSrcValue(),
-                              LD->getSrcValueOffset(), LD->getMemoryVT(),
-                              LD->isVolatile(), LD->isNonTemporal(), Align);
-    }
-  }
-
   // If load is not volatile and there are no uses of the loaded value (and
   // the updated indexed value in case of indexed loads), change uses of the
   // chain value into uses of the chain input (i.e. delete the dead load).
@@ -5097,6 +5382,18 @@
     }
   }
 
+  // Try to infer better alignment information than the load already has.
+  if (OptLevel != CodeGenOpt::None && LD->isUnindexed()) {
+    if (unsigned Align = DAG.InferPtrAlignment(Ptr)) {
+      if (Align > LD->getAlignment())
+        return DAG.getExtLoad(LD->getExtensionType(), N->getDebugLoc(),
+                              LD->getValueType(0),
+                              Chain, Ptr, LD->getSrcValue(),
+                              LD->getSrcValueOffset(), LD->getMemoryVT(),
+                              LD->isVolatile(), LD->isNonTemporal(), Align);
+    }
+  }
+
   if (CombinerAA) {
     // Walk up chain skipping non-aliasing memory nodes.
     SDValue BetterChain = FindBetterChain(N, Chain);
@@ -5140,9 +5437,141 @@
   if (CombineToPreIndexedLoadStore(N) || CombineToPostIndexedLoadStore(N))
     return SDValue(N, 0);
 
+  if (PromoteLoad(SDValue(N, 0)))
+    return SDValue(N, 0);
   return SDValue();
 }
 
+/// CheckForMaskedLoad - Check to see if V is (and load (ptr), imm), where the
+/// load is having specific bytes cleared out.  If so, return the byte size
+/// being masked out and the shift amount.
+static std::pair<unsigned, unsigned>
+CheckForMaskedLoad(SDValue V, SDValue Ptr, SDValue Chain) {
+  std::pair<unsigned, unsigned> Result(0, 0);
+  
+  // Check for the structure we're looking for.
+  if (V->getOpcode() != ISD::AND ||
+      !isa<ConstantSDNode>(V->getOperand(1)) ||
+      !ISD::isNormalLoad(V->getOperand(0).getNode()))
+    return Result;
+  
+  // Check the chain and pointer.
+  LoadSDNode *LD = cast<LoadSDNode>(V->getOperand(0));
+  if (LD->getBasePtr() != Ptr) return Result;  // Not from same pointer.
+  
+  // The store should be chained directly to the load or be an operand of a
+  // tokenfactor.
+  if (LD == Chain.getNode())
+    ; // ok.
+  else if (Chain->getOpcode() != ISD::TokenFactor)
+    return Result; // Fail.
+  else {
+    bool isOk = false;
+    for (unsigned i = 0, e = Chain->getNumOperands(); i != e; ++i)
+      if (Chain->getOperand(i).getNode() == LD) {
+        isOk = true;
+        break;
+      }
+    if (!isOk) return Result;
+  }
+  
+  // This only handles simple types.
+  if (V.getValueType() != MVT::i16 &&
+      V.getValueType() != MVT::i32 &&
+      V.getValueType() != MVT::i64)
+    return Result;
+
+  // Check the constant mask.  Invert it so that the bits being masked out are
+  // 0 and the bits being kept are 1.  Use getSExtValue so that leading bits
+  // follow the sign bit for uniformity.
+  uint64_t NotMask = ~cast<ConstantSDNode>(V->getOperand(1))->getSExtValue();
+  unsigned NotMaskLZ = CountLeadingZeros_64(NotMask);
+  if (NotMaskLZ & 7) return Result;  // Must be multiple of a byte.
+  unsigned NotMaskTZ = CountTrailingZeros_64(NotMask);
+  if (NotMaskTZ & 7) return Result;  // Must be multiple of a byte.
+  if (NotMaskLZ == 64) return Result;  // All zero mask.
+  
+  // See if we have a continuous run of bits.  If so, we have 0*1+0*
+  if (CountTrailingOnes_64(NotMask >> NotMaskTZ)+NotMaskTZ+NotMaskLZ != 64)
+    return Result;
+
+  // Adjust NotMaskLZ down to be from the actual size of the int instead of i64.
+  if (V.getValueType() != MVT::i64 && NotMaskLZ)
+    NotMaskLZ -= 64-V.getValueSizeInBits();
+  
+  unsigned MaskedBytes = (V.getValueSizeInBits()-NotMaskLZ-NotMaskTZ)/8;
+  switch (MaskedBytes) {
+  case 1: 
+  case 2: 
+  case 4: break;
+  default: return Result; // All one mask, or 5-byte mask.
+  }
+  
+  // Verify that the first bit starts at a multiple of mask so that the access
+  // is aligned the same as the access width.
+  if (NotMaskTZ && NotMaskTZ/8 % MaskedBytes) return Result;
+  
+  Result.first = MaskedBytes;
+  Result.second = NotMaskTZ/8;
+  return Result;
+}
+
+
+/// ShrinkLoadReplaceStoreWithStore - Check to see if IVal is something that
+/// provides a value as specified by MaskInfo.  If so, replace the specified
+/// store with a narrower store of truncated IVal.
+static SDNode *
+ShrinkLoadReplaceStoreWithStore(const std::pair<unsigned, unsigned> &MaskInfo,
+                                SDValue IVal, StoreSDNode *St,
+                                DAGCombiner *DC) {
+  unsigned NumBytes = MaskInfo.first;
+  unsigned ByteShift = MaskInfo.second;
+  SelectionDAG &DAG = DC->getDAG();
+  
+  // Check to see if IVal is all zeros in the part being masked in by the 'or'
+  // that uses this.  If not, this is not a replacement.
+  APInt Mask = ~APInt::getBitsSet(IVal.getValueSizeInBits(),
+                                  ByteShift*8, (ByteShift+NumBytes)*8);
+  if (!DAG.MaskedValueIsZero(IVal, Mask)) return 0;
+  
+  // Check that it is legal on the target to do this.  It is legal if the new
+  // VT we're shrinking to (i8/i16/i32) is legal or we're still before type
+  // legalization.
+  MVT VT = MVT::getIntegerVT(NumBytes*8);
+  if (!DC->isTypeLegal(VT))
+    return 0;
+  
+  // Okay, we can do this!  Replace the 'St' store with a store of IVal that is
+  // shifted by ByteShift and truncated down to NumBytes.
+  if (ByteShift)
+    IVal = DAG.getNode(ISD::SRL, IVal->getDebugLoc(), IVal.getValueType(), IVal,
+                       DAG.getConstant(ByteShift*8, DC->getShiftAmountTy()));
+
+  // Figure out the offset for the store and the alignment of the access.
+  unsigned StOffset;
+  unsigned NewAlign = St->getAlignment();
+
+  if (DAG.getTargetLoweringInfo().isLittleEndian())
+    StOffset = ByteShift;
+  else
+    StOffset = IVal.getValueType().getStoreSize() - ByteShift - NumBytes;
+  
+  SDValue Ptr = St->getBasePtr();
+  if (StOffset) {
+    Ptr = DAG.getNode(ISD::ADD, IVal->getDebugLoc(), Ptr.getValueType(),
+                      Ptr, DAG.getConstant(StOffset, Ptr.getValueType()));
+    NewAlign = MinAlign(NewAlign, StOffset);
+  }
+  
+  // Truncate down to the new size.
+  IVal = DAG.getNode(ISD::TRUNCATE, IVal->getDebugLoc(), VT, IVal);
+  
+  ++OpsNarrowed;
+  return DAG.getStore(St->getChain(), St->getDebugLoc(), IVal, Ptr, 
+                      St->getSrcValue(), St->getSrcValueOffset()+StOffset,
+                      false, false, NewAlign).getNode();
+}
+
 
 /// ReduceLoadOpStoreWidth - Look for sequence of load / op / store where op is
 /// one of 'or', 'xor', and 'and' of immediates. If 'op' is only touching some
@@ -5162,6 +5591,28 @@
     return SDValue();
 
   unsigned Opc = Value.getOpcode();
+  
+  // If this is "store (or X, Y), P" and X is "(and (load P), cst)", where cst
+  // is a byte mask indicating a consecutive number of bytes, check to see if
+  // Y is known to provide just those bytes.  If so, we try to replace the
+  // load + replace + store sequence with a single (narrower) store, which makes
+  // the load dead.
+  if (Opc == ISD::OR) {
+    std::pair<unsigned, unsigned> MaskedLoad;
+    MaskedLoad = CheckForMaskedLoad(Value.getOperand(0), Ptr, Chain);
+    if (MaskedLoad.first)
+      if (SDNode *NewST = ShrinkLoadReplaceStoreWithStore(MaskedLoad,
+                                                  Value.getOperand(1), ST,this))
+        return SDValue(NewST, 0);
+                                           
+    // Or is commutative, so try swapping X and Y.
+    MaskedLoad = CheckForMaskedLoad(Value.getOperand(1), Ptr, Chain);
+    if (MaskedLoad.first)
+      if (SDNode *NewST = ShrinkLoadReplaceStoreWithStore(MaskedLoad,
+                                                  Value.getOperand(0), ST,this))
+        return SDValue(NewST, 0);
+  }
+  
   if ((Opc != ISD::OR && Opc != ISD::XOR && Opc != ISD::AND) ||
       Value.getOperand(1).getOpcode() != ISD::Constant)
     return SDValue();
@@ -5209,8 +5660,8 @@
         PtrOff = (BitWidth + 7 - NewBW) / 8 - PtrOff;
 
       unsigned NewAlign = MinAlign(LD->getAlignment(), PtrOff);
-      if (NewAlign <
-          TLI.getTargetData()->getABITypeAlignment(NewVT.getTypeForEVT(*DAG.getContext())))
+      const Type *NewVTTy = NewVT.getTypeForEVT(*DAG.getContext());
+      if (NewAlign < TLI.getTargetData()->getABITypeAlignment(NewVTTy))
         return SDValue();
 
       SDValue NewPtr = DAG.getNode(ISD::ADD, LD->getDebugLoc(),
@@ -5248,17 +5699,6 @@
   SDValue Value = ST->getValue();
   SDValue Ptr   = ST->getBasePtr();
 
-  // Try to infer better alignment information than the store already has.
-  if (OptLevel != CodeGenOpt::None && ST->isUnindexed()) {
-    if (unsigned Align = DAG.InferPtrAlignment(Ptr)) {
-      if (Align > ST->getAlignment())
-        return DAG.getTruncStore(Chain, N->getDebugLoc(), Value,
-                                 Ptr, ST->getSrcValue(),
-                                 ST->getSrcValueOffset(), ST->getMemoryVT(),
-                                 ST->isVolatile(), ST->isNonTemporal(), Align);
-    }
-  }
-
   // If this is a store of a bit convert, store the input value if the
   // resultant store does not need a higher alignment than the original.
   if (Value.getOpcode() == ISD::BIT_CONVERT && !ST->isTruncatingStore() &&
@@ -5291,8 +5731,7 @@
       case MVT::ppcf128:
         break;
       case MVT::f32:
-        if (((TLI.isTypeLegal(MVT::i32) || !LegalTypes) && !LegalOperations &&
-             !ST->isVolatile()) ||
+        if ((isTypeLegal(MVT::i32) && !LegalOperations && !ST->isVolatile()) ||
             TLI.isOperationLegalOrCustom(ISD::STORE, MVT::i32)) {
           Tmp = DAG.getConstant((uint32_t)CFP->getValueAPF().
                               bitcastToAPInt().getZExtValue(), MVT::i32);
@@ -5303,7 +5742,7 @@
         }
         break;
       case MVT::f64:
-        if (((TLI.isTypeLegal(MVT::i64) || !LegalTypes) && !LegalOperations &&
+        if ((TLI.isTypeLegal(MVT::i64) && !LegalOperations &&
              !ST->isVolatile()) ||
             TLI.isOperationLegalOrCustom(ISD::STORE, MVT::i64)) {
           Tmp = DAG.getConstant(CFP->getValueAPF().bitcastToAPInt().
@@ -5349,6 +5788,17 @@
     }
   }
 
+  // Try to infer better alignment information than the store already has.
+  if (OptLevel != CodeGenOpt::None && ST->isUnindexed()) {
+    if (unsigned Align = DAG.InferPtrAlignment(Ptr)) {
+      if (Align > ST->getAlignment())
+        return DAG.getTruncStore(Chain, N->getDebugLoc(), Value,
+                                 Ptr, ST->getSrcValue(),
+                                 ST->getSrcValueOffset(), ST->getMemoryVT(),
+                                 ST->isVolatile(), ST->isNonTemporal(), Align);
+    }
+  }
+
   if (CombinerAA) {
     // Walk up chain skipping non-aliasing memory nodes.
     SDValue BetterChain = FindBetterChain(N, Chain);
@@ -5409,7 +5859,7 @@
     if (SimplifyDemandedBits(Value,
                              APInt::getLowBitsSet(
                                Value.getValueType().getScalarType().getSizeInBits(),
-                               ST->getMemoryVT().getSizeInBits())))
+                               ST->getMemoryVT().getScalarType().getSizeInBits())))
       return SDValue(N, 0);
   }
 
@@ -5549,7 +5999,7 @@
         InVec = InVec.getOperand(0);
       if (ISD::isNormalLoad(InVec.getNode())) {
         LN0 = cast<LoadSDNode>(InVec);
-        Elt = (Idx < (int)NumElems) ? Idx : Idx - NumElems;
+        Elt = (Idx < (int)NumElems) ? Idx : Idx - (int)NumElems;
       }
     }
 
@@ -5657,7 +6107,7 @@
     }
 
     // Add count and size info.
-    if (!TLI.isTypeLegal(VT) && LegalTypes)
+    if (!isTypeLegal(VT))
       return SDValue();
 
     // Return the new VECTOR_SHUFFLE node.
@@ -6285,7 +6735,7 @@
 /// FindBaseOffset - Return true if base is a frame index, which is known not
 // to alias with anything but itself.  Provides base object and offset as results.
 static bool FindBaseOffset(SDValue Ptr, SDValue &Base, int64_t &Offset,
-                           GlobalValue *&GV, void *&CV) {
+                           const GlobalValue *&GV, void *&CV) {
   // Assume it is a primitive operation.
   Base = Ptr; Offset = 0; GV = 0; CV = 0;
 
@@ -6333,7 +6783,7 @@
   // Gather base node and offset information.
   SDValue Base1, Base2;
   int64_t Offset1, Offset2;
-  GlobalValue *GV1, *GV2;
+  const GlobalValue *GV1, *GV2;
   void *CV1, *CV2;
   bool isFrameIndex1 = FindBaseOffset(Ptr1, Base1, Offset1, GV1, CV1);
   bool isFrameIndex2 = FindBaseOffset(Ptr2, Base2, Offset2, GV2, CV2);
diff --git a/lib/CodeGen/SelectionDAG/FastISel.cpp b/lib/CodeGen/SelectionDAG/FastISel.cpp
index 1d76c7c..8aa37a0 100644
--- a/lib/CodeGen/SelectionDAG/FastISel.cpp
+++ b/lib/CodeGen/SelectionDAG/FastISel.cpp
@@ -1,4 +1,4 @@
-///===-- FastISel.cpp - Implementation of the FastISel class --------------===//
+//===-- FastISel.cpp - Implementation of the FastISel class ---------------===//
 //
 //                     The LLVM Compiler Infrastructure
 //
@@ -47,17 +47,16 @@
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
-#include "llvm/CodeGen/DwarfWriter.h"
 #include "llvm/Analysis/DebugInfo.h"
 #include "llvm/Target/TargetData.h"
 #include "llvm/Target/TargetInstrInfo.h"
 #include "llvm/Target/TargetLowering.h"
 #include "llvm/Target/TargetMachine.h"
-#include "SelectionDAGBuilder.h"
+#include "llvm/Support/ErrorHandling.h"
 #include "FunctionLoweringInfo.h"
 using namespace llvm;
 
-unsigned FastISel::getRegForValue(Value *V) {
+unsigned FastISel::getRegForValue(const Value *V) {
   EVT RealVT = TLI.getValueType(V->getType(), /*AllowUnknown=*/true);
   // Don't handle non-simple values in FastISel.
   if (!RealVT.isSimple())
@@ -85,7 +84,7 @@
   if (Reg != 0)
     return Reg;
 
-  if (ConstantInt *CI = dyn_cast<ConstantInt>(V)) {
+  if (const ConstantInt *CI = dyn_cast<ConstantInt>(V)) {
     if (CI->getValue().getActiveBits() <= 64)
       Reg = FastEmit_i(VT, VT, ISD::Constant, CI->getZExtValue());
   } else if (isa<AllocaInst>(V)) {
@@ -95,10 +94,12 @@
     // local-CSE'd with actual integer zeros.
     Reg =
       getRegForValue(Constant::getNullValue(TD.getIntPtrType(V->getContext())));
-  } else if (ConstantFP *CF = dyn_cast<ConstantFP>(V)) {
+  } else if (const ConstantFP *CF = dyn_cast<ConstantFP>(V)) {
+    // Try to emit the constant directly.
     Reg = FastEmit_f(VT, VT, ISD::ConstantFP, CF);
 
     if (!Reg) {
+      // Try to emit the constant by using an integer constant with a cast.
       const APFloat &Flt = CF->getValueAPF();
       EVT IntVT = TLI.getPointerTy();
 
@@ -116,9 +117,9 @@
           Reg = FastEmit_r(IntVT.getSimpleVT(), VT, ISD::SINT_TO_FP, IntegerReg);
       }
     }
-  } else if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V)) {
-    if (!SelectOperator(CE, CE->getOpcode())) return 0;
-    Reg = LocalValueMap[CE];
+  } else if (const Operator *Op = dyn_cast<Operator>(V)) {
+    if (!SelectOperator(Op, Op->getOpcode())) return 0;
+    Reg = LocalValueMap[Op];
   } else if (isa<UndefValue>(V)) {
     Reg = createResultReg(TLI.getRegClassFor(VT));
     BuildMI(MBB, DL, TII.get(TargetOpcode::IMPLICIT_DEF), Reg);
@@ -136,7 +137,7 @@
   return Reg;
 }
 
-unsigned FastISel::lookUpRegForValue(Value *V) {
+unsigned FastISel::lookUpRegForValue(const Value *V) {
   // Look up the value to see if we already have a register for it. We
   // cache values defined by Instructions across blocks, and other values
   // only locally. This is because Instructions already have the SSA
@@ -152,7 +153,7 @@
 /// NOTE: This is only necessary because we might select a block that uses
 /// a value before we select the block that defines the value.  It might be
 /// possible to fix this by selecting blocks in reverse postorder.
-unsigned FastISel::UpdateValueMap(Value* I, unsigned Reg) {
+unsigned FastISel::UpdateValueMap(const Value *I, unsigned Reg) {
   if (!isa<Instruction>(I)) {
     LocalValueMap[I] = Reg;
     return Reg;
@@ -169,7 +170,7 @@
   return AssignedReg;
 }
 
-unsigned FastISel::getRegForGEPIndex(Value *Idx) {
+unsigned FastISel::getRegForGEPIndex(const Value *Idx) {
   unsigned IdxN = getRegForValue(Idx);
   if (IdxN == 0)
     // Unhandled operand. Halt "fast" selection and bail.
@@ -188,7 +189,7 @@
 /// SelectBinaryOp - Select and emit code for a binary operator instruction,
 /// which has an opcode which directly corresponds to the given ISD opcode.
 ///
-bool FastISel::SelectBinaryOp(User *I, unsigned ISDOpcode) {
+bool FastISel::SelectBinaryOp(const User *I, unsigned ISDOpcode) {
   EVT VT = EVT::getEVT(I->getType(), /*HandleUnknown=*/true);
   if (VT == MVT::Other || !VT.isSimple())
     // Unhandled type. Halt "fast" selection and bail.
@@ -254,7 +255,7 @@
   return true;
 }
 
-bool FastISel::SelectGetElementPtr(User *I) {
+bool FastISel::SelectGetElementPtr(const User *I) {
   unsigned N = getRegForValue(I->getOperand(0));
   if (N == 0)
     // Unhandled operand. Halt "fast" selection and bail.
@@ -262,9 +263,9 @@
 
   const Type *Ty = I->getOperand(0)->getType();
   MVT VT = TLI.getPointerTy();
-  for (GetElementPtrInst::op_iterator OI = I->op_begin()+1, E = I->op_end();
-       OI != E; ++OI) {
-    Value *Idx = *OI;
+  for (GetElementPtrInst::const_op_iterator OI = I->op_begin()+1,
+       E = I->op_end(); OI != E; ++OI) {
+    const Value *Idx = *OI;
     if (const StructType *StTy = dyn_cast<StructType>(Ty)) {
       unsigned Field = cast<ConstantInt>(Idx)->getZExtValue();
       if (Field) {
@@ -282,7 +283,7 @@
       Ty = cast<SequentialType>(Ty)->getElementType();
 
       // If this is a constant subscript, handle it quickly.
-      if (ConstantInt *CI = dyn_cast<ConstantInt>(Idx)) {
+      if (const ConstantInt *CI = dyn_cast<ConstantInt>(Idx)) {
         if (CI->getZExtValue() == 0) continue;
         uint64_t Offs = 
           TD.getTypeAllocSize(Ty)*cast<ConstantInt>(CI)->getSExtValue();
@@ -318,52 +319,56 @@
   return true;
 }
 
-bool FastISel::SelectCall(User *I) {
-  Function *F = cast<CallInst>(I)->getCalledFunction();
+bool FastISel::SelectCall(const User *I) {
+  const Function *F = cast<CallInst>(I)->getCalledFunction();
   if (!F) return false;
 
+  // Handle selected intrinsic function calls.
   unsigned IID = F->getIntrinsicID();
   switch (IID) {
   default: break;
   case Intrinsic::dbg_declare: {
-    DbgDeclareInst *DI = cast<DbgDeclareInst>(I);
-    if (!DIDescriptor::ValidDebugInfo(DI->getVariable(), CodeGenOpt::None)||!DW
-        || !DW->ShouldEmitDwarfDebug())
+    const DbgDeclareInst *DI = cast<DbgDeclareInst>(I);
+    if (!DIDescriptor::ValidDebugInfo(DI->getVariable(), CodeGenOpt::None) ||
+        !MF.getMMI().hasDebugInfo())
       return true;
 
-    Value *Address = DI->getAddress();
+    const Value *Address = DI->getAddress();
     if (!Address)
       return true;
-    AllocaInst *AI = dyn_cast<AllocaInst>(Address);
+    if (isa<UndefValue>(Address))
+      return true;
+    const AllocaInst *AI = dyn_cast<AllocaInst>(Address);
     // Don't handle byval struct arguments or VLAs, for example.
+    // Note that if we have a byval struct argument, fast ISel is turned off;
+    // those are handled in SelectionDAGBuilder.
     if (!AI) break;
     DenseMap<const AllocaInst*, int>::iterator SI =
       StaticAllocaMap.find(AI);
     if (SI == StaticAllocaMap.end()) break; // VLAs.
     int FI = SI->second;
-    if (MMI) {
-      if (MDNode *Dbg = DI->getMetadata("dbg"))
-        MMI->setVariableDbgInfo(DI->getVariable(), FI, Dbg);
-    }
+    if (!DI->getDebugLoc().isUnknown())
+      MF.getMMI().setVariableDbgInfo(DI->getVariable(), FI, DI->getDebugLoc());
+
     // Building the map above is target independent.  Generating DBG_VALUE
     // inline is target dependent; do this now.
     (void)TargetSelectInstruction(cast<Instruction>(I));
     return true;
   }
   case Intrinsic::dbg_value: {
-    // This requires target support, but right now X86 is the only Fast target.
-    DbgValueInst *DI = cast<DbgValueInst>(I);
+    // This form of DBG_VALUE is target-independent.
+    const DbgValueInst *DI = cast<DbgValueInst>(I);
     const TargetInstrDesc &II = TII.get(TargetOpcode::DBG_VALUE);
-    Value *V = DI->getValue();
+    const Value *V = DI->getValue();
     if (!V) {
       // Currently the optimizer can produce this; insert an undef to
       // help debugging.  Probably the optimizer should not do this.
       BuildMI(MBB, DL, II).addReg(0U).addImm(DI->getOffset()).
                                      addMetadata(DI->getVariable());
-    } else if (ConstantInt *CI = dyn_cast<ConstantInt>(V)) {
+    } else if (const ConstantInt *CI = dyn_cast<ConstantInt>(V)) {
       BuildMI(MBB, DL, II).addImm(CI->getZExtValue()).addImm(DI->getOffset()).
                                      addMetadata(DI->getVariable());
-    } else if (ConstantFP *CF = dyn_cast<ConstantFP>(V)) {
+    } else if (const ConstantFP *CF = dyn_cast<ConstantFP>(V)) {
       BuildMI(MBB, DL, II).addFPImm(CF).addImm(DI->getOffset()).
                                      addMetadata(DI->getVariable());
     } else if (unsigned Reg = lookUpRegForValue(V)) {
@@ -402,54 +407,51 @@
     switch (TLI.getOperationAction(ISD::EHSELECTION, VT)) {
     default: break;
     case TargetLowering::Expand: {
-      if (MMI) {
-        if (MBB->isLandingPad())
-          AddCatchInfo(*cast<CallInst>(I), MMI, MBB);
-        else {
+      if (MBB->isLandingPad())
+        AddCatchInfo(*cast<CallInst>(I), &MF.getMMI(), MBB);
+      else {
 #ifndef NDEBUG
-          CatchInfoLost.insert(cast<CallInst>(I));
+        CatchInfoLost.insert(cast<CallInst>(I));
 #endif
-          // FIXME: Mark exception selector register as live in.  Hack for PR1508.
-          unsigned Reg = TLI.getExceptionSelectorRegister();
-          if (Reg) MBB->addLiveIn(Reg);
-        }
-
+        // FIXME: Mark exception selector register as live in.  Hack for PR1508.
         unsigned Reg = TLI.getExceptionSelectorRegister();
-        EVT SrcVT = TLI.getPointerTy();
-        const TargetRegisterClass *RC = TLI.getRegClassFor(SrcVT);
-        unsigned ResultReg = createResultReg(RC);
-        bool InsertedCopy = TII.copyRegToReg(*MBB, MBB->end(), ResultReg, Reg,
-                                             RC, RC);
-        assert(InsertedCopy && "Can't copy address registers!");
-        InsertedCopy = InsertedCopy;
-
-        // Cast the register to the type of the selector.
-        if (SrcVT.bitsGT(MVT::i32))
-          ResultReg = FastEmit_r(SrcVT.getSimpleVT(), MVT::i32, ISD::TRUNCATE,
-                                 ResultReg);
-        else if (SrcVT.bitsLT(MVT::i32))
-          ResultReg = FastEmit_r(SrcVT.getSimpleVT(), MVT::i32,
-                                 ISD::SIGN_EXTEND, ResultReg);
-        if (ResultReg == 0)
-          // Unhandled operand. Halt "fast" selection and bail.
-          return false;
-
-        UpdateValueMap(I, ResultReg);
-      } else {
-        unsigned ResultReg =
-          getRegForValue(Constant::getNullValue(I->getType()));
-        UpdateValueMap(I, ResultReg);
+        if (Reg) MBB->addLiveIn(Reg);
       }
+
+      unsigned Reg = TLI.getExceptionSelectorRegister();
+      EVT SrcVT = TLI.getPointerTy();
+      const TargetRegisterClass *RC = TLI.getRegClassFor(SrcVT);
+      unsigned ResultReg = createResultReg(RC);
+      bool InsertedCopy = TII.copyRegToReg(*MBB, MBB->end(), ResultReg, Reg,
+                                           RC, RC);
+      assert(InsertedCopy && "Can't copy address registers!");
+      InsertedCopy = InsertedCopy;
+
+      // Cast the register to the type of the selector.
+      if (SrcVT.bitsGT(MVT::i32))
+        ResultReg = FastEmit_r(SrcVT.getSimpleVT(), MVT::i32, ISD::TRUNCATE,
+                               ResultReg);
+      else if (SrcVT.bitsLT(MVT::i32))
+        ResultReg = FastEmit_r(SrcVT.getSimpleVT(), MVT::i32,
+                               ISD::SIGN_EXTEND, ResultReg);
+      if (ResultReg == 0)
+        // Unhandled operand. Halt "fast" selection and bail.
+        return false;
+
+      UpdateValueMap(I, ResultReg);
+
       return true;
     }
     }
     break;
   }
   }
+
+  // An arbitrary call. Bail.
   return false;
 }
 
-bool FastISel::SelectCast(User *I, unsigned Opcode) {
+bool FastISel::SelectCast(const User *I, unsigned Opcode) {
   EVT SrcVT = TLI.getValueType(I->getOperand(0)->getType());
   EVT DstVT = TLI.getValueType(I->getType());
     
@@ -501,7 +503,7 @@
   return true;
 }
 
-bool FastISel::SelectBitCast(User *I) {
+bool FastISel::SelectBitCast(const User *I) {
   // If the bitcast doesn't change the type, just use the operand value.
   if (I->getType() == I->getOperand(0)->getType()) {
     unsigned Reg = getRegForValue(I->getOperand(0));
@@ -552,15 +554,28 @@
 }
 
 bool
-FastISel::SelectInstruction(Instruction *I) {
+FastISel::SelectInstruction(const Instruction *I) {
+  // Just before the terminator instruction, insert instructions to
+  // feed PHI nodes in successor blocks.
+  if (isa<TerminatorInst>(I))
+    if (!HandlePHINodesInSuccessorBlocks(I->getParent()))
+      return false;
+
+  DL = I->getDebugLoc();
+
   // First, try doing target-independent selection.
-  if (SelectOperator(I, I->getOpcode()))
+  if (SelectOperator(I, I->getOpcode())) {
+    DL = DebugLoc();
     return true;
+  }
 
   // Next, try calling the target to attempt to handle the instruction.
-  if (TargetSelectInstruction(I))
+  if (TargetSelectInstruction(I)) {
+    DL = DebugLoc();
     return true;
+  }
 
+  DL = DebugLoc();
   return false;
 }
 
@@ -581,7 +596,7 @@
 /// SelectFNeg - Emit an FNeg operation.
 ///
 bool
-FastISel::SelectFNeg(User *I) {
+FastISel::SelectFNeg(const User *I) {
   unsigned OpReg = getRegForValue(BinaryOperator::getFNegArgument(I));
   if (OpReg == 0) return false;
 
@@ -622,7 +637,7 @@
 }
 
 bool
-FastISel::SelectOperator(User *I, unsigned Opcode) {
+FastISel::SelectOperator(const User *I, unsigned Opcode) {
   switch (Opcode) {
   case Instruction::Add:
     return SelectBinaryOp(I, ISD::ADD);
@@ -668,10 +683,10 @@
     return SelectGetElementPtr(I);
 
   case Instruction::Br: {
-    BranchInst *BI = cast<BranchInst>(I);
+    const BranchInst *BI = cast<BranchInst>(I);
 
     if (BI->isUnconditional()) {
-      BasicBlock *LLVMSucc = BI->getSuccessor(0);
+      const BasicBlock *LLVMSucc = BI->getSuccessor(0);
       MachineBasicBlock *MSucc = MBBMap[LLVMSucc];
       FastEmitBranch(MSucc);
       return true;
@@ -686,10 +701,6 @@
     // Nothing to emit.
     return true;
 
-  case Instruction::PHI:
-    // PHI nodes are already emitted.
-    return true;
-
   case Instruction::Alloca:
     // FunctionLowering has the static-sized case covered.
     if (StaticAllocaMap.count(cast<AllocaInst>(I)))
@@ -729,6 +740,9 @@
     return true;
   }
 
+  case Instruction::PHI:
+    llvm_unreachable("FastISel shouldn't visit PHI nodes!");
+
   default:
     // Unhandled instruction. Halt "fast" selection and bail.
     return false;
@@ -736,25 +750,23 @@
 }
 
 FastISel::FastISel(MachineFunction &mf,
-                   MachineModuleInfo *mmi,
-                   DwarfWriter *dw,
                    DenseMap<const Value *, unsigned> &vm,
                    DenseMap<const BasicBlock *, MachineBasicBlock *> &bm,
-                   DenseMap<const AllocaInst *, int> &am
+                   DenseMap<const AllocaInst *, int> &am,
+                   std::vector<std::pair<MachineInstr*, unsigned> > &pn
 #ifndef NDEBUG
-                   , SmallSet<Instruction*, 8> &cil
+                   , SmallSet<const Instruction *, 8> &cil
 #endif
                    )
   : MBB(0),
     ValueMap(vm),
     MBBMap(bm),
     StaticAllocaMap(am),
+    PHINodesToUpdate(pn),
 #ifndef NDEBUG
     CatchInfoLost(cil),
 #endif
     MF(mf),
-    MMI(mmi),
-    DW(dw),
     MRI(MF.getRegInfo()),
     MFI(*MF.getFrameInfo()),
     MCP(*MF.getConstantPool()),
@@ -787,7 +799,7 @@
 }
 
 unsigned FastISel::FastEmit_f(MVT, MVT,
-                              unsigned, ConstantFP * /*FPImm*/) {
+                              unsigned, const ConstantFP * /*FPImm*/) {
   return 0;
 }
 
@@ -799,7 +811,7 @@
 
 unsigned FastISel::FastEmit_rf(MVT, MVT,
                                unsigned, unsigned /*Op0*/,
-                               ConstantFP * /*FPImm*/) {
+                               const ConstantFP * /*FPImm*/) {
   return 0;
 }
 
@@ -832,7 +844,7 @@
 /// FastEmit_rf. If that fails, it materializes the immediate into a register
 /// and try FastEmit_rr instead.
 unsigned FastISel::FastEmit_rf_(MVT VT, unsigned Opcode,
-                                unsigned Op0, ConstantFP *FPImm,
+                                unsigned Op0, const ConstantFP *FPImm,
                                 MVT ImmType) {
   // First check if immediate type is legal. If not, we can't use the rf form.
   unsigned ResultReg = FastEmit_rf(VT, VT, Opcode, Op0, FPImm);
@@ -942,7 +954,7 @@
 
 unsigned FastISel::FastEmitInst_rf(unsigned MachineInstOpcode,
                                    const TargetRegisterClass *RC,
-                                   unsigned Op0, ConstantFP *FPImm) {
+                                   unsigned Op0, const ConstantFP *FPImm) {
   unsigned ResultReg = createResultReg(RC);
   const TargetInstrDesc &II = TII.get(MachineInstOpcode);
 
@@ -1018,3 +1030,67 @@
 unsigned FastISel::FastEmitZExtFromI1(MVT VT, unsigned Op) {
   return FastEmit_ri(VT, VT, ISD::AND, Op, 1);
 }
+
+/// HandlePHINodesInSuccessorBlocks - Handle PHI nodes in successor blocks.
+/// Emit code to ensure constants are copied into registers when needed.
+/// Remember the virtual registers that need to be added to the Machine PHI
+/// nodes as input.  We cannot just directly add them, because expansion
+/// might result in multiple MBB's for one BB.  As such, the start of the
+/// BB might correspond to a different MBB than the end.
+bool FastISel::HandlePHINodesInSuccessorBlocks(const BasicBlock *LLVMBB) {
+  const TerminatorInst *TI = LLVMBB->getTerminator();
+
+  SmallPtrSet<MachineBasicBlock *, 4> SuccsHandled;
+  unsigned OrigNumPHINodesToUpdate = PHINodesToUpdate.size();
+
+  // Check successor nodes' PHI nodes that expect a constant to be available
+  // from this block.
+  for (unsigned succ = 0, e = TI->getNumSuccessors(); succ != e; ++succ) {
+    const BasicBlock *SuccBB = TI->getSuccessor(succ);
+    if (!isa<PHINode>(SuccBB->begin())) continue;
+    MachineBasicBlock *SuccMBB = MBBMap[SuccBB];
+
+    // If this terminator has multiple identical successors (common for
+    // switches), only handle each succ once.
+    if (!SuccsHandled.insert(SuccMBB)) continue;
+
+    MachineBasicBlock::iterator MBBI = SuccMBB->begin();
+
+    // At this point we know that there is a 1-1 correspondence between LLVM PHI
+    // nodes and Machine PHI nodes, but the incoming operands have not been
+    // emitted yet.
+    for (BasicBlock::const_iterator I = SuccBB->begin();
+         const PHINode *PN = dyn_cast<PHINode>(I); ++I) {
+      // Ignore dead phi's.
+      if (PN->use_empty()) continue;
+
+      // Only handle legal types. Two interesting things to note here. First,
+      // by bailing out early, we may leave behind some dead instructions,
+      // since SelectionDAG's HandlePHINodesInSuccessorBlocks will insert its
+      // own moves. Second, this check is necessary becuase FastISel doesn't
+      // use CreateRegForValue to create registers, so it always creates
+      // exactly one register for each non-void instruction.
+      EVT VT = TLI.getValueType(PN->getType(), /*AllowUnknown=*/true);
+      if (VT == MVT::Other || !TLI.isTypeLegal(VT)) {
+        // Promote MVT::i1.
+        if (VT == MVT::i1)
+          VT = TLI.getTypeToTransformTo(LLVMBB->getContext(), VT);
+        else {
+          PHINodesToUpdate.resize(OrigNumPHINodesToUpdate);
+          return false;
+        }
+      }
+
+      const Value *PHIOp = PN->getIncomingValueForBlock(LLVMBB);
+
+      unsigned Reg = getRegForValue(PHIOp);
+      if (Reg == 0) {
+        PHINodesToUpdate.resize(OrigNumPHINodesToUpdate);
+        return false;
+      }
+      PHINodesToUpdate.push_back(std::make_pair(MBBI++, Reg));
+    }
+  }
+
+  return true;
+}
diff --git a/lib/CodeGen/SelectionDAG/FunctionLoweringInfo.cpp b/lib/CodeGen/SelectionDAG/FunctionLoweringInfo.cpp
index 50f4c32..8fe619e 100644
--- a/lib/CodeGen/SelectionDAG/FunctionLoweringInfo.cpp
+++ b/lib/CodeGen/SelectionDAG/FunctionLoweringInfo.cpp
@@ -14,19 +14,18 @@
 
 #define DEBUG_TYPE "function-lowering-info"
 #include "FunctionLoweringInfo.h"
-#include "llvm/CallingConv.h"
 #include "llvm/DerivedTypes.h"
 #include "llvm/Function.h"
 #include "llvm/Instructions.h"
 #include "llvm/IntrinsicInst.h"
 #include "llvm/LLVMContext.h"
 #include "llvm/Module.h"
+#include "llvm/CodeGen/Analysis.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
-#include "llvm/Analysis/DebugInfo.h"
 #include "llvm/Target/TargetRegisterInfo.h"
 #include "llvm/Target/TargetData.h"
 #include "llvm/Target/TargetFrameInfo.h"
@@ -34,99 +33,21 @@
 #include "llvm/Target/TargetIntrinsicInfo.h"
 #include "llvm/Target/TargetLowering.h"
 #include "llvm/Target/TargetOptions.h"
-#include "llvm/Support/Compiler.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/MathExtras.h"
-#include "llvm/Support/raw_ostream.h"
 #include <algorithm>
 using namespace llvm;
 
-/// ComputeLinearIndex - Given an LLVM IR aggregate type and a sequence
-/// of insertvalue or extractvalue indices that identify a member, return
-/// the linearized index of the start of the member.
-///
-unsigned llvm::ComputeLinearIndex(const TargetLowering &TLI, const Type *Ty,
-                                  const unsigned *Indices,
-                                  const unsigned *IndicesEnd,
-                                  unsigned CurIndex) {
-  // Base case: We're done.
-  if (Indices && Indices == IndicesEnd)
-    return CurIndex;
-
-  // Given a struct type, recursively traverse the elements.
-  if (const StructType *STy = dyn_cast<StructType>(Ty)) {
-    for (StructType::element_iterator EB = STy->element_begin(),
-                                      EI = EB,
-                                      EE = STy->element_end();
-        EI != EE; ++EI) {
-      if (Indices && *Indices == unsigned(EI - EB))
-        return ComputeLinearIndex(TLI, *EI, Indices+1, IndicesEnd, CurIndex);
-      CurIndex = ComputeLinearIndex(TLI, *EI, 0, 0, CurIndex);
-    }
-    return CurIndex;
-  }
-  // Given an array type, recursively traverse the elements.
-  else if (const ArrayType *ATy = dyn_cast<ArrayType>(Ty)) {
-    const Type *EltTy = ATy->getElementType();
-    for (unsigned i = 0, e = ATy->getNumElements(); i != e; ++i) {
-      if (Indices && *Indices == i)
-        return ComputeLinearIndex(TLI, EltTy, Indices+1, IndicesEnd, CurIndex);
-      CurIndex = ComputeLinearIndex(TLI, EltTy, 0, 0, CurIndex);
-    }
-    return CurIndex;
-  }
-  // We haven't found the type we're looking for, so keep searching.
-  return CurIndex + 1;
-}
-
-/// ComputeValueVTs - Given an LLVM IR type, compute a sequence of
-/// EVTs that represent all the individual underlying
-/// non-aggregate types that comprise it.
-///
-/// If Offsets is non-null, it points to a vector to be filled in
-/// with the in-memory offsets of each of the individual values.
-///
-void llvm::ComputeValueVTs(const TargetLowering &TLI, const Type *Ty,
-                           SmallVectorImpl<EVT> &ValueVTs,
-                           SmallVectorImpl<uint64_t> *Offsets,
-                           uint64_t StartingOffset) {
-  // Given a struct type, recursively traverse the elements.
-  if (const StructType *STy = dyn_cast<StructType>(Ty)) {
-    const StructLayout *SL = TLI.getTargetData()->getStructLayout(STy);
-    for (StructType::element_iterator EB = STy->element_begin(),
-                                      EI = EB,
-                                      EE = STy->element_end();
-         EI != EE; ++EI)
-      ComputeValueVTs(TLI, *EI, ValueVTs, Offsets,
-                      StartingOffset + SL->getElementOffset(EI - EB));
-    return;
-  }
-  // Given an array type, recursively traverse the elements.
-  if (const ArrayType *ATy = dyn_cast<ArrayType>(Ty)) {
-    const Type *EltTy = ATy->getElementType();
-    uint64_t EltSize = TLI.getTargetData()->getTypeAllocSize(EltTy);
-    for (unsigned i = 0, e = ATy->getNumElements(); i != e; ++i)
-      ComputeValueVTs(TLI, EltTy, ValueVTs, Offsets,
-                      StartingOffset + i * EltSize);
-    return;
-  }
-  // Interpret void as zero return values.
-  if (Ty->isVoidTy())
-    return;
-  // Base case: we can get an EVT for this LLVM IR type.
-  ValueVTs.push_back(TLI.getValueType(Ty));
-  if (Offsets)
-    Offsets->push_back(StartingOffset);
-}
-
 /// isUsedOutsideOfDefiningBlock - Return true if this instruction is used by
 /// PHI nodes or outside of the basic block that defines it, or used by a
 /// switch or atomic instruction, which may expand to multiple basic blocks.
-static bool isUsedOutsideOfDefiningBlock(Instruction *I) {
+static bool isUsedOutsideOfDefiningBlock(const Instruction *I) {
+  if (I->use_empty()) return false;
   if (isa<PHINode>(I)) return true;
-  BasicBlock *BB = I->getParent();
-  for (Value::use_iterator UI = I->use_begin(), E = I->use_end(); UI != E; ++UI)
+  const BasicBlock *BB = I->getParent();
+  for (Value::const_use_iterator UI = I->use_begin(), E = I->use_end();
+        UI != E; ++UI)
     if (cast<Instruction>(*UI)->getParent() != BB || isa<PHINode>(*UI))
       return true;
   return false;
@@ -135,7 +56,7 @@
 /// isOnlyUsedInEntryBlock - If the specified argument is only used in the
 /// entry block, return true.  This includes arguments used by switches, since
 /// the switch may expand into multiple basic blocks.
-static bool isOnlyUsedInEntryBlock(Argument *A, bool EnableFastISel) {
+static bool isOnlyUsedInEntryBlock(const Argument *A, bool EnableFastISel) {
   // With FastISel active, we may be splitting blocks, so force creation
   // of virtual registers for all non-dead arguments.
   // Don't force virtual registers for byval arguments though, because
@@ -143,18 +64,19 @@
   if (EnableFastISel && !A->hasByValAttr())
     return A->use_empty();
 
-  BasicBlock *Entry = A->getParent()->begin();
-  for (Value::use_iterator UI = A->use_begin(), E = A->use_end(); UI != E; ++UI)
+  const BasicBlock *Entry = A->getParent()->begin();
+  for (Value::const_use_iterator UI = A->use_begin(), E = A->use_end();
+       UI != E; ++UI)
     if (cast<Instruction>(*UI)->getParent() != Entry || isa<SwitchInst>(*UI))
       return false;  // Use not in entry block.
   return true;
 }
 
-FunctionLoweringInfo::FunctionLoweringInfo(TargetLowering &tli)
+FunctionLoweringInfo::FunctionLoweringInfo(const TargetLowering &tli)
   : TLI(tli) {
 }
 
-void FunctionLoweringInfo::set(Function &fn, MachineFunction &mf,
+void FunctionLoweringInfo::set(const Function &fn, MachineFunction &mf,
                                bool EnableFastISel) {
   Fn = &fn;
   MF = &mf;
@@ -162,7 +84,7 @@
 
   // Create a vreg for each argument register that is not dead and is used
   // outside of the entry block for the function.
-  for (Function::arg_iterator AI = Fn->arg_begin(), E = Fn->arg_end();
+  for (Function::const_arg_iterator AI = Fn->arg_begin(), E = Fn->arg_end();
        AI != E; ++AI)
     if (!isOnlyUsedInEntryBlock(AI, EnableFastISel))
       InitializeRegForValue(AI);
@@ -170,10 +92,10 @@
   // Initialize the mapping of values to registers.  This is only set up for
   // instruction values that are used outside of the block that defines
   // them.
-  Function::iterator BB = Fn->begin(), EB = Fn->end();
-  for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
-    if (AllocaInst *AI = dyn_cast<AllocaInst>(I))
-      if (ConstantInt *CUI = dyn_cast<ConstantInt>(AI->getArraySize())) {
+  Function::const_iterator BB = Fn->begin(), EB = Fn->end();
+  for (BasicBlock::const_iterator I = BB->begin(), E = BB->end(); I != E; ++I)
+    if (const AllocaInst *AI = dyn_cast<AllocaInst>(I))
+      if (const ConstantInt *CUI = dyn_cast<ConstantInt>(AI->getArraySize())) {
         const Type *Ty = AI->getAllocatedType();
         uint64_t TySize = TLI.getTargetData()->getTypeAllocSize(Ty);
         unsigned Align =
@@ -187,8 +109,8 @@
       }
 
   for (; BB != EB; ++BB)
-    for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
-      if (!I->use_empty() && isUsedOutsideOfDefiningBlock(I))
+    for (BasicBlock::const_iterator I = BB->begin(), E = BB->end(); I != E; ++I)
+      if (isUsedOutsideOfDefiningBlock(I))
         if (!isa<AllocaInst>(I) ||
             !StaticAllocaMap.count(cast<AllocaInst>(I)))
           InitializeRegForValue(I);
@@ -196,7 +118,7 @@
   // Create an initial MachineBasicBlock for each LLVM BasicBlock in F.  This
   // also creates the initial PHI MachineInstrs, though none of the input
   // operands are populated.
-  for (BB = Fn->begin(), EB = Fn->end(); BB != EB; ++BB) {
+  for (BB = Fn->begin(); BB != EB; ++BB) {
     MachineBasicBlock *MBB = mf.CreateMachineBasicBlock(BB);
     MBBMap[BB] = MBB;
     MF->push_back(MBB);
@@ -209,14 +131,11 @@
 
     // Create Machine PHI nodes for LLVM PHI nodes, lowering them as
     // appropriate.
-    PHINode *PN;
-    DebugLoc DL;
-    for (BasicBlock::iterator
-           I = BB->begin(), E = BB->end(); I != E; ++I) {
+    for (BasicBlock::const_iterator I = BB->begin();
+         const PHINode *PN = dyn_cast<PHINode>(I); ++I) {
+      if (PN->use_empty()) continue;
 
-      PN = dyn_cast<PHINode>(I);
-      if (!PN || PN->use_empty()) continue;
-
+      DebugLoc DL = PN->getDebugLoc();
       unsigned PHIReg = ValueMap[PN];
       assert(PHIReg && "PHI node does not have an assigned virtual register!");
 
@@ -232,12 +151,20 @@
       }
     }
   }
+
+  // Mark landing pad blocks.
+  for (BB = Fn->begin(); BB != EB; ++BB)
+    if (const InvokeInst *Invoke = dyn_cast<InvokeInst>(BB->getTerminator()))
+      MBBMap[Invoke->getSuccessor(1)]->setIsLandingPad();
 }
 
 /// clear - Clear out all the function-specific state. This returns this
 /// FunctionLoweringInfo to an empty state, ready to be used for a
 /// different function.
 void FunctionLoweringInfo::clear() {
+  assert(CatchInfoFound.size() == CatchInfoLost.size() &&
+         "Not all catch info was assigned to a landing pad!");
+
   MBBMap.clear();
   ValueMap.clear();
   StaticAllocaMap.clear();
@@ -277,21 +204,12 @@
   return FirstReg;
 }
 
-/// ExtractTypeInfo - Returns the type info, possibly bitcast, encoded in V.
-GlobalVariable *llvm::ExtractTypeInfo(Value *V) {
-  V = V->stripPointerCasts();
-  GlobalVariable *GV = dyn_cast<GlobalVariable>(V);
-  assert ((GV || isa<ConstantPointerNull>(V)) &&
-          "TypeInfo must be a global variable or NULL");
-  return GV;
-}
-
 /// AddCatchInfo - Extract the personality and type infos from an eh.selector
 /// call, and add them to the specified machine basic block.
-void llvm::AddCatchInfo(CallInst &I, MachineModuleInfo *MMI,
+void llvm::AddCatchInfo(const CallInst &I, MachineModuleInfo *MMI,
                         MachineBasicBlock *MBB) {
   // Inform the MachineModuleInfo of the personality for this landing pad.
-  ConstantExpr *CE = cast<ConstantExpr>(I.getOperand(2));
+  const ConstantExpr *CE = cast<ConstantExpr>(I.getOperand(2));
   assert(CE->getOpcode() == Instruction::BitCast &&
          isa<Function>(CE->getOperand(0)) &&
          "Personality should be a function");
@@ -299,11 +217,11 @@
 
   // Gather all the type infos for this landing pad and pass them along to
   // MachineModuleInfo.
-  std::vector<GlobalVariable *> TyInfo;
+  std::vector<const GlobalVariable *> TyInfo;
   unsigned N = I.getNumOperands();
 
   for (unsigned i = N - 1; i > 2; --i) {
-    if (ConstantInt *CI = dyn_cast<ConstantInt>(I.getOperand(i))) {
+    if (const ConstantInt *CI = dyn_cast<ConstantInt>(I.getOperand(i))) {
       unsigned FilterLength = CI->getZExtValue();
       unsigned FirstCatch = i + FilterLength + !FilterLength;
       assert (FirstCatch <= N && "Invalid filter length");
@@ -340,10 +258,11 @@
   }
 }
 
-void llvm::CopyCatchInfo(BasicBlock *SrcBB, BasicBlock *DestBB,
+void llvm::CopyCatchInfo(const BasicBlock *SrcBB, const BasicBlock *DestBB,
                          MachineModuleInfo *MMI, FunctionLoweringInfo &FLI) {
-  for (BasicBlock::iterator I = SrcBB->begin(), E = --SrcBB->end(); I != E; ++I)
-    if (EHSelectorInst *EHSel = dyn_cast<EHSelectorInst>(I)) {
+  for (BasicBlock::const_iterator I = SrcBB->begin(), E = --SrcBB->end();
+       I != E; ++I)
+    if (const EHSelectorInst *EHSel = dyn_cast<EHSelectorInst>(I)) {
       // Apply the catch info to DestBB.
       AddCatchInfo(*EHSel, MMI, FLI.MBBMap[DestBB]);
 #ifndef NDEBUG
diff --git a/lib/CodeGen/SelectionDAG/FunctionLoweringInfo.h b/lib/CodeGen/SelectionDAG/FunctionLoweringInfo.h
index d851e64..c855db8 100644
--- a/lib/CodeGen/SelectionDAG/FunctionLoweringInfo.h
+++ b/lib/CodeGen/SelectionDAG/FunctionLoweringInfo.h
@@ -15,12 +15,16 @@
 #ifndef FUNCTIONLOWERINGINFO_H
 #define FUNCTIONLOWERINGINFO_H
 
+#include "llvm/InlineAsm.h"
+#include "llvm/Instructions.h"
 #include "llvm/ADT/APInt.h"
 #include "llvm/ADT/DenseMap.h"
 #ifndef NDEBUG
 #include "llvm/ADT/SmallSet.h"
 #endif
 #include "llvm/CodeGen/ValueTypes.h"
+#include "llvm/CodeGen/ISDOpcodes.h"
+#include "llvm/Support/CallSite.h"
 #include <vector>
 
 namespace llvm {
@@ -31,6 +35,7 @@
 class Function;
 class GlobalVariable;
 class Instruction;
+class MachineInstr;
 class MachineBasicBlock;
 class MachineFunction;
 class MachineModuleInfo;
@@ -44,8 +49,8 @@
 ///
 class FunctionLoweringInfo {
 public:
-  TargetLowering &TLI;
-  Function *Fn;
+  const TargetLowering &TLI;
+  const Function *Fn;
   MachineFunction *MF;
   MachineRegisterInfo *RegInfo;
 
@@ -57,13 +62,6 @@
   /// allocated to hold a pointer to the hidden sret parameter.
   unsigned DemoteRegister;
 
-  explicit FunctionLoweringInfo(TargetLowering &TLI);
-
-  /// set - Initialize this FunctionLoweringInfo with the given Function
-  /// and its associated MachineFunction.
-  ///
-  void set(Function &Fn, MachineFunction &MF, bool EnableFastISel);
-
   /// MBBMap - A mapping from LLVM basic blocks to their machine code entry.
   DenseMap<const BasicBlock*, MachineBasicBlock *> MBBMap;
 
@@ -78,10 +76,38 @@
   DenseMap<const AllocaInst*, int> StaticAllocaMap;
 
 #ifndef NDEBUG
-  SmallSet<Instruction*, 8> CatchInfoLost;
-  SmallSet<Instruction*, 8> CatchInfoFound;
+  SmallSet<const Instruction *, 8> CatchInfoLost;
+  SmallSet<const Instruction *, 8> CatchInfoFound;
 #endif
 
+  struct LiveOutInfo {
+    unsigned NumSignBits;
+    APInt KnownOne, KnownZero;
+    LiveOutInfo() : NumSignBits(0), KnownOne(1, 0), KnownZero(1, 0) {}
+  };
+  
+  /// LiveOutRegInfo - Information about live out vregs, indexed by their
+  /// register number offset by 'FirstVirtualRegister'.
+  std::vector<LiveOutInfo> LiveOutRegInfo;
+
+  /// PHINodesToUpdate - A list of phi instructions whose operand list will
+  /// be updated after processing the current basic block.
+  /// TODO: This isn't per-function state, it's per-basic-block state. But
+  /// there's no other convenient place for it to live right now.
+  std::vector<std::pair<MachineInstr*, unsigned> > PHINodesToUpdate;
+
+  explicit FunctionLoweringInfo(const TargetLowering &TLI);
+
+  /// set - Initialize this FunctionLoweringInfo with the given Function
+  /// and its associated MachineFunction.
+  ///
+  void set(const Function &Fn, MachineFunction &MF, bool EnableFastISel);
+
+  /// clear - Clear out all the function-specific state. This returns this
+  /// FunctionLoweringInfo to an empty state, ready to be used for a
+  /// different function.
+  void clear();
+
   unsigned MakeReg(EVT VT);
   
   /// isExportedInst - Return true if the specified value is an instruction
@@ -97,53 +123,15 @@
     assert(R == 0 && "Already initialized this value register!");
     return R = CreateRegForValue(V);
   }
-  
-  struct LiveOutInfo {
-    unsigned NumSignBits;
-    APInt KnownOne, KnownZero;
-    LiveOutInfo() : NumSignBits(0), KnownOne(1, 0), KnownZero(1, 0) {}
-  };
-  
-  /// LiveOutRegInfo - Information about live out vregs, indexed by their
-  /// register number offset by 'FirstVirtualRegister'.
-  std::vector<LiveOutInfo> LiveOutRegInfo;
-
-  /// clear - Clear out all the function-specific state. This returns this
-  /// FunctionLoweringInfo to an empty state, ready to be used for a
-  /// different function.
-  void clear();
 };
 
-/// ComputeLinearIndex - Given an LLVM IR aggregate type and a sequence
-/// of insertvalue or extractvalue indices that identify a member, return
-/// the linearized index of the start of the member.
-///
-unsigned ComputeLinearIndex(const TargetLowering &TLI, const Type *Ty,
-                            const unsigned *Indices,
-                            const unsigned *IndicesEnd,
-                            unsigned CurIndex = 0);
-
-/// ComputeValueVTs - Given an LLVM IR type, compute a sequence of
-/// EVTs that represent all the individual underlying
-/// non-aggregate types that comprise it.
-///
-/// If Offsets is non-null, it points to a vector to be filled in
-/// with the in-memory offsets of each of the individual values.
-///
-void ComputeValueVTs(const TargetLowering &TLI, const Type *Ty,
-                     SmallVectorImpl<EVT> &ValueVTs,
-                     SmallVectorImpl<uint64_t> *Offsets = 0,
-                     uint64_t StartingOffset = 0);
-
-/// ExtractTypeInfo - Returns the type info, possibly bitcast, encoded in V.
-GlobalVariable *ExtractTypeInfo(Value *V);
-
 /// AddCatchInfo - Extract the personality and type infos from an eh.selector
 /// call, and add them to the specified machine basic block.
-void AddCatchInfo(CallInst &I, MachineModuleInfo *MMI, MachineBasicBlock *MBB);
+void AddCatchInfo(const CallInst &I,
+                  MachineModuleInfo *MMI, MachineBasicBlock *MBB);
 
 /// CopyCatchInfo - Copy catch information from DestBB to SrcBB.
-void CopyCatchInfo(BasicBlock *SrcBB, BasicBlock *DestBB,
+void CopyCatchInfo(const BasicBlock *SrcBB, const BasicBlock *DestBB,
                    MachineModuleInfo *MMI, FunctionLoweringInfo &FLI);
 
 } // end namespace llvm
diff --git a/lib/CodeGen/SelectionDAG/InstrEmitter.cpp b/lib/CodeGen/SelectionDAG/InstrEmitter.cpp
index 625de11..f857a1c 100644
--- a/lib/CodeGen/SelectionDAG/InstrEmitter.cpp
+++ b/lib/CodeGen/SelectionDAG/InstrEmitter.cpp
@@ -15,7 +15,7 @@
 
 #define DEBUG_TYPE "instr-emitter"
 #include "InstrEmitter.h"
-#include "SDDbgValue.h"
+#include "SDNodeDbgValue.h"
 #include "llvm/CodeGen/MachineConstantPool.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
@@ -264,7 +264,8 @@
 InstrEmitter::AddRegisterOperand(MachineInstr *MI, SDValue Op,
                                  unsigned IIOpNum,
                                  const TargetInstrDesc *II,
-                                 DenseMap<SDValue, unsigned> &VRBaseMap) {
+                                 DenseMap<SDValue, unsigned> &VRBaseMap,
+                                 bool IsDebug) {
   assert(Op.getValueType() != MVT::Other &&
          Op.getValueType() != MVT::Flag &&
          "Chain and flag operands should occur at end of operand list!");
@@ -295,7 +296,11 @@
     }
   }
 
-  MI->addOperand(MachineOperand::CreateReg(VReg, isOptDef));
+  MI->addOperand(MachineOperand::CreateReg(VReg, isOptDef,
+                                           false/*isImp*/, false/*isKill*/,
+                                           false/*isDead*/, false/*isUndef*/,
+                                           false/*isEarlyClobber*/,
+                                           0/*SubReg*/, IsDebug));
 }
 
 /// AddOperand - Add the specified operand to the specified machine instr.  II
@@ -305,9 +310,10 @@
 void InstrEmitter::AddOperand(MachineInstr *MI, SDValue Op,
                               unsigned IIOpNum,
                               const TargetInstrDesc *II,
-                              DenseMap<SDValue, unsigned> &VRBaseMap) {
+                              DenseMap<SDValue, unsigned> &VRBaseMap,
+                              bool IsDebug) {
   if (Op.isMachineOpcode()) {
-    AddRegisterOperand(MI, Op, IIOpNum, II, VRBaseMap);
+    AddRegisterOperand(MI, Op, IIOpNum, II, VRBaseMap, IsDebug);
   } else if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op)) {
     MI->addOperand(MachineOperand::CreateImm(C->getSExtValue()));
   } else if (ConstantFPSDNode *F = dyn_cast<ConstantFPSDNode>(Op)) {
@@ -356,7 +362,7 @@
     assert(Op.getValueType() != MVT::Other &&
            Op.getValueType() != MVT::Flag &&
            "Chain and flag operands should occur at end of operand list!");
-    AddRegisterOperand(MI, Op, IIOpNum, II, VRBaseMap);
+    AddRegisterOperand(MI, Op, IIOpNum, II, VRBaseMap, IsDebug);
   }
 }
 
@@ -498,143 +504,163 @@
   assert(isNew && "Node emitted out of order - early");
 }
 
-/// EmitDbgValue - Generate any debug info that refers to this Node.  Constant
-/// dbg_value is not handled here.
-void
-InstrEmitter::EmitDbgValue(SDNode *Node,
-                           DenseMap<SDValue, unsigned> &VRBaseMap,
-                           SDDbgValue *sd) {
-  if (!Node->getHasDebugValue())
-    return;
-  if (!sd)
-    return;
-  unsigned VReg = getVR(SDValue(sd->getSDNode(), sd->getResNo()), VRBaseMap);
-  const TargetInstrDesc &II = TII->get(TargetOpcode::DBG_VALUE);
-  DebugLoc DL = sd->getDebugLoc();
-  MachineInstr *MI;
-  if (VReg) {
-    MI = BuildMI(*MF, DL, II).addReg(VReg, RegState::Debug).
-                              addImm(sd->getOffset()).
-                              addMetadata(sd->getMDPtr());
-  } else {
-    // Insert an Undef so we can see what we dropped.
-    MI = BuildMI(*MF, DL, II).addReg(0U).addImm(sd->getOffset()).
-                                    addMetadata(sd->getMDPtr());
-  }
-  MBB->insert(InsertPos, MI);
-}
-
-/// EmitDbgValue - Generate constant debug info.  No SDNode is involved.
-void
-InstrEmitter::EmitDbgValue(SDDbgValue *sd) {
-  if (!sd)
-    return;
-  const TargetInstrDesc &II = TII->get(TargetOpcode::DBG_VALUE);
-  DebugLoc DL = sd->getDebugLoc();
-  MachineInstr *MI;
-  Value *V = sd->getConst();
-  if (ConstantInt *CI = dyn_cast<ConstantInt>(V)) {
-    MI = BuildMI(*MF, DL, II).addImm(CI->getZExtValue()).
-                                   addImm(sd->getOffset()).
-                                   addMetadata(sd->getMDPtr());
-  } else if (ConstantFP *CF = dyn_cast<ConstantFP>(V)) {
-    MI = BuildMI(*MF, DL, II).addFPImm(CF).addImm(sd->getOffset()).
-                                   addMetadata(sd->getMDPtr());
-  } else {
-    // Insert an Undef so we can see what we dropped.
-    MI = BuildMI(*MF, DL, II).addReg(0U).addImm(sd->getOffset()).
-                                    addMetadata(sd->getMDPtr());
-  }
-  MBB->insert(InsertPos, MI);
-}
-
-/// EmitNode - Generate machine code for a node and needed dependencies.
+/// EmitDbgValue - Generate machine instruction for a dbg_value node.
 ///
-void InstrEmitter::EmitNode(SDNode *Node, bool IsClone, bool IsCloned,
-                            DenseMap<SDValue, unsigned> &VRBaseMap,
+MachineInstr *InstrEmitter::EmitDbgValue(SDDbgValue *SD,
+                                         DenseMap<SDValue, unsigned> &VRBaseMap,
                          DenseMap<MachineBasicBlock*, MachineBasicBlock*> *EM) {
-  // If machine instruction
-  if (Node->isMachineOpcode()) {
-    unsigned Opc = Node->getMachineOpcode();
-    
-    // Handle subreg insert/extract specially
-    if (Opc == TargetOpcode::EXTRACT_SUBREG || 
-        Opc == TargetOpcode::INSERT_SUBREG ||
-        Opc == TargetOpcode::SUBREG_TO_REG) {
-      EmitSubregNode(Node, VRBaseMap);
-      return;
-    }
+  uint64_t Offset = SD->getOffset();
+  MDNode* MDPtr = SD->getMDPtr();
+  DebugLoc DL = SD->getDebugLoc();
 
-    // Handle COPY_TO_REGCLASS specially.
-    if (Opc == TargetOpcode::COPY_TO_REGCLASS) {
-      EmitCopyToRegClassNode(Node, VRBaseMap);
-      return;
+  if (SD->getKind() == SDDbgValue::FRAMEIX) {
+    // Stack address; this needs to be lowered in target-dependent fashion.
+    // EmitTargetCodeForFrameDebugValue is responsible for allocation.
+    unsigned FrameIx = SD->getFrameIx();
+    return TII->emitFrameIndexDebugValue(*MF, FrameIx, Offset, MDPtr, DL);
+  }
+  // Otherwise, we're going to create an instruction here.
+  const TargetInstrDesc &II = TII->get(TargetOpcode::DBG_VALUE);
+  MachineInstrBuilder MIB = BuildMI(*MF, DL, II);
+  if (SD->getKind() == SDDbgValue::SDNODE) {
+    SDNode *Node = SD->getSDNode();
+    SDValue Op = SDValue(Node, SD->getResNo());
+    // It's possible we replaced this SDNode with other(s) and therefore
+    // didn't generate code for it.  It's better to catch these cases where
+    // they happen and transfer the debug info, but trying to guarantee that
+    // in all cases would be very fragile; this is a safeguard for any
+    // that were missed.
+    DenseMap<SDValue, unsigned>::iterator I = VRBaseMap.find(Op);
+    if (I==VRBaseMap.end())
+      MIB.addReg(0U);       // undef
+    else
+      AddOperand(&*MIB, Op, (*MIB).getNumOperands(), &II, VRBaseMap,
+                 true /*IsDebug*/);
+  } else if (SD->getKind() == SDDbgValue::CONST) {
+    const Value *V = SD->getConst();
+    if (const ConstantInt *CI = dyn_cast<ConstantInt>(V)) {
+      MIB.addImm(CI->getSExtValue());
+    } else if (const ConstantFP *CF = dyn_cast<ConstantFP>(V)) {
+      MIB.addFPImm(CF);
+    } else {
+      // Could be an Undef.  In any case insert an Undef so we can see what we
+      // dropped.
+      MIB.addReg(0U);
     }
+  } else {
+    // Insert an Undef so we can see what we dropped.
+    MIB.addReg(0U);
+  }
 
-    if (Opc == TargetOpcode::IMPLICIT_DEF)
-      // We want a unique VR for each IMPLICIT_DEF use.
-      return;
-    
-    const TargetInstrDesc &II = TII->get(Opc);
-    unsigned NumResults = CountResults(Node);
-    unsigned NodeOperands = CountOperands(Node);
-    bool HasPhysRegOuts = (NumResults > II.getNumDefs()) &&
-                          II.getImplicitDefs() != 0;
+  MIB.addImm(Offset).addMetadata(MDPtr);
+  return &*MIB;
+}
+
+/// EmitMachineNode - Generate machine code for a target-specific node and
+/// needed dependencies.
+///
+void InstrEmitter::
+EmitMachineNode(SDNode *Node, bool IsClone, bool IsCloned,
+                DenseMap<SDValue, unsigned> &VRBaseMap,
+                DenseMap<MachineBasicBlock*, MachineBasicBlock*> *EM) {
+  unsigned Opc = Node->getMachineOpcode();
+  
+  // Handle subreg insert/extract specially
+  if (Opc == TargetOpcode::EXTRACT_SUBREG || 
+      Opc == TargetOpcode::INSERT_SUBREG ||
+      Opc == TargetOpcode::SUBREG_TO_REG) {
+    EmitSubregNode(Node, VRBaseMap);
+    return;
+  }
+
+  // Handle COPY_TO_REGCLASS specially.
+  if (Opc == TargetOpcode::COPY_TO_REGCLASS) {
+    EmitCopyToRegClassNode(Node, VRBaseMap);
+    return;
+  }
+
+  if (Opc == TargetOpcode::IMPLICIT_DEF)
+    // We want a unique VR for each IMPLICIT_DEF use.
+    return;
+  
+  const TargetInstrDesc &II = TII->get(Opc);
+  unsigned NumResults = CountResults(Node);
+  unsigned NodeOperands = CountOperands(Node);
+  bool HasPhysRegOuts = NumResults > II.getNumDefs() && II.getImplicitDefs()!=0;
 #ifndef NDEBUG
-    unsigned NumMIOperands = NodeOperands + NumResults;
-    assert((II.getNumOperands() == NumMIOperands ||
-            HasPhysRegOuts || II.isVariadic()) &&
-           "#operands for dag node doesn't match .td file!"); 
+  unsigned NumMIOperands = NodeOperands + NumResults;
+  if (II.isVariadic())
+    assert(NumMIOperands >= II.getNumOperands() &&
+           "Too few operands for a variadic node!");
+  else
+    assert(NumMIOperands >= II.getNumOperands() &&
+           NumMIOperands <= II.getNumOperands()+II.getNumImplicitDefs() &&
+           "#operands for dag node doesn't match .td file!");
 #endif
 
-    // Create the new machine instruction.
-    MachineInstr *MI = BuildMI(*MF, Node->getDebugLoc(), II);
-    
-    // Add result register values for things that are defined by this
-    // instruction.
-    if (NumResults)
-      CreateVirtualRegisters(Node, MI, II, IsClone, IsCloned, VRBaseMap);
-    
-    // Emit all of the actual operands of this instruction, adding them to the
-    // instruction as appropriate.
-    bool HasOptPRefs = II.getNumDefs() > NumResults;
-    assert((!HasOptPRefs || !HasPhysRegOuts) &&
-           "Unable to cope with optional defs and phys regs defs!");
-    unsigned NumSkip = HasOptPRefs ? II.getNumDefs() - NumResults : 0;
-    for (unsigned i = NumSkip; i != NodeOperands; ++i)
-      AddOperand(MI, Node->getOperand(i), i-NumSkip+II.getNumDefs(), &II,
-                 VRBaseMap);
+  // Create the new machine instruction.
+  MachineInstr *MI = BuildMI(*MF, Node->getDebugLoc(), II);
+  
+  // Add result register values for things that are defined by this
+  // instruction.
+  if (NumResults)
+    CreateVirtualRegisters(Node, MI, II, IsClone, IsCloned, VRBaseMap);
+  
+  // Emit all of the actual operands of this instruction, adding them to the
+  // instruction as appropriate.
+  bool HasOptPRefs = II.getNumDefs() > NumResults;
+  assert((!HasOptPRefs || !HasPhysRegOuts) &&
+         "Unable to cope with optional defs and phys regs defs!");
+  unsigned NumSkip = HasOptPRefs ? II.getNumDefs() - NumResults : 0;
+  for (unsigned i = NumSkip; i != NodeOperands; ++i)
+    AddOperand(MI, Node->getOperand(i), i-NumSkip+II.getNumDefs(), &II,
+               VRBaseMap);
 
-    // Transfer all of the memory reference descriptions of this instruction.
-    MI->setMemRefs(cast<MachineSDNode>(Node)->memoperands_begin(),
-                   cast<MachineSDNode>(Node)->memoperands_end());
+  // Transfer all of the memory reference descriptions of this instruction.
+  MI->setMemRefs(cast<MachineSDNode>(Node)->memoperands_begin(),
+                 cast<MachineSDNode>(Node)->memoperands_end());
 
-    if (II.usesCustomInsertionHook()) {
-      // Insert this instruction into the basic block using a target
-      // specific inserter which may returns a new basic block.
-      MBB = TLI->EmitInstrWithCustomInserter(MI, MBB, EM);
-      InsertPos = MBB->end();
-    } else {
-      MBB->insert(InsertPos, MI);
-    }
-
-    // Additional results must be an physical register def.
-    if (HasPhysRegOuts) {
-      for (unsigned i = II.getNumDefs(); i < NumResults; ++i) {
-        unsigned Reg = II.getImplicitDefs()[i - II.getNumDefs()];
-        if (Node->hasAnyUseOfValue(i))
-          EmitCopyFromReg(Node, i, IsClone, IsCloned, Reg, VRBaseMap);
-        // If there are no uses, mark the register as dead now, so that
-        // MachineLICM/Sink can see that it's dead. Don't do this if the
-        // node has a Flag value, for the benefit of targets still using
-        // Flag for values in physregs.
-        else if (Node->getValueType(Node->getNumValues()-1) != MVT::Flag)
-          MI->addRegisterDead(Reg, TRI);
-      }
-    }
+  if (II.usesCustomInsertionHook()) {
+    // Insert this instruction into the basic block using a target
+    // specific inserter which may returns a new basic block.
+    MBB = TLI->EmitInstrWithCustomInserter(MI, MBB, EM);
+    InsertPos = MBB->end();
     return;
   }
+  
+  MBB->insert(InsertPos, MI);
 
+  // Additional results must be an physical register def.
+  if (HasPhysRegOuts) {
+    for (unsigned i = II.getNumDefs(); i < NumResults; ++i) {
+      unsigned Reg = II.getImplicitDefs()[i - II.getNumDefs()];
+      if (Node->hasAnyUseOfValue(i))
+        EmitCopyFromReg(Node, i, IsClone, IsCloned, Reg, VRBaseMap);
+      // If there are no uses, mark the register as dead now, so that
+      // MachineLICM/Sink can see that it's dead. Don't do this if the
+      // node has a Flag value, for the benefit of targets still using
+      // Flag for values in physregs.
+      else if (Node->getValueType(Node->getNumValues()-1) != MVT::Flag)
+        MI->addRegisterDead(Reg, TRI);
+    }
+  }
+  
+  // If the instruction has implicit defs and the node doesn't, mark the
+  // implicit def as dead.  If the node has any flag outputs, we don't do this
+  // because we don't know what implicit defs are being used by flagged nodes.
+  if (Node->getValueType(Node->getNumValues()-1) != MVT::Flag)
+    if (const unsigned *IDList = II.getImplicitDefs()) {
+      for (unsigned i = NumResults, e = II.getNumDefs()+II.getNumImplicitDefs();
+           i != e; ++i)
+        MI->addRegisterDead(IDList[i-II.getNumDefs()], TRI);
+    }
+}
+
+/// EmitSpecialNode - Generate machine code for a target-independent node and
+/// needed dependencies.
+void InstrEmitter::
+EmitSpecialNode(SDNode *Node, bool IsClone, bool IsCloned,
+                DenseMap<SDValue, unsigned> &VRBaseMap) {
   switch (Node->getOpcode()) {
   default:
 #ifndef NDEBUG
@@ -684,6 +710,13 @@
     EmitCopyFromReg(Node, 0, IsClone, IsCloned, SrcReg, VRBaseMap);
     break;
   }
+  case ISD::EH_LABEL: {
+    MCSymbol *S = cast<EHLabelSDNode>(Node)->getLabel();
+    BuildMI(*MBB, InsertPos, Node->getDebugLoc(),
+            TII->get(TargetOpcode::EH_LABEL)).addSym(S);
+    break;
+  }
+      
   case ISD::INLINEASM: {
     unsigned NumOps = Node->getNumOperands();
     if (Node->getOperand(NumOps-1).getValueType() == MVT::Flag)
@@ -694,12 +727,12 @@
                                TII->get(TargetOpcode::INLINEASM));
 
     // Add the asm string as an external symbol operand.
-    const char *AsmStr =
-      cast<ExternalSymbolSDNode>(Node->getOperand(1))->getSymbol();
+    SDValue AsmStrV = Node->getOperand(InlineAsm::Op_AsmString);
+    const char *AsmStr = cast<ExternalSymbolSDNode>(AsmStrV)->getSymbol();
     MI->addOperand(MachineOperand::CreateES(AsmStr));
       
     // Add all of the operand registers to the instruction.
-    for (unsigned i = 2; i != NumOps;) {
+    for (unsigned i = InlineAsm::Op_FirstOperand; i != NumOps;) {
       unsigned Flags =
         cast<ConstantSDNode>(Node->getOperand(i))->getZExtValue();
       unsigned NumVals = InlineAsm::getNumOperandRegisters(Flags);
@@ -707,24 +740,24 @@
       MI->addOperand(MachineOperand::CreateImm(Flags));
       ++i;  // Skip the ID value.
         
-      switch (Flags & 7) {
+      switch (InlineAsm::getKind(Flags)) {
       default: llvm_unreachable("Bad flags!");
-      case 2:   // Def of register.
+        case InlineAsm::Kind_RegDef:
         for (; NumVals; --NumVals, ++i) {
           unsigned Reg = cast<RegisterSDNode>(Node->getOperand(i))->getReg();
           MI->addOperand(MachineOperand::CreateReg(Reg, true));
         }
         break;
-      case 6:   // Def of earlyclobber register.
+      case InlineAsm::Kind_RegDefEarlyClobber:
         for (; NumVals; --NumVals, ++i) {
           unsigned Reg = cast<RegisterSDNode>(Node->getOperand(i))->getReg();
           MI->addOperand(MachineOperand::CreateReg(Reg, true, false, false, 
                                                    false, false, true));
         }
         break;
-      case 1:  // Use of register.
-      case 3:  // Immediate.
-      case 4:  // Addressing mode.
+      case InlineAsm::Kind_RegUse:  // Use of register.
+      case InlineAsm::Kind_Imm:  // Immediate.
+      case InlineAsm::Kind_Mem:  // Addressing mode.
         // The addressing mode has been selected, just add all of the
         // operands to the machine instruction.
         for (; NumVals; --NumVals, ++i)
@@ -732,6 +765,13 @@
         break;
       }
     }
+    
+    // Get the mdnode from the asm if it exists and add it to the instruction.
+    SDValue MDV = Node->getOperand(InlineAsm::Op_MDNode);
+    const MDNode *MD = cast<MDNodeSDNode>(MDV)->getMD();
+    if (MD)
+      MI->addOperand(MachineOperand::CreateMetadata(MD));
+    
     MBB->insert(InsertPos, MI);
     break;
   }
diff --git a/lib/CodeGen/SelectionDAG/InstrEmitter.h b/lib/CodeGen/SelectionDAG/InstrEmitter.h
index 4fe9f19..f8fd6ef 100644
--- a/lib/CodeGen/SelectionDAG/InstrEmitter.h
+++ b/lib/CodeGen/SelectionDAG/InstrEmitter.h
@@ -64,7 +64,8 @@
   void AddRegisterOperand(MachineInstr *MI, SDValue Op,
                           unsigned IIOpNum,
                           const TargetInstrDesc *II,
-                          DenseMap<SDValue, unsigned> &VRBaseMap);
+                          DenseMap<SDValue, unsigned> &VRBaseMap,
+                          bool IsDebug = false);
 
   /// AddOperand - Add the specified operand to the specified machine instr.  II
   /// specifies the instruction information for the node, and IIOpNum is the
@@ -73,7 +74,8 @@
   void AddOperand(MachineInstr *MI, SDValue Op,
                   unsigned IIOpNum,
                   const TargetInstrDesc *II,
-                  DenseMap<SDValue, unsigned> &VRBaseMap);
+                  DenseMap<SDValue, unsigned> &VRBaseMap,
+                  bool IsDebug = false);
 
   /// EmitSubregNode - Generate machine code for subreg nodes.
   ///
@@ -98,21 +100,22 @@
   /// MachineInstr.
   static unsigned CountOperands(SDNode *Node);
 
-  /// EmitDbgValue - Generate any debug info that refers to this Node.  Constant
-  /// dbg_value is not handled here.
-  void EmitDbgValue(SDNode *Node,
-                    DenseMap<SDValue, unsigned> &VRBaseMap,
-                    SDDbgValue* sd);
-
-
-  /// EmitDbgValue - Generate a constant DBG_VALUE.  No node is involved.
-  void EmitDbgValue(SDDbgValue* sd);
+  /// EmitDbgValue - Generate machine instruction for a dbg_value node.
+  ///
+  MachineInstr *EmitDbgValue(SDDbgValue *SD,
+                          DenseMap<SDValue, unsigned> &VRBaseMap,
+                          DenseMap<MachineBasicBlock*, MachineBasicBlock*> *EM);
 
   /// EmitNode - Generate machine code for a node and needed dependencies.
   ///
   void EmitNode(SDNode *Node, bool IsClone, bool IsCloned,
                 DenseMap<SDValue, unsigned> &VRBaseMap,
-                DenseMap<MachineBasicBlock*, MachineBasicBlock*> *EM);
+                DenseMap<MachineBasicBlock*, MachineBasicBlock*> *EM) {
+    if (Node->isMachineOpcode())
+      EmitMachineNode(Node, IsClone, IsCloned, VRBaseMap, EM);
+    else
+      EmitSpecialNode(Node, IsClone, IsCloned, VRBaseMap);
+  }
 
   /// getBlock - Return the current basic block.
   MachineBasicBlock *getBlock() { return MBB; }
@@ -123,6 +126,13 @@
   /// InstrEmitter - Construct an InstrEmitter and set it to start inserting
   /// at the given position in the given block.
   InstrEmitter(MachineBasicBlock *mbb, MachineBasicBlock::iterator insertpos);
+  
+private:
+  void EmitMachineNode(SDNode *Node, bool IsClone, bool IsCloned,
+                       DenseMap<SDValue, unsigned> &VRBaseMap,
+                       DenseMap<MachineBasicBlock*, MachineBasicBlock*> *EM);
+  void EmitSpecialNode(SDNode *Node, bool IsClone, bool IsCloned,
+                       DenseMap<SDValue, unsigned> &VRBaseMap);
 };
 
 }
diff --git a/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp b/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
index f498263..bedfa57 100644
--- a/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
+++ b/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
@@ -16,7 +16,6 @@
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineJumpTableInfo.h"
 #include "llvm/CodeGen/MachineModuleInfo.h"
-#include "llvm/CodeGen/DwarfWriter.h"
 #include "llvm/Analysis/DebugInfo.h"
 #include "llvm/CodeGen/PseudoSourceValue.h"
 #include "llvm/Target/TargetFrameInfo.h"
@@ -33,13 +32,11 @@
 #include "llvm/LLVMContext.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
-#include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/MathExtras.h"
 #include "llvm/Support/raw_ostream.h"
 #include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/SmallPtrSet.h"
-#include <map>
 using namespace llvm;
 
 //===----------------------------------------------------------------------===//
@@ -56,7 +53,8 @@
 ///
 namespace {
 class SelectionDAGLegalize {
-  TargetLowering &TLI;
+  const TargetMachine &TM;
+  const TargetLowering &TLI;
   SelectionDAG &DAG;
   CodeGenOpt::Level OptLevel;
 
@@ -214,7 +212,8 @@
 
 SelectionDAGLegalize::SelectionDAGLegalize(SelectionDAG &dag,
                                            CodeGenOpt::Level ol)
-  : TLI(dag.getTargetLoweringInfo()), DAG(dag), OptLevel(ol),
+  : TM(dag.getTarget()), TLI(dag.getTargetLoweringInfo()),
+    DAG(dag), OptLevel(ol),
     ValueTypeActions(TLI.getValueTypeActions()) {
   assert(MVT::LAST_VALUETYPE <= MVT::MAX_ALLOWED_VALUETYPE &&
          "Too many value types for ValueTypeActions to hold!");
@@ -410,7 +409,9 @@
       // to the final destination using (unaligned) integer loads and stores.
       EVT StoredVT = ST->getMemoryVT();
       EVT RegVT =
-        TLI.getRegisterType(*DAG.getContext(), EVT::getIntegerVT(*DAG.getContext(), StoredVT.getSizeInBits()));
+        TLI.getRegisterType(*DAG.getContext(),
+                            EVT::getIntegerVT(*DAG.getContext(),
+                                              StoredVT.getSizeInBits()));
       unsigned StoredBytes = StoredVT.getSizeInBits() / 8;
       unsigned RegBytes = RegVT.getSizeInBits() / 8;
       unsigned NumRegs = (StoredBytes + RegBytes - 1) / RegBytes;
@@ -446,7 +447,8 @@
       // The last store may be partial.  Do a truncating store.  On big-endian
       // machines this requires an extending load from the stack slot to ensure
       // that the bits are in the right place.
-      EVT MemVT = EVT::getIntegerVT(*DAG.getContext(), 8 * (StoredBytes - Offset));
+      EVT MemVT = EVT::getIntegerVT(*DAG.getContext(),
+                                    8 * (StoredBytes - Offset));
 
       // Load from the stack slot.
       SDValue Load = DAG.getExtLoad(ISD::EXTLOAD, dl, RegVT, Store, StackPtr,
@@ -549,7 +551,8 @@
       }
 
       // The last copy may be partial.  Do an extending load.
-      EVT MemVT = EVT::getIntegerVT(*DAG.getContext(), 8 * (LoadedBytes - Offset));
+      EVT MemVT = EVT::getIntegerVT(*DAG.getContext(),
+                                    8 * (LoadedBytes - Offset));
       SDValue Load = DAG.getExtLoad(ISD::EXTLOAD, dl, RegVT, Chain, Ptr,
                                     LD->getSrcValue(), SVOffset + Offset,
                                     MemVT, LD->isVolatile(),
@@ -969,11 +972,11 @@
     Node->dump( &DAG);
     dbgs() << "\n";
 #endif
-    llvm_unreachable("Do not know how to legalize this operator!");
+    assert(0 && "Do not know how to legalize this operator!");
 
   case ISD::BUILD_VECTOR:
     switch (TLI.getOperationAction(ISD::BUILD_VECTOR, Node->getValueType(0))) {
-    default: llvm_unreachable("This action is not supported yet!");
+    default: assert(0 && "This action is not supported yet!");
     case TargetLowering::Custom:
       Tmp3 = TLI.LowerOperation(Result, DAG);
       if (Tmp3.getNode()) {
@@ -1090,7 +1093,7 @@
       Tmp4 = Result.getValue(1);
 
       switch (TLI.getOperationAction(Node->getOpcode(), VT)) {
-      default: llvm_unreachable("This action is not supported yet!");
+      default: assert(0 && "This action is not supported yet!");
       case TargetLowering::Legal:
         // If this is an unaligned load and the target doesn't support it,
         // expand it.
@@ -1260,7 +1263,7 @@
         Tmp2 = LegalizeOp(Ch);
       } else {
         switch (TLI.getLoadExtAction(ExtType, SrcVT)) {
-        default: llvm_unreachable("This action is not supported yet!");
+        default: assert(0 && "This action is not supported yet!");
         case TargetLowering::Custom:
           isCustom = true;
           // FALLTHROUGH
@@ -1358,7 +1361,7 @@
 
         EVT VT = Tmp3.getValueType();
         switch (TLI.getOperationAction(ISD::STORE, VT)) {
-        default: llvm_unreachable("This action is not supported yet!");
+        default: assert(0 && "This action is not supported yet!");
         case TargetLowering::Legal:
           // If this is an unaligned store and the target doesn't support it,
           // expand it.
@@ -1395,7 +1398,8 @@
         // Promote to a byte-sized store with upper bits zero if not
         // storing an integral number of bytes.  For example, promote
         // TRUNCSTORE:i1 X -> TRUNCSTORE:i8 (and X, 1)
-        EVT NVT = EVT::getIntegerVT(*DAG.getContext(), StVT.getStoreSizeInBits());
+        EVT NVT = EVT::getIntegerVT(*DAG.getContext(),
+                                    StVT.getStoreSizeInBits());
         Tmp3 = DAG.getZeroExtendInReg(Tmp3, dl, StVT);
         Result = DAG.getTruncStore(Tmp1, dl, Tmp3, Tmp2, ST->getSrcValue(),
                                    SVOffset, NVT, isVolatile, isNonTemporal,
@@ -1460,7 +1464,7 @@
                                           ST->getOffset());
 
         switch (TLI.getTruncStoreAction(ST->getValue().getValueType(), StVT)) {
-        default: llvm_unreachable("This action is not supported yet!");
+        default: assert(0 && "This action is not supported yet!");
         case TargetLowering::Legal:
           // If this is an unaligned store and the target doesn't support it,
           // expand it.
@@ -1585,35 +1589,51 @@
   DebugLoc dl = Node->getDebugLoc();
   SDValue Tmp1 = Node->getOperand(0);
   SDValue Tmp2 = Node->getOperand(1);
-  assert((Tmp2.getValueType() == MVT::f32 ||
-          Tmp2.getValueType() == MVT::f64) &&
-          "Ugly special-cased code!");
-  // Get the sign bit of the RHS.
+
+  // Get the sign bit of the RHS.  First obtain a value that has the same
+  // sign as the sign bit, i.e. negative if and only if the sign bit is 1.
   SDValue SignBit;
-  EVT IVT = Tmp2.getValueType() == MVT::f64 ? MVT::i64 : MVT::i32;
+  EVT FloatVT = Tmp2.getValueType();
+  EVT IVT = EVT::getIntegerVT(*DAG.getContext(), FloatVT.getSizeInBits());
   if (isTypeLegal(IVT)) {
+    // Convert to an integer with the same sign bit.
     SignBit = DAG.getNode(ISD::BIT_CONVERT, dl, IVT, Tmp2);
   } else {
-    assert(isTypeLegal(TLI.getPointerTy()) &&
-            (TLI.getPointerTy() == MVT::i32 || 
-            TLI.getPointerTy() == MVT::i64) &&
-            "Legal type for load?!");
-    SDValue StackPtr = DAG.CreateStackTemporary(Tmp2.getValueType());
-    SDValue StorePtr = StackPtr, LoadPtr = StackPtr;
+    // Store the float to memory, then load the sign part out as an integer.
+    MVT LoadTy = TLI.getPointerTy();
+    // First create a temporary that is aligned for both the load and store.
+    SDValue StackPtr = DAG.CreateStackTemporary(FloatVT, LoadTy);
+    // Then store the float to it.
     SDValue Ch =
-      DAG.getStore(DAG.getEntryNode(), dl, Tmp2, StorePtr, NULL, 0,
+      DAG.getStore(DAG.getEntryNode(), dl, Tmp2, StackPtr, NULL, 0,
                    false, false, 0);
-    if (Tmp2.getValueType() == MVT::f64 && TLI.isLittleEndian())
-      LoadPtr = DAG.getNode(ISD::ADD, dl, StackPtr.getValueType(),
-                            LoadPtr, DAG.getIntPtrConstant(4));
-    SignBit = DAG.getExtLoad(ISD::SEXTLOAD, dl, TLI.getPointerTy(),
-                             Ch, LoadPtr, NULL, 0, MVT::i32,
-                             false, false, 0);
+    if (TLI.isBigEndian()) {
+      assert(FloatVT.isByteSized() && "Unsupported floating point type!");
+      // Load out a legal integer with the same sign bit as the float.
+      SignBit = DAG.getLoad(LoadTy, dl, Ch, StackPtr, NULL, 0, false, false, 0);
+    } else { // Little endian
+      SDValue LoadPtr = StackPtr;
+      // The float may be wider than the integer we are going to load.  Advance
+      // the pointer so that the loaded integer will contain the sign bit.
+      unsigned Strides = (FloatVT.getSizeInBits()-1)/LoadTy.getSizeInBits();
+      unsigned ByteOffset = (Strides * LoadTy.getSizeInBits()) / 8;
+      LoadPtr = DAG.getNode(ISD::ADD, dl, LoadPtr.getValueType(),
+                            LoadPtr, DAG.getIntPtrConstant(ByteOffset));
+      // Load a legal integer containing the sign bit.
+      SignBit = DAG.getLoad(LoadTy, dl, Ch, LoadPtr, NULL, 0, false, false, 0);
+      // Move the sign bit to the top bit of the loaded integer.
+      unsigned BitShift = LoadTy.getSizeInBits() -
+        (FloatVT.getSizeInBits() - 8 * ByteOffset);
+      assert(BitShift < LoadTy.getSizeInBits() && "Pointer advanced wrong?");
+      if (BitShift)
+        SignBit = DAG.getNode(ISD::SHL, dl, LoadTy, SignBit,
+                              DAG.getConstant(BitShift,TLI.getShiftAmountTy()));
+    }
   }
-  SignBit =
-      DAG.getSetCC(dl, TLI.getSetCCResultType(SignBit.getValueType()),
-                    SignBit, DAG.getConstant(0, SignBit.getValueType()),
-                    ISD::SETLT);
+  // Now get the sign bit proper, by seeing whether the value is negative.
+  SignBit = DAG.getSetCC(dl, TLI.getSetCCResultType(SignBit.getValueType()),
+                         SignBit, DAG.getConstant(0, SignBit.getValueType()),
+                         ISD::SETLT);
   // Get the absolute value of the result.
   SDValue AbsVal = DAG.getNode(ISD::FABS, dl, Tmp1.getValueType(), Tmp1);
   // Select between the nabs and abs value based on the sign bit of
@@ -1643,8 +1663,7 @@
   SDValue SP = DAG.getCopyFromReg(Chain, dl, SPReg, VT);
   Chain = SP.getValue(1);
   unsigned Align = cast<ConstantSDNode>(Tmp3)->getZExtValue();
-  unsigned StackAlign =
-    TLI.getTargetMachine().getFrameInfo()->getStackAlignment();
+  unsigned StackAlign = TM.getFrameInfo()->getStackAlignment();
   if (Align > StackAlign)
     SP = DAG.getNode(ISD::AND, dl, VT, SP,
                       DAG.getConstant(-(uint64_t)Align, VT));
@@ -1668,7 +1687,7 @@
   EVT OpVT = LHS.getValueType();
   ISD::CondCode CCCode = cast<CondCodeSDNode>(CC)->get();
   switch (TLI.getCondCodeAction(CCCode, OpVT)) {
-  default: llvm_unreachable("Unknown condition code action!");
+  default: assert(0 && "Unknown condition code action!");
   case TargetLowering::Legal:
     // Nothing to do.
     break;
@@ -1676,7 +1695,7 @@
     ISD::CondCode CC1 = ISD::SETCC_INVALID, CC2 = ISD::SETCC_INVALID;
     unsigned Opc = 0;
     switch (CCCode) {
-    default: llvm_unreachable("Don't know how to expand this condition!");
+    default: assert(0 && "Don't know how to expand this condition!");
     case ISD::SETOEQ: CC1 = ISD::SETEQ; CC2 = ISD::SETO;  Opc = ISD::AND; break;
     case ISD::SETOGT: CC1 = ISD::SETGT; CC2 = ISD::SETO;  Opc = ISD::AND; break;
     case ISD::SETOGE: CC1 = ISD::SETGE; CC2 = ISD::SETO;  Opc = ISD::AND; break;
@@ -1723,8 +1742,8 @@
   unsigned SrcSize = SrcOp.getValueType().getSizeInBits();
   unsigned SlotSize = SlotVT.getSizeInBits();
   unsigned DestSize = DestVT.getSizeInBits();
-  unsigned DestAlign =
-    TLI.getTargetData()->getPrefTypeAlignment(DestVT.getTypeForEVT(*DAG.getContext()));
+  const Type *DestType = DestVT.getTypeForEVT(*DAG.getContext());
+  unsigned DestAlign = TLI.getTargetData()->getPrefTypeAlignment(DestType);
 
   // Emit a store to the stack slot.  Use a truncstore if the input value is
   // later than DestVT.
@@ -1915,7 +1934,7 @@
                                               RTLIB::Libcall Call_PPCF128) {
   RTLIB::Libcall LC;
   switch (Node->getValueType(0).getSimpleVT().SimpleTy) {
-  default: llvm_unreachable("Unexpected request for libcall!");
+  default: assert(0 && "Unexpected request for libcall!");
   case MVT::f32: LC = Call_F32; break;
   case MVT::f64: LC = Call_F64; break;
   case MVT::f80: LC = Call_F80; break;
@@ -1932,7 +1951,7 @@
                                                RTLIB::Libcall Call_I128) {
   RTLIB::Libcall LC;
   switch (Node->getValueType(0).getSimpleVT().SimpleTy) {
-  default: llvm_unreachable("Unexpected request for libcall!");
+  default: assert(0 && "Unexpected request for libcall!");
   case MVT::i8:   LC = Call_I8; break;
   case MVT::i16:  LC = Call_I16; break;
   case MVT::i32:  LC = Call_I32; break;
@@ -2047,7 +2066,7 @@
   // offset depending on the data type.
   uint64_t FF;
   switch (Op0.getValueType().getSimpleVT().SimpleTy) {
-  default: llvm_unreachable("Unsupported integer type!");
+  default: assert(0 && "Unsupported integer type!");
   case MVT::i8 : FF = 0x43800000ULL; break;  // 2^8  (as a float)
   case MVT::i16: FF = 0x47800000ULL; break;  // 2^16 (as a float)
   case MVT::i32: FF = 0x4F800000ULL; break;  // 2^32 (as a float)
@@ -2167,7 +2186,7 @@
   EVT SHVT = TLI.getShiftAmountTy();
   SDValue Tmp1, Tmp2, Tmp3, Tmp4, Tmp5, Tmp6, Tmp7, Tmp8;
   switch (VT.getSimpleVT().SimpleTy) {
-  default: llvm_unreachable("Unhandled Expand type in BSWAP!");
+  default: assert(0 && "Unhandled Expand type in BSWAP!");
   case MVT::i16:
     Tmp2 = DAG.getNode(ISD::SHL, dl, VT, Op, DAG.getConstant(8, SHVT));
     Tmp1 = DAG.getNode(ISD::SRL, dl, VT, Op, DAG.getConstant(8, SHVT));
@@ -2212,7 +2231,7 @@
 SDValue SelectionDAGLegalize::ExpandBitCount(unsigned Opc, SDValue Op,
                                              DebugLoc dl) {
   switch (Opc) {
-  default: llvm_unreachable("Cannot expand this yet!");
+  default: assert(0 && "Cannot expand this yet!");
   case ISD::CTPOP: {
     static const uint64_t mask[6] = {
       0x5555555555555555ULL, 0x3333333333333333ULL,
@@ -2318,10 +2337,10 @@
     EVT VT = Node->getValueType(0);
     if (VT.isInteger())
       Results.push_back(DAG.getConstant(0, VT));
-    else if (VT.isFloatingPoint())
+    else {
+      assert(VT.isFloatingPoint() && "Unknown value type!");
       Results.push_back(DAG.getConstantFP(0, VT));
-    else
-      llvm_unreachable("Unknown value type!");
+    }
     break;
   }
   case ISD::TRAP: {
@@ -2416,7 +2435,7 @@
     // Increment the pointer, VAList, to the next vaarg
     Tmp3 = DAG.getNode(ISD::ADD, dl, TLI.getPointerTy(), VAList,
                        DAG.getConstant(TLI.getTargetData()->
-                                       getTypeAllocSize(VT.getTypeForEVT(*DAG.getContext())),
+                          getTypeAllocSize(VT.getTypeForEVT(*DAG.getContext())),
                                        TLI.getPointerTy()));
     // Store the incremented VAList to the legalized pointer
     Tmp3 = DAG.getStore(VAList.getValue(1), dl, Tmp3, Tmp2, V, 0,
@@ -2620,6 +2639,12 @@
     Results.push_back(ExpandFPLibCall(Node, RTLIB::REM_F32, RTLIB::REM_F64,
                                       RTLIB::REM_F80, RTLIB::REM_PPCF128));
     break;
+  case ISD::FP16_TO_FP32:
+    Results.push_back(ExpandLibCall(RTLIB::FPEXT_F16_F32, Node, false));
+    break;
+  case ISD::FP32_TO_FP16:
+    Results.push_back(ExpandLibCall(RTLIB::FPROUND_F32_F16, Node, false));
+    break;
   case ISD::ConstantFP: {
     ConstantFPSDNode *CFP = cast<ConstantFPSDNode>(Node);
     // Check to see if this FP immediate is already legal.
@@ -2821,7 +2846,14 @@
       BottomHalf = DAG.getNode(Ops[isSigned][1], dl, DAG.getVTList(VT, VT), LHS,
                                RHS);
       TopHalf = BottomHalf.getValue(1);
-    } else if (TLI.isTypeLegal(EVT::getIntegerVT(*DAG.getContext(), VT.getSizeInBits() * 2))) {
+    } else {
+      // FIXME: We should be able to fall back to a libcall with an illegal
+      // type in some cases.
+      // Also, we can fall back to a division in some cases, but that's a big
+      // performance hit in the general case.
+      assert(TLI.isTypeLegal(EVT::getIntegerVT(*DAG.getContext(),
+                                               VT.getSizeInBits() * 2)) &&
+             "Don't know how to expand this operation yet!");
       EVT WideVT = EVT::getIntegerVT(*DAG.getContext(), VT.getSizeInBits() * 2);
       LHS = DAG.getNode(Ops[isSigned][2], dl, WideVT, LHS);
       RHS = DAG.getNode(Ops[isSigned][2], dl, WideVT, RHS);
@@ -2830,12 +2862,6 @@
                                DAG.getIntPtrConstant(0));
       TopHalf = DAG.getNode(ISD::EXTRACT_ELEMENT, dl, VT, Tmp1,
                             DAG.getIntPtrConstant(1));
-    } else {
-      // FIXME: We should be able to fall back to a libcall with an illegal
-      // type in some cases.
-      // Also, we can fall back to a division in some cases, but that's a big
-      // performance hit in the general case.
-      llvm_unreachable("Don't know how to expand this operation yet!");
     }
     if (isSigned) {
       Tmp1 = DAG.getConstant(VT.getSizeInBits() - 1, TLI.getShiftAmountTy());
@@ -2895,7 +2921,7 @@
                                 PseudoSourceValue::getJumpTable(), 0, MemVT,
                                 false, false, 0);
     Addr = LD;
-    if (TLI.getTargetMachine().getRelocationModel() == Reloc::PIC_) {
+    if (TM.getRelocationModel() == Reloc::PIC_) {
       // For PIC, the sequence is:
       // BRIND(load(Jumptable + index) + RelocBase)
       // RelocBase can be JumpTable, GOT or some sort of global base.
@@ -3057,11 +3083,10 @@
     if (OVT.isVector()) {
       ExtOp   = ISD::BIT_CONVERT;
       TruncOp = ISD::BIT_CONVERT;
-    } else if (OVT.isInteger()) {
+    } else {
+      assert(OVT.isInteger() && "Cannot promote logic operation");
       ExtOp   = ISD::ANY_EXTEND;
       TruncOp = ISD::TRUNCATE;
-    } else {
-      llvm_report_error("Cannot promote logic operation");
     }
     // Promote each of the values to the new type.
     Tmp1 = DAG.getNode(ExtOp, dl, NVT, Node->getOperand(0));
diff --git a/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp b/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp
index 35a7c7c..e3eb949 100644
--- a/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp
+++ b/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp
@@ -79,6 +79,7 @@
     case ISD::FNEG:        R = SoftenFloatRes_FNEG(N); break;
     case ISD::FP_EXTEND:   R = SoftenFloatRes_FP_EXTEND(N); break;
     case ISD::FP_ROUND:    R = SoftenFloatRes_FP_ROUND(N); break;
+    case ISD::FP16_TO_FP32:R = SoftenFloatRes_FP16_TO_FP32(N); break;
     case ISD::FPOW:        R = SoftenFloatRes_FPOW(N); break;
     case ISD::FPOWI:       R = SoftenFloatRes_FPOWI(N); break;
     case ISD::FREM:        R = SoftenFloatRes_FREM(N); break;
@@ -108,14 +109,16 @@
 SDValue DAGTypeLegalizer::SoftenFloatRes_BUILD_PAIR(SDNode *N) {
   // Convert the inputs to integers, and build a new pair out of them.
   return DAG.getNode(ISD::BUILD_PAIR, N->getDebugLoc(),
-                     TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0)),
+                     TLI.getTypeToTransformTo(*DAG.getContext(),
+                                              N->getValueType(0)),
                      BitConvertToInteger(N->getOperand(0)),
                      BitConvertToInteger(N->getOperand(1)));
 }
 
 SDValue DAGTypeLegalizer::SoftenFloatRes_ConstantFP(ConstantFPSDNode *N) {
   return DAG.getConstant(N->getValueAPF().bitcastToAPInt(),
-                         TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0)));
+                         TLI.getTypeToTransformTo(*DAG.getContext(),
+                                                  N->getValueType(0)));
 }
 
 SDValue DAGTypeLegalizer::SoftenFloatRes_EXTRACT_VECTOR_ELT(SDNode *N) {
@@ -332,6 +335,15 @@
   return MakeLibCall(LC, NVT, &Op, 1, false, N->getDebugLoc());
 }
 
+// FIXME: Should we just use 'normal' FP_EXTEND / FP_TRUNC instead of special
+// nodes?
+SDValue DAGTypeLegalizer::SoftenFloatRes_FP16_TO_FP32(SDNode *N) {
+  EVT NVT = TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0));
+  SDValue Op = N->getOperand(0);
+  return MakeLibCall(RTLIB::FPEXT_F16_F32, NVT, &Op, 1, false,
+                     N->getDebugLoc());
+}
+
 SDValue DAGTypeLegalizer::SoftenFloatRes_FP_ROUND(SDNode *N) {
   EVT NVT = TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0));
   SDValue Op = N->getOperand(0);
@@ -480,7 +492,8 @@
 }
 
 SDValue DAGTypeLegalizer::SoftenFloatRes_UNDEF(SDNode *N) {
-  return DAG.getUNDEF(TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0)));
+  return DAG.getUNDEF(TLI.getTypeToTransformTo(*DAG.getContext(),
+                                               N->getValueType(0)));
 }
 
 SDValue DAGTypeLegalizer::SoftenFloatRes_VAARG(SDNode *N) {
@@ -522,7 +535,8 @@
   // Sign/zero extend the argument if the libcall takes a larger type.
   SDValue Op = DAG.getNode(Signed ? ISD::SIGN_EXTEND : ISD::ZERO_EXTEND, dl,
                            NVT, N->getOperand(0));
-  return MakeLibCall(LC, TLI.getTypeToTransformTo(*DAG.getContext(), RVT), &Op, 1, false, dl);
+  return MakeLibCall(LC, TLI.getTypeToTransformTo(*DAG.getContext(), RVT),
+                     &Op, 1, false, dl);
 }
 
 
@@ -548,6 +562,7 @@
   case ISD::FP_ROUND:    Res = SoftenFloatOp_FP_ROUND(N); break;
   case ISD::FP_TO_SINT:  Res = SoftenFloatOp_FP_TO_SINT(N); break;
   case ISD::FP_TO_UINT:  Res = SoftenFloatOp_FP_TO_UINT(N); break;
+  case ISD::FP32_TO_FP16:Res = SoftenFloatOp_FP32_TO_FP16(N); break;
   case ISD::SELECT_CC:   Res = SoftenFloatOp_SELECT_CC(N); break;
   case ISD::SETCC:       Res = SoftenFloatOp_SETCC(N); break;
   case ISD::STORE:       Res = SoftenFloatOp_STORE(N, OpNo); break;
@@ -704,6 +719,13 @@
   return MakeLibCall(LC, RVT, &Op, 1, false, N->getDebugLoc());
 }
 
+SDValue DAGTypeLegalizer::SoftenFloatOp_FP32_TO_FP16(SDNode *N) {
+  EVT RVT = N->getValueType(0);
+  RTLIB::Libcall LC = RTLIB::FPROUND_F32_F16;
+  SDValue Op = GetSoftenedFloat(N->getOperand(0));
+  return MakeLibCall(LC, RVT, &Op, 1, false, N->getDebugLoc());
+}
+
 SDValue DAGTypeLegalizer::SoftenFloatOp_SELECT_CC(SDNode *N) {
   SDValue NewLHS = N->getOperand(0), NewRHS = N->getOperand(1);
   ISD::CondCode CCCode = cast<CondCodeSDNode>(N->getOperand(4))->get();
@@ -800,6 +822,7 @@
   case ISD::FABS:       ExpandFloatRes_FABS(N, Lo, Hi); break;
   case ISD::FADD:       ExpandFloatRes_FADD(N, Lo, Hi); break;
   case ISD::FCEIL:      ExpandFloatRes_FCEIL(N, Lo, Hi); break;
+  case ISD::FCOPYSIGN:  ExpandFloatRes_FCOPYSIGN(N, Lo, Hi); break;
   case ISD::FCOS:       ExpandFloatRes_FCOS(N, Lo, Hi); break;
   case ISD::FDIV:       ExpandFloatRes_FDIV(N, Lo, Hi); break;
   case ISD::FEXP:       ExpandFloatRes_FEXP(N, Lo, Hi); break;
@@ -873,6 +896,17 @@
   GetPairElements(Call, Lo, Hi);
 }
 
+void DAGTypeLegalizer::ExpandFloatRes_FCOPYSIGN(SDNode *N,
+                                                SDValue &Lo, SDValue &Hi) {
+  SDValue Call = LibCallify(GetFPLibCall(N->getValueType(0),
+                                         RTLIB::COPYSIGN_F32,
+                                         RTLIB::COPYSIGN_F64,
+                                         RTLIB::COPYSIGN_F80,
+                                         RTLIB::COPYSIGN_PPCF128),
+                            N, false);
+  GetPairElements(Call, Lo, Hi);
+}
+
 void DAGTypeLegalizer::ExpandFloatRes_FCOS(SDNode *N,
                                            SDValue &Lo, SDValue &Hi) {
   SDValue Call = LibCallify(GetFPLibCall(N->getValueType(0),
@@ -1374,7 +1408,8 @@
   SDValue Chain = ST->getChain();
   SDValue Ptr = ST->getBasePtr();
 
-  EVT NVT = TLI.getTypeToTransformTo(*DAG.getContext(), ST->getValue().getValueType());
+  EVT NVT = TLI.getTypeToTransformTo(*DAG.getContext(),
+                                     ST->getValue().getValueType());
   assert(NVT.isByteSized() && "Expanded type not byte sized!");
   assert(ST->getMemoryVT().bitsLE(NVT) && "Float type not round?");
 
diff --git a/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp b/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
index 81f28ad..548454c 100644
--- a/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
+++ b/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
@@ -80,6 +80,8 @@
   case ISD::FP_TO_SINT:
   case ISD::FP_TO_UINT:  Res = PromoteIntRes_FP_TO_XINT(N); break;
 
+  case ISD::FP32_TO_FP16:Res = PromoteIntRes_FP32_TO_FP16(N); break;
+
   case ISD::AND:
   case ISD::OR:
   case ISD::XOR:
@@ -202,7 +204,8 @@
       std::swap(Lo, Hi);
 
     InOp = DAG.getNode(ISD::ANY_EXTEND, dl,
-                       EVT::getIntegerVT(*DAG.getContext(), NOutVT.getSizeInBits()),
+                       EVT::getIntegerVT(*DAG.getContext(),
+                                         NOutVT.getSizeInBits()),
                        JoinIntegers(Lo, Hi));
     return DAG.getNode(ISD::BIT_CONVERT, dl, NOutVT, InOp);
   }
@@ -324,6 +327,16 @@
                      NVT, Res, DAG.getValueType(N->getValueType(0)));
 }
 
+SDValue DAGTypeLegalizer::PromoteIntRes_FP32_TO_FP16(SDNode *N) {
+  EVT NVT = TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0));
+  DebugLoc dl = N->getDebugLoc();
+
+  SDValue Res = DAG.getNode(N->getOpcode(), dl, NVT, N->getOperand(0));
+
+  return DAG.getNode(ISD::AssertZext, dl,
+                     NVT, Res, DAG.getValueType(N->getValueType(0)));
+}
+
 SDValue DAGTypeLegalizer::PromoteIntRes_INT_EXTEND(SDNode *N) {
   EVT NVT = TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0));
   DebugLoc dl = N->getDebugLoc();
@@ -452,7 +465,7 @@
 
 SDValue DAGTypeLegalizer::PromoteIntRes_SHL(SDNode *N) {
   return DAG.getNode(ISD::SHL, N->getDebugLoc(),
-                     TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0)),
+                TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0)),
                      GetPromotedInteger(N->getOperand(0)), N->getOperand(1));
 }
 
@@ -543,7 +556,8 @@
 }
 
 SDValue DAGTypeLegalizer::PromoteIntRes_UNDEF(SDNode *N) {
-  return DAG.getUNDEF(TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0)));
+  return DAG.getUNDEF(TLI.getTypeToTransformTo(*DAG.getContext(),
+                                               N->getValueType(0)));
 }
 
 SDValue DAGTypeLegalizer::PromoteIntRes_VAARG(SDNode *N) {
@@ -634,6 +648,7 @@
   case ISD::STORE:        Res = PromoteIntOp_STORE(cast<StoreSDNode>(N),
                                                    OpNo); break;
   case ISD::TRUNCATE:     Res = PromoteIntOp_TRUNCATE(N); break;
+  case ISD::FP16_TO_FP32:
   case ISD::UINT_TO_FP:   Res = PromoteIntOp_UINT_TO_FP(N); break;
   case ISD::ZERO_EXTEND:  Res = PromoteIntOp_ZERO_EXTEND(N); break;
 
@@ -1370,7 +1385,8 @@
 
   if (NVTBits < EVTBits) {
     Hi = DAG.getNode(ISD::AssertSext, dl, NVT, Hi,
-                     DAG.getValueType(EVT::getIntegerVT(*DAG.getContext(), EVTBits - NVTBits)));
+                     DAG.getValueType(EVT::getIntegerVT(*DAG.getContext(),
+                                                        EVTBits - NVTBits)));
   } else {
     Lo = DAG.getNode(ISD::AssertSext, dl, NVT, Lo, DAG.getValueType(EVT));
     // The high part replicates the sign bit of Lo, make it explicit.
@@ -1390,7 +1406,8 @@
 
   if (NVTBits < EVTBits) {
     Hi = DAG.getNode(ISD::AssertZext, dl, NVT, Hi,
-                     DAG.getValueType(EVT::getIntegerVT(*DAG.getContext(), EVTBits - NVTBits)));
+                     DAG.getValueType(EVT::getIntegerVT(*DAG.getContext(),
+                                                        EVTBits - NVTBits)));
   } else {
     Lo = DAG.getNode(ISD::AssertZext, dl, NVT, Lo, DAG.getValueType(EVT));
     // The high part must be zero, make it explicit.
@@ -1833,7 +1850,8 @@
     unsigned ExcessBits =
       Op.getValueType().getSizeInBits() - NVT.getSizeInBits();
     Hi = DAG.getNode(ISD::SIGN_EXTEND_INREG, dl, Hi.getValueType(), Hi,
-                     DAG.getValueType(EVT::getIntegerVT(*DAG.getContext(), ExcessBits)));
+                     DAG.getValueType(EVT::getIntegerVT(*DAG.getContext(),
+                                                        ExcessBits)));
   }
 }
 
@@ -1955,7 +1973,8 @@
     SplitInteger(Res, Lo, Hi);
     unsigned ExcessBits =
       Op.getValueType().getSizeInBits() - NVT.getSizeInBits();
-    Hi = DAG.getZeroExtendInReg(Hi, dl, EVT::getIntegerVT(*DAG.getContext(), ExcessBits));
+    Hi = DAG.getZeroExtendInReg(Hi, dl,
+                                EVT::getIntegerVT(*DAG.getContext(), ExcessBits));
   }
 }
 
@@ -2256,7 +2275,8 @@
     unsigned EBytes = ExtVT.getStoreSize();
     unsigned IncrementSize = NVT.getSizeInBits()/8;
     unsigned ExcessBits = (EBytes - IncrementSize)*8;
-    EVT HiVT = EVT::getIntegerVT(*DAG.getContext(), ExtVT.getSizeInBits() - ExcessBits);
+    EVT HiVT = EVT::getIntegerVT(*DAG.getContext(),
+                                 ExtVT.getSizeInBits() - ExcessBits);
 
     if (ExcessBits < NVT.getSizeInBits()) {
       // Transfer high bits from the top of Lo to the bottom of Hi.
diff --git a/lib/CodeGen/SelectionDAG/LegalizeTypes.cpp b/lib/CodeGen/SelectionDAG/LegalizeTypes.cpp
index f3e7ca4..17f131b 100644
--- a/lib/CodeGen/SelectionDAG/LegalizeTypes.cpp
+++ b/lib/CodeGen/SelectionDAG/LegalizeTypes.cpp
@@ -721,7 +721,8 @@
 }
 
 void DAGTypeLegalizer::SetPromotedInteger(SDValue Op, SDValue Result) {
-  assert(Result.getValueType() == TLI.getTypeToTransformTo(*DAG.getContext(), Op.getValueType()) &&
+  assert(Result.getValueType() ==
+         TLI.getTypeToTransformTo(*DAG.getContext(), Op.getValueType()) &&
          "Invalid type for promoted integer");
   AnalyzeNewValue(Result);
 
@@ -731,7 +732,8 @@
 }
 
 void DAGTypeLegalizer::SetSoftenedFloat(SDValue Op, SDValue Result) {
-  assert(Result.getValueType() == TLI.getTypeToTransformTo(*DAG.getContext(), Op.getValueType()) &&
+  assert(Result.getValueType() ==
+         TLI.getTypeToTransformTo(*DAG.getContext(), Op.getValueType()) &&
          "Invalid type for softened float");
   AnalyzeNewValue(Result);
 
@@ -762,7 +764,8 @@
 
 void DAGTypeLegalizer::SetExpandedInteger(SDValue Op, SDValue Lo,
                                           SDValue Hi) {
-  assert(Lo.getValueType() == TLI.getTypeToTransformTo(*DAG.getContext(), Op.getValueType()) &&
+  assert(Lo.getValueType() ==
+         TLI.getTypeToTransformTo(*DAG.getContext(), Op.getValueType()) &&
          Hi.getValueType() == Lo.getValueType() &&
          "Invalid type for expanded integer");
   // Lo/Hi may have been newly allocated, if so, add nodeid's as relevant.
@@ -788,7 +791,8 @@
 
 void DAGTypeLegalizer::SetExpandedFloat(SDValue Op, SDValue Lo,
                                         SDValue Hi) {
-  assert(Lo.getValueType() == TLI.getTypeToTransformTo(*DAG.getContext(), Op.getValueType()) &&
+  assert(Lo.getValueType() ==
+         TLI.getTypeToTransformTo(*DAG.getContext(), Op.getValueType()) &&
          Hi.getValueType() == Lo.getValueType() &&
          "Invalid type for expanded float");
   // Lo/Hi may have been newly allocated, if so, add nodeid's as relevant.
@@ -832,7 +836,8 @@
 }
 
 void DAGTypeLegalizer::SetWidenedVector(SDValue Op, SDValue Result) {
-  assert(Result.getValueType() == TLI.getTypeToTransformTo(*DAG.getContext(), Op.getValueType()) &&
+  assert(Result.getValueType() ==
+         TLI.getTypeToTransformTo(*DAG.getContext(), Op.getValueType()) &&
          "Invalid type for widened vector");
   AnalyzeNewValue(Result);
 
@@ -940,7 +945,8 @@
   } else {
     unsigned NumElements = InVT.getVectorNumElements();
     assert(!(NumElements & 1) && "Splitting vector, but not in half!");
-    LoVT = HiVT = EVT::getVectorVT(*DAG.getContext(), InVT.getVectorElementType(), NumElements/2);
+    LoVT = HiVT = EVT::getVectorVT(*DAG.getContext(),
+                                   InVT.getVectorElementType(), NumElements/2);
   }
 }
 
@@ -980,7 +986,8 @@
   DebugLoc dlLo = Lo.getDebugLoc();
   EVT LVT = Lo.getValueType();
   EVT HVT = Hi.getValueType();
-  EVT NVT = EVT::getIntegerVT(*DAG.getContext(), LVT.getSizeInBits() + HVT.getSizeInBits());
+  EVT NVT = EVT::getIntegerVT(*DAG.getContext(),
+                              LVT.getSizeInBits() + HVT.getSizeInBits());
 
   Lo = DAG.getNode(ISD::ZERO_EXTEND, dlLo, NVT, Lo);
   Hi = DAG.getNode(ISD::ANY_EXTEND, dlHi, NVT, Hi);
@@ -1082,7 +1089,8 @@
 /// type half the size of Op's.
 void DAGTypeLegalizer::SplitInteger(SDValue Op,
                                     SDValue &Lo, SDValue &Hi) {
-  EVT HalfVT = EVT::getIntegerVT(*DAG.getContext(), Op.getValueType().getSizeInBits()/2);
+  EVT HalfVT = EVT::getIntegerVT(*DAG.getContext(),
+                                 Op.getValueType().getSizeInBits()/2);
   SplitInteger(Op, HalfVT, HalfVT, Lo, Hi);
 }
 
diff --git a/lib/CodeGen/SelectionDAG/LegalizeTypes.h b/lib/CodeGen/SelectionDAG/LegalizeTypes.h
index b0af357..d60ad60 100644
--- a/lib/CodeGen/SelectionDAG/LegalizeTypes.h
+++ b/lib/CodeGen/SelectionDAG/LegalizeTypes.h
@@ -33,7 +33,7 @@
 /// into small values.
 ///
 class VISIBILITY_HIDDEN DAGTypeLegalizer {
-  TargetLowering &TLI;
+  const TargetLowering &TLI;
   SelectionDAG &DAG;
 public:
   // NodeIdFlags - This pass uses the NodeId on the SDNodes to hold information
@@ -257,6 +257,7 @@
   SDValue PromoteIntRes_CTTZ(SDNode *N);
   SDValue PromoteIntRes_EXTRACT_VECTOR_ELT(SDNode *N);
   SDValue PromoteIntRes_FP_TO_XINT(SDNode *N);
+  SDValue PromoteIntRes_FP32_TO_FP16(SDNode *N);
   SDValue PromoteIntRes_INT_EXTEND(SDNode *N);
   SDValue PromoteIntRes_LOAD(LoadSDNode *N);
   SDValue PromoteIntRes_Overflow(SDNode *N);
@@ -406,6 +407,7 @@
   SDValue SoftenFloatRes_FNEARBYINT(SDNode *N);
   SDValue SoftenFloatRes_FNEG(SDNode *N);
   SDValue SoftenFloatRes_FP_EXTEND(SDNode *N);
+  SDValue SoftenFloatRes_FP16_TO_FP32(SDNode *N);
   SDValue SoftenFloatRes_FP_ROUND(SDNode *N);
   SDValue SoftenFloatRes_FPOW(SDNode *N);
   SDValue SoftenFloatRes_FPOWI(SDNode *N);
@@ -429,6 +431,7 @@
   SDValue SoftenFloatOp_FP_ROUND(SDNode *N);
   SDValue SoftenFloatOp_FP_TO_SINT(SDNode *N);
   SDValue SoftenFloatOp_FP_TO_UINT(SDNode *N);
+  SDValue SoftenFloatOp_FP32_TO_FP16(SDNode *N);
   SDValue SoftenFloatOp_SELECT_CC(SDNode *N);
   SDValue SoftenFloatOp_SETCC(SDNode *N);
   SDValue SoftenFloatOp_STORE(SDNode *N, unsigned OpNo);
@@ -455,6 +458,7 @@
   void ExpandFloatRes_FABS      (SDNode *N, SDValue &Lo, SDValue &Hi);
   void ExpandFloatRes_FADD      (SDNode *N, SDValue &Lo, SDValue &Hi);
   void ExpandFloatRes_FCEIL     (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_FCOPYSIGN (SDNode *N, SDValue &Lo, SDValue &Hi);
   void ExpandFloatRes_FCOS      (SDNode *N, SDValue &Lo, SDValue &Hi);
   void ExpandFloatRes_FDIV      (SDNode *N, SDValue &Lo, SDValue &Hi);
   void ExpandFloatRes_FEXP      (SDNode *N, SDValue &Lo, SDValue &Hi);
diff --git a/lib/CodeGen/SelectionDAG/LegalizeTypesGeneric.cpp b/lib/CodeGen/SelectionDAG/LegalizeTypesGeneric.cpp
index 5e83b4b..88e1e62 100644
--- a/lib/CodeGen/SelectionDAG/LegalizeTypesGeneric.cpp
+++ b/lib/CodeGen/SelectionDAG/LegalizeTypesGeneric.cpp
@@ -173,8 +173,9 @@
   EVT NewVT = TLI.getTypeToTransformTo(*DAG.getContext(), OldVT);
 
   SDValue NewVec = DAG.getNode(ISD::BIT_CONVERT, dl,
-                                 EVT::getVectorVT(*DAG.getContext(), NewVT, 2*OldElts),
-                                 OldVec);
+                               EVT::getVectorVT(*DAG.getContext(),
+                                                NewVT, 2*OldElts),
+                               OldVec);
 
   // Extract the elements at 2 * Idx and 2 * Idx + 1 from the new vector.
   SDValue Idx = N->getOperand(1);
@@ -268,7 +269,9 @@
     // is no point, and it might create expansion loops).  For example, on
     // x86 this turns v1i64 = BIT_CONVERT i64 into v1i64 = BIT_CONVERT v2i32.
     EVT OVT = N->getOperand(0).getValueType();
-    EVT NVT = EVT::getVectorVT(*DAG.getContext(), TLI.getTypeToTransformTo(*DAG.getContext(), OVT), 2);
+    EVT NVT = EVT::getVectorVT(*DAG.getContext(),
+                               TLI.getTypeToTransformTo(*DAG.getContext(), OVT),
+                               2);
 
     if (isTypeLegal(NVT)) {
       SDValue Parts[2];
@@ -312,8 +315,9 @@
   }
 
   SDValue NewVec = DAG.getNode(ISD::BUILD_VECTOR, dl,
-                                 EVT::getVectorVT(*DAG.getContext(), NewVT, NewElts.size()),
-                                 &NewElts[0], NewElts.size());
+                               EVT::getVectorVT(*DAG.getContext(),
+                                                NewVT, NewElts.size()),
+                               &NewElts[0], NewElts.size());
 
   // Convert the new vector to the old vector type.
   return DAG.getNode(ISD::BIT_CONVERT, dl, VecVT, NewVec);
@@ -380,7 +384,8 @@
   DebugLoc dl = N->getDebugLoc();
 
   StoreSDNode *St = cast<StoreSDNode>(N);
-  EVT NVT = TLI.getTypeToTransformTo(*DAG.getContext(), St->getValue().getValueType());
+  EVT NVT = TLI.getTypeToTransformTo(*DAG.getContext(),
+                                     St->getValue().getValueType());
   SDValue Chain = St->getChain();
   SDValue Ptr = St->getBasePtr();
   int SVOffset = St->getSrcValueOffset();
diff --git a/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp b/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp
index b5f84c0..0e2bd02 100644
--- a/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp
+++ b/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp
@@ -34,7 +34,7 @@
 namespace {
 class VectorLegalizer {
   SelectionDAG& DAG;
-  TargetLowering& TLI;
+  const TargetLowering &TLI;
   bool Changed; // Keep track of whether anything changed
 
   /// LegalizedNodes - For nodes that are of legal width, and that have more
diff --git a/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp b/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
index 8363c3a..7efeea1 100644
--- a/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
+++ b/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
@@ -705,8 +705,9 @@
   // Store the new element.  This may be larger than the vector element type,
   // so use a truncating store.
   SDValue EltPtr = GetVectorElementPointer(StackPtr, EltVT, Idx);
+  const Type *VecType = VecVT.getTypeForEVT(*DAG.getContext());
   unsigned Alignment =
-    TLI.getTargetData()->getPrefTypeAlignment(VecVT.getTypeForEVT(*DAG.getContext()));
+    TLI.getTargetData()->getPrefTypeAlignment(VecType);
   Store = DAG.getTruncStore(Store, dl, Elt, EltPtr, NULL, 0, EltVT,
                             false, false, 0);
 
@@ -1419,7 +1420,8 @@
     ShOp = GetWidenedVector(ShOp);
     ShVT = ShOp.getValueType();
   }
-  EVT ShWidenVT = EVT::getVectorVT(*DAG.getContext(), ShVT.getVectorElementType(),
+  EVT ShWidenVT = EVT::getVectorVT(*DAG.getContext(),
+                                   ShVT.getVectorElementType(),
                                    WidenVT.getVectorNumElements());
   if (ShVT != ShWidenVT)
     ShOp = ModifyToType(ShOp, ShWidenVT);
@@ -1493,7 +1495,8 @@
     unsigned NewNumElts = WidenSize / InSize;
     if (InVT.isVector()) {
       EVT InEltVT = InVT.getVectorElementType();
-      NewInVT= EVT::getVectorVT(*DAG.getContext(), InEltVT, WidenSize / InEltVT.getSizeInBits());
+      NewInVT= EVT::getVectorVT(*DAG.getContext(), InEltVT,
+                                WidenSize / InEltVT.getSizeInBits());
     } else {
       NewInVT = EVT::getVectorVT(*DAG.getContext(), InVT, NewNumElts);
     }
@@ -1617,7 +1620,8 @@
   SDValue RndOp = N->getOperand(3);
   SDValue SatOp = N->getOperand(4);
 
-  EVT      WidenVT = TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0));
+  EVT      WidenVT = TLI.getTypeToTransformTo(*DAG.getContext(),
+                                              N->getValueType(0));
   unsigned WidenNumElts = WidenVT.getVectorNumElements();
 
   EVT InVT = InOp.getValueType();
@@ -1791,7 +1795,8 @@
   EVT CondVT = Cond1.getValueType();
   if (CondVT.isVector()) {
     EVT CondEltVT = CondVT.getVectorElementType();
-    EVT CondWidenVT =  EVT::getVectorVT(*DAG.getContext(), CondEltVT, WidenNumElts);
+    EVT CondWidenVT =  EVT::getVectorVT(*DAG.getContext(),
+                                        CondEltVT, WidenNumElts);
     if (getTypeAction(CondVT) == WidenVector)
       Cond1 = GetWidenedVector(Cond1);
 
@@ -1859,7 +1864,8 @@
   SDValue InOp1 = N->getOperand(0);
   EVT InVT = InOp1.getValueType();
   assert(InVT.isVector() && "can not widen non vector type");
-  EVT WidenInVT = EVT::getVectorVT(*DAG.getContext(), InVT.getVectorElementType(), WidenNumElts);
+  EVT WidenInVT = EVT::getVectorVT(*DAG.getContext(),
+                                   InVT.getVectorElementType(), WidenNumElts);
   InOp1 = GetWidenedVector(InOp1);
   SDValue InOp2 = GetWidenedVector(N->getOperand(1));
 
@@ -2124,7 +2130,7 @@
   // The routines chops the vector into the largest vector loads with the same
   // element type or scalar loads and then recombines it to the widen vector
   // type.
-  EVT WidenVT = TLI.getTypeToTransformTo(*DAG.getContext(), LD->getValueType(0));
+  EVT WidenVT = TLI.getTypeToTransformTo(*DAG.getContext(),LD->getValueType(0));
   unsigned WidenWidth = WidenVT.getSizeInBits();
   EVT LdVT    = LD->getMemoryVT();
   DebugLoc dl = LD->getDebugLoc();
@@ -2167,7 +2173,7 @@
       } else
         return LdOp;
     } else {
-      unsigned NumElts = WidenWidth / LdWidth;
+      unsigned NumElts = WidenWidth / NewVTWidth;
       EVT NewVecVT = EVT::getVectorVT(*DAG.getContext(), NewVT, NumElts);
       SDValue VecOp = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, NewVecVT, LdOp);
       return DAG.getNode(ISD::BIT_CONVERT, dl, WidenVT, VecOp);
diff --git a/lib/CodeGen/SelectionDAG/SDDbgValue.h b/lib/CodeGen/SelectionDAG/SDDbgValue.h
deleted file mode 100644
index 9e15fc9..0000000
--- a/lib/CodeGen/SelectionDAG/SDDbgValue.h
+++ /dev/null
@@ -1,67 +0,0 @@
-//===-- llvm/CodeGen/SDDbgValue.h - SD dbg_value handling--------*- C++ -*-===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This file declares the SDDbgValue class.
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef LLVM_CODEGEN_SDDBGVALUE_H
-#define LLVM_CODEGEN_SDDBGVALUE_H
-
-#include "llvm/ADT/SmallVector.h"
-#include "llvm/Support/DebugLoc.h"
-
-namespace llvm {
-
-class MDNode;
-class SDNode;
-class Value;
-
-/// SDDbgValue - Holds the information from a dbg_value node through SDISel.
-/// Either Const or Node is nonzero, but not both.
-/// We do not use SDValue here to avoid including its header.
-
-class SDDbgValue {
-  SDNode *Node;           // valid for non-constants
-  unsigned ResNo;         // valid for non-constants
-  Value *Const;           // valid for constants
-  MDNode *mdPtr;
-  uint64_t Offset;
-  DebugLoc DL;
-public:
-  // Constructor for non-constants.
-  SDDbgValue(MDNode *mdP, SDNode *N, unsigned R, uint64_t off, DebugLoc dl) :
-    Node(N), ResNo(R), Const(0), mdPtr(mdP), Offset(off), DL(dl) {}
-
-  // Constructor for constants.
-  SDDbgValue(MDNode *mdP, Value *C, uint64_t off, DebugLoc dl) : Node(0),
-    ResNo(0), Const(C), mdPtr(mdP), Offset(off), DL(dl) {}
-
-  // Returns the MDNode pointer.
-  MDNode *getMDPtr() { return mdPtr; }
-
-  // Returns the SDNode* (valid for non-constants only).
-  SDNode *getSDNode() { assert (!Const); return Node; }
-
-  // Returns the ResNo (valid for non-constants only).
-  unsigned getResNo() { assert (!Const); return ResNo; }
-
-  // Returns the Value* for a constant (invalid for non-constants).
-  Value *getConst() { assert (!Node); return Const; }
-
-  // Returns the offset.
-  uint64_t getOffset() { return Offset; }
-
-  // Returns the DebugLoc.
-  DebugLoc getDebugLoc() { return DL; }
-};
-
-} // end llvm namespace
-
-#endif
diff --git a/lib/CodeGen/SelectionDAG/SDNodeDbgValue.h b/lib/CodeGen/SelectionDAG/SDNodeDbgValue.h
new file mode 100644
index 0000000..ac2d338
--- /dev/null
+++ b/lib/CodeGen/SelectionDAG/SDNodeDbgValue.h
@@ -0,0 +1,114 @@
+//===-- llvm/CodeGen/SDNodeDbgValue.h - SelectionDAG dbg_value --*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file declares the SDDbgValue class.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_CODEGEN_SDNODEDBGVALUE_H
+#define LLVM_CODEGEN_SDNODEDBGVALUE_H
+
+#include "llvm/ADT/SmallVector.h"
+#include "llvm/Support/DebugLoc.h"
+#include "llvm/System/DataTypes.h"
+
+namespace llvm {
+
+class MDNode;
+class SDNode;
+class Value;
+
+/// SDDbgValue - Holds the information from a dbg_value node through SDISel.
+/// We do not use SDValue here to avoid including its header.
+
+class SDDbgValue {
+public:
+  enum DbgValueKind {
+    SDNODE = 0,             // value is the result of an expression
+    CONST = 1,              // value is a constant
+    FRAMEIX = 2             // value is contents of a stack location
+  };
+private:
+  enum DbgValueKind kind;
+  union {
+    struct {
+      SDNode *Node;         // valid for expressions
+      unsigned ResNo;       // valid for expressions
+    } s;
+    const Value *Const;     // valid for constants
+    unsigned FrameIx;       // valid for stack objects
+  } u;
+  MDNode *mdPtr;
+  uint64_t Offset;
+  DebugLoc DL;
+  unsigned Order;
+  bool Invalid;
+public:
+  // Constructor for non-constants.
+  SDDbgValue(MDNode *mdP, SDNode *N, unsigned R, uint64_t off, DebugLoc dl,
+             unsigned O) : mdPtr(mdP), Offset(off), DL(dl), Order(O),
+                           Invalid(false) {
+    kind = SDNODE;
+    u.s.Node = N;
+    u.s.ResNo = R;
+  }
+
+  // Constructor for constants.
+  SDDbgValue(MDNode *mdP, const Value *C, uint64_t off, DebugLoc dl,
+             unsigned O) : 
+    mdPtr(mdP), Offset(off), DL(dl), Order(O), Invalid(false) {
+    kind = CONST;
+    u.Const = C;
+  }
+
+  // Constructor for frame indices.
+  SDDbgValue(MDNode *mdP, unsigned FI, uint64_t off, DebugLoc dl, unsigned O) : 
+    mdPtr(mdP), Offset(off), DL(dl), Order(O), Invalid(false) {
+    kind = FRAMEIX;
+    u.FrameIx = FI;
+  }
+
+  // Returns the kind.
+  DbgValueKind getKind() { return kind; }
+
+  // Returns the MDNode pointer.
+  MDNode *getMDPtr() { return mdPtr; }
+
+  // Returns the SDNode* for a register ref
+  SDNode *getSDNode() { assert (kind==SDNODE); return u.s.Node; }
+
+  // Returns the ResNo for a register ref
+  unsigned getResNo() { assert (kind==SDNODE); return u.s.ResNo; }
+
+  // Returns the Value* for a constant
+  const Value *getConst() { assert (kind==CONST); return u.Const; }
+
+  // Returns the FrameIx for a stack object
+  unsigned getFrameIx() { assert (kind==FRAMEIX); return u.FrameIx; }
+
+  // Returns the offset.
+  uint64_t getOffset() { return Offset; }
+
+  // Returns the DebugLoc.
+  DebugLoc getDebugLoc() { return DL; }
+
+  // Returns the SDNodeOrder.  This is the order of the preceding node in the
+  // input.
+  unsigned getOrder() { return Order; }
+
+  // setIsInvalidated / isInvalidated - Setter / getter of the "Invalidated"
+  // property. A SDDbgValue is invalid if the SDNode that produces the value is
+  // deleted.
+  void setIsInvalidated() { Invalid = true; }
+  bool isInvalidated() { return Invalid; }
+};
+
+} // end llvm namespace
+
+#endif
diff --git a/lib/CodeGen/SelectionDAG/ScheduleDAGRRList.cpp b/lib/CodeGen/SelectionDAG/ScheduleDAGRRList.cpp
index 3f1766d..da02850 100644
--- a/lib/CodeGen/SelectionDAG/ScheduleDAGRRList.cpp
+++ b/lib/CodeGen/SelectionDAG/ScheduleDAGRRList.cpp
@@ -17,18 +17,19 @@
 
 #define DEBUG_TYPE "pre-RA-sched"
 #include "ScheduleDAGSDNodes.h"
+#include "llvm/InlineAsm.h"
 #include "llvm/CodeGen/SchedulerRegistry.h"
 #include "llvm/CodeGen/SelectionDAGISel.h"
 #include "llvm/Target/TargetRegisterInfo.h"
 #include "llvm/Target/TargetData.h"
 #include "llvm/Target/TargetMachine.h"
 #include "llvm/Target/TargetInstrInfo.h"
-#include "llvm/Support/Debug.h"
-#include "llvm/Support/ErrorHandling.h"
 #include "llvm/ADT/PriorityQueue.h"
 #include "llvm/ADT/SmallSet.h"
 #include "llvm/ADT/Statistic.h"
 #include "llvm/ADT/STLExtras.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/raw_ostream.h"
 #include <climits>
 using namespace llvm;
@@ -647,13 +648,14 @@
       if (Node->getOperand(NumOps-1).getValueType() == MVT::Flag)
         --NumOps;  // Ignore the flag operand.
 
-      for (unsigned i = 2; i != NumOps;) {
+      for (unsigned i = InlineAsm::Op_FirstOperand; i != NumOps;) {
         unsigned Flags =
           cast<ConstantSDNode>(Node->getOperand(i))->getZExtValue();
-        unsigned NumVals = (Flags & 0xffff) >> 3;
+        unsigned NumVals = InlineAsm::getNumOperandRegisters(Flags);
 
         ++i; // Skip the ID value.
-        if ((Flags & 7) == 2 || (Flags & 7) == 6) {
+        if (InlineAsm::isRegDefKind(Flags) ||
+            InlineAsm::isRegDefEarlyClobberKind(Flags)) {
           // Check for def of register or earlyclobber register.
           for (; NumVals; --NumVals, ++i) {
             unsigned Reg = cast<RegisterSDNode>(Node->getOperand(i))->getReg();
diff --git a/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp b/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp
index 06e7b8c..c3e7854 100644
--- a/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp
+++ b/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp
@@ -13,6 +13,7 @@
 //===----------------------------------------------------------------------===//
 
 #define DEBUG_TYPE "pre-RA-sched"
+#include "SDNodeDbgValue.h"
 #include "ScheduleDAGSDNodes.h"
 #include "InstrEmitter.h"
 #include "llvm/CodeGen/SelectionDAG.h"
@@ -22,6 +23,7 @@
 #include "llvm/Target/TargetSubtarget.h"
 #include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/ADT/SmallSet.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/Statistic.h"
 #include "llvm/Support/Debug.h"
@@ -351,8 +353,8 @@
         const SDep& dep = SDep(OpSU, isChain ? SDep::Order : SDep::Data,
                                OpSU->Latency, PhysReg);
         if (!isChain && !UnitLatencies) {
-          ComputeOperandLatency(OpSU, SU, (SDep &)dep);
-          ST.adjustSchedDependency(OpSU, SU, (SDep &)dep);
+          ComputeOperandLatency(OpSU, SU, const_cast<SDep &>(dep));
+          ST.adjustSchedDependency(OpSU, SU, const_cast<SDep &>(dep));
         }
 
         SU->addPred(dep);
@@ -406,12 +408,79 @@
   }
 }
 
+namespace {
+  struct OrderSorter {
+    bool operator()(const std::pair<unsigned, MachineInstr*> &A,
+                    const std::pair<unsigned, MachineInstr*> &B) {
+      return A.first < B.first;
+    }
+  };
+}
+
+// ProcessSourceNode - Process nodes with source order numbers. These are added
+// to a vector which EmitSchedule use to determine how to insert dbg_value
+// instructions in the right order.
+static void ProcessSourceNode(SDNode *N, SelectionDAG *DAG,
+                           InstrEmitter &Emitter,
+                           DenseMap<MachineBasicBlock*, MachineBasicBlock*> *EM,
+                           DenseMap<SDValue, unsigned> &VRBaseMap,
+                    SmallVector<std::pair<unsigned, MachineInstr*>, 32> &Orders,
+                           SmallSet<unsigned, 8> &Seen) {
+  unsigned Order = DAG->GetOrdering(N);
+  if (!Order || !Seen.insert(Order))
+    return;
+
+  MachineBasicBlock *BB = Emitter.getBlock();
+  if (BB->empty() || BB->back().isPHI()) {
+    // Did not insert any instruction.
+    Orders.push_back(std::make_pair(Order, (MachineInstr*)0));
+    return;
+  }
+
+  Orders.push_back(std::make_pair(Order, &BB->back()));
+  if (!N->getHasDebugValue())
+    return;
+  // Opportunistically insert immediate dbg_value uses, i.e. those with source
+  // order number right after the N.
+  MachineBasicBlock::iterator InsertPos = Emitter.getInsertPos();
+  SmallVector<SDDbgValue*,2> &DVs = DAG->GetDbgValues(N);
+  for (unsigned i = 0, e = DVs.size(); i != e; ++i) {
+    if (DVs[i]->isInvalidated())
+      continue;
+    unsigned DVOrder = DVs[i]->getOrder();
+    if (DVOrder == ++Order) {
+      MachineInstr *DbgMI = Emitter.EmitDbgValue(DVs[i], VRBaseMap, EM);
+      if (DbgMI) {
+        Orders.push_back(std::make_pair(DVOrder, DbgMI));
+        BB->insert(InsertPos, DbgMI);
+      }
+      DVs[i]->setIsInvalidated();
+    }
+  }
+}
+
+
 /// EmitSchedule - Emit the machine code in scheduled order.
 MachineBasicBlock *ScheduleDAGSDNodes::
 EmitSchedule(DenseMap<MachineBasicBlock*, MachineBasicBlock*> *EM) {
   InstrEmitter Emitter(BB, InsertPos);
   DenseMap<SDValue, unsigned> VRBaseMap;
   DenseMap<SUnit*, unsigned> CopyVRBaseMap;
+  SmallVector<std::pair<unsigned, MachineInstr*>, 32> Orders;
+  SmallSet<unsigned, 8> Seen;
+  bool HasDbg = DAG->hasDebugValues();
+
+  // If this is the first BB, emit byval parameter dbg_value's.
+  if (HasDbg && BB->getParent()->begin() == MachineFunction::iterator(BB)) {
+    SDDbgInfo::DbgIterator PDI = DAG->ByvalParmDbgBegin();
+    SDDbgInfo::DbgIterator PDE = DAG->ByvalParmDbgEnd();
+    for (; PDI != PDE; ++PDI) {
+      MachineInstr *DbgMI= Emitter.EmitDbgValue(*PDI, VRBaseMap, EM);
+      if (DbgMI)
+        BB->insert(BB->end(), DbgMI);
+    }
+  }
+
   for (unsigned i = 0, e = Sequence.size(); i != e; i++) {
     SUnit *SU = Sequence[i];
     if (!SU) {
@@ -433,12 +502,83 @@
          N = N->getFlaggedNode())
       FlaggedNodes.push_back(N);
     while (!FlaggedNodes.empty()) {
+      SDNode *N = FlaggedNodes.back();
       Emitter.EmitNode(FlaggedNodes.back(), SU->OrigNode != SU, SU->isCloned,
                        VRBaseMap, EM);
+      // Remember the source order of the inserted instruction.
+      if (HasDbg)
+        ProcessSourceNode(N, DAG, Emitter, EM, VRBaseMap, Orders, Seen);
       FlaggedNodes.pop_back();
     }
     Emitter.EmitNode(SU->getNode(), SU->OrigNode != SU, SU->isCloned,
                      VRBaseMap, EM);
+    // Remember the source order of the inserted instruction.
+    if (HasDbg)
+      ProcessSourceNode(SU->getNode(), DAG, Emitter, EM, VRBaseMap, Orders,
+                        Seen);
+  }
+
+  // Insert all the dbg_values which have not already been inserted in source
+  // order sequence.
+  if (HasDbg) {
+    MachineBasicBlock::iterator BBBegin = BB->empty() ? BB->end() : BB->begin();
+    while (BBBegin != BB->end() && BBBegin->isPHI())
+      ++BBBegin;
+
+    // Sort the source order instructions and use the order to insert debug
+    // values.
+    std::sort(Orders.begin(), Orders.end(), OrderSorter());
+
+    SDDbgInfo::DbgIterator DI = DAG->DbgBegin();
+    SDDbgInfo::DbgIterator DE = DAG->DbgEnd();
+    // Now emit the rest according to source order.
+    unsigned LastOrder = 0;
+    MachineInstr *LastMI = 0;
+    for (unsigned i = 0, e = Orders.size(); i != e && DI != DE; ++i) {
+      unsigned Order = Orders[i].first;
+      MachineInstr *MI = Orders[i].second;
+      // Insert all SDDbgValue's whose order(s) are before "Order".
+      if (!MI)
+        continue;
+      MachineBasicBlock *MIBB = MI->getParent();
+#ifndef NDEBUG
+      unsigned LastDIOrder = 0;
+#endif
+      for (; DI != DE &&
+             (*DI)->getOrder() >= LastOrder && (*DI)->getOrder() < Order; ++DI) {
+#ifndef NDEBUG
+        assert((*DI)->getOrder() >= LastDIOrder &&
+               "SDDbgValue nodes must be in source order!");
+        LastDIOrder = (*DI)->getOrder();
+#endif
+        if ((*DI)->isInvalidated())
+          continue;
+        MachineInstr *DbgMI = Emitter.EmitDbgValue(*DI, VRBaseMap, EM);
+        if (DbgMI) {
+          if (!LastOrder)
+            // Insert to start of the BB (after PHIs).
+            BB->insert(BBBegin, DbgMI);
+          else {
+            MachineBasicBlock::iterator Pos = MI;
+            MIBB->insert(llvm::next(Pos), DbgMI);
+          }
+        }
+      }
+      LastOrder = Order;
+      LastMI = MI;
+    }
+    // Add trailing DbgValue's before the terminator. FIXME: May want to add
+    // some of them before one or more conditional branches?
+    while (DI != DE) {
+      MachineBasicBlock *InsertBB = Emitter.getBlock();
+      MachineBasicBlock::iterator Pos= Emitter.getBlock()->getFirstTerminator();
+      if (!(*DI)->isInvalidated()) {
+        MachineInstr *DbgMI= Emitter.EmitDbgValue(*DI, VRBaseMap, EM);
+        if (DbgMI)
+          InsertBB->insert(Pos, DbgMI);
+      }
+      ++DI;
+    }
   }
 
   BB = Emitter.getBlock();
diff --git a/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.h b/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.h
index 6b829b6..43d8333 100644
--- a/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.h
+++ b/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.h
@@ -59,7 +59,8 @@
       if (isa<JumpTableSDNode>(Node))      return true;
       if (isa<ExternalSymbolSDNode>(Node)) return true;
       if (isa<BlockAddressSDNode>(Node))   return true;
-      if (Node->getOpcode() == ISD::EntryToken) return true;
+      if (Node->getOpcode() == ISD::EntryToken ||
+          isa<MDNodeSDNode>(Node)) return true;
       return false;
     }
 
diff --git a/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index 746d4e2..c479f65 100644
--- a/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -13,6 +13,7 @@
 
 #include "llvm/CodeGen/SelectionDAG.h"
 #include "SDNodeOrdering.h"
+#include "SDNodeDbgValue.h"
 #include "llvm/Constants.h"
 #include "llvm/Analysis/ValueTracking.h"
 #include "llvm/Function.h"
@@ -303,10 +304,6 @@
   return Result;
 }
 
-const TargetMachine &SelectionDAG::getTarget() const {
-  return MF->getTarget();
-}
-
 //===----------------------------------------------------------------------===//
 //                           SDNode Profile Support
 //===----------------------------------------------------------------------===//
@@ -596,6 +593,11 @@
 
   // Remove the ordering of this node.
   Ordering->remove(N);
+
+  // If any of the SDDbgValue nodes refer to this SDNode, invalidate them.
+  SmallVector<SDDbgValue*, 2> &DbgVals = DbgInfo->getSDDbgValues(N);
+  for (unsigned i = 0, e = DbgVals.size(); i != e; ++i)
+    DbgVals[i]->setIsInvalidated();
 }
 
 /// RemoveNodeFromCSEMaps - Take the specified node out of the CSE map that
@@ -786,26 +788,25 @@
 }
 
 // EntryNode could meaningfully have debug info if we can find it...
-SelectionDAG::SelectionDAG(TargetLowering &tli, FunctionLoweringInfo &fli)
-  : TLI(tli), FLI(fli), DW(0),
-    EntryNode(ISD::EntryToken, DebugLoc::getUnknownLoc(),
-              getVTList(MVT::Other)),
+SelectionDAG::SelectionDAG(const TargetMachine &tm, FunctionLoweringInfo &fli)
+  : TM(tm), TLI(*tm.getTargetLowering()), FLI(fli),
+    EntryNode(ISD::EntryToken, DebugLoc(), getVTList(MVT::Other)),
     Root(getEntryNode()), Ordering(0) {
   AllNodes.push_back(&EntryNode);
   Ordering = new SDNodeOrdering();
+  DbgInfo = new SDDbgInfo();
 }
 
-void SelectionDAG::init(MachineFunction &mf, MachineModuleInfo *mmi,
-                        DwarfWriter *dw) {
+void SelectionDAG::init(MachineFunction &mf) {
   MF = &mf;
-  MMI = mmi;
-  DW = dw;
   Context = &mf.getFunction()->getContext();
 }
 
 SelectionDAG::~SelectionDAG() {
   allnodes_clear();
   delete Ordering;
+  DbgInfo->clear();
+  delete DbgInfo;
 }
 
 void SelectionDAG::allnodes_clear() {
@@ -833,6 +834,9 @@
   Root = getEntryNode();
   delete Ordering;
   Ordering = new SDNodeOrdering();
+  DbgInfo->clear();
+  delete DbgInfo;
+  DbgInfo = new SDDbgInfo();
 }
 
 SDValue SelectionDAG::getSExtOrTrunc(SDValue Op, DebugLoc DL, EVT VT) {
@@ -898,8 +902,7 @@
       return SDValue(N, 0);
 
   if (!N) {
-    N = NodeAllocator.Allocate<ConstantSDNode>();
-    new (N) ConstantSDNode(isT, &Val, EltVT);
+    N = new (NodeAllocator) ConstantSDNode(isT, &Val, EltVT);
     CSEMap.InsertNode(N, IP);
     AllNodes.push_back(N);
   }
@@ -908,8 +911,7 @@
   if (VT.isVector()) {
     SmallVector<SDValue, 8> Ops;
     Ops.assign(VT.getVectorNumElements(), Result);
-    Result = getNode(ISD::BUILD_VECTOR, DebugLoc::getUnknownLoc(),
-                     VT, &Ops[0], Ops.size());
+    Result = getNode(ISD::BUILD_VECTOR, DebugLoc(), VT, &Ops[0], Ops.size());
   }
   return Result;
 }
@@ -942,8 +944,7 @@
       return SDValue(N, 0);
 
   if (!N) {
-    N = NodeAllocator.Allocate<ConstantFPSDNode>();
-    new (N) ConstantFPSDNode(isTarget, &V, EltVT);
+    N = new (NodeAllocator) ConstantFPSDNode(isTarget, &V, EltVT);
     CSEMap.InsertNode(N, IP);
     AllNodes.push_back(N);
   }
@@ -953,8 +954,7 @@
     SmallVector<SDValue, 8> Ops;
     Ops.assign(VT.getVectorNumElements(), Result);
     // FIXME DebugLoc info might be appropriate here
-    Result = getNode(ISD::BUILD_VECTOR, DebugLoc::getUnknownLoc(),
-                     VT, &Ops[0], Ops.size());
+    Result = getNode(ISD::BUILD_VECTOR, DebugLoc(), VT, &Ops[0], Ops.size());
   }
   return Result;
 }
@@ -1002,8 +1002,8 @@
   if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
     return SDValue(E, 0);
 
-  SDNode *N = NodeAllocator.Allocate<GlobalAddressSDNode>();
-  new (N) GlobalAddressSDNode(Opc, GV, VT, Offset, TargetFlags);
+  SDNode *N = new (NodeAllocator) GlobalAddressSDNode(Opc, GV, VT,
+                                                      Offset, TargetFlags);
   CSEMap.InsertNode(N, IP);
   AllNodes.push_back(N);
   return SDValue(N, 0);
@@ -1018,8 +1018,7 @@
   if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
     return SDValue(E, 0);
 
-  SDNode *N = NodeAllocator.Allocate<FrameIndexSDNode>();
-  new (N) FrameIndexSDNode(FI, VT, isTarget);
+  SDNode *N = new (NodeAllocator) FrameIndexSDNode(FI, VT, isTarget);
   CSEMap.InsertNode(N, IP);
   AllNodes.push_back(N);
   return SDValue(N, 0);
@@ -1038,14 +1037,14 @@
   if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
     return SDValue(E, 0);
 
-  SDNode *N = NodeAllocator.Allocate<JumpTableSDNode>();
-  new (N) JumpTableSDNode(JTI, VT, isTarget, TargetFlags);
+  SDNode *N = new (NodeAllocator) JumpTableSDNode(JTI, VT, isTarget,
+                                                  TargetFlags);
   CSEMap.InsertNode(N, IP);
   AllNodes.push_back(N);
   return SDValue(N, 0);
 }
 
-SDValue SelectionDAG::getConstantPool(Constant *C, EVT VT,
+SDValue SelectionDAG::getConstantPool(const Constant *C, EVT VT,
                                       unsigned Alignment, int Offset,
                                       bool isTarget,
                                       unsigned char TargetFlags) {
@@ -1064,8 +1063,8 @@
   if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
     return SDValue(E, 0);
 
-  SDNode *N = NodeAllocator.Allocate<ConstantPoolSDNode>();
-  new (N) ConstantPoolSDNode(isTarget, C, VT, Offset, Alignment, TargetFlags);
+  SDNode *N = new (NodeAllocator) ConstantPoolSDNode(isTarget, C, VT, Offset,
+                                                     Alignment, TargetFlags);
   CSEMap.InsertNode(N, IP);
   AllNodes.push_back(N);
   return SDValue(N, 0);
@@ -1091,8 +1090,8 @@
   if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
     return SDValue(E, 0);
 
-  SDNode *N = NodeAllocator.Allocate<ConstantPoolSDNode>();
-  new (N) ConstantPoolSDNode(isTarget, C, VT, Offset, Alignment, TargetFlags);
+  SDNode *N = new (NodeAllocator) ConstantPoolSDNode(isTarget, C, VT, Offset,
+                                                     Alignment, TargetFlags);
   CSEMap.InsertNode(N, IP);
   AllNodes.push_back(N);
   return SDValue(N, 0);
@@ -1106,8 +1105,7 @@
   if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
     return SDValue(E, 0);
 
-  SDNode *N = NodeAllocator.Allocate<BasicBlockSDNode>();
-  new (N) BasicBlockSDNode(MBB);
+  SDNode *N = new (NodeAllocator) BasicBlockSDNode(MBB);
   CSEMap.InsertNode(N, IP);
   AllNodes.push_back(N);
   return SDValue(N, 0);
@@ -1122,8 +1120,7 @@
     ExtendedValueTypeNodes[VT] : ValueTypeNodes[VT.getSimpleVT().SimpleTy];
 
   if (N) return SDValue(N, 0);
-  N = NodeAllocator.Allocate<VTSDNode>();
-  new (N) VTSDNode(VT);
+  N = new (NodeAllocator) VTSDNode(VT);
   AllNodes.push_back(N);
   return SDValue(N, 0);
 }
@@ -1131,8 +1128,7 @@
 SDValue SelectionDAG::getExternalSymbol(const char *Sym, EVT VT) {
   SDNode *&N = ExternalSymbols[Sym];
   if (N) return SDValue(N, 0);
-  N = NodeAllocator.Allocate<ExternalSymbolSDNode>();
-  new (N) ExternalSymbolSDNode(false, Sym, 0, VT);
+  N = new (NodeAllocator) ExternalSymbolSDNode(false, Sym, 0, VT);
   AllNodes.push_back(N);
   return SDValue(N, 0);
 }
@@ -1143,8 +1139,7 @@
     TargetExternalSymbols[std::pair<std::string,unsigned char>(Sym,
                                                                TargetFlags)];
   if (N) return SDValue(N, 0);
-  N = NodeAllocator.Allocate<ExternalSymbolSDNode>();
-  new (N) ExternalSymbolSDNode(true, Sym, TargetFlags, VT);
+  N = new (NodeAllocator) ExternalSymbolSDNode(true, Sym, TargetFlags, VT);
   AllNodes.push_back(N);
   return SDValue(N, 0);
 }
@@ -1154,8 +1149,7 @@
     CondCodeNodes.resize(Cond+1);
 
   if (CondCodeNodes[Cond] == 0) {
-    CondCodeSDNode *N = NodeAllocator.Allocate<CondCodeSDNode>();
-    new (N) CondCodeSDNode(Cond);
+    CondCodeSDNode *N = new (NodeAllocator) CondCodeSDNode(Cond);
     CondCodeNodes[Cond] = N;
     AllNodes.push_back(N);
   }
@@ -1260,8 +1254,8 @@
   int *MaskAlloc = OperandAllocator.Allocate<int>(NElts);
   memcpy(MaskAlloc, &MaskVec[0], NElts * sizeof(int));
 
-  ShuffleVectorSDNode *N = NodeAllocator.Allocate<ShuffleVectorSDNode>();
-  new (N) ShuffleVectorSDNode(VT, dl, N1, N2, MaskAlloc);
+  ShuffleVectorSDNode *N =
+    new (NodeAllocator) ShuffleVectorSDNode(VT, dl, N1, N2, MaskAlloc);
   CSEMap.InsertNode(N, IP);
   AllNodes.push_back(N);
   return SDValue(N, 0);
@@ -1284,8 +1278,8 @@
   if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
     return SDValue(E, 0);
 
-  CvtRndSatSDNode *N = NodeAllocator.Allocate<CvtRndSatSDNode>();
-  new (N) CvtRndSatSDNode(VT, dl, Ops, 5, Code);
+  CvtRndSatSDNode *N = new (NodeAllocator) CvtRndSatSDNode(VT, dl, Ops, 5,
+                                                           Code);
   CSEMap.InsertNode(N, IP);
   AllNodes.push_back(N);
   return SDValue(N, 0);
@@ -1299,32 +1293,29 @@
   if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
     return SDValue(E, 0);
 
-  SDNode *N = NodeAllocator.Allocate<RegisterSDNode>();
-  new (N) RegisterSDNode(RegNo, VT);
+  SDNode *N = new (NodeAllocator) RegisterSDNode(RegNo, VT);
   CSEMap.InsertNode(N, IP);
   AllNodes.push_back(N);
   return SDValue(N, 0);
 }
 
-SDValue SelectionDAG::getLabel(unsigned Opcode, DebugLoc dl,
-                               SDValue Root,
-                               unsigned LabelID) {
+SDValue SelectionDAG::getEHLabel(DebugLoc dl, SDValue Root, MCSymbol *Label) {
   FoldingSetNodeID ID;
   SDValue Ops[] = { Root };
-  AddNodeIDNode(ID, Opcode, getVTList(MVT::Other), &Ops[0], 1);
-  ID.AddInteger(LabelID);
+  AddNodeIDNode(ID, ISD::EH_LABEL, getVTList(MVT::Other), &Ops[0], 1);
+  ID.AddPointer(Label);
   void *IP = 0;
   if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
     return SDValue(E, 0);
-
-  SDNode *N = NodeAllocator.Allocate<LabelSDNode>();
-  new (N) LabelSDNode(Opcode, dl, Root, LabelID);
+  
+  SDNode *N = new (NodeAllocator) EHLabelSDNode(dl, Root, Label);
   CSEMap.InsertNode(N, IP);
   AllNodes.push_back(N);
   return SDValue(N, 0);
 }
 
-SDValue SelectionDAG::getBlockAddress(BlockAddress *BA, EVT VT,
+
+SDValue SelectionDAG::getBlockAddress(const BlockAddress *BA, EVT VT,
                                       bool isTarget,
                                       unsigned char TargetFlags) {
   unsigned Opc = isTarget ? ISD::TargetBlockAddress : ISD::BlockAddress;
@@ -1337,8 +1328,7 @@
   if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
     return SDValue(E, 0);
 
-  SDNode *N = NodeAllocator.Allocate<BlockAddressSDNode>();
-  new (N) BlockAddressSDNode(Opc, VT, BA, TargetFlags);
+  SDNode *N = new (NodeAllocator) BlockAddressSDNode(Opc, VT, BA, TargetFlags);
   CSEMap.InsertNode(N, IP);
   AllNodes.push_back(N);
   return SDValue(N, 0);
@@ -1356,13 +1346,29 @@
   if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
     return SDValue(E, 0);
 
-  SDNode *N = NodeAllocator.Allocate<SrcValueSDNode>();
-  new (N) SrcValueSDNode(V);
+  SDNode *N = new (NodeAllocator) SrcValueSDNode(V);
   CSEMap.InsertNode(N, IP);
   AllNodes.push_back(N);
   return SDValue(N, 0);
 }
 
+/// getMDNode - Return an MDNodeSDNode which holds an MDNode.
+SDValue SelectionDAG::getMDNode(const MDNode *MD) {
+  FoldingSetNodeID ID;
+  AddNodeIDNode(ID, ISD::MDNODE_SDNODE, getVTList(MVT::Other), 0, 0);
+  ID.AddPointer(MD);
+  
+  void *IP = 0;
+  if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
+    return SDValue(E, 0);
+  
+  SDNode *N = new (NodeAllocator) MDNodeSDNode(MD);
+  CSEMap.InsertNode(N, IP);
+  AllNodes.push_back(N);
+  return SDValue(N, 0);
+}
+
+
 /// getShiftAmountOperand - Return the specified value casted to
 /// the target's desired shift amount type.
 SDValue SelectionDAG::getShiftAmountOperand(SDValue Op) {
@@ -1911,7 +1917,8 @@
     // Output known-0 bits are known if clear or set in both the low clear bits
     // common to both LHS & RHS.  For example, 8+(X<<3) is known to have the
     // low 3 bits clear.
-    APInt Mask2 = APInt::getLowBitsSet(BitWidth, Mask.countTrailingOnes());
+    APInt Mask2 = APInt::getLowBitsSet(BitWidth,
+                                       BitWidth - Mask.countLeadingZeros());
     ComputeMaskedBits(Op.getOperand(0), Mask2, KnownZero2, KnownOne2, Depth+1);
     assert((KnownZero2 & KnownOne2) == 0 && "Bits known to be one AND zero?");
     unsigned KnownZeroOut = KnownZero2.countTrailingOnes();
@@ -2260,10 +2267,9 @@
   GlobalAddressSDNode *GA = dyn_cast<GlobalAddressSDNode>(Op);
   if (!GA) return false;
   if (GA->getOffset() != 0) return false;
-  GlobalVariable *GV = dyn_cast<GlobalVariable>(GA->getGlobal());
+  const GlobalVariable *GV = dyn_cast<GlobalVariable>(GA->getGlobal());
   if (!GV) return false;
-  MachineModuleInfo *MMI = getMachineModuleInfo();
-  return MMI && MMI->hasDebugInfo();
+  return MF->getMMI().hasDebugInfo();
 }
 
 
@@ -2306,8 +2312,7 @@
   if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
     return SDValue(E, 0);
 
-  SDNode *N = NodeAllocator.Allocate<SDNode>();
-  new (N) SDNode(Opcode, DL, getVTList(VT));
+  SDNode *N = new (NodeAllocator) SDNode(Opcode, DL, getVTList(VT));
   CSEMap.InsertNode(N, IP);
 
   AllNodes.push_back(N);
@@ -2322,22 +2327,20 @@
   // Constant fold unary operations with an integer constant operand.
   if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Operand.getNode())) {
     const APInt &Val = C->getAPIntValue();
-    unsigned BitWidth = VT.getSizeInBits();
     switch (Opcode) {
     default: break;
     case ISD::SIGN_EXTEND:
-      return getConstant(APInt(Val).sextOrTrunc(BitWidth), VT);
+      return getConstant(APInt(Val).sextOrTrunc(VT.getSizeInBits()), VT);
     case ISD::ANY_EXTEND:
     case ISD::ZERO_EXTEND:
     case ISD::TRUNCATE:
-      return getConstant(APInt(Val).zextOrTrunc(BitWidth), VT);
+      return getConstant(APInt(Val).zextOrTrunc(VT.getSizeInBits()), VT);
     case ISD::UINT_TO_FP:
     case ISD::SINT_TO_FP: {
       const uint64_t zero[] = {0, 0};
-      // No compile time operations on this type.
-      if (VT==MVT::ppcf128)
-        break;
-      APFloat apf = APFloat(APInt(BitWidth, 2, zero));
+      // No compile time operations on ppcf128.
+      if (VT == MVT::ppcf128) break;
+      APFloat apf = APFloat(APInt(VT.getSizeInBits(), 2, zero));
       (void)apf.convertFromAPInt(Val,
                                  Opcode==ISD::SINT_TO_FP,
                                  APFloat::rmNearestTiesToEven);
@@ -2537,12 +2540,10 @@
     if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
       return SDValue(E, 0);
 
-    N = NodeAllocator.Allocate<UnarySDNode>();
-    new (N) UnarySDNode(Opcode, DL, VTs, Operand);
+    N = new (NodeAllocator) UnarySDNode(Opcode, DL, VTs, Operand);
     CSEMap.InsertNode(N, IP);
   } else {
-    N = NodeAllocator.Allocate<UnarySDNode>();
-    new (N) UnarySDNode(Opcode, DL, VTs, Operand);
+    N = new (NodeAllocator) UnarySDNode(Opcode, DL, VTs, Operand);
   }
 
   AllNodes.push_back(N);
@@ -2970,12 +2971,10 @@
     if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
       return SDValue(E, 0);
 
-    N = NodeAllocator.Allocate<BinarySDNode>();
-    new (N) BinarySDNode(Opcode, DL, VTs, N1, N2);
+    N = new (NodeAllocator) BinarySDNode(Opcode, DL, VTs, N1, N2);
     CSEMap.InsertNode(N, IP);
   } else {
-    N = NodeAllocator.Allocate<BinarySDNode>();
-    new (N) BinarySDNode(Opcode, DL, VTs, N1, N2);
+    N = new (NodeAllocator) BinarySDNode(Opcode, DL, VTs, N1, N2);
   }
 
   AllNodes.push_back(N);
@@ -3048,12 +3047,10 @@
     if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
       return SDValue(E, 0);
 
-    N = NodeAllocator.Allocate<TernarySDNode>();
-    new (N) TernarySDNode(Opcode, DL, VTs, N1, N2, N3);
+    N = new (NodeAllocator) TernarySDNode(Opcode, DL, VTs, N1, N2, N3);
     CSEMap.InsertNode(N, IP);
   } else {
-    N = NodeAllocator.Allocate<TernarySDNode>();
-    new (N) TernarySDNode(Opcode, DL, VTs, N1, N2, N3);
+    N = new (NodeAllocator) TernarySDNode(Opcode, DL, VTs, N1, N2, N3);
   }
 
   AllNodes.push_back(N);
@@ -3104,6 +3101,8 @@
 /// operand.
 static SDValue getMemsetValue(SDValue Value, EVT VT, SelectionDAG &DAG,
                               DebugLoc dl) {
+  assert(Value.getOpcode() != ISD::UNDEF);
+
   unsigned NumBits = VT.getScalarType().getSizeInBits();
   if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Value)) {
     APInt Val = APInt(NumBits, C->getZExtValue() & 255);
@@ -3142,11 +3141,17 @@
   if (Str.empty()) {
     if (VT.isInteger())
       return DAG.getConstant(0, VT);
-    unsigned NumElts = VT.getVectorNumElements();
-    MVT EltVT = (VT.getVectorElementType() == MVT::f32) ? MVT::i32 : MVT::i64;
-    return DAG.getNode(ISD::BIT_CONVERT, dl, VT,
-                       DAG.getConstant(0,
-                       EVT::getVectorVT(*DAG.getContext(), EltVT, NumElts)));
+    else if (VT.getSimpleVT().SimpleTy == MVT::f32 ||
+             VT.getSimpleVT().SimpleTy == MVT::f64)
+      return DAG.getConstantFP(0.0, VT);
+    else if (VT.isVector()) {
+      unsigned NumElts = VT.getVectorNumElements();
+      MVT EltVT = (VT.getVectorElementType() == MVT::f32) ? MVT::i32 : MVT::i64;
+      return DAG.getNode(ISD::BIT_CONVERT, dl, VT,
+                         DAG.getConstant(0, EVT::getVectorVT(*DAG.getContext(),
+                                                             EltVT, NumElts)));
+    } else
+      llvm_unreachable("Expected type!");
   }
 
   assert(!VT.isVector() && "Can't handle vector type here!");
@@ -3187,58 +3192,43 @@
   if (!G)
     return false;
 
-  GlobalVariable *GV = dyn_cast<GlobalVariable>(G->getGlobal());
+  const GlobalVariable *GV = dyn_cast<GlobalVariable>(G->getGlobal());
   if (GV && GetConstantStringInfo(GV, Str, SrcDelta, false))
     return true;
 
   return false;
 }
 
-/// MeetsMaxMemopRequirement - Determines if the number of memory ops required
-/// to replace the memset / memcpy is below the threshold. It also returns the
-/// types of the sequence of memory ops to perform memset / memcpy.
-static
-bool MeetsMaxMemopRequirement(std::vector<EVT> &MemOps,
-                              SDValue Dst, SDValue Src,
-                              unsigned Limit, uint64_t Size, unsigned &Align,
-                              std::string &Str, bool &isSrcStr,
-                              SelectionDAG &DAG,
-                              const TargetLowering &TLI) {
-  isSrcStr = isMemSrcFromString(Src, Str);
-  bool isSrcConst = isa<ConstantSDNode>(Src);
-  EVT VT = TLI.getOptimalMemOpType(Size, Align, isSrcConst, isSrcStr, DAG);
-  bool AllowUnalign = TLI.allowsUnalignedMemoryAccesses(VT);
-  if (VT != MVT::Other) {
-    const Type *Ty = VT.getTypeForEVT(*DAG.getContext());
-    unsigned NewAlign = (unsigned) TLI.getTargetData()->getABITypeAlignment(Ty);
-    // If source is a string constant, this will require an unaligned load.
-    if (NewAlign > Align && (isSrcConst || AllowUnalign)) {
-      if (Dst.getOpcode() != ISD::FrameIndex) {
-        // Can't change destination alignment. It requires a unaligned store.
-        if (AllowUnalign)
-          VT = MVT::Other;
-      } else {
-        int FI = cast<FrameIndexSDNode>(Dst)->getIndex();
-        MachineFrameInfo *MFI = DAG.getMachineFunction().getFrameInfo();
-        if (MFI->isFixedObjectIndex(FI)) {
-          // Can't change destination alignment. It requires a unaligned store.
-          if (AllowUnalign)
-            VT = MVT::Other;
-        } else {
-          // Give the stack frame object a larger alignment if needed.
-          if (MFI->getObjectAlignment(FI) < NewAlign)
-            MFI->setObjectAlignment(FI, NewAlign);
-          Align = NewAlign;
-        }
-      }
-    }
-  }
+/// FindOptimalMemOpLowering - Determines the optimial series memory ops
+/// to replace the memset / memcpy. Return true if the number of memory ops
+/// is below the threshold. It returns the types of the sequence of
+/// memory ops to perform memset / memcpy by reference.
+static bool FindOptimalMemOpLowering(std::vector<EVT> &MemOps,
+                                     unsigned Limit, uint64_t Size,
+                                     unsigned DstAlign, unsigned SrcAlign,
+                                     bool NonScalarIntSafe,
+                                     bool MemcpyStrSrc,
+                                     SelectionDAG &DAG,
+                                     const TargetLowering &TLI) {
+  assert((SrcAlign == 0 || SrcAlign >= DstAlign) &&
+         "Expecting memcpy / memset source to meet alignment requirement!");
+  // If 'SrcAlign' is zero, that means the memory operation does not need load
+  // the value, i.e. memset or memcpy from constant string. Otherwise, it's
+  // the inferred alignment of the source. 'DstAlign', on the other hand, is the
+  // specified alignment of the memory operation. If it is zero, that means
+  // it's possible to change the alignment of the destination. 'MemcpyStrSrc'
+  // indicates whether the memcpy source is constant so it does not need to be
+  // loaded.
+  EVT VT = TLI.getOptimalMemOpType(Size, DstAlign, SrcAlign,
+                                   NonScalarIntSafe, MemcpyStrSrc,
+                                   DAG.getMachineFunction());
 
   if (VT == MVT::Other) {
-    if (TLI.allowsUnalignedMemoryAccesses(MVT::i64)) {
-      VT = MVT::i64;
+    if (DstAlign >= TLI.getTargetData()->getPointerPrefAlignment() ||
+        TLI.allowsUnalignedMemoryAccesses(VT)) {
+      VT = TLI.getPointerTy();
     } else {
-      switch (Align & 7) {
+      switch (DstAlign & 7) {
       case 0:  VT = MVT::i64; break;
       case 4:  VT = MVT::i32; break;
       case 2:  VT = MVT::i16; break;
@@ -3260,7 +3250,7 @@
     unsigned VTSize = VT.getSizeInBits() / 8;
     while (VTSize > Size) {
       // For now, only use non-vector load / store's for the left-over pieces.
-      if (VT.isVector()) {
+      if (VT.isVector() || VT.isFloatingPoint()) {
         VT = MVT::i64;
         while (!TLI.isTypeLegal(VT))
           VT = (MVT::SimpleValueType)(VT.getSimpleVT().SimpleTy - 1);
@@ -3283,28 +3273,51 @@
 }
 
 static SDValue getMemcpyLoadsAndStores(SelectionDAG &DAG, DebugLoc dl,
-                                         SDValue Chain, SDValue Dst,
-                                         SDValue Src, uint64_t Size,
-                                         unsigned Align, bool AlwaysInline,
-                                         const Value *DstSV, uint64_t DstSVOff,
-                                         const Value *SrcSV, uint64_t SrcSVOff){
-  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+                                       SDValue Chain, SDValue Dst,
+                                       SDValue Src, uint64_t Size,
+                                       unsigned Align, bool isVol,
+                                       bool AlwaysInline,
+                                       const Value *DstSV, uint64_t DstSVOff,
+                                       const Value *SrcSV, uint64_t SrcSVOff) {
+  // Turn a memcpy of undef to nop.
+  if (Src.getOpcode() == ISD::UNDEF)
+    return Chain;
 
   // Expand memcpy to a series of load and store ops if the size operand falls
   // below a certain threshold.
+  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
   std::vector<EVT> MemOps;
+  bool DstAlignCanChange = false;
+  MachineFrameInfo *MFI = DAG.getMachineFunction().getFrameInfo();
+  FrameIndexSDNode *FI = dyn_cast<FrameIndexSDNode>(Dst);
+  if (FI && !MFI->isFixedObjectIndex(FI->getIndex()))
+    DstAlignCanChange = true;
+  unsigned SrcAlign = DAG.InferPtrAlignment(Src);
+  if (Align > SrcAlign)
+    SrcAlign = Align;
+  std::string Str;
+  bool CopyFromStr = isMemSrcFromString(Src, Str);
+  bool isZeroStr = CopyFromStr && Str.empty();
   uint64_t Limit = -1ULL;
   if (!AlwaysInline)
     Limit = TLI.getMaxStoresPerMemcpy();
-  unsigned DstAlign = Align;  // Destination alignment can change.
-  std::string Str;
-  bool CopyFromStr;
-  if (!MeetsMaxMemopRequirement(MemOps, Dst, Src, Limit, Size, DstAlign,
-                                Str, CopyFromStr, DAG, TLI))
+  if (!FindOptimalMemOpLowering(MemOps, Limit, Size,
+                                (DstAlignCanChange ? 0 : Align),
+                                (isZeroStr ? 0 : SrcAlign),
+                                true, CopyFromStr, DAG, TLI))
     return SDValue();
 
+  if (DstAlignCanChange) {
+    const Type *Ty = MemOps[0].getTypeForEVT(*DAG.getContext());
+    unsigned NewAlign = (unsigned) TLI.getTargetData()->getABITypeAlignment(Ty);
+    if (NewAlign > Align) {
+      // Give the stack frame object a larger alignment if needed.
+      if (MFI->getObjectAlignment(FI->getIndex()) < NewAlign)
+        MFI->setObjectAlignment(FI->getIndex(), NewAlign);
+      Align = NewAlign;
+    }
+  }
 
-  bool isZeroStr = CopyFromStr && Str.empty();
   SmallVector<SDValue, 8> OutChains;
   unsigned NumMemOps = MemOps.size();
   uint64_t SrcOff = 0, DstOff = 0;
@@ -3313,16 +3326,17 @@
     unsigned VTSize = VT.getSizeInBits() / 8;
     SDValue Value, Store;
 
-    if (CopyFromStr && (isZeroStr || !VT.isVector())) {
+    if (CopyFromStr &&
+        (isZeroStr || (VT.isInteger() && !VT.isVector()))) {
       // It's unlikely a store of a vector immediate can be done in a single
       // instruction. It would require a load from a constantpool first.
-      // We also handle store a vector with all zero's.
+      // We only handle zero vectors here.
       // FIXME: Handle other cases where store of vector immediate is done in
       // a single instruction.
       Value = getMemsetStringVal(VT, dl, DAG, TLI, Str, SrcOff);
       Store = DAG.getStore(Chain, dl, Value,
                            getMemBasePlusOffset(Dst, DstOff, DAG),
-                           DstSV, DstSVOff + DstOff, false, false, DstAlign);
+                           DstSV, DstSVOff + DstOff, isVol, false, Align);
     } else {
       // The type might not be legal for the target.  This should only happen
       // if the type is smaller than a legal type, as on PPC, so the right
@@ -3333,11 +3347,12 @@
       assert(NVT.bitsGE(VT));
       Value = DAG.getExtLoad(ISD::EXTLOAD, dl, NVT, Chain,
                              getMemBasePlusOffset(Src, SrcOff, DAG),
-                             SrcSV, SrcSVOff + SrcOff, VT, false, false, Align);
+                             SrcSV, SrcSVOff + SrcOff, VT, isVol, false,
+                             MinAlign(SrcAlign, SrcOff));
       Store = DAG.getTruncStore(Chain, dl, Value,
                                 getMemBasePlusOffset(Dst, DstOff, DAG),
-                                DstSV, DstSVOff + DstOff, VT, false, false,
-                                DstAlign);
+                                DstSV, DstSVOff + DstOff, VT, isVol, false,
+                                Align);
     }
     OutChains.push_back(Store);
     SrcOff += VTSize;
@@ -3349,28 +3364,49 @@
 }
 
 static SDValue getMemmoveLoadsAndStores(SelectionDAG &DAG, DebugLoc dl,
-                                          SDValue Chain, SDValue Dst,
-                                          SDValue Src, uint64_t Size,
-                                          unsigned Align, bool AlwaysInline,
-                                          const Value *DstSV, uint64_t DstSVOff,
-                                          const Value *SrcSV, uint64_t SrcSVOff){
-  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+                                        SDValue Chain, SDValue Dst,
+                                        SDValue Src, uint64_t Size,
+                                        unsigned Align,  bool isVol,
+                                        bool AlwaysInline,
+                                        const Value *DstSV, uint64_t DstSVOff,
+                                        const Value *SrcSV, uint64_t SrcSVOff) {
+  // Turn a memmove of undef to nop.
+  if (Src.getOpcode() == ISD::UNDEF)
+    return Chain;
 
   // Expand memmove to a series of load and store ops if the size operand falls
   // below a certain threshold.
+  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
   std::vector<EVT> MemOps;
   uint64_t Limit = -1ULL;
   if (!AlwaysInline)
     Limit = TLI.getMaxStoresPerMemmove();
-  unsigned DstAlign = Align;  // Destination alignment can change.
-  std::string Str;
-  bool CopyFromStr;
-  if (!MeetsMaxMemopRequirement(MemOps, Dst, Src, Limit, Size, DstAlign,
-                                Str, CopyFromStr, DAG, TLI))
+  bool DstAlignCanChange = false;
+  MachineFrameInfo *MFI = DAG.getMachineFunction().getFrameInfo();
+  FrameIndexSDNode *FI = dyn_cast<FrameIndexSDNode>(Dst);
+  if (FI && !MFI->isFixedObjectIndex(FI->getIndex()))
+    DstAlignCanChange = true;
+  unsigned SrcAlign = DAG.InferPtrAlignment(Src);
+  if (Align > SrcAlign)
+    SrcAlign = Align;
+
+  if (!FindOptimalMemOpLowering(MemOps, Limit, Size,
+                                (DstAlignCanChange ? 0 : Align),
+                                SrcAlign, true, false, DAG, TLI))
     return SDValue();
 
-  uint64_t SrcOff = 0, DstOff = 0;
+  if (DstAlignCanChange) {
+    const Type *Ty = MemOps[0].getTypeForEVT(*DAG.getContext());
+    unsigned NewAlign = (unsigned) TLI.getTargetData()->getABITypeAlignment(Ty);
+    if (NewAlign > Align) {
+      // Give the stack frame object a larger alignment if needed.
+      if (MFI->getObjectAlignment(FI->getIndex()) < NewAlign)
+        MFI->setObjectAlignment(FI->getIndex(), NewAlign);
+      Align = NewAlign;
+    }
+  }
 
+  uint64_t SrcOff = 0, DstOff = 0;
   SmallVector<SDValue, 8> LoadValues;
   SmallVector<SDValue, 8> LoadChains;
   SmallVector<SDValue, 8> OutChains;
@@ -3382,7 +3418,7 @@
 
     Value = DAG.getLoad(VT, dl, Chain,
                         getMemBasePlusOffset(Src, SrcOff, DAG),
-                        SrcSV, SrcSVOff + SrcOff, false, false, Align);
+                        SrcSV, SrcSVOff + SrcOff, isVol, false, SrcAlign);
     LoadValues.push_back(Value);
     LoadChains.push_back(Value.getValue(1));
     SrcOff += VTSize;
@@ -3397,7 +3433,7 @@
 
     Store = DAG.getStore(Chain, dl, LoadValues[i],
                          getMemBasePlusOffset(Dst, DstOff, DAG),
-                         DstSV, DstSVOff + DstOff, false, false, DstAlign);
+                         DstSV, DstSVOff + DstOff, isVol, false, Align);
     OutChains.push_back(Store);
     DstOff += VTSize;
   }
@@ -3407,24 +3443,43 @@
 }
 
 static SDValue getMemsetStores(SelectionDAG &DAG, DebugLoc dl,
-                                 SDValue Chain, SDValue Dst,
-                                 SDValue Src, uint64_t Size,
-                                 unsigned Align,
-                                 const Value *DstSV, uint64_t DstSVOff) {
-  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+                               SDValue Chain, SDValue Dst,
+                               SDValue Src, uint64_t Size,
+                               unsigned Align, bool isVol,
+                               const Value *DstSV, uint64_t DstSVOff) {
+  // Turn a memset of undef to nop.
+  if (Src.getOpcode() == ISD::UNDEF)
+    return Chain;
 
   // Expand memset to a series of load/store ops if the size operand
   // falls below a certain threshold.
+  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
   std::vector<EVT> MemOps;
-  std::string Str;
-  bool CopyFromStr;
-  if (!MeetsMaxMemopRequirement(MemOps, Dst, Src, TLI.getMaxStoresPerMemset(),
-                                Size, Align, Str, CopyFromStr, DAG, TLI))
+  bool DstAlignCanChange = false;
+  MachineFrameInfo *MFI = DAG.getMachineFunction().getFrameInfo();
+  FrameIndexSDNode *FI = dyn_cast<FrameIndexSDNode>(Dst);
+  if (FI && !MFI->isFixedObjectIndex(FI->getIndex()))
+    DstAlignCanChange = true;
+  bool NonScalarIntSafe =
+    isa<ConstantSDNode>(Src) && cast<ConstantSDNode>(Src)->isNullValue();
+  if (!FindOptimalMemOpLowering(MemOps, TLI.getMaxStoresPerMemset(),
+                                Size, (DstAlignCanChange ? 0 : Align), 0,
+                                NonScalarIntSafe, false, DAG, TLI))
     return SDValue();
 
+  if (DstAlignCanChange) {
+    const Type *Ty = MemOps[0].getTypeForEVT(*DAG.getContext());
+    unsigned NewAlign = (unsigned) TLI.getTargetData()->getABITypeAlignment(Ty);
+    if (NewAlign > Align) {
+      // Give the stack frame object a larger alignment if needed.
+      if (MFI->getObjectAlignment(FI->getIndex()) < NewAlign)
+        MFI->setObjectAlignment(FI->getIndex(), NewAlign);
+      Align = NewAlign;
+    }
+  }
+
   SmallVector<SDValue, 8> OutChains;
   uint64_t DstOff = 0;
-
   unsigned NumMemOps = MemOps.size();
   for (unsigned i = 0; i < NumMemOps; i++) {
     EVT VT = MemOps[i];
@@ -3432,7 +3487,7 @@
     SDValue Value = getMemsetValue(Src, VT, DAG, dl);
     SDValue Store = DAG.getStore(Chain, dl, Value,
                                  getMemBasePlusOffset(Dst, DstOff, DAG),
-                                 DstSV, DstSVOff + DstOff, false, false, 0);
+                                 DstSV, DstSVOff + DstOff, isVol, false, 0);
     OutChains.push_back(Store);
     DstOff += VTSize;
   }
@@ -3443,7 +3498,7 @@
 
 SDValue SelectionDAG::getMemcpy(SDValue Chain, DebugLoc dl, SDValue Dst,
                                 SDValue Src, SDValue Size,
-                                unsigned Align, bool AlwaysInline,
+                                unsigned Align, bool isVol, bool AlwaysInline,
                                 const Value *DstSV, uint64_t DstSVOff,
                                 const Value *SrcSV, uint64_t SrcSVOff) {
 
@@ -3455,10 +3510,9 @@
     if (ConstantSize->isNullValue())
       return Chain;
 
-    SDValue Result =
-      getMemcpyLoadsAndStores(*this, dl, Chain, Dst, Src,
-                              ConstantSize->getZExtValue(),
-                              Align, false, DstSV, DstSVOff, SrcSV, SrcSVOff);
+    SDValue Result = getMemcpyLoadsAndStores(*this, dl, Chain, Dst, Src,
+                                             ConstantSize->getZExtValue(),Align,
+                                isVol, false, DstSV, DstSVOff, SrcSV, SrcSVOff);
     if (Result.getNode())
       return Result;
   }
@@ -3467,7 +3521,7 @@
   // code. If the target chooses to do this, this is the next best.
   SDValue Result =
     TLI.EmitTargetCodeForMemcpy(*this, dl, Chain, Dst, Src, Size, Align,
-                                AlwaysInline,
+                                isVol, AlwaysInline,
                                 DstSV, DstSVOff, SrcSV, SrcSVOff);
   if (Result.getNode())
     return Result;
@@ -3477,10 +3531,16 @@
   if (AlwaysInline) {
     assert(ConstantSize && "AlwaysInline requires a constant size!");
     return getMemcpyLoadsAndStores(*this, dl, Chain, Dst, Src,
-                                   ConstantSize->getZExtValue(), Align, true,
-                                   DstSV, DstSVOff, SrcSV, SrcSVOff);
+                                   ConstantSize->getZExtValue(), Align, isVol,
+                                   true, DstSV, DstSVOff, SrcSV, SrcSVOff);
   }
 
+  // FIXME: If the memcpy is volatile (isVol), lowering it to a plain libc
+  // memcpy is not guaranteed to be safe. libc memcpys aren't required to
+  // respect volatile, so they may do things like read or write memory
+  // beyond the given memory regions. But fixing this isn't easy, and most
+  // people don't care.
+
   // Emit a library call.
   TargetLowering::ArgListTy Args;
   TargetLowering::ArgListEntry Entry;
@@ -3502,7 +3562,7 @@
 
 SDValue SelectionDAG::getMemmove(SDValue Chain, DebugLoc dl, SDValue Dst,
                                  SDValue Src, SDValue Size,
-                                 unsigned Align,
+                                 unsigned Align, bool isVol,
                                  const Value *DstSV, uint64_t DstSVOff,
                                  const Value *SrcSV, uint64_t SrcSVOff) {
 
@@ -3516,8 +3576,8 @@
 
     SDValue Result =
       getMemmoveLoadsAndStores(*this, dl, Chain, Dst, Src,
-                               ConstantSize->getZExtValue(),
-                               Align, false, DstSV, DstSVOff, SrcSV, SrcSVOff);
+                               ConstantSize->getZExtValue(), Align, isVol,
+                               false, DstSV, DstSVOff, SrcSV, SrcSVOff);
     if (Result.getNode())
       return Result;
   }
@@ -3525,11 +3585,14 @@
   // Then check to see if we should lower the memmove with target-specific
   // code. If the target chooses to do this, this is the next best.
   SDValue Result =
-    TLI.EmitTargetCodeForMemmove(*this, dl, Chain, Dst, Src, Size, Align,
+    TLI.EmitTargetCodeForMemmove(*this, dl, Chain, Dst, Src, Size, Align, isVol,
                                  DstSV, DstSVOff, SrcSV, SrcSVOff);
   if (Result.getNode())
     return Result;
 
+  // FIXME: If the memmove is volatile, lowering it to plain libc memmove may
+  // not be safe.  See memcpy above for more details.
+
   // Emit a library call.
   TargetLowering::ArgListTy Args;
   TargetLowering::ArgListEntry Entry;
@@ -3551,7 +3614,7 @@
 
 SDValue SelectionDAG::getMemset(SDValue Chain, DebugLoc dl, SDValue Dst,
                                 SDValue Src, SDValue Size,
-                                unsigned Align,
+                                unsigned Align, bool isVol,
                                 const Value *DstSV, uint64_t DstSVOff) {
 
   // Check to see if we should lower the memset to stores first.
@@ -3564,7 +3627,8 @@
 
     SDValue Result =
       getMemsetStores(*this, dl, Chain, Dst, Src, ConstantSize->getZExtValue(),
-                      Align, DstSV, DstSVOff);
+                      Align, isVol, DstSV, DstSVOff);
+
     if (Result.getNode())
       return Result;
   }
@@ -3572,12 +3636,12 @@
   // Then check to see if we should lower the memset with target-specific
   // code. If the target chooses to do this, this is the next best.
   SDValue Result =
-    TLI.EmitTargetCodeForMemset(*this, dl, Chain, Dst, Src, Size, Align,
+    TLI.EmitTargetCodeForMemset(*this, dl, Chain, Dst, Src, Size, Align, isVol,
                                 DstSV, DstSVOff);
   if (Result.getNode())
     return Result;
 
-  // Emit a library call.
+  // Emit a library call.  
   const Type *IntPtrTy = TLI.getTargetData()->getIntPtrType(*getContext());
   TargetLowering::ArgListTy Args;
   TargetLowering::ArgListEntry Entry;
@@ -3654,8 +3718,8 @@
     cast<AtomicSDNode>(E)->refineAlignment(MMO);
     return SDValue(E, 0);
   }
-  SDNode* N = NodeAllocator.Allocate<AtomicSDNode>();
-  new (N) AtomicSDNode(Opcode, dl, VTs, MemVT, Chain, Ptr, Cmp, Swp, MMO);
+  SDNode *N = new (NodeAllocator) AtomicSDNode(Opcode, dl, VTs, MemVT, Chain,
+                                               Ptr, Cmp, Swp, MMO);
   CSEMap.InsertNode(N, IP);
   AllNodes.push_back(N);
   return SDValue(N, 0);
@@ -3717,8 +3781,8 @@
     cast<AtomicSDNode>(E)->refineAlignment(MMO);
     return SDValue(E, 0);
   }
-  SDNode* N = NodeAllocator.Allocate<AtomicSDNode>();
-  new (N) AtomicSDNode(Opcode, dl, VTs, MemVT, Chain, Ptr, Val, MMO);
+  SDNode *N = new (NodeAllocator) AtomicSDNode(Opcode, dl, VTs, MemVT, Chain,
+                                               Ptr, Val, MMO);
   CSEMap.InsertNode(N, IP);
   AllNodes.push_back(N);
   return SDValue(N, 0);
@@ -3796,12 +3860,12 @@
       return SDValue(E, 0);
     }
 
-    N = NodeAllocator.Allocate<MemIntrinsicSDNode>();
-    new (N) MemIntrinsicSDNode(Opcode, dl, VTList, Ops, NumOps, MemVT, MMO);
+    N = new (NodeAllocator) MemIntrinsicSDNode(Opcode, dl, VTList, Ops, NumOps,
+                                               MemVT, MMO);
     CSEMap.InsertNode(N, IP);
   } else {
-    N = NodeAllocator.Allocate<MemIntrinsicSDNode>();
-    new (N) MemIntrinsicSDNode(Opcode, dl, VTList, Ops, NumOps, MemVT, MMO);
+    N = new (NodeAllocator) MemIntrinsicSDNode(Opcode, dl, VTList, Ops, NumOps,
+                                               MemVT, MMO);
   }
   AllNodes.push_back(N);
   return SDValue(N, 0);
@@ -3874,8 +3938,8 @@
     cast<LoadSDNode>(E)->refineAlignment(MMO);
     return SDValue(E, 0);
   }
-  SDNode *N = NodeAllocator.Allocate<LoadSDNode>();
-  new (N) LoadSDNode(Ops, dl, VTs, AM, ExtType, MemVT, MMO);
+  SDNode *N = new (NodeAllocator) LoadSDNode(Ops, dl, VTs, AM, ExtType,
+                                             MemVT, MMO);
   CSEMap.InsertNode(N, IP);
   AllNodes.push_back(N);
   return SDValue(N, 0);
@@ -3956,8 +4020,8 @@
     cast<StoreSDNode>(E)->refineAlignment(MMO);
     return SDValue(E, 0);
   }
-  SDNode *N = NodeAllocator.Allocate<StoreSDNode>();
-  new (N) StoreSDNode(Ops, dl, VTs, ISD::UNINDEXED, false, VT, MMO);
+  SDNode *N = new (NodeAllocator) StoreSDNode(Ops, dl, VTs, ISD::UNINDEXED,
+                                              false, VT, MMO);
   CSEMap.InsertNode(N, IP);
   AllNodes.push_back(N);
   return SDValue(N, 0);
@@ -4020,8 +4084,8 @@
     cast<StoreSDNode>(E)->refineAlignment(MMO);
     return SDValue(E, 0);
   }
-  SDNode *N = NodeAllocator.Allocate<StoreSDNode>();
-  new (N) StoreSDNode(Ops, dl, VTs, ISD::UNINDEXED, true, SVT, MMO);
+  SDNode *N = new (NodeAllocator) StoreSDNode(Ops, dl, VTs, ISD::UNINDEXED,
+                                              true, SVT, MMO);
   CSEMap.InsertNode(N, IP);
   AllNodes.push_back(N);
   return SDValue(N, 0);
@@ -4043,10 +4107,10 @@
   if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
     return SDValue(E, 0);
 
-  SDNode *N = NodeAllocator.Allocate<StoreSDNode>();
-  new (N) StoreSDNode(Ops, dl, VTs, AM,
-                      ST->isTruncatingStore(), ST->getMemoryVT(),
-                      ST->getMemOperand());
+  SDNode *N = new (NodeAllocator) StoreSDNode(Ops, dl, VTs, AM,
+                                              ST->isTruncatingStore(),
+                                              ST->getMemoryVT(),
+                                              ST->getMemOperand());
   CSEMap.InsertNode(N, IP);
   AllNodes.push_back(N);
   return SDValue(N, 0);
@@ -4117,12 +4181,10 @@
     if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
       return SDValue(E, 0);
 
-    N = NodeAllocator.Allocate<SDNode>();
-    new (N) SDNode(Opcode, DL, VTs, Ops, NumOps);
+    N = new (NodeAllocator) SDNode(Opcode, DL, VTs, Ops, NumOps);
     CSEMap.InsertNode(N, IP);
   } else {
-    N = NodeAllocator.Allocate<SDNode>();
-    new (N) SDNode(Opcode, DL, VTs, Ops, NumOps);
+    N = new (NodeAllocator) SDNode(Opcode, DL, VTs, Ops, NumOps);
   }
 
   AllNodes.push_back(N);
@@ -4185,32 +4247,26 @@
       return SDValue(E, 0);
 
     if (NumOps == 1) {
-      N = NodeAllocator.Allocate<UnarySDNode>();
-      new (N) UnarySDNode(Opcode, DL, VTList, Ops[0]);
+      N = new (NodeAllocator) UnarySDNode(Opcode, DL, VTList, Ops[0]);
     } else if (NumOps == 2) {
-      N = NodeAllocator.Allocate<BinarySDNode>();
-      new (N) BinarySDNode(Opcode, DL, VTList, Ops[0], Ops[1]);
+      N = new (NodeAllocator) BinarySDNode(Opcode, DL, VTList, Ops[0], Ops[1]);
     } else if (NumOps == 3) {
-      N = NodeAllocator.Allocate<TernarySDNode>();
-      new (N) TernarySDNode(Opcode, DL, VTList, Ops[0], Ops[1], Ops[2]);
+      N = new (NodeAllocator) TernarySDNode(Opcode, DL, VTList, Ops[0], Ops[1],
+                                            Ops[2]);
     } else {
-      N = NodeAllocator.Allocate<SDNode>();
-      new (N) SDNode(Opcode, DL, VTList, Ops, NumOps);
+      N = new (NodeAllocator) SDNode(Opcode, DL, VTList, Ops, NumOps);
     }
     CSEMap.InsertNode(N, IP);
   } else {
     if (NumOps == 1) {
-      N = NodeAllocator.Allocate<UnarySDNode>();
-      new (N) UnarySDNode(Opcode, DL, VTList, Ops[0]);
+      N = new (NodeAllocator) UnarySDNode(Opcode, DL, VTList, Ops[0]);
     } else if (NumOps == 2) {
-      N = NodeAllocator.Allocate<BinarySDNode>();
-      new (N) BinarySDNode(Opcode, DL, VTList, Ops[0], Ops[1]);
+      N = new (NodeAllocator) BinarySDNode(Opcode, DL, VTList, Ops[0], Ops[1]);
     } else if (NumOps == 3) {
-      N = NodeAllocator.Allocate<TernarySDNode>();
-      new (N) TernarySDNode(Opcode, DL, VTList, Ops[0], Ops[1], Ops[2]);
+      N = new (NodeAllocator) TernarySDNode(Opcode, DL, VTList, Ops[0], Ops[1],
+                                            Ops[2]);
     } else {
-      N = NodeAllocator.Allocate<SDNode>();
-      new (N) SDNode(Opcode, DL, VTList, Ops, NumOps);
+      N = new (NodeAllocator) SDNode(Opcode, DL, VTList, Ops, NumOps);
     }
   }
   AllNodes.push_back(N);
@@ -4635,7 +4691,7 @@
         // remainder of the current SelectionDAG iteration, so we can allocate
         // the operands directly out of a pool with no recycling metadata.
         MN->InitOperands(OperandAllocator.Allocate<SDUse>(NumOps),
-                        Ops, NumOps);
+                         Ops, NumOps);
       else
         MN->InitOperands(MN->LocalOperands, Ops, NumOps);
       MN->OperandsNeedDelete = false;
@@ -4809,8 +4865,7 @@
   }
 
   // Allocate a new MachineSDNode.
-  N = NodeAllocator.Allocate<MachineSDNode>();
-  new (N) MachineSDNode(~Opcode, DL, VTs);
+  N = new (NodeAllocator) MachineSDNode(~Opcode, DL, VTs);
 
   // Initialize the operands list.
   if (NumOps > array_lengthof(N->LocalOperands))
@@ -4869,6 +4924,26 @@
   return NULL;
 }
 
+/// getDbgValue - Creates a SDDbgValue node.
+///
+SDDbgValue *
+SelectionDAG::getDbgValue(MDNode *MDPtr, SDNode *N, unsigned R, uint64_t Off,
+                          DebugLoc DL, unsigned O) {
+  return new (Allocator) SDDbgValue(MDPtr, N, R, Off, DL, O);
+}
+
+SDDbgValue *
+SelectionDAG::getDbgValue(MDNode *MDPtr, const Value *C, uint64_t Off,
+                          DebugLoc DL, unsigned O) {
+  return new (Allocator) SDDbgValue(MDPtr, C, Off, DL, O);
+}
+
+SDDbgValue *
+SelectionDAG::getDbgValue(MDNode *MDPtr, unsigned FI, uint64_t Off,
+                          DebugLoc DL, unsigned O) {
+  return new (Allocator) SDDbgValue(MDPtr, FI, Off, DL, O);
+}
+
 namespace {
 
 /// RAUWUpdateListener - Helper for ReplaceAllUsesWith - When the node
@@ -5264,6 +5339,13 @@
   return Ordering->getOrder(SD);
 }
 
+/// AddDbgValue - Add a dbg_value SDNode. If SD is non-null that means the
+/// value is produced by SD.
+void SelectionDAG::AddDbgValue(SDDbgValue *DB, SDNode *SD, bool isParameter) {
+  DbgInfo->add(DB, SD, isParameter);
+  if (SD)
+    SD->setHasDebugValue(true);
+}
 
 //===----------------------------------------------------------------------===//
 //                              SDNode Class
@@ -5275,9 +5357,8 @@
 
 GlobalAddressSDNode::GlobalAddressSDNode(unsigned Opc, const GlobalValue *GA,
                                          EVT VT, int64_t o, unsigned char TF)
-  : SDNode(Opc, DebugLoc::getUnknownLoc(), getSDVTList(VT)),
-    Offset(o), TargetFlags(TF) {
-  TheGlobal = const_cast<GlobalValue*>(GA);
+  : SDNode(Opc, DebugLoc(), getSDVTList(VT)), Offset(o), TargetFlags(TF) {
+  TheGlobal = GA;
 }
 
 MemSDNode::MemSDNode(unsigned Opc, DebugLoc dl, SDVTList VTs, EVT memvt,
@@ -5497,6 +5578,7 @@
   case ISD::PCMARKER:      return "PCMarker";
   case ISD::READCYCLECOUNTER: return "ReadCycleCounter";
   case ISD::SRCVALUE:      return "SrcValue";
+  case ISD::MDNODE_SDNODE: return "MDNode";
   case ISD::EntryToken:    return "EntryToken";
   case ISD::TokenFactor:   return "TokenFactor";
   case ISD::AssertSext:    return "AssertSext";
@@ -5639,6 +5721,8 @@
   case ISD::FP_TO_SINT:  return "fp_to_sint";
   case ISD::FP_TO_UINT:  return "fp_to_uint";
   case ISD::BIT_CONVERT: return "bit_convert";
+  case ISD::FP16_TO_FP32: return "fp16_to_fp32";
+  case ISD::FP32_TO_FP16: return "fp32_to_fp16";
 
   case ISD::CONVERT_RNDSAT: {
     switch (cast<CvtRndSatSDNode>(this)->getCvtCode()) {
@@ -5863,6 +5947,11 @@
       OS << "<" << M->getValue() << ">";
     else
       OS << "<null>";
+  } else if (const MDNodeSDNode *MD = dyn_cast<MDNodeSDNode>(this)) {
+    if (MD->getMD())
+      OS << "<" << MD->getMD() << ">";
+    else
+      OS << "<null>";
   } else if (const VTSDNode *N = dyn_cast<VTSDNode>(this)) {
     OS << ":" << N->getVT().getEVTString();
   }
@@ -5911,7 +6000,7 @@
   if (G)
     if (unsigned Order = G->GetOrdering(this))
       OS << " [ORD=" << Order << ']';
-  
+
   if (getNodeId() != -1)
     OS << " [ID=" << getNodeId() << ']';
 }
@@ -6078,8 +6167,8 @@
       return true;
   }
 
-  GlobalValue *GV1 = NULL;
-  GlobalValue *GV2 = NULL;
+  const GlobalValue *GV1 = NULL;
+  const GlobalValue *GV2 = NULL;
   int64_t Offset1 = 0;
   int64_t Offset2 = 0;
   bool isGA1 = TLI.isGAPlusOffset(Loc.getNode(), GV1, Offset1);
@@ -6094,10 +6183,22 @@
 /// it cannot be inferred.
 unsigned SelectionDAG::InferPtrAlignment(SDValue Ptr) const {
   // If this is a GlobalAddress + cst, return the alignment.
-  GlobalValue *GV;
+  const GlobalValue *GV;
   int64_t GVOffset = 0;
-  if (TLI.isGAPlusOffset(Ptr.getNode(), GV, GVOffset))
-    return MinAlign(GV->getAlignment(), GVOffset);
+  if (TLI.isGAPlusOffset(Ptr.getNode(), GV, GVOffset)) {
+    // If GV has specified alignment, then use it. Otherwise, use the preferred
+    // alignment.
+    unsigned Align = GV->getAlignment();
+    if (!Align) {
+      if (const GlobalVariable *GVar = dyn_cast<GlobalVariable>(GV)) {
+        if (GVar->hasInitializer()) {
+          const TargetData *TD = TLI.getTargetData();
+          Align = TD->getPreferredAlignment(GVar);
+        }
+      }
+    }
+    return MinAlign(Align, GVOffset);
+  }
 
   // If this is a direct reference to a stack slot, use information about the
   // stack slot's alignment.
@@ -6251,8 +6352,8 @@
     if (OpVal.getOpcode() == ISD::UNDEF)
       SplatUndef |= APInt::getBitsSet(sz, BitPos, BitPos + EltBitSize);
     else if (ConstantSDNode *CN = dyn_cast<ConstantSDNode>(OpVal))
-      SplatValue |= (APInt(CN->getAPIntValue()).zextOrTrunc(EltBitSize).
-                     zextOrTrunc(sz) << BitPos);
+      SplatValue |= APInt(CN->getAPIntValue()).zextOrTrunc(EltBitSize).
+                    zextOrTrunc(sz) << BitPos;
     else if (ConstantFPSDNode *CN = dyn_cast<ConstantFPSDNode>(OpVal))
       SplatValue |= CN->getValueAPF().bitcastToAPInt().zextOrTrunc(sz) <<BitPos;
      else
diff --git a/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp b/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
index 05be9a1..ade3836 100644
--- a/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
+++ b/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
@@ -12,6 +12,7 @@
 //===----------------------------------------------------------------------===//
 
 #define DEBUG_TYPE "isel"
+#include "SDNodeDbgValue.h"
 #include "SelectionDAGBuilder.h"
 #include "FunctionLoweringInfo.h"
 #include "llvm/ADT/BitVector.h"
@@ -27,7 +28,9 @@
 #include "llvm/Instructions.h"
 #include "llvm/Intrinsics.h"
 #include "llvm/IntrinsicInst.h"
+#include "llvm/LLVMContext.h"
 #include "llvm/Module.h"
+#include "llvm/CodeGen/Analysis.h"
 #include "llvm/CodeGen/FastISel.h"
 #include "llvm/CodeGen/GCStrategy.h"
 #include "llvm/CodeGen/GCMetadata.h"
@@ -39,7 +42,6 @@
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/PseudoSourceValue.h"
 #include "llvm/CodeGen/SelectionDAG.h"
-#include "llvm/CodeGen/DwarfWriter.h"
 #include "llvm/Analysis/DebugInfo.h"
 #include "llvm/Target/TargetRegisterInfo.h"
 #include "llvm/Target/TargetData.h"
@@ -168,7 +170,7 @@
     /// AddInlineAsmOperands - Add this value to the specified inlineasm node
     /// operand list.  This adds the code marker, matching input operand index
     /// (if applicable), and includes the number of values added into it.
-    void AddInlineAsmOperands(unsigned Code,
+    void AddInlineAsmOperands(unsigned Kind,
                               bool HasMatching, unsigned MatchingIdx,
                               SelectionDAG &DAG,
                               std::vector<SDValue> &Ops) const;
@@ -533,7 +535,7 @@
   TD = DAG.getTarget().getTargetData();
 }
 
-/// clear - Clear out the curret SelectionDAG and the associated
+/// clear - Clear out the current SelectionDAG and the associated
 /// state and prepare this SelectionDAGBuilder object to be used
 /// for a new block. This doesn't clear out information about
 /// additional blocks that are needed to complete switch lowering
@@ -545,7 +547,7 @@
   PendingExports.clear();
   EdgeMapping.clear();
   DAG.clear();
-  CurDebugLoc = DebugLoc::getUnknownLoc();
+  CurDebugLoc = DebugLoc();
   HasTailCall = false;
 }
 
@@ -612,11 +614,26 @@
     AssignOrderingToNode(Node->getOperand(I).getNode());
 }
 
-void SelectionDAGBuilder::visit(Instruction &I) {
+void SelectionDAGBuilder::visit(const Instruction &I) {
+  // Set up outgoing PHI node register values before emitting the terminator.
+  if (isa<TerminatorInst>(&I))
+    HandlePHINodesInSuccessorBlocks(I.getParent());
+
+  CurDebugLoc = I.getDebugLoc();
+
   visit(I.getOpcode(), I);
+
+  if (!isa<TerminatorInst>(&I) && !HasTailCall)
+    CopyToExportRegsIfNeeded(&I);
+
+  CurDebugLoc = DebugLoc();
 }
 
-void SelectionDAGBuilder::visit(unsigned Opcode, User &I) {
+void SelectionDAGBuilder::visitPHI(const PHINode &) {
+  llvm_unreachable("SelectionDAGBuilder shouldn't visit PHI nodes!");
+}
+
+void SelectionDAGBuilder::visit(unsigned Opcode, const User &I) {
   // Note: this doesn't use InstVisitor, because it has to work with
   // ConstantExpr's in addition to instructions.
   switch (Opcode) {
@@ -638,28 +655,28 @@
   SDValue &N = NodeMap[V];
   if (N.getNode()) return N;
 
-  if (Constant *C = const_cast<Constant*>(dyn_cast<Constant>(V))) {
+  if (const Constant *C = dyn_cast<Constant>(V)) {
     EVT VT = TLI.getValueType(V->getType(), true);
 
-    if (ConstantInt *CI = dyn_cast<ConstantInt>(C))
+    if (const ConstantInt *CI = dyn_cast<ConstantInt>(C))
       return N = DAG.getConstant(*CI, VT);
 
-    if (GlobalValue *GV = dyn_cast<GlobalValue>(C))
+    if (const GlobalValue *GV = dyn_cast<GlobalValue>(C))
       return N = DAG.getGlobalAddress(GV, VT);
 
     if (isa<ConstantPointerNull>(C))
       return N = DAG.getConstant(0, TLI.getPointerTy());
 
-    if (ConstantFP *CFP = dyn_cast<ConstantFP>(C))
+    if (const ConstantFP *CFP = dyn_cast<ConstantFP>(C))
       return N = DAG.getConstantFP(*CFP, VT);
 
     if (isa<UndefValue>(C) && !V->getType()->isAggregateType())
       return N = DAG.getUNDEF(VT);
 
-    if (ConstantExpr *CE = dyn_cast<ConstantExpr>(C)) {
+    if (const ConstantExpr *CE = dyn_cast<ConstantExpr>(C)) {
       visit(CE->getOpcode(), *CE);
       SDValue N1 = NodeMap[V];
-      assert(N1.getNode() && "visit didn't populate the ValueMap!");
+      assert(N1.getNode() && "visit didn't populate the NodeMap!");
       return N1;
     }
 
@@ -704,7 +721,7 @@
                                 getCurDebugLoc());
     }
 
-    if (BlockAddress *BA = dyn_cast<BlockAddress>(C))
+    if (const BlockAddress *BA = dyn_cast<BlockAddress>(C))
       return DAG.getBlockAddress(BA, VT);
 
     const VectorType *VecTy = cast<VectorType>(V->getType());
@@ -713,7 +730,7 @@
     // Now that we know the number and type of the elements, get that number of
     // elements into the Ops array based on what kind of constant it is.
     SmallVector<SDValue, 16> Ops;
-    if (ConstantVector *CP = dyn_cast<ConstantVector>(C)) {
+    if (const ConstantVector *CP = dyn_cast<ConstantVector>(C)) {
       for (unsigned i = 0; i != NumElements; ++i)
         Ops.push_back(getValue(CP->getOperand(i)));
     } else {
@@ -756,7 +773,7 @@
 static void getReturnInfo(const Type* ReturnType,
                    Attributes attr, SmallVectorImpl<EVT> &OutVTs,
                    SmallVectorImpl<ISD::ArgFlagsTy> &OutFlags,
-                   TargetLowering &TLI,
+                   const TargetLowering &TLI,
                    SmallVectorImpl<uint64_t> *Offsets = 0) {
   SmallVector<EVT, 4> ValueVTs;
   ComputeValueVTs(TLI, ReturnType, ValueVTs);
@@ -811,7 +828,7 @@
   }
 }
 
-void SelectionDAGBuilder::visitRet(ReturnInst &I) {
+void SelectionDAGBuilder::visitRet(const ReturnInst &I) {
   SDValue Chain = getControlRoot();
   SmallVector<ISD::OutputArg, 8> Outs;
   FunctionLoweringInfo &FLI = DAG.getFunctionLoweringInfo();
@@ -916,18 +933,18 @@
 /// CopyToExportRegsIfNeeded - If the given value has virtual registers
 /// created for it, emit nodes to copy the value into the virtual
 /// registers.
-void SelectionDAGBuilder::CopyToExportRegsIfNeeded(Value *V) {
-  if (!V->use_empty()) {
-    DenseMap<const Value *, unsigned>::iterator VMI = FuncInfo.ValueMap.find(V);
-    if (VMI != FuncInfo.ValueMap.end())
-      CopyValueToVirtualRegister(V, VMI->second);
+void SelectionDAGBuilder::CopyToExportRegsIfNeeded(const Value *V) {
+  DenseMap<const Value *, unsigned>::iterator VMI = FuncInfo.ValueMap.find(V);
+  if (VMI != FuncInfo.ValueMap.end()) {
+    assert(!V->use_empty() && "Unused value assigned virtual registers!");
+    CopyValueToVirtualRegister(V, VMI->second);
   }
 }
 
 /// ExportFromCurrentBlock - If this condition isn't known to be exported from
 /// the current basic block, add it to ValueMap now so that we'll get a
 /// CopyTo/FromReg.
-void SelectionDAGBuilder::ExportFromCurrentBlock(Value *V) {
+void SelectionDAGBuilder::ExportFromCurrentBlock(const Value *V) {
   // No need to export constants.
   if (!isa<Instruction>(V) && !isa<Argument>(V)) return;
 
@@ -938,11 +955,11 @@
   CopyValueToVirtualRegister(V, Reg);
 }
 
-bool SelectionDAGBuilder::isExportableFromCurrentBlock(Value *V,
+bool SelectionDAGBuilder::isExportableFromCurrentBlock(const Value *V,
                                                      const BasicBlock *FromBB) {
   // The operands of the setcc have to be in this block.  We don't know
   // how to export them from some other block.
-  if (Instruction *VI = dyn_cast<Instruction>(V)) {
+  if (const Instruction *VI = dyn_cast<Instruction>(V)) {
     // Can export from current BB.
     if (VI->getParent() == FromBB)
       return true;
@@ -971,85 +988,31 @@
   return true;
 }
 
-/// getFCmpCondCode - Return the ISD condition code corresponding to
-/// the given LLVM IR floating-point condition code.  This includes
-/// consideration of global floating-point math flags.
-///
-static ISD::CondCode getFCmpCondCode(FCmpInst::Predicate Pred) {
-  ISD::CondCode FPC, FOC;
-  switch (Pred) {
-  case FCmpInst::FCMP_FALSE: FOC = FPC = ISD::SETFALSE; break;
-  case FCmpInst::FCMP_OEQ:   FOC = ISD::SETEQ; FPC = ISD::SETOEQ; break;
-  case FCmpInst::FCMP_OGT:   FOC = ISD::SETGT; FPC = ISD::SETOGT; break;
-  case FCmpInst::FCMP_OGE:   FOC = ISD::SETGE; FPC = ISD::SETOGE; break;
-  case FCmpInst::FCMP_OLT:   FOC = ISD::SETLT; FPC = ISD::SETOLT; break;
-  case FCmpInst::FCMP_OLE:   FOC = ISD::SETLE; FPC = ISD::SETOLE; break;
-  case FCmpInst::FCMP_ONE:   FOC = ISD::SETNE; FPC = ISD::SETONE; break;
-  case FCmpInst::FCMP_ORD:   FOC = FPC = ISD::SETO;   break;
-  case FCmpInst::FCMP_UNO:   FOC = FPC = ISD::SETUO;  break;
-  case FCmpInst::FCMP_UEQ:   FOC = ISD::SETEQ; FPC = ISD::SETUEQ; break;
-  case FCmpInst::FCMP_UGT:   FOC = ISD::SETGT; FPC = ISD::SETUGT; break;
-  case FCmpInst::FCMP_UGE:   FOC = ISD::SETGE; FPC = ISD::SETUGE; break;
-  case FCmpInst::FCMP_ULT:   FOC = ISD::SETLT; FPC = ISD::SETULT; break;
-  case FCmpInst::FCMP_ULE:   FOC = ISD::SETLE; FPC = ISD::SETULE; break;
-  case FCmpInst::FCMP_UNE:   FOC = ISD::SETNE; FPC = ISD::SETUNE; break;
-  case FCmpInst::FCMP_TRUE:  FOC = FPC = ISD::SETTRUE; break;
-  default:
-    llvm_unreachable("Invalid FCmp predicate opcode!");
-    FOC = FPC = ISD::SETFALSE;
-    break;
-  }
-  if (FiniteOnlyFPMath())
-    return FOC;
-  else
-    return FPC;
-}
-
-/// getICmpCondCode - Return the ISD condition code corresponding to
-/// the given LLVM IR integer condition code.
-///
-static ISD::CondCode getICmpCondCode(ICmpInst::Predicate Pred) {
-  switch (Pred) {
-  case ICmpInst::ICMP_EQ:  return ISD::SETEQ;
-  case ICmpInst::ICMP_NE:  return ISD::SETNE;
-  case ICmpInst::ICMP_SLE: return ISD::SETLE;
-  case ICmpInst::ICMP_ULE: return ISD::SETULE;
-  case ICmpInst::ICMP_SGE: return ISD::SETGE;
-  case ICmpInst::ICMP_UGE: return ISD::SETUGE;
-  case ICmpInst::ICMP_SLT: return ISD::SETLT;
-  case ICmpInst::ICMP_ULT: return ISD::SETULT;
-  case ICmpInst::ICMP_SGT: return ISD::SETGT;
-  case ICmpInst::ICMP_UGT: return ISD::SETUGT;
-  default:
-    llvm_unreachable("Invalid ICmp predicate opcode!");
-    return ISD::SETNE;
-  }
-}
-
 /// EmitBranchForMergedCondition - Helper method for FindMergedConditions.
 /// This function emits a branch and is used at the leaves of an OR or an
 /// AND operator tree.
 ///
 void
-SelectionDAGBuilder::EmitBranchForMergedCondition(Value *Cond,
+SelectionDAGBuilder::EmitBranchForMergedCondition(const Value *Cond,
                                                   MachineBasicBlock *TBB,
                                                   MachineBasicBlock *FBB,
-                                                  MachineBasicBlock *CurBB) {
+                                                  MachineBasicBlock *CurBB,
+                                                  MachineBasicBlock *SwitchBB) {
   const BasicBlock *BB = CurBB->getBasicBlock();
 
   // If the leaf of the tree is a comparison, merge the condition into
   // the caseblock.
-  if (CmpInst *BOp = dyn_cast<CmpInst>(Cond)) {
+  if (const CmpInst *BOp = dyn_cast<CmpInst>(Cond)) {
     // The operands of the cmp have to be in this block.  We don't know
     // how to export them from some other block.  If this is the first block
     // of the sequence, no exporting is needed.
-    if (CurBB == CurMBB ||
+    if (CurBB == SwitchBB ||
         (isExportableFromCurrentBlock(BOp->getOperand(0), BB) &&
          isExportableFromCurrentBlock(BOp->getOperand(1), BB))) {
       ISD::CondCode Condition;
-      if (ICmpInst *IC = dyn_cast<ICmpInst>(Cond)) {
+      if (const ICmpInst *IC = dyn_cast<ICmpInst>(Cond)) {
         Condition = getICmpCondCode(IC->getPredicate());
-      } else if (FCmpInst *FC = dyn_cast<FCmpInst>(Cond)) {
+      } else if (const FCmpInst *FC = dyn_cast<FCmpInst>(Cond)) {
         Condition = getFCmpCondCode(FC->getPredicate());
       } else {
         Condition = ISD::SETEQ; // silence warning.
@@ -1070,19 +1033,20 @@
 }
 
 /// FindMergedConditions - If Cond is an expression like
-void SelectionDAGBuilder::FindMergedConditions(Value *Cond,
+void SelectionDAGBuilder::FindMergedConditions(const Value *Cond,
                                                MachineBasicBlock *TBB,
                                                MachineBasicBlock *FBB,
                                                MachineBasicBlock *CurBB,
+                                               MachineBasicBlock *SwitchBB,
                                                unsigned Opc) {
   // If this node is not part of the or/and tree, emit it as a branch.
-  Instruction *BOp = dyn_cast<Instruction>(Cond);
+  const Instruction *BOp = dyn_cast<Instruction>(Cond);
   if (!BOp || !(isa<BinaryOperator>(BOp) || isa<CmpInst>(BOp)) ||
       (unsigned)BOp->getOpcode() != Opc || !BOp->hasOneUse() ||
       BOp->getParent() != CurBB->getBasicBlock() ||
       !InBlock(BOp->getOperand(0), CurBB->getBasicBlock()) ||
       !InBlock(BOp->getOperand(1), CurBB->getBasicBlock())) {
-    EmitBranchForMergedCondition(Cond, TBB, FBB, CurBB);
+    EmitBranchForMergedCondition(Cond, TBB, FBB, CurBB, SwitchBB);
     return;
   }
 
@@ -1102,10 +1066,10 @@
     //
 
     // Emit the LHS condition.
-    FindMergedConditions(BOp->getOperand(0), TBB, TmpBB, CurBB, Opc);
+    FindMergedConditions(BOp->getOperand(0), TBB, TmpBB, CurBB, SwitchBB, Opc);
 
     // Emit the RHS condition into TmpBB.
-    FindMergedConditions(BOp->getOperand(1), TBB, FBB, TmpBB, Opc);
+    FindMergedConditions(BOp->getOperand(1), TBB, FBB, TmpBB, SwitchBB, Opc);
   } else {
     assert(Opc == Instruction::And && "Unknown merge op!");
     // Codegen X & Y as:
@@ -1118,10 +1082,10 @@
     //  This requires creation of TmpBB after CurBB.
 
     // Emit the LHS condition.
-    FindMergedConditions(BOp->getOperand(0), TmpBB, FBB, CurBB, Opc);
+    FindMergedConditions(BOp->getOperand(0), TmpBB, FBB, CurBB, SwitchBB, Opc);
 
     // Emit the RHS condition into TmpBB.
-    FindMergedConditions(BOp->getOperand(1), TBB, FBB, TmpBB, Opc);
+    FindMergedConditions(BOp->getOperand(1), TBB, FBB, TmpBB, SwitchBB, Opc);
   }
 }
 
@@ -1156,19 +1120,21 @@
   return true;
 }
 
-void SelectionDAGBuilder::visitBr(BranchInst &I) {
+void SelectionDAGBuilder::visitBr(const BranchInst &I) {
+  MachineBasicBlock *BrMBB = FuncInfo.MBBMap[I.getParent()];
+
   // Update machine-CFG edges.
   MachineBasicBlock *Succ0MBB = FuncInfo.MBBMap[I.getSuccessor(0)];
 
   // Figure out which block is immediately after the current one.
   MachineBasicBlock *NextBlock = 0;
-  MachineFunction::iterator BBI = CurMBB;
+  MachineFunction::iterator BBI = BrMBB;
   if (++BBI != FuncInfo.MF->end())
     NextBlock = BBI;
 
   if (I.isUnconditional()) {
     // Update machine-CFG edges.
-    CurMBB->addSuccessor(Succ0MBB);
+    BrMBB->addSuccessor(Succ0MBB);
 
     // If this is not a fall-through branch, emit the branch.
     if (Succ0MBB != NextBlock)
@@ -1181,7 +1147,7 @@
 
   // If this condition is one of the special cases we handle, do special stuff
   // now.
-  Value *CondVal = I.getCondition();
+  const Value *CondVal = I.getCondition();
   MachineBasicBlock *Succ1MBB = FuncInfo.MBBMap[I.getSuccessor(1)];
 
   // If this is a series of conditions that are or'd or and'd together, emit
@@ -1199,15 +1165,16 @@
   //     cmp D, E
   //     jle foo
   //
-  if (BinaryOperator *BOp = dyn_cast<BinaryOperator>(CondVal)) {
+  if (const BinaryOperator *BOp = dyn_cast<BinaryOperator>(CondVal)) {
     if (BOp->hasOneUse() &&
         (BOp->getOpcode() == Instruction::And ||
          BOp->getOpcode() == Instruction::Or)) {
-      FindMergedConditions(BOp, Succ0MBB, Succ1MBB, CurMBB, BOp->getOpcode());
+      FindMergedConditions(BOp, Succ0MBB, Succ1MBB, BrMBB, BrMBB,
+                           BOp->getOpcode());
       // If the compares in later blocks need to use values not currently
       // exported from this block, export them now.  This block should always
       // be the first entry.
-      assert(SwitchCases[0].ThisBB == CurMBB && "Unexpected lowering!");
+      assert(SwitchCases[0].ThisBB == BrMBB && "Unexpected lowering!");
 
       // Allow some cases to be rejected.
       if (ShouldEmitAsBranches(SwitchCases)) {
@@ -1217,7 +1184,7 @@
         }
 
         // Emit the branch for this block.
-        visitSwitchCase(SwitchCases[0]);
+        visitSwitchCase(SwitchCases[0], BrMBB);
         SwitchCases.erase(SwitchCases.begin());
         return;
       }
@@ -1233,16 +1200,17 @@
 
   // Create a CaseBlock record representing this branch.
   CaseBlock CB(ISD::SETEQ, CondVal, ConstantInt::getTrue(*DAG.getContext()),
-               NULL, Succ0MBB, Succ1MBB, CurMBB);
+               NULL, Succ0MBB, Succ1MBB, BrMBB);
 
   // Use visitSwitchCase to actually insert the fast branch sequence for this
   // cond branch.
-  visitSwitchCase(CB);
+  visitSwitchCase(CB, BrMBB);
 }
 
 /// visitSwitchCase - Emits the necessary code to represent a single node in
 /// the binary search tree resulting from lowering a switch instruction.
-void SelectionDAGBuilder::visitSwitchCase(CaseBlock &CB) {
+void SelectionDAGBuilder::visitSwitchCase(CaseBlock &CB,
+                                          MachineBasicBlock *SwitchBB) {
   SDValue Cond;
   SDValue CondLHS = getValue(CB.CmpLHS);
   DebugLoc dl = getCurDebugLoc();
@@ -1281,13 +1249,13 @@
   }
 
   // Update successor info
-  CurMBB->addSuccessor(CB.TrueBB);
-  CurMBB->addSuccessor(CB.FalseBB);
+  SwitchBB->addSuccessor(CB.TrueBB);
+  SwitchBB->addSuccessor(CB.FalseBB);
 
   // Set NextBlock to be the MBB immediately after the current one, if any.
   // This is used to avoid emitting unnecessary branches to the next block.
   MachineBasicBlock *NextBlock = 0;
-  MachineFunction::iterator BBI = CurMBB;
+  MachineFunction::iterator BBI = SwitchBB;
   if (++BBI != FuncInfo.MF->end())
     NextBlock = BBI;
 
@@ -1305,11 +1273,11 @@
 
   // If the branch was constant folded, fix up the CFG.
   if (BrCond.getOpcode() == ISD::BR) {
-    CurMBB->removeSuccessor(CB.FalseBB);
+    SwitchBB->removeSuccessor(CB.FalseBB);
   } else {
     // Otherwise, go ahead and insert the false branch.
     if (BrCond == getControlRoot())
-      CurMBB->removeSuccessor(CB.TrueBB);
+      SwitchBB->removeSuccessor(CB.TrueBB);
 
     if (CB.FalseBB != NextBlock)
       BrCond = DAG.getNode(ISD::BR, dl, MVT::Other, BrCond,
@@ -1336,7 +1304,8 @@
 /// visitJumpTableHeader - This function emits necessary code to produce index
 /// in the JumpTable from switch case.
 void SelectionDAGBuilder::visitJumpTableHeader(JumpTable &JT,
-                                               JumpTableHeader &JTH) {
+                                               JumpTableHeader &JTH,
+                                               MachineBasicBlock *SwitchBB) {
   // Subtract the lowest switch case value from the value being switched on and
   // conditional branch to default mbb if the result is greater than the
   // difference between smallest and largest cases.
@@ -1368,7 +1337,7 @@
   // Set NextBlock to be the MBB immediately after the current one, if any.
   // This is used to avoid emitting unnecessary branches to the next block.
   MachineBasicBlock *NextBlock = 0;
-  MachineFunction::iterator BBI = CurMBB;
+  MachineFunction::iterator BBI = SwitchBB;
 
   if (++BBI != FuncInfo.MF->end())
     NextBlock = BBI;
@@ -1386,7 +1355,8 @@
 
 /// visitBitTestHeader - This function emits necessary code to produce value
 /// suitable for "bit tests"
-void SelectionDAGBuilder::visitBitTestHeader(BitTestBlock &B) {
+void SelectionDAGBuilder::visitBitTestHeader(BitTestBlock &B,
+                                             MachineBasicBlock *SwitchBB) {
   // Subtract the minimum value
   SDValue SwitchOp = getValue(B.SValue);
   EVT VT = SwitchOp.getValueType();
@@ -1409,14 +1379,14 @@
   // Set NextBlock to be the MBB immediately after the current one, if any.
   // This is used to avoid emitting unnecessary branches to the next block.
   MachineBasicBlock *NextBlock = 0;
-  MachineFunction::iterator BBI = CurMBB;
+  MachineFunction::iterator BBI = SwitchBB;
   if (++BBI != FuncInfo.MF->end())
     NextBlock = BBI;
 
   MachineBasicBlock* MBB = B.Cases[0].ThisBB;
 
-  CurMBB->addSuccessor(B.Default);
-  CurMBB->addSuccessor(MBB);
+  SwitchBB->addSuccessor(B.Default);
+  SwitchBB->addSuccessor(MBB);
 
   SDValue BrRange = DAG.getNode(ISD::BRCOND, getCurDebugLoc(),
                                 MVT::Other, CopyTo, RangeCmp,
@@ -1432,7 +1402,8 @@
 /// visitBitTestCase - this function produces one "bit test"
 void SelectionDAGBuilder::visitBitTestCase(MachineBasicBlock* NextMBB,
                                            unsigned Reg,
-                                           BitTestCase &B) {
+                                           BitTestCase &B,
+                                           MachineBasicBlock *SwitchBB) {
   // Make desired shift
   SDValue ShiftOp = DAG.getCopyFromReg(getControlRoot(), getCurDebugLoc(), Reg,
                                        TLI.getPointerTy());
@@ -1450,8 +1421,8 @@
                                 AndOp, DAG.getConstant(0, TLI.getPointerTy()),
                                 ISD::SETNE);
 
-  CurMBB->addSuccessor(B.TargetBB);
-  CurMBB->addSuccessor(NextMBB);
+  SwitchBB->addSuccessor(B.TargetBB);
+  SwitchBB->addSuccessor(NextMBB);
 
   SDValue BrAnd = DAG.getNode(ISD::BRCOND, getCurDebugLoc(),
                               MVT::Other, getControlRoot(),
@@ -1460,7 +1431,7 @@
   // Set NextBlock to be the MBB immediately after the current one, if any.
   // This is used to avoid emitting unnecessary branches to the next block.
   MachineBasicBlock *NextBlock = 0;
-  MachineFunction::iterator BBI = CurMBB;
+  MachineFunction::iterator BBI = SwitchBB;
   if (++BBI != FuncInfo.MF->end())
     NextBlock = BBI;
 
@@ -1471,7 +1442,9 @@
   DAG.setRoot(BrAnd);
 }
 
-void SelectionDAGBuilder::visitInvoke(InvokeInst &I) {
+void SelectionDAGBuilder::visitInvoke(const InvokeInst &I) {
+  MachineBasicBlock *InvokeMBB = FuncInfo.MBBMap[I.getParent()];
+
   // Retrieve successors.
   MachineBasicBlock *Return = FuncInfo.MBBMap[I.getSuccessor(0)];
   MachineBasicBlock *LandingPad = FuncInfo.MBBMap[I.getSuccessor(1)];
@@ -1487,8 +1460,8 @@
   CopyToExportRegsIfNeeded(&I);
 
   // Update successor info
-  CurMBB->addSuccessor(Return);
-  CurMBB->addSuccessor(LandingPad);
+  InvokeMBB->addSuccessor(Return);
+  InvokeMBB->addSuccessor(LandingPad);
 
   // Drop into normal successor.
   DAG.setRoot(DAG.getNode(ISD::BR, getCurDebugLoc(),
@@ -1496,15 +1469,16 @@
                           DAG.getBasicBlock(Return)));
 }
 
-void SelectionDAGBuilder::visitUnwind(UnwindInst &I) {
+void SelectionDAGBuilder::visitUnwind(const UnwindInst &I) {
 }
 
 /// handleSmallSwitchCaseRange - Emit a series of specific tests (suitable for
 /// small case ranges).
 bool SelectionDAGBuilder::handleSmallSwitchRange(CaseRec& CR,
                                                  CaseRecVector& WorkList,
-                                                 Value* SV,
-                                                 MachineBasicBlock* Default) {
+                                                 const Value* SV,
+                                                 MachineBasicBlock *Default,
+                                                 MachineBasicBlock *SwitchBB) {
   Case& BackCase  = *(CR.Range.second-1);
 
   // Size is the number of Cases represented by this range.
@@ -1557,7 +1531,7 @@
       FallThrough = Default;
     }
 
-    Value *RHS, *LHS, *MHS;
+    const Value *RHS, *LHS, *MHS;
     ISD::CondCode CC;
     if (I->High == I->Low) {
       // This is just small small case range :) containing exactly 1 case
@@ -1573,8 +1547,8 @@
     // code into the current block.  Otherwise, push the CaseBlock onto the
     // vector to be later processed by SDISel, and insert the node's MBB
     // before the next MBB.
-    if (CurBlock == CurMBB)
-      visitSwitchCase(CB);
+    if (CurBlock == SwitchBB)
+      visitSwitchCase(CB, SwitchBB);
     else
       SwitchCases.push_back(CB);
 
@@ -1600,8 +1574,9 @@
 /// handleJTSwitchCase - Emit jumptable for current switch case range
 bool SelectionDAGBuilder::handleJTSwitchCase(CaseRec& CR,
                                              CaseRecVector& WorkList,
-                                             Value* SV,
-                                             MachineBasicBlock* Default) {
+                                             const Value* SV,
+                                             MachineBasicBlock* Default,
+                                             MachineBasicBlock *SwitchBB) {
   Case& FrontCase = *CR.Range.first;
   Case& BackCase  = *(CR.Range.second-1);
 
@@ -1613,7 +1588,7 @@
        I!=E; ++I)
     TSize += I->size();
 
-  if (!areJTsAllowed(TLI) || TSize.ult(APInt(First.getBitWidth(), 4)))
+  if (!areJTsAllowed(TLI) || TSize.ult(4))
     return false;
 
   APInt Range = ComputeRange(First, Last);
@@ -1674,18 +1649,17 @@
     }
   }
 
-  // Create a jump table index for this jump table, or return an existing
-  // one.
+  // Create a jump table index for this jump table.
   unsigned JTEncoding = TLI.getJumpTableEncoding();
   unsigned JTI = CurMF->getOrCreateJumpTableInfo(JTEncoding)
-                       ->getJumpTableIndex(DestBBs);
+                       ->createJumpTableIndex(DestBBs);
 
   // Set the jump table information so that we can codegen it as a second
   // MachineBasicBlock
   JumpTable JT(-1U, JTI, JumpTableBB, Default);
-  JumpTableHeader JTH(First, Last, SV, CR.CaseBB, (CR.CaseBB == CurMBB));
-  if (CR.CaseBB == CurMBB)
-    visitJumpTableHeader(JT, JTH);
+  JumpTableHeader JTH(First, Last, SV, CR.CaseBB, (CR.CaseBB == SwitchBB));
+  if (CR.CaseBB == SwitchBB)
+    visitJumpTableHeader(JT, JTH, SwitchBB);
 
   JTCases.push_back(JumpTableBlock(JTH, JT));
 
@@ -1696,8 +1670,9 @@
 /// 2 subtrees.
 bool SelectionDAGBuilder::handleBTSplitSwitchCase(CaseRec& CR,
                                                   CaseRecVector& WorkList,
-                                                  Value* SV,
-                                                  MachineBasicBlock* Default) {
+                                                  const Value* SV,
+                                                  MachineBasicBlock *Default,
+                                                  MachineBasicBlock *SwitchBB) {
   // Get the MachineFunction which holds the current MBB.  This is used when
   // inserting any additional MBBs necessary to represent the switch.
   MachineFunction *CurMF = FuncInfo.MF;
@@ -1811,8 +1786,8 @@
   // Otherwise, branch to LHS.
   CaseBlock CB(ISD::SETLT, SV, C, NULL, TrueBB, FalseBB, CR.CaseBB);
 
-  if (CR.CaseBB == CurMBB)
-    visitSwitchCase(CB);
+  if (CR.CaseBB == SwitchBB)
+    visitSwitchCase(CB, SwitchBB);
   else
     SwitchCases.push_back(CB);
 
@@ -1824,8 +1799,9 @@
 /// of masks and emit bit tests with these masks.
 bool SelectionDAGBuilder::handleBitTestsSwitchCase(CaseRec& CR,
                                                    CaseRecVector& WorkList,
-                                                   Value* SV,
-                                                   MachineBasicBlock* Default){
+                                                   const Value* SV,
+                                                   MachineBasicBlock* Default,
+                                                   MachineBasicBlock *SwitchBB){
   EVT PTy = TLI.getPointerTy();
   unsigned IntPtrBits = PTy.getSizeInBits();
 
@@ -1868,7 +1844,7 @@
                << "Low bound: " << minValue << '\n'
                << "High bound: " << maxValue << '\n');
 
-  if (cmpRange.uge(APInt(cmpRange.getBitWidth(), IntPtrBits)) ||
+  if (cmpRange.uge(IntPtrBits) ||
       (!(Dests.size() == 1 && numCmps >= 3) &&
        !(Dests.size() == 2 && numCmps >= 5) &&
        !(Dests.size() >= 3 && numCmps >= 6)))
@@ -1880,8 +1856,7 @@
   // Optimize the case where all the case values fit in a
   // word without having to subtract minValue. In this case,
   // we can optimize away the subtraction.
-  if (minValue.isNonNegative() &&
-      maxValue.slt(APInt(maxValue.getBitWidth(), IntPtrBits))) {
+  if (minValue.isNonNegative() && maxValue.slt(IntPtrBits)) {
     cmpRange = maxValue;
   } else {
     lowBound = minValue;
@@ -1941,11 +1916,11 @@
   }
 
   BitTestBlock BTB(lowBound, cmpRange, SV,
-                   -1U, (CR.CaseBB == CurMBB),
+                   -1U, (CR.CaseBB == SwitchBB),
                    CR.CaseBB, Default, BTC);
 
-  if (CR.CaseBB == CurMBB)
-    visitBitTestHeader(BTB);
+  if (CR.CaseBB == SwitchBB)
+    visitBitTestHeader(BTB, SwitchBB);
 
   BitTestCases.push_back(BTB);
 
@@ -1995,7 +1970,9 @@
   return numCmps;
 }
 
-void SelectionDAGBuilder::visitSwitch(SwitchInst &SI) {
+void SelectionDAGBuilder::visitSwitch(const SwitchInst &SI) {
+  MachineBasicBlock *SwitchMBB = FuncInfo.MBBMap[SI.getParent()];
+
   // Figure out which block is immediately after the current one.
   MachineBasicBlock *NextBlock = 0;
   MachineBasicBlock *Default = FuncInfo.MBBMap[SI.getDefaultDest()];
@@ -2006,7 +1983,7 @@
     // Update machine-CFG edges.
 
     // If this is not a fall-through branch, emit the branch.
-    CurMBB->addSuccessor(Default);
+    SwitchMBB->addSuccessor(Default);
     if (Default != NextBlock)
       DAG.setRoot(DAG.getNode(ISD::BR, getCurDebugLoc(),
                               MVT::Other, getControlRoot(),
@@ -2027,38 +2004,41 @@
   // Get the Value to be switched on and default basic blocks, which will be
   // inserted into CaseBlock records, representing basic blocks in the binary
   // search tree.
-  Value *SV = SI.getOperand(0);
+  const Value *SV = SI.getOperand(0);
 
   // Push the initial CaseRec onto the worklist
   CaseRecVector WorkList;
-  WorkList.push_back(CaseRec(CurMBB,0,0,CaseRange(Cases.begin(),Cases.end())));
+  WorkList.push_back(CaseRec(SwitchMBB,0,0,
+                             CaseRange(Cases.begin(),Cases.end())));
 
   while (!WorkList.empty()) {
     // Grab a record representing a case range to process off the worklist
     CaseRec CR = WorkList.back();
     WorkList.pop_back();
 
-    if (handleBitTestsSwitchCase(CR, WorkList, SV, Default))
+    if (handleBitTestsSwitchCase(CR, WorkList, SV, Default, SwitchMBB))
       continue;
 
     // If the range has few cases (two or less) emit a series of specific
     // tests.
-    if (handleSmallSwitchRange(CR, WorkList, SV, Default))
+    if (handleSmallSwitchRange(CR, WorkList, SV, Default, SwitchMBB))
       continue;
 
     // If the switch has more than 5 blocks, and at least 40% dense, and the
     // target supports indirect branches, then emit a jump table rather than
     // lowering the switch to a binary tree of conditional branches.
-    if (handleJTSwitchCase(CR, WorkList, SV, Default))
+    if (handleJTSwitchCase(CR, WorkList, SV, Default, SwitchMBB))
       continue;
 
     // Emit binary tree. We need to pick a pivot, and push left and right ranges
     // onto the worklist. Leafs are handled via handleSmallSwitchRange() call.
-    handleBTSplitSwitchCase(CR, WorkList, SV, Default);
+    handleBTSplitSwitchCase(CR, WorkList, SV, Default, SwitchMBB);
   }
 }
 
-void SelectionDAGBuilder::visitIndirectBr(IndirectBrInst &I) {
+void SelectionDAGBuilder::visitIndirectBr(const IndirectBrInst &I) {
+  MachineBasicBlock *IndirectBrMBB = FuncInfo.MBBMap[I.getParent()];
+
   // Update machine-CFG edges with unique successors.
   SmallVector<BasicBlock*, 32> succs;
   succs.reserve(I.getNumSuccessors());
@@ -2067,14 +2047,14 @@
   array_pod_sort(succs.begin(), succs.end());
   succs.erase(std::unique(succs.begin(), succs.end()), succs.end());
   for (unsigned i = 0, e = succs.size(); i != e; ++i)
-    CurMBB->addSuccessor(FuncInfo.MBBMap[succs[i]]);
+    IndirectBrMBB->addSuccessor(FuncInfo.MBBMap[succs[i]]);
 
   DAG.setRoot(DAG.getNode(ISD::BRIND, getCurDebugLoc(),
                           MVT::Other, getControlRoot(),
                           getValue(I.getAddress())));
 }
 
-void SelectionDAGBuilder::visitFSub(User &I) {
+void SelectionDAGBuilder::visitFSub(const User &I) {
   // -0.0 - X --> fneg
   const Type *Ty = I.getType();
   if (Ty->isVectorTy()) {
@@ -2104,14 +2084,14 @@
   visitBinary(I, ISD::FSUB);
 }
 
-void SelectionDAGBuilder::visitBinary(User &I, unsigned OpCode) {
+void SelectionDAGBuilder::visitBinary(const User &I, unsigned OpCode) {
   SDValue Op1 = getValue(I.getOperand(0));
   SDValue Op2 = getValue(I.getOperand(1));
   setValue(&I, DAG.getNode(OpCode, getCurDebugLoc(),
                            Op1.getValueType(), Op1, Op2));
 }
 
-void SelectionDAGBuilder::visitShift(User &I, unsigned Opcode) {
+void SelectionDAGBuilder::visitShift(const User &I, unsigned Opcode) {
   SDValue Op1 = getValue(I.getOperand(0));
   SDValue Op2 = getValue(I.getOperand(1));
   if (!I.getType()->isVectorTy() &&
@@ -2145,11 +2125,11 @@
                            Op1.getValueType(), Op1, Op2));
 }
 
-void SelectionDAGBuilder::visitICmp(User &I) {
+void SelectionDAGBuilder::visitICmp(const User &I) {
   ICmpInst::Predicate predicate = ICmpInst::BAD_ICMP_PREDICATE;
-  if (ICmpInst *IC = dyn_cast<ICmpInst>(&I))
+  if (const ICmpInst *IC = dyn_cast<ICmpInst>(&I))
     predicate = IC->getPredicate();
-  else if (ConstantExpr *IC = dyn_cast<ConstantExpr>(&I))
+  else if (const ConstantExpr *IC = dyn_cast<ConstantExpr>(&I))
     predicate = ICmpInst::Predicate(IC->getPredicate());
   SDValue Op1 = getValue(I.getOperand(0));
   SDValue Op2 = getValue(I.getOperand(1));
@@ -2159,11 +2139,11 @@
   setValue(&I, DAG.getSetCC(getCurDebugLoc(), DestVT, Op1, Op2, Opcode));
 }
 
-void SelectionDAGBuilder::visitFCmp(User &I) {
+void SelectionDAGBuilder::visitFCmp(const User &I) {
   FCmpInst::Predicate predicate = FCmpInst::BAD_FCMP_PREDICATE;
-  if (FCmpInst *FC = dyn_cast<FCmpInst>(&I))
+  if (const FCmpInst *FC = dyn_cast<FCmpInst>(&I))
     predicate = FC->getPredicate();
-  else if (ConstantExpr *FC = dyn_cast<ConstantExpr>(&I))
+  else if (const ConstantExpr *FC = dyn_cast<ConstantExpr>(&I))
     predicate = FCmpInst::Predicate(FC->getPredicate());
   SDValue Op1 = getValue(I.getOperand(0));
   SDValue Op2 = getValue(I.getOperand(1));
@@ -2172,7 +2152,7 @@
   setValue(&I, DAG.getSetCC(getCurDebugLoc(), DestVT, Op1, Op2, Condition));
 }
 
-void SelectionDAGBuilder::visitSelect(User &I) {
+void SelectionDAGBuilder::visitSelect(const User &I) {
   SmallVector<EVT, 4> ValueVTs;
   ComputeValueVTs(TLI, I.getType(), ValueVTs);
   unsigned NumValues = ValueVTs.size();
@@ -2185,7 +2165,8 @@
 
   for (unsigned i = 0; i != NumValues; ++i)
     Values[i] = DAG.getNode(ISD::SELECT, getCurDebugLoc(),
-                            TrueVal.getNode()->getValueType(i), Cond,
+                          TrueVal.getNode()->getValueType(TrueVal.getResNo()+i),
+                            Cond,
                             SDValue(TrueVal.getNode(),
                                     TrueVal.getResNo() + i),
                             SDValue(FalseVal.getNode(),
@@ -2196,14 +2177,14 @@
                            &Values[0], NumValues));
 }
 
-void SelectionDAGBuilder::visitTrunc(User &I) {
+void SelectionDAGBuilder::visitTrunc(const User &I) {
   // TruncInst cannot be a no-op cast because sizeof(src) > sizeof(dest).
   SDValue N = getValue(I.getOperand(0));
   EVT DestVT = TLI.getValueType(I.getType());
   setValue(&I, DAG.getNode(ISD::TRUNCATE, getCurDebugLoc(), DestVT, N));
 }
 
-void SelectionDAGBuilder::visitZExt(User &I) {
+void SelectionDAGBuilder::visitZExt(const User &I) {
   // ZExt cannot be a no-op cast because sizeof(src) < sizeof(dest).
   // ZExt also can't be a cast to bool for same reason. So, nothing much to do
   SDValue N = getValue(I.getOperand(0));
@@ -2211,7 +2192,7 @@
   setValue(&I, DAG.getNode(ISD::ZERO_EXTEND, getCurDebugLoc(), DestVT, N));
 }
 
-void SelectionDAGBuilder::visitSExt(User &I) {
+void SelectionDAGBuilder::visitSExt(const User &I) {
   // SExt cannot be a no-op cast because sizeof(src) < sizeof(dest).
   // SExt also can't be a cast to bool for same reason. So, nothing much to do
   SDValue N = getValue(I.getOperand(0));
@@ -2219,7 +2200,7 @@
   setValue(&I, DAG.getNode(ISD::SIGN_EXTEND, getCurDebugLoc(), DestVT, N));
 }
 
-void SelectionDAGBuilder::visitFPTrunc(User &I) {
+void SelectionDAGBuilder::visitFPTrunc(const User &I) {
   // FPTrunc is never a no-op cast, no need to check
   SDValue N = getValue(I.getOperand(0));
   EVT DestVT = TLI.getValueType(I.getType());
@@ -2227,42 +2208,42 @@
                            DestVT, N, DAG.getIntPtrConstant(0)));
 }
 
-void SelectionDAGBuilder::visitFPExt(User &I){
+void SelectionDAGBuilder::visitFPExt(const User &I){
   // FPTrunc is never a no-op cast, no need to check
   SDValue N = getValue(I.getOperand(0));
   EVT DestVT = TLI.getValueType(I.getType());
   setValue(&I, DAG.getNode(ISD::FP_EXTEND, getCurDebugLoc(), DestVT, N));
 }
 
-void SelectionDAGBuilder::visitFPToUI(User &I) {
+void SelectionDAGBuilder::visitFPToUI(const User &I) {
   // FPToUI is never a no-op cast, no need to check
   SDValue N = getValue(I.getOperand(0));
   EVT DestVT = TLI.getValueType(I.getType());
   setValue(&I, DAG.getNode(ISD::FP_TO_UINT, getCurDebugLoc(), DestVT, N));
 }
 
-void SelectionDAGBuilder::visitFPToSI(User &I) {
+void SelectionDAGBuilder::visitFPToSI(const User &I) {
   // FPToSI is never a no-op cast, no need to check
   SDValue N = getValue(I.getOperand(0));
   EVT DestVT = TLI.getValueType(I.getType());
   setValue(&I, DAG.getNode(ISD::FP_TO_SINT, getCurDebugLoc(), DestVT, N));
 }
 
-void SelectionDAGBuilder::visitUIToFP(User &I) {
+void SelectionDAGBuilder::visitUIToFP(const User &I) {
   // UIToFP is never a no-op cast, no need to check
   SDValue N = getValue(I.getOperand(0));
   EVT DestVT = TLI.getValueType(I.getType());
   setValue(&I, DAG.getNode(ISD::UINT_TO_FP, getCurDebugLoc(), DestVT, N));
 }
 
-void SelectionDAGBuilder::visitSIToFP(User &I){
+void SelectionDAGBuilder::visitSIToFP(const User &I){
   // SIToFP is never a no-op cast, no need to check
   SDValue N = getValue(I.getOperand(0));
   EVT DestVT = TLI.getValueType(I.getType());
   setValue(&I, DAG.getNode(ISD::SINT_TO_FP, getCurDebugLoc(), DestVT, N));
 }
 
-void SelectionDAGBuilder::visitPtrToInt(User &I) {
+void SelectionDAGBuilder::visitPtrToInt(const User &I) {
   // What to do depends on the size of the integer and the size of the pointer.
   // We can either truncate, zero extend, or no-op, accordingly.
   SDValue N = getValue(I.getOperand(0));
@@ -2271,7 +2252,7 @@
   setValue(&I, DAG.getZExtOrTrunc(N, getCurDebugLoc(), DestVT));
 }
 
-void SelectionDAGBuilder::visitIntToPtr(User &I) {
+void SelectionDAGBuilder::visitIntToPtr(const User &I) {
   // What to do depends on the size of the integer and the size of the pointer.
   // We can either truncate, zero extend, or no-op, accordingly.
   SDValue N = getValue(I.getOperand(0));
@@ -2280,7 +2261,7 @@
   setValue(&I, DAG.getZExtOrTrunc(N, getCurDebugLoc(), DestVT));
 }
 
-void SelectionDAGBuilder::visitBitCast(User &I) {
+void SelectionDAGBuilder::visitBitCast(const User &I) {
   SDValue N = getValue(I.getOperand(0));
   EVT DestVT = TLI.getValueType(I.getType());
 
@@ -2293,7 +2274,7 @@
     setValue(&I, N);            // noop cast.
 }
 
-void SelectionDAGBuilder::visitInsertElement(User &I) {
+void SelectionDAGBuilder::visitInsertElement(const User &I) {
   SDValue InVec = getValue(I.getOperand(0));
   SDValue InVal = getValue(I.getOperand(1));
   SDValue InIdx = DAG.getNode(ISD::ZERO_EXTEND, getCurDebugLoc(),
@@ -2304,7 +2285,7 @@
                            InVec, InVal, InIdx));
 }
 
-void SelectionDAGBuilder::visitExtractElement(User &I) {
+void SelectionDAGBuilder::visitExtractElement(const User &I) {
   SDValue InVec = getValue(I.getOperand(0));
   SDValue InIdx = DAG.getNode(ISD::ZERO_EXTEND, getCurDebugLoc(),
                               TLI.getPointerTy(),
@@ -2323,7 +2304,7 @@
   return true;
 }
 
-void SelectionDAGBuilder::visitShuffleVector(User &I) {
+void SelectionDAGBuilder::visitShuffleVector(const User &I) {
   SmallVector<int, 8> Mask;
   SDValue Src1 = getValue(I.getOperand(0));
   SDValue Src2 = getValue(I.getOperand(1));
@@ -2504,7 +2485,7 @@
                            VT, &Ops[0], Ops.size()));
 }
 
-void SelectionDAGBuilder::visitInsertValue(InsertValueInst &I) {
+void SelectionDAGBuilder::visitInsertValue(const InsertValueInst &I) {
   const Value *Op0 = I.getOperand(0);
   const Value *Op1 = I.getOperand(1);
   const Type *AggTy = I.getType();
@@ -2545,7 +2526,7 @@
                            &Values[0], NumAggValues));
 }
 
-void SelectionDAGBuilder::visitExtractValue(ExtractValueInst &I) {
+void SelectionDAGBuilder::visitExtractValue(const ExtractValueInst &I) {
   const Value *Op0 = I.getOperand(0);
   const Type *AggTy = Op0->getType();
   const Type *ValTy = I.getType();
@@ -2573,13 +2554,13 @@
                            &Values[0], NumValValues));
 }
 
-void SelectionDAGBuilder::visitGetElementPtr(User &I) {
+void SelectionDAGBuilder::visitGetElementPtr(const User &I) {
   SDValue N = getValue(I.getOperand(0));
   const Type *Ty = I.getOperand(0)->getType();
 
-  for (GetElementPtrInst::op_iterator OI = I.op_begin()+1, E = I.op_end();
+  for (GetElementPtrInst::const_op_iterator OI = I.op_begin()+1, E = I.op_end();
        OI != E; ++OI) {
-    Value *Idx = *OI;
+    const Value *Idx = *OI;
     if (const StructType *StTy = dyn_cast<StructType>(Ty)) {
       unsigned Field = cast<ConstantInt>(Idx)->getZExtValue();
       if (Field) {
@@ -2590,11 +2571,16 @@
       }
 
       Ty = StTy->getElementType(Field);
+    } else if (const UnionType *UnTy = dyn_cast<UnionType>(Ty)) {
+      unsigned Field = cast<ConstantInt>(Idx)->getZExtValue();
+      
+      // Offset canonically 0 for unions, but type changes
+      Ty = UnTy->getElementType(Field);
     } else {
       Ty = cast<SequentialType>(Ty)->getElementType();
 
       // If this is a constant subscript, handle it quickly.
-      if (ConstantInt *CI = dyn_cast<ConstantInt>(Idx)) {
+      if (const ConstantInt *CI = dyn_cast<ConstantInt>(Idx)) {
         if (CI->getZExtValue() == 0) continue;
         uint64_t Offs =
             TD->getTypeAllocSize(Ty)*cast<ConstantInt>(CI)->getSExtValue();
@@ -2645,7 +2631,7 @@
   setValue(&I, N);
 }
 
-void SelectionDAGBuilder::visitAlloca(AllocaInst &I) {
+void SelectionDAGBuilder::visitAlloca(const AllocaInst &I) {
   // If this is a fixed sized alloca in the entry block of the function,
   // allocate it statically on the stack.
   if (FuncInfo.StaticAllocaMap.count(&I))
@@ -2669,8 +2655,7 @@
   // Handle alignment.  If the requested alignment is less than or equal to
   // the stack alignment, ignore it.  If the size is greater than or equal to
   // the stack alignment, we note this in the DYNAMIC_STACKALLOC node.
-  unsigned StackAlign =
-    TLI.getTargetMachine().getFrameInfo()->getStackAlignment();
+  unsigned StackAlign = TM.getFrameInfo()->getStackAlignment();
   if (Align <= StackAlign)
     Align = 0;
 
@@ -2697,7 +2682,7 @@
   FuncInfo.MF->getFrameInfo()->CreateVariableSizedObject();
 }
 
-void SelectionDAGBuilder::visitLoad(LoadInst &I) {
+void SelectionDAGBuilder::visitLoad(const LoadInst &I) {
   const Value *SV = I.getOperand(0);
   SDValue Ptr = getValue(SV);
 
@@ -2757,9 +2742,9 @@
                            &Values[0], NumValues));
 }
 
-void SelectionDAGBuilder::visitStore(StoreInst &I) {
-  Value *SrcV = I.getOperand(0);
-  Value *PtrV = I.getOperand(1);
+void SelectionDAGBuilder::visitStore(const StoreInst &I) {
+  const Value *SrcV = I.getOperand(0);
+  const Value *PtrV = I.getOperand(1);
 
   SmallVector<EVT, 4> ValueVTs;
   SmallVector<uint64_t, 4> Offsets;
@@ -2796,7 +2781,7 @@
 
 /// visitTargetIntrinsic - Lower a call of a target intrinsic to an INTRINSIC
 /// node.
-void SelectionDAGBuilder::visitTargetIntrinsic(CallInst &I,
+void SelectionDAGBuilder::visitTargetIntrinsic(const CallInst &I,
                                                unsigned Intrinsic) {
   bool HasChain = !I.doesNotAccessMemory();
   bool OnlyLoad = HasChain && I.onlyReadsMemory();
@@ -2922,7 +2907,8 @@
 /// visitIntrinsicCall: I is a call instruction
 ///                     Op is the associated NodeType for I
 const char *
-SelectionDAGBuilder::implVisitBinaryAtomic(CallInst& I, ISD::NodeType Op) {
+SelectionDAGBuilder::implVisitBinaryAtomic(const CallInst& I,
+                                           ISD::NodeType Op) {
   SDValue Root = getRoot();
   SDValue L =
     DAG.getAtomic(Op, getCurDebugLoc(),
@@ -2938,7 +2924,7 @@
 
 // implVisitAluOverflow - Lower arithmetic overflow instrinsics.
 const char *
-SelectionDAGBuilder::implVisitAluOverflow(CallInst &I, ISD::NodeType Op) {
+SelectionDAGBuilder::implVisitAluOverflow(const CallInst &I, ISD::NodeType Op) {
   SDValue Op1 = getValue(I.getOperand(1));
   SDValue Op2 = getValue(I.getOperand(2));
 
@@ -2950,7 +2936,7 @@
 /// visitExp - Lower an exp intrinsic. Handles the special sequences for
 /// limited-precision mode.
 void
-SelectionDAGBuilder::visitExp(CallInst &I) {
+SelectionDAGBuilder::visitExp(const CallInst &I) {
   SDValue result;
   DebugLoc dl = getCurDebugLoc();
 
@@ -3076,7 +3062,7 @@
 /// visitLog - Lower a log intrinsic. Handles the special sequences for
 /// limited-precision mode.
 void
-SelectionDAGBuilder::visitLog(CallInst &I) {
+SelectionDAGBuilder::visitLog(const CallInst &I) {
   SDValue result;
   DebugLoc dl = getCurDebugLoc();
 
@@ -3186,7 +3172,7 @@
 /// visitLog2 - Lower a log2 intrinsic. Handles the special sequences for
 /// limited-precision mode.
 void
-SelectionDAGBuilder::visitLog2(CallInst &I) {
+SelectionDAGBuilder::visitLog2(const CallInst &I) {
   SDValue result;
   DebugLoc dl = getCurDebugLoc();
 
@@ -3295,7 +3281,7 @@
 /// visitLog10 - Lower a log10 intrinsic. Handles the special sequences for
 /// limited-precision mode.
 void
-SelectionDAGBuilder::visitLog10(CallInst &I) {
+SelectionDAGBuilder::visitLog10(const CallInst &I) {
   SDValue result;
   DebugLoc dl = getCurDebugLoc();
 
@@ -3397,7 +3383,7 @@
 /// visitExp2 - Lower an exp2 intrinsic. Handles the special sequences for
 /// limited-precision mode.
 void
-SelectionDAGBuilder::visitExp2(CallInst &I) {
+SelectionDAGBuilder::visitExp2(const CallInst &I) {
   SDValue result;
   DebugLoc dl = getCurDebugLoc();
 
@@ -3511,9 +3497,9 @@
 /// visitPow - Lower a pow intrinsic. Handles the special sequences for
 /// limited-precision mode with x == 10.0f.
 void
-SelectionDAGBuilder::visitPow(CallInst &I) {
+SelectionDAGBuilder::visitPow(const CallInst &I) {
   SDValue result;
-  Value *Val = I.getOperand(1);
+  const Value *Val = I.getOperand(1);
   DebugLoc dl = getCurDebugLoc();
   bool IsExp10 = false;
 
@@ -3659,7 +3645,7 @@
     if (Val == 0)
       return DAG.getConstantFP(1.0, LHS.getValueType());
 
-    Function *F = DAG.getMachineFunction().getFunction();
+    const Function *F = DAG.getMachineFunction().getFunction();
     if (!F->hasFnAttr(Attribute::OptimizeForSize) ||
         // If optimizing for size, don't insert too many multiplies.  This
         // inserts up to 5 multiplies.
@@ -3700,7 +3686,7 @@
 /// we want to emit this as a call to a named external function, return the name
 /// otherwise lower it and return null.
 const char *
-SelectionDAGBuilder::visitIntrinsicCall(CallInst &I, unsigned Intrinsic) {
+SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) {
   DebugLoc dl = getCurDebugLoc();
   SDValue Res;
 
@@ -3725,28 +3711,50 @@
   case Intrinsic::longjmp:
     return "_longjmp"+!TLI.usesUnderscoreLongJmp();
   case Intrinsic::memcpy: {
+    // Assert for address < 256 since we support only user defined address
+    // spaces.
+    assert(cast<PointerType>(I.getOperand(1)->getType())->getAddressSpace()
+           < 256 &&
+           cast<PointerType>(I.getOperand(2)->getType())->getAddressSpace()
+           < 256 &&
+           "Unknown address space");
     SDValue Op1 = getValue(I.getOperand(1));
     SDValue Op2 = getValue(I.getOperand(2));
     SDValue Op3 = getValue(I.getOperand(3));
     unsigned Align = cast<ConstantInt>(I.getOperand(4))->getZExtValue();
-    DAG.setRoot(DAG.getMemcpy(getRoot(), dl, Op1, Op2, Op3, Align, false,
+    bool isVol = cast<ConstantInt>(I.getOperand(5))->getZExtValue();
+    DAG.setRoot(DAG.getMemcpy(getRoot(), dl, Op1, Op2, Op3, Align, isVol, false,
                               I.getOperand(1), 0, I.getOperand(2), 0));
     return 0;
   }
   case Intrinsic::memset: {
+    // Assert for address < 256 since we support only user defined address
+    // spaces.
+    assert(cast<PointerType>(I.getOperand(1)->getType())->getAddressSpace()
+           < 256 &&
+           "Unknown address space");
     SDValue Op1 = getValue(I.getOperand(1));
     SDValue Op2 = getValue(I.getOperand(2));
     SDValue Op3 = getValue(I.getOperand(3));
     unsigned Align = cast<ConstantInt>(I.getOperand(4))->getZExtValue();
-    DAG.setRoot(DAG.getMemset(getRoot(), dl, Op1, Op2, Op3, Align,
+    bool isVol = cast<ConstantInt>(I.getOperand(5))->getZExtValue();
+    DAG.setRoot(DAG.getMemset(getRoot(), dl, Op1, Op2, Op3, Align, isVol,
                               I.getOperand(1), 0));
     return 0;
   }
   case Intrinsic::memmove: {
+    // Assert for address < 256 since we support only user defined address
+    // spaces.
+    assert(cast<PointerType>(I.getOperand(1)->getType())->getAddressSpace()
+           < 256 &&
+           cast<PointerType>(I.getOperand(2)->getType())->getAddressSpace()
+           < 256 &&
+           "Unknown address space");
     SDValue Op1 = getValue(I.getOperand(1));
     SDValue Op2 = getValue(I.getOperand(2));
     SDValue Op3 = getValue(I.getOperand(3));
     unsigned Align = cast<ConstantInt>(I.getOperand(4))->getZExtValue();
+    bool isVol = cast<ConstantInt>(I.getOperand(5))->getZExtValue();
 
     // If the source and destination are known to not be aliases, we can
     // lower memmove as memcpy.
@@ -3755,69 +3763,128 @@
       Size = C->getZExtValue();
     if (AA->alias(I.getOperand(1), Size, I.getOperand(2), Size) ==
         AliasAnalysis::NoAlias) {
-      DAG.setRoot(DAG.getMemcpy(getRoot(), dl, Op1, Op2, Op3, Align, false,
-                                I.getOperand(1), 0, I.getOperand(2), 0));
+      DAG.setRoot(DAG.getMemcpy(getRoot(), dl, Op1, Op2, Op3, Align, isVol, 
+                                false, I.getOperand(1), 0, I.getOperand(2), 0));
       return 0;
     }
 
-    DAG.setRoot(DAG.getMemmove(getRoot(), dl, Op1, Op2, Op3, Align,
+    DAG.setRoot(DAG.getMemmove(getRoot(), dl, Op1, Op2, Op3, Align, isVol,
                                I.getOperand(1), 0, I.getOperand(2), 0));
     return 0;
   }
   case Intrinsic::dbg_declare: {
-    // FIXME: currently, we get here only if OptLevel != CodeGenOpt::None.
-    // The real handling of this intrinsic is in FastISel.
-    if (OptLevel != CodeGenOpt::None)
-      // FIXME: Variable debug info is not supported here.
-      return 0;
-    DwarfWriter *DW = DAG.getDwarfWriter();
-    if (!DW)
-      return 0;
-    DbgDeclareInst &DI = cast<DbgDeclareInst>(I);
+    const DbgDeclareInst &DI = cast<DbgDeclareInst>(I);
     if (!DIDescriptor::ValidDebugInfo(DI.getVariable(), CodeGenOpt::None))
       return 0;
 
     MDNode *Variable = DI.getVariable();
-    Value *Address = DI.getAddress();
+    // Parameters are handled specially.
+    bool isParameter = false;
+    ConstantInt *CI = dyn_cast_or_null<ConstantInt>(Variable->getOperand(0));
+    if (CI) {
+      unsigned Val = CI->getZExtValue();
+      unsigned Tag = Val & ~LLVMDebugVersionMask;
+      if (Tag == dwarf::DW_TAG_arg_variable)
+        isParameter = true;
+    }
+    const Value *Address = DI.getAddress();
     if (!Address)
       return 0;
-    if (BitCastInst *BCI = dyn_cast<BitCastInst>(Address))
+    if (const BitCastInst *BCI = dyn_cast<BitCastInst>(Address))
       Address = BCI->getOperand(0);
-    AllocaInst *AI = dyn_cast<AllocaInst>(Address);
-    // Don't handle byval struct arguments or VLAs, for example.
-    if (!AI)
-      return 0;
-    DenseMap<const AllocaInst*, int>::iterator SI =
-      FuncInfo.StaticAllocaMap.find(AI);
-    if (SI == FuncInfo.StaticAllocaMap.end())
-      return 0; // VLAs.
-    int FI = SI->second;
+    const AllocaInst *AI = dyn_cast<AllocaInst>(Address);
+    if (AI) {
+      // Don't handle byval arguments or VLAs, for example.
+      // Non-byval arguments are handled here (they refer to the stack temporary
+      // alloca at this point).
+      DenseMap<const AllocaInst*, int>::iterator SI =
+        FuncInfo.StaticAllocaMap.find(AI);
+      if (SI == FuncInfo.StaticAllocaMap.end())
+        return 0; // VLAs.
+      int FI = SI->second;
 
-    if (MachineModuleInfo *MMI = DAG.getMachineModuleInfo())
-      if (MDNode *Dbg = DI.getMetadata("dbg"))
-        MMI->setVariableDbgInfo(Variable, FI, Dbg);
+      MachineModuleInfo &MMI = DAG.getMachineFunction().getMMI();
+      if (!DI.getDebugLoc().isUnknown() && MMI.hasDebugInfo())
+        MMI.setVariableDbgInfo(Variable, FI, DI.getDebugLoc());
+    }
+
+    // Build an entry in DbgOrdering.  Debug info input nodes get an SDNodeOrder
+    // but do not always have a corresponding SDNode built.  The SDNodeOrder
+    // absolute, but not relative, values are different depending on whether
+    // debug info exists.
+    ++SDNodeOrder;
+    SDValue &N = NodeMap[Address];
+    SDDbgValue *SDV;
+    if (N.getNode()) {
+      if (isParameter && !AI) {
+        FrameIndexSDNode *FINode = dyn_cast<FrameIndexSDNode>(N.getNode());
+        if (FINode)
+          // Byval parameter.  We have a frame index at this point.
+          SDV = DAG.getDbgValue(Variable, FINode->getIndex(),
+                                0, dl, SDNodeOrder);
+        else
+          // Can't do anything with other non-AI cases yet.  This might be a
+          // parameter of a callee function that got inlined, for example.
+          return 0;
+      } else if (AI)
+        SDV = DAG.getDbgValue(Variable, N.getNode(), N.getResNo(),
+                              0, dl, SDNodeOrder);
+      else
+        // Can't do anything with other non-AI cases yet.
+        return 0;
+      DAG.AddDbgValue(SDV, N.getNode(), isParameter);
+    } else {
+      // Generating Undefs here seems to be actively harmful because it
+      // affects the line numbers.
+      return 0;
+#if 0
+      // This isn't useful, but it shows what we're missing.
+      SDV = DAG.getDbgValue(Variable, UndefValue::get(Address->getType()),
+                            0, dl, SDNodeOrder);
+      DAG.AddDbgValue(SDV, 0, isParameter);
+#endif
+    }
     return 0;
   }
   case Intrinsic::dbg_value: {
-    // FIXME: currently, we get here only if OptLevel != CodeGenOpt::None.
-    // The real handling of this intrinsic is in FastISel.
-    if (OptLevel != CodeGenOpt::None)
-      // FIXME: Variable debug info is not supported here.
-      return 0;
-    DwarfWriter *DW = DAG.getDwarfWriter();
-    if (!DW)
-      return 0;
-    DbgValueInst &DI = cast<DbgValueInst>(I);
+    const DbgValueInst &DI = cast<DbgValueInst>(I);
     if (!DIDescriptor::ValidDebugInfo(DI.getVariable(), CodeGenOpt::None))
       return 0;
 
     MDNode *Variable = DI.getVariable();
-    Value *V = DI.getValue();
+    uint64_t Offset = DI.getOffset();
+    const Value *V = DI.getValue();
     if (!V)
       return 0;
-    if (BitCastInst *BCI = dyn_cast<BitCastInst>(V))
+
+    // Build an entry in DbgOrdering.  Debug info input nodes get an SDNodeOrder
+    // but do not always have a corresponding SDNode built.  The SDNodeOrder
+    // absolute, but not relative, values are different depending on whether
+    // debug info exists.
+    ++SDNodeOrder;
+    SDDbgValue *SDV;
+    if (isa<ConstantInt>(V) || isa<ConstantFP>(V)) {
+      SDV = DAG.getDbgValue(Variable, V, Offset, dl, SDNodeOrder);
+      DAG.AddDbgValue(SDV, 0, false);
+    } else {
+      SDValue &N = NodeMap[V];
+      if (N.getNode()) {
+        SDV = DAG.getDbgValue(Variable, N.getNode(),
+                              N.getResNo(), Offset, dl, SDNodeOrder);
+        DAG.AddDbgValue(SDV, N.getNode(), false);
+      } else {
+        // We may expand this to cover more cases.  One case where we have no
+        // data available is an unreferenced parameter; we need this fallback.
+        SDV = DAG.getDbgValue(Variable, UndefValue::get(V->getType()),
+                              Offset, dl, SDNodeOrder);
+        DAG.AddDbgValue(SDV, 0, false);
+      }
+    }
+
+    // Build a debug info table entry.
+    if (const BitCastInst *BCI = dyn_cast<BitCastInst>(V))
       V = BCI->getOperand(0);
-    AllocaInst *AI = dyn_cast<AllocaInst>(V);
+    const AllocaInst *AI = dyn_cast<AllocaInst>(V);
     // Don't handle byval struct arguments or VLAs, for example.
     if (!AI)
       return 0;
@@ -3826,14 +3893,16 @@
     if (SI == FuncInfo.StaticAllocaMap.end())
       return 0; // VLAs.
     int FI = SI->second;
-    if (MachineModuleInfo *MMI = DAG.getMachineModuleInfo())
-      if (MDNode *Dbg = DI.getMetadata("dbg"))
-        MMI->setVariableDbgInfo(Variable, FI, Dbg);
+    
+    MachineModuleInfo &MMI = DAG.getMachineFunction().getMMI();
+    if (!DI.getDebugLoc().isUnknown() && MMI.hasDebugInfo())
+      MMI.setVariableDbgInfo(Variable, FI, DI.getDebugLoc());
     return 0;
   }
   case Intrinsic::eh_exception: {
     // Insert the EXCEPTIONADDR instruction.
-    assert(CurMBB->isLandingPad() &&"Call to eh.exception not in landing pad!");
+    assert(FuncInfo.MBBMap[I.getParent()]->isLandingPad() &&
+           "Call to eh.exception not in landing pad!");
     SDVTList VTs = DAG.getVTList(TLI.getPointerTy(), MVT::Other);
     SDValue Ops[1];
     Ops[0] = DAG.getRoot();
@@ -3844,17 +3913,17 @@
   }
 
   case Intrinsic::eh_selector: {
-    MachineModuleInfo *MMI = DAG.getMachineModuleInfo();
-
-    if (CurMBB->isLandingPad())
-      AddCatchInfo(I, MMI, CurMBB);
+    MachineBasicBlock *CallMBB = FuncInfo.MBBMap[I.getParent()];
+    MachineModuleInfo &MMI = DAG.getMachineFunction().getMMI();
+    if (CallMBB->isLandingPad())
+      AddCatchInfo(I, &MMI, CallMBB);
     else {
 #ifndef NDEBUG
       FuncInfo.CatchInfoLost.insert(&I);
 #endif
       // FIXME: Mark exception selector register as live in.  Hack for PR1508.
       unsigned Reg = TLI.getExceptionSelectorRegister();
-      if (Reg) CurMBB->addLiveIn(Reg);
+      if (Reg) FuncInfo.MBBMap[I.getParent()]->addLiveIn(Reg);
     }
 
     // Insert the EHSELECTION instruction.
@@ -3869,40 +3938,25 @@
   }
 
   case Intrinsic::eh_typeid_for: {
-    MachineModuleInfo *MMI = DAG.getMachineModuleInfo();
-
-    if (MMI) {
-      // Find the type id for the given typeinfo.
-      GlobalVariable *GV = ExtractTypeInfo(I.getOperand(1));
-      unsigned TypeID = MMI->getTypeIDFor(GV);
-      Res = DAG.getConstant(TypeID, MVT::i32);
-    } else {
-      // Return something different to eh_selector.
-      Res = DAG.getConstant(1, MVT::i32);
-    }
-
+    // Find the type id for the given typeinfo.
+    GlobalVariable *GV = ExtractTypeInfo(I.getOperand(1));
+    unsigned TypeID = DAG.getMachineFunction().getMMI().getTypeIDFor(GV);
+    Res = DAG.getConstant(TypeID, MVT::i32);
     setValue(&I, Res);
     return 0;
   }
 
   case Intrinsic::eh_return_i32:
   case Intrinsic::eh_return_i64:
-    if (MachineModuleInfo *MMI = DAG.getMachineModuleInfo()) {
-      MMI->setCallsEHReturn(true);
-      DAG.setRoot(DAG.getNode(ISD::EH_RETURN, dl,
-                              MVT::Other,
-                              getControlRoot(),
-                              getValue(I.getOperand(1)),
-                              getValue(I.getOperand(2))));
-    } else {
-      setValue(&I, DAG.getConstant(0, TLI.getPointerTy()));
-    }
-
+    DAG.getMachineFunction().getMMI().setCallsEHReturn(true);
+    DAG.setRoot(DAG.getNode(ISD::EH_RETURN, dl,
+                            MVT::Other,
+                            getControlRoot(),
+                            getValue(I.getOperand(1)),
+                            getValue(I.getOperand(2))));
     return 0;
   case Intrinsic::eh_unwind_init:
-    if (MachineModuleInfo *MMI = DAG.getMachineModuleInfo()) {
-      MMI->setCallsUnwindInit(true);
-    }
+    DAG.getMachineFunction().getMMI().setCallsUnwindInit(true);
     return 0;
   case Intrinsic::eh_dwarf_cfa: {
     EVT VT = getValue(I.getOperand(1)).getValueType();
@@ -3921,12 +3975,12 @@
     return 0;
   }
   case Intrinsic::eh_sjlj_callsite: {
-    MachineModuleInfo *MMI = DAG.getMachineModuleInfo();
+    MachineModuleInfo &MMI = DAG.getMachineFunction().getMMI();
     ConstantInt *CI = dyn_cast<ConstantInt>(I.getOperand(1));
     assert(CI && "Non-constant call site value in eh.sjlj.callsite!");
-    assert(MMI->getCurrentCallSite() == 0 && "Overlapping call sites!");
+    assert(MMI.getCurrentCallSite() == 0 && "Overlapping call sites!");
 
-    MMI->setCurrentCallSite(CI->getZExtValue());
+    MMI.setCurrentCallSite(CI->getZExtValue());
     return 0;
   }
 
@@ -3952,7 +4006,7 @@
     case Intrinsic::convertuu:  Code = ISD::CVT_UU; break;
     }
     EVT DestVT = TLI.getValueType(I.getType());
-    Value *Op1 = I.getOperand(1);
+    const Value *Op1 = I.getOperand(1);
     Res = DAG.getConvertRndSat(DestVT, getCurDebugLoc(), getValue(Op1),
                                DAG.getValueType(DestVT),
                                DAG.getValueType(getValue(Op1).getValueType()),
@@ -3999,6 +4053,14 @@
   case Intrinsic::pow:
     visitPow(I);
     return 0;
+  case Intrinsic::convert_to_fp16:
+    setValue(&I, DAG.getNode(ISD::FP32_TO_FP16, dl,
+                             MVT::i16, getValue(I.getOperand(1))));
+    return 0;
+  case Intrinsic::convert_from_fp16:
+    setValue(&I, DAG.getNode(ISD::FP16_TO_FP32, dl,
+                             MVT::f32, getValue(I.getOperand(1))));
+    return 0;
   case Intrinsic::pcmarker: {
     SDValue Tmp = getValue(I.getOperand(1));
     DAG.setRoot(DAG.getNode(ISD::PCMARKER, dl, MVT::Other, getRoot(), Tmp));
@@ -4113,8 +4175,8 @@
   }
   case Intrinsic::gcroot:
     if (GFI) {
-      Value *Alloca = I.getOperand(1);
-      Constant *TypeMap = cast<Constant>(I.getOperand(2));
+      const Value *Alloca = I.getOperand(1);
+      const Constant *TypeMap = cast<Constant>(I.getOperand(2));
 
       FrameIndexSDNode *FI = cast<FrameIndexSDNode>(getValue(Alloca).getNode());
       GFI->addStackRoot(FI->getIndex(), TypeMap);
@@ -4211,97 +4273,14 @@
   }
 }
 
-/// Test if the given instruction is in a position to be optimized
-/// with a tail-call. This roughly means that it's in a block with
-/// a return and there's nothing that needs to be scheduled
-/// between it and the return.
-///
-/// This function only tests target-independent requirements.
-static bool
-isInTailCallPosition(CallSite CS, Attributes CalleeRetAttr,
-                     const TargetLowering &TLI) {
-  const Instruction *I = CS.getInstruction();
-  const BasicBlock *ExitBB = I->getParent();
-  const TerminatorInst *Term = ExitBB->getTerminator();
-  const ReturnInst *Ret = dyn_cast<ReturnInst>(Term);
-  const Function *F = ExitBB->getParent();
-
-  // The block must end in a return statement or unreachable.
-  //
-  // FIXME: Decline tailcall if it's not guaranteed and if the block ends in
-  // an unreachable, for now. The way tailcall optimization is currently
-  // implemented means it will add an epilogue followed by a jump. That is
-  // not profitable. Also, if the callee is a special function (e.g.
-  // longjmp on x86), it can end up causing miscompilation that has not
-  // been fully understood.
-  if (!Ret &&
-      (!GuaranteedTailCallOpt || !isa<UnreachableInst>(Term))) return false;
-
-  // If I will have a chain, make sure no other instruction that will have a
-  // chain interposes between I and the return.
-  if (I->mayHaveSideEffects() || I->mayReadFromMemory() ||
-      !I->isSafeToSpeculativelyExecute())
-    for (BasicBlock::const_iterator BBI = prior(prior(ExitBB->end())); ;
-         --BBI) {
-      if (&*BBI == I)
-        break;
-      if (BBI->mayHaveSideEffects() || BBI->mayReadFromMemory() ||
-          !BBI->isSafeToSpeculativelyExecute())
-        return false;
-    }
-
-  // If the block ends with a void return or unreachable, it doesn't matter
-  // what the call's return type is.
-  if (!Ret || Ret->getNumOperands() == 0) return true;
-
-  // If the return value is undef, it doesn't matter what the call's
-  // return type is.
-  if (isa<UndefValue>(Ret->getOperand(0))) return true;
-
-  // Conservatively require the attributes of the call to match those of
-  // the return. Ignore noalias because it doesn't affect the call sequence.
-  unsigned CallerRetAttr = F->getAttributes().getRetAttributes();
-  if ((CalleeRetAttr ^ CallerRetAttr) & ~Attribute::NoAlias)
-    return false;
-
-  // It's not safe to eliminate the sign / zero extension of the return value.
-  if ((CallerRetAttr & Attribute::ZExt) || (CallerRetAttr & Attribute::SExt))
-    return false;
-
-  // Otherwise, make sure the unmodified return value of I is the return value.
-  for (const Instruction *U = dyn_cast<Instruction>(Ret->getOperand(0)); ;
-       U = dyn_cast<Instruction>(U->getOperand(0))) {
-    if (!U)
-      return false;
-    if (!U->hasOneUse())
-      return false;
-    if (U == I)
-      break;
-    // Check for a truly no-op truncate.
-    if (isa<TruncInst>(U) &&
-        TLI.isTruncateFree(U->getOperand(0)->getType(), U->getType()))
-      continue;
-    // Check for a truly no-op bitcast.
-    if (isa<BitCastInst>(U) &&
-        (U->getOperand(0)->getType() == U->getType() ||
-         (U->getOperand(0)->getType()->isPointerTy() &&
-          U->getType()->isPointerTy())))
-      continue;
-    // Otherwise it's not a true no-op.
-    return false;
-  }
-
-  return true;
-}
-
-void SelectionDAGBuilder::LowerCallTo(CallSite CS, SDValue Callee,
+void SelectionDAGBuilder::LowerCallTo(ImmutableCallSite CS, SDValue Callee,
                                       bool isTailCall,
                                       MachineBasicBlock *LandingPad) {
   const PointerType *PT = cast<PointerType>(CS.getCalledValue()->getType());
   const FunctionType *FTy = cast<FunctionType>(PT->getElementType());
   const Type *RetTy = FTy->getReturnType();
-  MachineModuleInfo *MMI = DAG.getMachineModuleInfo();
-  unsigned BeginLabel = 0, EndLabel = 0;
+  MachineModuleInfo &MMI = DAG.getMachineFunction().getMMI();
+  MCSymbol *BeginLabel = 0;
 
   TargetLowering::ArgListTy Args;
   TargetLowering::ArgListEntry Entry;
@@ -4342,7 +4321,7 @@
     RetTy = Type::getVoidTy(FTy->getContext());
   }
 
-  for (CallSite::arg_iterator i = CS.arg_begin(), e = CS.arg_end();
+  for (ImmutableCallSite::arg_iterator i = CS.arg_begin(), e = CS.arg_end();
        i != e; ++i) {
     SDValue ArgNode = getValue(*i);
     Entry.Node = ArgNode; Entry.Ty = (*i)->getType();
@@ -4358,25 +4337,24 @@
     Args.push_back(Entry);
   }
 
-  if (LandingPad && MMI) {
+  if (LandingPad) {
     // Insert a label before the invoke call to mark the try range.  This can be
     // used to detect deletion of the invoke via the MachineModuleInfo.
-    BeginLabel = MMI->NextLabelID();
+    BeginLabel = MMI.getContext().CreateTempSymbol();
 
     // For SjLj, keep track of which landing pads go with which invokes
     // so as to maintain the ordering of pads in the LSDA.
-    unsigned CallSiteIndex = MMI->getCurrentCallSite();
+    unsigned CallSiteIndex = MMI.getCurrentCallSite();
     if (CallSiteIndex) {
-      MMI->setCallSiteBeginLabel(BeginLabel, CallSiteIndex);
+      MMI.setCallSiteBeginLabel(BeginLabel, CallSiteIndex);
       // Now that the call site is handled, stop tracking it.
-      MMI->setCurrentCallSite(0);
+      MMI.setCurrentCallSite(0);
     }
 
     // Both PendingLoads and PendingExports must be flushed here;
     // this call might not return.
     (void)getRoot();
-    DAG.setRoot(DAG.getLabel(ISD::EH_LABEL, getCurDebugLoc(),
-                             getControlRoot(), BeginLabel));
+    DAG.setRoot(DAG.getEHLabel(getCurDebugLoc(), getControlRoot(), BeginLabel));
   }
 
   // Check if target-independent constraints permit a tail call here.
@@ -4461,26 +4439,25 @@
   else
     HasTailCall = true;
 
-  if (LandingPad && MMI) {
+  if (LandingPad) {
     // Insert a label at the end of the invoke call to mark the try range.  This
     // can be used to detect deletion of the invoke via the MachineModuleInfo.
-    EndLabel = MMI->NextLabelID();
-    DAG.setRoot(DAG.getLabel(ISD::EH_LABEL, getCurDebugLoc(),
-                             getRoot(), EndLabel));
+    MCSymbol *EndLabel = MMI.getContext().CreateTempSymbol();
+    DAG.setRoot(DAG.getEHLabel(getCurDebugLoc(), getRoot(), EndLabel));
 
     // Inform MachineModuleInfo of range.
-    MMI->addInvoke(LandingPad, BeginLabel, EndLabel);
+    MMI.addInvoke(LandingPad, BeginLabel, EndLabel);
   }
 }
 
 /// IsOnlyUsedInZeroEqualityComparison - Return true if it only matters that the
 /// value is equal or not-equal to zero.
-static bool IsOnlyUsedInZeroEqualityComparison(Value *V) {
-  for (Value::use_iterator UI = V->use_begin(), E = V->use_end();
+static bool IsOnlyUsedInZeroEqualityComparison(const Value *V) {
+  for (Value::const_use_iterator UI = V->use_begin(), E = V->use_end();
        UI != E; ++UI) {
-    if (ICmpInst *IC = dyn_cast<ICmpInst>(*UI))
+    if (const ICmpInst *IC = dyn_cast<ICmpInst>(*UI))
       if (IC->isEquality())
-        if (Constant *C = dyn_cast<Constant>(IC->getOperand(1)))
+        if (const Constant *C = dyn_cast<Constant>(IC->getOperand(1)))
           if (C->isNullValue())
             continue;
     // Unknown instruction.
@@ -4489,17 +4466,20 @@
   return true;
 }
 
-static SDValue getMemCmpLoad(Value *PtrVal, MVT LoadVT, const Type *LoadTy,
+static SDValue getMemCmpLoad(const Value *PtrVal, MVT LoadVT,
+                             const Type *LoadTy,
                              SelectionDAGBuilder &Builder) {
 
   // Check to see if this load can be trivially constant folded, e.g. if the
   // input is from a string literal.
-  if (Constant *LoadInput = dyn_cast<Constant>(PtrVal)) {
+  if (const Constant *LoadInput = dyn_cast<Constant>(PtrVal)) {
     // Cast pointer to the type we really want to load.
-    LoadInput = ConstantExpr::getBitCast(LoadInput,
+    LoadInput = ConstantExpr::getBitCast(const_cast<Constant *>(LoadInput),
                                          PointerType::getUnqual(LoadTy));
 
-    if (Constant *LoadCst = ConstantFoldLoadFromConstPtr(LoadInput, Builder.TD))
+    if (const Constant *LoadCst =
+          ConstantFoldLoadFromConstPtr(const_cast<Constant *>(LoadInput),
+                                       Builder.TD))
       return Builder.getValue(LoadCst);
   }
 
@@ -4532,18 +4512,18 @@
 /// visitMemCmpCall - See if we can lower a call to memcmp in an optimized form.
 /// If so, return true and lower it, otherwise return false and it will be
 /// lowered like a normal call.
-bool SelectionDAGBuilder::visitMemCmpCall(CallInst &I) {
+bool SelectionDAGBuilder::visitMemCmpCall(const CallInst &I) {
   // Verify that the prototype makes sense.  int memcmp(void*,void*,size_t)
   if (I.getNumOperands() != 4)
     return false;
 
-  Value *LHS = I.getOperand(1), *RHS = I.getOperand(2);
+  const Value *LHS = I.getOperand(1), *RHS = I.getOperand(2);
   if (!LHS->getType()->isPointerTy() || !RHS->getType()->isPointerTy() ||
       !I.getOperand(3)->getType()->isIntegerTy() ||
       !I.getType()->isIntegerTy())
     return false;
 
-  ConstantInt *Size = dyn_cast<ConstantInt>(I.getOperand(3));
+  const ConstantInt *Size = dyn_cast<ConstantInt>(I.getOperand(3));
 
   // memcmp(S1,S2,2) != 0 -> (*(short*)LHS != *(short*)RHS)  != 0
   // memcmp(S1,S2,4) != 0 -> (*(int*)LHS != *(int*)RHS)  != 0
@@ -4609,11 +4589,11 @@
 }
 
 
-void SelectionDAGBuilder::visitCall(CallInst &I) {
+void SelectionDAGBuilder::visitCall(const CallInst &I) {
   const char *RenameFn = 0;
   if (Function *F = I.getCalledFunction()) {
     if (F->isDeclaration()) {
-      const TargetIntrinsicInfo *II = TLI.getTargetMachine().getIntrinsicInfo();
+      const TargetIntrinsicInfo *II = TM.getIntrinsicInfo();
       if (II) {
         if (unsigned IID = II->getIntrinsicID(F)) {
           RenameFn = visitIntrinsicCall(I, IID);
@@ -4632,7 +4612,7 @@
     // can't be a library call.
     if (!F->hasLocalLinkage() && F->hasName()) {
       StringRef Name = F->getName();
-      if (Name == "copysign" || Name == "copysignf") {
+      if (Name == "copysign" || Name == "copysignf" || Name == "copysignl") {
         if (I.getNumOperands() == 3 &&   // Basic sanity checks.
             I.getOperand(1)->getType()->isFloatingPointTy() &&
             I.getType() == I.getOperand(1)->getType() &&
@@ -4837,14 +4817,13 @@
 /// AddInlineAsmOperands - Add this value to the specified inlineasm node
 /// operand list.  This adds the code marker and includes the number of
 /// values added into it.
-void RegsForValue::AddInlineAsmOperands(unsigned Code,
-                                        bool HasMatching,unsigned MatchingIdx,
+void RegsForValue::AddInlineAsmOperands(unsigned Code, bool HasMatching,
+                                        unsigned MatchingIdx,
                                         SelectionDAG &DAG,
                                         std::vector<SDValue> &Ops) const {
-  assert(Regs.size() < (1 << 13) && "Too many inline asm outputs!");
-  unsigned Flag = Code | (Regs.size() << 3);
+  unsigned Flag = InlineAsm::getFlagWord(Code, Regs.size());
   if (HasMatching)
-    Flag |= 0x80000000 | (MatchingIdx << 16);
+    Flag = InlineAsm::getFlagWordForMatchingOp(Flag, MatchingIdx);
   SDValue Res = DAG.getTargetConstant(Flag, MVT::i32);
   Ops.push_back(Res);
 
@@ -4960,7 +4939,7 @@
     if (isIndirect) {
       const llvm::PointerType *PtrTy = dyn_cast<PointerType>(OpTy);
       if (!PtrTy)
-        llvm_report_error("Indirect operand for inline asm not a pointer!");
+        report_fatal_error("Indirect operand for inline asm not a pointer!");
       OpTy = PtrTy->getElementType();
     }
 
@@ -5180,31 +5159,10 @@
   // Otherwise, we couldn't allocate enough registers for this.
 }
 
-/// hasInlineAsmMemConstraint - Return true if the inline asm instruction being
-/// processed uses a memory 'm' constraint.
-static bool
-hasInlineAsmMemConstraint(std::vector<InlineAsm::ConstraintInfo> &CInfos,
-                          const TargetLowering &TLI) {
-  for (unsigned i = 0, e = CInfos.size(); i != e; ++i) {
-    InlineAsm::ConstraintInfo &CI = CInfos[i];
-    for (unsigned j = 0, ee = CI.Codes.size(); j != ee; ++j) {
-      TargetLowering::ConstraintType CType = TLI.getConstraintType(CI.Codes[j]);
-      if (CType == TargetLowering::C_Memory)
-        return true;
-    }
-
-    // Indirect operand accesses access memory.
-    if (CI.isIndirect)
-      return true;
-  }
-
-  return false;
-}
-
 /// visitInlineAsm - Handle a call to an InlineAsm object.
 ///
-void SelectionDAGBuilder::visitInlineAsm(CallSite CS) {
-  InlineAsm *IA = cast<InlineAsm>(CS.getCalledValue());
+void SelectionDAGBuilder::visitInlineAsm(ImmutableCallSite CS) {
+  const InlineAsm *IA = cast<InlineAsm>(CS.getCalledValue());
 
   /// ConstraintOperands - Information about all of the constraints.
   std::vector<SDISelAsmOperandInfo> ConstraintOperands;
@@ -5240,7 +5198,7 @@
     case InlineAsm::isOutput:
       // Indirect outputs just consume an argument.
       if (OpInfo.isIndirect) {
-        OpInfo.CallOperandVal = CS.getArgument(ArgNo++);
+        OpInfo.CallOperandVal = const_cast<Value *>(CS.getArgument(ArgNo++));
         break;
       }
 
@@ -5257,7 +5215,7 @@
       ++ResNo;
       break;
     case InlineAsm::isInput:
-      OpInfo.CallOperandVal = CS.getArgument(ArgNo++);
+      OpInfo.CallOperandVal = const_cast<Value *>(CS.getArgument(ArgNo++));
       break;
     case InlineAsm::isClobber:
       // Nothing to do.
@@ -5270,7 +5228,7 @@
       // Strip bitcasts, if any.  This mostly comes up for functions.
       OpInfo.CallOperandVal = OpInfo.CallOperandVal->stripPointerCasts();
 
-      if (BasicBlock *BB = dyn_cast<BasicBlock>(OpInfo.CallOperandVal)) {
+      if (const BasicBlock *BB = dyn_cast<BasicBlock>(OpInfo.CallOperandVal)) {
         OpInfo.CallOperand = DAG.getBasicBlock(FuncInfo.MBBMap[BB]);
       } else {
         OpInfo.CallOperand = getValue(OpInfo.CallOperandVal);
@@ -5293,14 +5251,15 @@
     // error.
     if (OpInfo.hasMatchingInput()) {
       SDISelAsmOperandInfo &Input = ConstraintOperands[OpInfo.MatchingInput];
+      
       if (OpInfo.ConstraintVT != Input.ConstraintVT) {
         if ((OpInfo.ConstraintVT.isInteger() !=
              Input.ConstraintVT.isInteger()) ||
             (OpInfo.ConstraintVT.getSizeInBits() !=
              Input.ConstraintVT.getSizeInBits())) {
-          llvm_report_error("Unsupported asm: input constraint"
-                            " with a matching output constraint of incompatible"
-                            " type!");
+          report_fatal_error("Unsupported asm: input constraint"
+                             " with a matching output constraint of"
+                             " incompatible type!");
         }
         Input.ConstraintVT = OpInfo.ConstraintVT;
       }
@@ -5322,7 +5281,7 @@
 
       // If the operand is a float, integer, or vector constant, spill to a
       // constant pool entry to get its address.
-      Value *OpVal = OpInfo.CallOperandVal;
+      const Value *OpVal = OpInfo.CallOperandVal;
       if (isa<ConstantFP>(OpVal) || isa<ConstantInt>(OpVal) ||
           isa<ConstantVector>(OpVal)) {
         OpInfo.CallOperand = DAG.getConstantPool(cast<Constant>(OpVal),
@@ -5375,6 +5334,11 @@
           DAG.getTargetExternalSymbol(IA->getAsmString().c_str(),
                                       TLI.getPointerTy()));
 
+  // If we have a !srcloc metadata node associated with it, we want to attach
+  // this to the ultimately generated inline asm machineinstr.  To do this, we
+  // pass in the third operand as this (potentially null) inline asm MDNode.
+  const MDNode *SrcLoc = CS.getInstruction()->getMetadata("srcloc");
+  AsmNodeOperands.push_back(DAG.getMDNode(SrcLoc));
 
   // Loop over all of the inputs, copying the operand values into the
   // appropriate registers and processing the output regs.
@@ -5394,8 +5358,8 @@
         assert(OpInfo.isIndirect && "Memory output must be indirect operand");
 
         // Add information to the INLINEASM node to know about this output.
-        unsigned ResOpType = 4/*MEM*/ | (1<<3);
-        AsmNodeOperands.push_back(DAG.getTargetConstant(ResOpType,
+        unsigned OpFlags = InlineAsm::getFlagWord(InlineAsm::Kind_Mem, 1);
+        AsmNodeOperands.push_back(DAG.getTargetConstant(OpFlags,
                                                         TLI.getPointerTy()));
         AsmNodeOperands.push_back(OpInfo.CallOperand);
         break;
@@ -5405,10 +5369,9 @@
 
       // Copy the output from the appropriate register.  Find a register that
       // we can use.
-      if (OpInfo.AssignedRegs.Regs.empty()) {
-        llvm_report_error("Couldn't allocate output reg for"
-                          " constraint '" + OpInfo.ConstraintCode + "'!");
-      }
+      if (OpInfo.AssignedRegs.Regs.empty())
+        report_fatal_error("Couldn't allocate output reg for constraint '" +
+                           Twine(OpInfo.ConstraintCode) + "'!");
 
       // If this is an indirect operand, store through the pointer after the
       // asm.
@@ -5425,8 +5388,8 @@
       // Add information to the INLINEASM node to know that this register is
       // set.
       OpInfo.AssignedRegs.AddInlineAsmOperands(OpInfo.isEarlyClobber ?
-                                               6 /* EARLYCLOBBER REGDEF */ :
-                                               2 /* REGDEF */ ,
+                                           InlineAsm::Kind_RegDefEarlyClobber :
+                                               InlineAsm::Kind_RegDef,
                                                false,
                                                0,
                                                DAG,
@@ -5443,27 +5406,30 @@
 
         // Scan until we find the definition we already emitted of this operand.
         // When we find it, create a RegsForValue operand.
-        unsigned CurOp = 2;  // The first operand.
+        unsigned CurOp = InlineAsm::Op_FirstOperand;
         for (; OperandNo; --OperandNo) {
           // Advance to the next operand.
           unsigned OpFlag =
             cast<ConstantSDNode>(AsmNodeOperands[CurOp])->getZExtValue();
-          assert(((OpFlag & 7) == 2 /*REGDEF*/ ||
-                  (OpFlag & 7) == 6 /*EARLYCLOBBER REGDEF*/ ||
-                  (OpFlag & 7) == 4 /*MEM*/) &&
-                 "Skipped past definitions?");
+          assert((InlineAsm::isRegDefKind(OpFlag) ||
+                  InlineAsm::isRegDefEarlyClobberKind(OpFlag) ||
+                  InlineAsm::isMemKind(OpFlag)) && "Skipped past definitions?");
           CurOp += InlineAsm::getNumOperandRegisters(OpFlag)+1;
         }
 
         unsigned OpFlag =
           cast<ConstantSDNode>(AsmNodeOperands[CurOp])->getZExtValue();
-        if ((OpFlag & 7) == 2 /*REGDEF*/
-            || (OpFlag & 7) == 6 /* EARLYCLOBBER REGDEF */) {
+        if (InlineAsm::isRegDefKind(OpFlag) ||
+            InlineAsm::isRegDefEarlyClobberKind(OpFlag)) {
           // Add (OpFlag&0xffff)>>3 registers to MatchedRegs.
           if (OpInfo.isIndirect) {
-            llvm_report_error("Don't know how to handle tied indirect "
-                              "register inputs yet!");
+            // This happens on gcc/testsuite/gcc.dg/pr8788-1.c
+            LLVMContext &Ctx = *DAG.getContext();
+            Ctx.emitError(CS.getInstruction(),  "inline asm not supported yet:"
+                          " don't know how to handle tied "
+                          "indirect register inputs");
           }
+          
           RegsForValue MatchedRegs;
           MatchedRegs.TLI = &TLI;
           MatchedRegs.ValueVTs.push_back(InOperandVal.getValueType());
@@ -5478,22 +5444,23 @@
           // Use the produced MatchedRegs object to
           MatchedRegs.getCopyToRegs(InOperandVal, DAG, getCurDebugLoc(),
                                     Chain, &Flag);
-          MatchedRegs.AddInlineAsmOperands(1 /*REGUSE*/,
+          MatchedRegs.AddInlineAsmOperands(InlineAsm::Kind_RegUse,
                                            true, OpInfo.getMatchedOperand(),
                                            DAG, AsmNodeOperands);
           break;
-        } else {
-          assert(((OpFlag & 7) == 4) && "Unknown matching constraint!");
-          assert((InlineAsm::getNumOperandRegisters(OpFlag)) == 1 &&
-                 "Unexpected number of operands");
-          // Add information to the INLINEASM node to know about this input.
-          // See InlineAsm.h isUseOperandTiedToDef.
-          OpFlag |= 0x80000000 | (OpInfo.getMatchedOperand() << 16);
-          AsmNodeOperands.push_back(DAG.getTargetConstant(OpFlag,
-                                                          TLI.getPointerTy()));
-          AsmNodeOperands.push_back(AsmNodeOperands[CurOp+1]);
-          break;
         }
+        
+        assert(InlineAsm::isMemKind(OpFlag) && "Unknown matching constraint!");
+        assert(InlineAsm::getNumOperandRegisters(OpFlag) == 1 &&
+               "Unexpected number of operands");
+        // Add information to the INLINEASM node to know about this input.
+        // See InlineAsm.h isUseOperandTiedToDef.
+        OpFlag = InlineAsm::getFlagWordForMatchingOp(OpFlag,
+                                                    OpInfo.getMatchedOperand());
+        AsmNodeOperands.push_back(DAG.getTargetConstant(OpFlag,
+                                                        TLI.getPointerTy()));
+        AsmNodeOperands.push_back(AsmNodeOperands[CurOp+1]);
+        break;
       }
 
       if (OpInfo.ConstraintType == TargetLowering::C_Other) {
@@ -5503,24 +5470,26 @@
         std::vector<SDValue> Ops;
         TLI.LowerAsmOperandForConstraint(InOperandVal, OpInfo.ConstraintCode[0],
                                          hasMemory, Ops, DAG);
-        if (Ops.empty()) {
-          llvm_report_error("Invalid operand for inline asm"
-                            " constraint '" + OpInfo.ConstraintCode + "'!");
-        }
+        if (Ops.empty())
+          report_fatal_error("Invalid operand for inline asm constraint '" +
+                             Twine(OpInfo.ConstraintCode) + "'!");
 
         // Add information to the INLINEASM node to know about this input.
-        unsigned ResOpType = 3 /*IMM*/ | (Ops.size() << 3);
+        unsigned ResOpType =
+          InlineAsm::getFlagWord(InlineAsm::Kind_Imm, Ops.size());
         AsmNodeOperands.push_back(DAG.getTargetConstant(ResOpType,
                                                         TLI.getPointerTy()));
         AsmNodeOperands.insert(AsmNodeOperands.end(), Ops.begin(), Ops.end());
         break;
-      } else if (OpInfo.ConstraintType == TargetLowering::C_Memory) {
+      }
+      
+      if (OpInfo.ConstraintType == TargetLowering::C_Memory) {
         assert(OpInfo.isIndirect && "Operand must be indirect to be a mem!");
         assert(InOperandVal.getValueType() == TLI.getPointerTy() &&
                "Memory operands expect pointer values");
 
         // Add information to the INLINEASM node to know about this input.
-        unsigned ResOpType = 4/*MEM*/ | (1<<3);
+        unsigned ResOpType = InlineAsm::getFlagWord(InlineAsm::Kind_Mem, 1);
         AsmNodeOperands.push_back(DAG.getTargetConstant(ResOpType,
                                                         TLI.getPointerTy()));
         AsmNodeOperands.push_back(InOperandVal);
@@ -5535,15 +5504,14 @@
 
       // Copy the input into the appropriate registers.
       if (OpInfo.AssignedRegs.Regs.empty() ||
-          !OpInfo.AssignedRegs.areValueTypesLegal()) {
-        llvm_report_error("Couldn't allocate input reg for"
-                          " constraint '"+ OpInfo.ConstraintCode +"'!");
-      }
+          !OpInfo.AssignedRegs.areValueTypesLegal())
+        report_fatal_error("Couldn't allocate input reg for constraint '" +
+                           Twine(OpInfo.ConstraintCode) + "'!");
 
       OpInfo.AssignedRegs.getCopyToRegs(InOperandVal, DAG, getCurDebugLoc(),
                                         Chain, &Flag);
 
-      OpInfo.AssignedRegs.AddInlineAsmOperands(1/*REGUSE*/, false, 0,
+      OpInfo.AssignedRegs.AddInlineAsmOperands(InlineAsm::Kind_RegUse, false, 0,
                                                DAG, AsmNodeOperands);
       break;
     }
@@ -5551,7 +5519,8 @@
       // Add the clobbered value to the operand list, so that the register
       // allocator is aware that the physreg got clobbered.
       if (!OpInfo.AssignedRegs.Regs.empty())
-        OpInfo.AssignedRegs.AddInlineAsmOperands(6 /* EARLYCLOBBER REGDEF */,
+        OpInfo.AssignedRegs.AddInlineAsmOperands(
+                                            InlineAsm::Kind_RegDefEarlyClobber,
                                                  false, 0, DAG,
                                                  AsmNodeOperands);
       break;
@@ -5559,7 +5528,7 @@
     }
   }
 
-  // Finish up input operands.
+  // Finish up input operands.  Set the input chain and add the flag last.
   AsmNodeOperands[0] = Chain;
   if (Flag.getNode()) AsmNodeOperands.push_back(Flag);
 
@@ -5604,17 +5573,16 @@
       return;
   }
 
-  std::vector<std::pair<SDValue, Value*> > StoresToEmit;
+  std::vector<std::pair<SDValue, const Value *> > StoresToEmit;
 
   // Process indirect outputs, first output all of the flagged copies out of
   // physregs.
   for (unsigned i = 0, e = IndirectStoresToEmit.size(); i != e; ++i) {
     RegsForValue &OutRegs = IndirectStoresToEmit[i].first;
-    Value *Ptr = IndirectStoresToEmit[i].second;
+    const Value *Ptr = IndirectStoresToEmit[i].second;
     SDValue OutVal = OutRegs.getCopyFromRegs(DAG, getCurDebugLoc(),
                                              Chain, &Flag);
     StoresToEmit.push_back(std::make_pair(OutVal, Ptr));
-
   }
 
   // Emit the non-flagged stores from the physregs.
@@ -5635,14 +5603,14 @@
   DAG.setRoot(Chain);
 }
 
-void SelectionDAGBuilder::visitVAStart(CallInst &I) {
+void SelectionDAGBuilder::visitVAStart(const CallInst &I) {
   DAG.setRoot(DAG.getNode(ISD::VASTART, getCurDebugLoc(),
                           MVT::Other, getRoot(),
                           getValue(I.getOperand(1)),
                           DAG.getSrcValue(I.getOperand(1))));
 }
 
-void SelectionDAGBuilder::visitVAArg(VAArgInst &I) {
+void SelectionDAGBuilder::visitVAArg(const VAArgInst &I) {
   SDValue V = DAG.getVAArg(TLI.getValueType(I.getType()), getCurDebugLoc(),
                            getRoot(), getValue(I.getOperand(0)),
                            DAG.getSrcValue(I.getOperand(0)));
@@ -5650,14 +5618,14 @@
   DAG.setRoot(V.getValue(1));
 }
 
-void SelectionDAGBuilder::visitVAEnd(CallInst &I) {
+void SelectionDAGBuilder::visitVAEnd(const CallInst &I) {
   DAG.setRoot(DAG.getNode(ISD::VAEND, getCurDebugLoc(),
                           MVT::Other, getRoot(),
                           getValue(I.getOperand(1)),
                           DAG.getSrcValue(I.getOperand(1))));
 }
 
-void SelectionDAGBuilder::visitVACopy(CallInst &I) {
+void SelectionDAGBuilder::visitVACopy(const CallInst &I) {
   DAG.setRoot(DAG.getNode(ISD::VACOPY, getCurDebugLoc(),
                           MVT::Other, getRoot(),
                           getValue(I.getOperand(1)),
@@ -5677,7 +5645,8 @@
                             CallingConv::ID CallConv, bool isTailCall,
                             bool isReturnValueUsed,
                             SDValue Callee,
-                            ArgListTy &Args, SelectionDAG &DAG, DebugLoc dl) {
+                            ArgListTy &Args, SelectionDAG &DAG,
+                            DebugLoc dl) const {
   // Handle all of the outgoing arguments.
   SmallVector<ISD::OutputArg, 32> Outs;
   for (unsigned i = 0, e = Args.size(); i != e; ++i) {
@@ -5777,12 +5746,6 @@
          "LowerCall emitted a return value for a tail call!");
   assert((isTailCall || InVals.size() == Ins.size()) &&
          "LowerCall didn't emit the correct number of values!");
-  DEBUG(for (unsigned i = 0, e = Ins.size(); i != e; ++i) {
-          assert(InVals[i].getNode() &&
-                 "LowerCall emitted a null value!");
-          assert(Ins[i].VT == InVals[i].getValueType() &&
-                 "LowerCall emitted a value with the wrong type!");
-        });
 
   // For a tail call, the return value is merely live-out and there aren't
   // any nodes in the DAG representing it. Return a special value to
@@ -5793,6 +5756,13 @@
     return std::make_pair(SDValue(), SDValue());
   }
 
+  DEBUG(for (unsigned i = 0, e = Ins.size(); i != e; ++i) {
+          assert(InVals[i].getNode() &&
+                 "LowerCall emitted a null value!");
+          assert(Ins[i].VT == InVals[i].getValueType() &&
+                 "LowerCall emitted a value with the wrong type!");
+        });
+
   // Collect the legal value parts into potentially illegal values
   // that correspond to the original function's return values.
   ISD::NodeType AssertOp = ISD::DELETED_NODE;
@@ -5827,18 +5797,19 @@
 
 void TargetLowering::LowerOperationWrapper(SDNode *N,
                                            SmallVectorImpl<SDValue> &Results,
-                                           SelectionDAG &DAG) {
+                                           SelectionDAG &DAG) const {
   SDValue Res = LowerOperation(SDValue(N, 0), DAG);
   if (Res.getNode())
     Results.push_back(Res);
 }
 
-SDValue TargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) {
+SDValue TargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
   llvm_unreachable("LowerOperation not implemented for this target!");
   return SDValue();
 }
 
-void SelectionDAGBuilder::CopyValueToVirtualRegister(Value *V, unsigned Reg) {
+void
+SelectionDAGBuilder::CopyValueToVirtualRegister(const Value *V, unsigned Reg) {
   SDValue Op = getValue(V);
   assert((Op.getOpcode() != ISD::CopyFromReg ||
           cast<RegisterSDNode>(Op.getOperand(1))->getReg() != Reg) &&
@@ -5853,9 +5824,9 @@
 
 #include "llvm/CodeGen/SelectionDAGISel.h"
 
-void SelectionDAGISel::LowerArguments(BasicBlock *LLVMBB) {
+void SelectionDAGISel::LowerArguments(const BasicBlock *LLVMBB) {
   // If this is the entry block, emit arguments.
-  Function &F = *LLVMBB->getParent();
+  const Function &F = *LLVMBB->getParent();
   SelectionDAG &DAG = SDB->DAG;
   SDValue OldRoot = DAG.getRoot();
   DebugLoc dl = SDB->getCurDebugLoc();
@@ -5880,14 +5851,14 @@
     // or one register.
     ISD::ArgFlagsTy Flags;
     Flags.setSRet();
-    EVT RegisterVT = TLI.getRegisterType(*CurDAG->getContext(), ValueVTs[0]);
+    EVT RegisterVT = TLI.getRegisterType(*DAG.getContext(), ValueVTs[0]);
     ISD::InputArg RetArg(Flags, RegisterVT, true);
     Ins.push_back(RetArg);
   }
 
   // Set up the incoming argument description vector.
   unsigned Idx = 1;
-  for (Function::arg_iterator I = F.arg_begin(), E = F.arg_end();
+  for (Function::const_arg_iterator I = F.arg_begin(), E = F.arg_end();
        I != E; ++I, ++Idx) {
     SmallVector<EVT, 4> ValueVTs;
     ComputeValueVTs(TLI, I->getType(), ValueVTs);
@@ -5989,7 +5960,7 @@
     ++i;
   }
 
-  for (Function::arg_iterator I = F.arg_begin(), E = F.arg_end(); I != E;
+  for (Function::const_arg_iterator I = F.arg_begin(), E = F.arg_end(); I != E;
       ++I, ++Idx) {
     SmallVector<SDValue, 4> ArgValues;
     SmallVector<EVT, 4> ValueVTs;
@@ -6016,8 +5987,10 @@
     }
 
     if (!I->use_empty()) {
-      SDValue Res = DAG.getMergeValues(&ArgValues[0], NumValues,
-                                       SDB->getCurDebugLoc());
+      SDValue Res;
+      if (!ArgValues.empty())
+        Res = DAG.getMergeValues(&ArgValues[0], NumValues,
+                                 SDB->getCurDebugLoc());
       SDB->setValue(I, Res);
 
       // If this argument is live outside of the entry block, insert a copy from
@@ -6030,7 +6003,7 @@
 
   // Finally, if the target has anything special to do, allow it to do so.
   // FIXME: this should insert code into the DAG!
-  EmitFunctionEntryCode(F, SDB->DAG.getMachineFunction());
+  EmitFunctionEntryCode();
 }
 
 /// Handle PHI nodes in successor blocks.  Emit code into the SelectionDAG to
@@ -6041,51 +6014,50 @@
 /// the end.
 ///
 void
-SelectionDAGISel::HandlePHINodesInSuccessorBlocks(BasicBlock *LLVMBB) {
-  TerminatorInst *TI = LLVMBB->getTerminator();
+SelectionDAGBuilder::HandlePHINodesInSuccessorBlocks(const BasicBlock *LLVMBB) {
+  const TerminatorInst *TI = LLVMBB->getTerminator();
 
   SmallPtrSet<MachineBasicBlock *, 4> SuccsHandled;
 
   // Check successor nodes' PHI nodes that expect a constant to be available
   // from this block.
   for (unsigned succ = 0, e = TI->getNumSuccessors(); succ != e; ++succ) {
-    BasicBlock *SuccBB = TI->getSuccessor(succ);
+    const BasicBlock *SuccBB = TI->getSuccessor(succ);
     if (!isa<PHINode>(SuccBB->begin())) continue;
-    MachineBasicBlock *SuccMBB = FuncInfo->MBBMap[SuccBB];
+    MachineBasicBlock *SuccMBB = FuncInfo.MBBMap[SuccBB];
 
     // If this terminator has multiple identical successors (common for
     // switches), only handle each succ once.
     if (!SuccsHandled.insert(SuccMBB)) continue;
 
     MachineBasicBlock::iterator MBBI = SuccMBB->begin();
-    PHINode *PN;
 
     // At this point we know that there is a 1-1 correspondence between LLVM PHI
     // nodes and Machine PHI nodes, but the incoming operands have not been
     // emitted yet.
-    for (BasicBlock::iterator I = SuccBB->begin();
-         (PN = dyn_cast<PHINode>(I)); ++I) {
+    for (BasicBlock::const_iterator I = SuccBB->begin();
+         const PHINode *PN = dyn_cast<PHINode>(I); ++I) {
       // Ignore dead phi's.
       if (PN->use_empty()) continue;
 
       unsigned Reg;
-      Value *PHIOp = PN->getIncomingValueForBlock(LLVMBB);
+      const Value *PHIOp = PN->getIncomingValueForBlock(LLVMBB);
 
-      if (Constant *C = dyn_cast<Constant>(PHIOp)) {
-        unsigned &RegOut = SDB->ConstantsOut[C];
+      if (const Constant *C = dyn_cast<Constant>(PHIOp)) {
+        unsigned &RegOut = ConstantsOut[C];
         if (RegOut == 0) {
-          RegOut = FuncInfo->CreateRegForValue(C);
-          SDB->CopyValueToVirtualRegister(C, RegOut);
+          RegOut = FuncInfo.CreateRegForValue(C);
+          CopyValueToVirtualRegister(C, RegOut);
         }
         Reg = RegOut;
       } else {
-        Reg = FuncInfo->ValueMap[PHIOp];
+        Reg = FuncInfo.ValueMap[PHIOp];
         if (Reg == 0) {
           assert(isa<AllocaInst>(PHIOp) &&
-                 FuncInfo->StaticAllocaMap.count(cast<AllocaInst>(PHIOp)) &&
+                 FuncInfo.StaticAllocaMap.count(cast<AllocaInst>(PHIOp)) &&
                  "Didn't codegen value into a register!??");
-          Reg = FuncInfo->CreateRegForValue(PHIOp);
-          SDB->CopyValueToVirtualRegister(PHIOp, Reg);
+          Reg = FuncInfo.CreateRegForValue(PHIOp);
+          CopyValueToVirtualRegister(PHIOp, Reg);
         }
       }
 
@@ -6095,77 +6067,12 @@
       ComputeValueVTs(TLI, PN->getType(), ValueVTs);
       for (unsigned vti = 0, vte = ValueVTs.size(); vti != vte; ++vti) {
         EVT VT = ValueVTs[vti];
-        unsigned NumRegisters = TLI.getNumRegisters(*CurDAG->getContext(), VT);
+        unsigned NumRegisters = TLI.getNumRegisters(*DAG.getContext(), VT);
         for (unsigned i = 0, e = NumRegisters; i != e; ++i)
-          SDB->PHINodesToUpdate.push_back(std::make_pair(MBBI++, Reg+i));
+          FuncInfo.PHINodesToUpdate.push_back(std::make_pair(MBBI++, Reg+i));
         Reg += NumRegisters;
       }
     }
   }
-  SDB->ConstantsOut.clear();
-}
-
-/// This is the Fast-ISel version of HandlePHINodesInSuccessorBlocks. It only
-/// supports legal types, and it emits MachineInstrs directly instead of
-/// creating SelectionDAG nodes.
-///
-bool
-SelectionDAGISel::HandlePHINodesInSuccessorBlocksFast(BasicBlock *LLVMBB,
-                                                      FastISel *F) {
-  TerminatorInst *TI = LLVMBB->getTerminator();
-
-  SmallPtrSet<MachineBasicBlock *, 4> SuccsHandled;
-  unsigned OrigNumPHINodesToUpdate = SDB->PHINodesToUpdate.size();
-
-  // Check successor nodes' PHI nodes that expect a constant to be available
-  // from this block.
-  for (unsigned succ = 0, e = TI->getNumSuccessors(); succ != e; ++succ) {
-    BasicBlock *SuccBB = TI->getSuccessor(succ);
-    if (!isa<PHINode>(SuccBB->begin())) continue;
-    MachineBasicBlock *SuccMBB = FuncInfo->MBBMap[SuccBB];
-
-    // If this terminator has multiple identical successors (common for
-    // switches), only handle each succ once.
-    if (!SuccsHandled.insert(SuccMBB)) continue;
-
-    MachineBasicBlock::iterator MBBI = SuccMBB->begin();
-    PHINode *PN;
-
-    // At this point we know that there is a 1-1 correspondence between LLVM PHI
-    // nodes and Machine PHI nodes, but the incoming operands have not been
-    // emitted yet.
-    for (BasicBlock::iterator I = SuccBB->begin();
-         (PN = dyn_cast<PHINode>(I)); ++I) {
-      // Ignore dead phi's.
-      if (PN->use_empty()) continue;
-
-      // Only handle legal types. Two interesting things to note here. First,
-      // by bailing out early, we may leave behind some dead instructions,
-      // since SelectionDAG's HandlePHINodesInSuccessorBlocks will insert its
-      // own moves. Second, this check is necessary becuase FastISel doesn't
-      // use CreateRegForValue to create registers, so it always creates
-      // exactly one register for each non-void instruction.
-      EVT VT = TLI.getValueType(PN->getType(), /*AllowUnknown=*/true);
-      if (VT == MVT::Other || !TLI.isTypeLegal(VT)) {
-        // Promote MVT::i1.
-        if (VT == MVT::i1)
-          VT = TLI.getTypeToTransformTo(*CurDAG->getContext(), VT);
-        else {
-          SDB->PHINodesToUpdate.resize(OrigNumPHINodesToUpdate);
-          return false;
-        }
-      }
-
-      Value *PHIOp = PN->getIncomingValueForBlock(LLVMBB);
-
-      unsigned Reg = F->getRegForValue(PHIOp);
-      if (Reg == 0) {
-        SDB->PHINodesToUpdate.resize(OrigNumPHINodesToUpdate);
-        return false;
-      }
-      SDB->PHINodesToUpdate.push_back(std::make_pair(MBBI++, Reg));
-    }
-  }
-
-  return true;
+  ConstantsOut.clear();
 }
diff --git a/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h b/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h
index bc4b33d..d9495fb 100644
--- a/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h
+++ b/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h
@@ -56,7 +56,6 @@
 class Instruction;
 class LoadInst;
 class MachineBasicBlock;
-class MachineFunction;
 class MachineInstr;
 class MachineRegisterInfo;
 class PHINode;
@@ -81,11 +80,8 @@
 //===----------------------------------------------------------------------===//
 /// SelectionDAGBuilder - This is the common target-independent lowering
 /// implementation that is parameterized by a TargetLowering object.
-/// Also, targets can overload any lowering method.
 ///
 class SelectionDAGBuilder {
-  MachineBasicBlock *CurMBB;
-
   /// CurDebugLoc - current file + line number.  Changes as we build the DAG.
   DebugLoc CurDebugLoc;
 
@@ -144,15 +140,16 @@
   /// CaseRec - A struct with ctor used in lowering switches to a binary tree
   /// of conditional branches.
   struct CaseRec {
-    CaseRec(MachineBasicBlock *bb, Constant *lt, Constant *ge, CaseRange r) :
+    CaseRec(MachineBasicBlock *bb, const Constant *lt, const Constant *ge,
+            CaseRange r) :
     CaseBB(bb), LT(lt), GE(ge), Range(r) {}
 
     /// CaseBB - The MBB in which to emit the compare and branch
     MachineBasicBlock *CaseBB;
     /// LT, GE - If nonzero, we know the current case value must be less-than or
     /// greater-than-or-equal-to these Constants.
-    Constant *LT;
-    Constant *GE;
+    const Constant *LT;
+    const Constant *GE;
     /// Range - A pair of iterators representing the range of case values to be
     /// processed at this point in the binary search tree.
     CaseRange Range;
@@ -183,7 +180,8 @@
   /// SelectionDAGBuilder and SDISel for the code generation of additional basic
   /// blocks needed by multi-case switch statements.
   struct CaseBlock {
-    CaseBlock(ISD::CondCode cc, Value *cmplhs, Value *cmprhs, Value *cmpmiddle,
+    CaseBlock(ISD::CondCode cc, const Value *cmplhs, const Value *cmprhs,
+              const Value *cmpmiddle,
               MachineBasicBlock *truebb, MachineBasicBlock *falsebb,
               MachineBasicBlock *me)
       : CC(cc), CmpLHS(cmplhs), CmpMHS(cmpmiddle), CmpRHS(cmprhs),
@@ -193,7 +191,7 @@
     // CmpLHS/CmpRHS/CmpMHS - The LHS/MHS/RHS of the comparison to emit.
     // Emit by default LHS op RHS. MHS is used for range comparisons:
     // If MHS is not null: (LHS <= MHS) and (MHS <= RHS).
-    Value *CmpLHS, *CmpMHS, *CmpRHS;
+    const Value *CmpLHS, *CmpMHS, *CmpRHS;
     // TrueBB/FalseBB - the block to branch to if the setcc is true/false.
     MachineBasicBlock *TrueBB, *FalseBB;
     // ThisBB - the block into which to emit the code for the setcc and branches
@@ -215,12 +213,12 @@
     MachineBasicBlock *Default;
   };
   struct JumpTableHeader {
-    JumpTableHeader(APInt F, APInt L, Value *SV, MachineBasicBlock *H,
+    JumpTableHeader(APInt F, APInt L, const Value *SV, MachineBasicBlock *H,
                     bool E = false):
       First(F), Last(L), SValue(SV), HeaderBB(H), Emitted(E) {}
     APInt First;
     APInt Last;
-    Value *SValue;
+    const Value *SValue;
     MachineBasicBlock *HeaderBB;
     bool Emitted;
   };
@@ -237,7 +235,7 @@
   typedef SmallVector<BitTestCase, 3> BitTestInfo;
 
   struct BitTestBlock {
-    BitTestBlock(APInt F, APInt R, Value* SV,
+    BitTestBlock(APInt F, APInt R, const Value* SV,
                  unsigned Rg, bool E,
                  MachineBasicBlock* P, MachineBasicBlock* D,
                  const BitTestInfo& C):
@@ -245,7 +243,7 @@
       Parent(P), Default(D), Cases(C) { }
     APInt First;
     APInt Range;
-    Value  *SValue;
+    const Value *SValue;
     unsigned Reg;
     bool Emitted;
     MachineBasicBlock *Parent;
@@ -257,7 +255,8 @@
   // TLI - This is information that describes the available target features we
   // need for lowering.  This indicates when operations are unavailable,
   // implemented with a libcall, etc.
-  TargetLowering &TLI;
+  const TargetMachine &TM;
+  const TargetLowering &TLI;
   SelectionDAG &DAG;
   const TargetData *TD;
   AliasAnalysis *AA;
@@ -272,17 +271,13 @@
   /// SwitchInst code generation information.
   std::vector<BitTestBlock> BitTestCases;
 
-  /// PHINodesToUpdate - A list of phi instructions whose operand list will
-  /// be updated after processing the current basic block.
-  std::vector<std::pair<MachineInstr*, unsigned> > PHINodesToUpdate;
-
   /// EdgeMapping - If an edge from CurMBB to any MBB is changed (e.g. due to
   /// scheduler custom lowering), track the change here.
   DenseMap<MachineBasicBlock*, MachineBasicBlock*> EdgeMapping;
 
   // Emit PHI-node-operand constants only once even if used by multiple
   // PHI nodes.
-  DenseMap<Constant*, unsigned> ConstantsOut;
+  DenseMap<const Constant *, unsigned> ConstantsOut;
 
   /// FuncInfo - Information about the function as a whole.
   ///
@@ -303,18 +298,16 @@
 
   LLVMContext *Context;
 
-  SelectionDAGBuilder(SelectionDAG &dag, TargetLowering &tli,
-                      FunctionLoweringInfo &funcinfo,
+  SelectionDAGBuilder(SelectionDAG &dag, FunctionLoweringInfo &funcinfo,
                       CodeGenOpt::Level ol)
-    : CurDebugLoc(DebugLoc::getUnknownLoc()), SDNodeOrder(0),
-      TLI(tli), DAG(dag), FuncInfo(funcinfo), OptLevel(ol),
-      HasTailCall(false),
-      Context(dag.getContext()) {
+    : SDNodeOrder(0), TM(dag.getTarget()), TLI(dag.getTargetLoweringInfo()),
+      DAG(dag), FuncInfo(funcinfo), OptLevel(ol),
+      HasTailCall(false), Context(dag.getContext()) {
   }
 
   void init(GCFunctionInfo *gfi, AliasAnalysis &aa);
 
-  /// clear - Clear out the curret SelectionDAG and the associated
+  /// clear - Clear out the current SelectionDAG and the associated
   /// state and prepare this SelectionDAGBuilder object to be used
   /// for a new block. This doesn't clear out information about
   /// additional blocks that are needed to complete switch lowering
@@ -336,22 +329,19 @@
   SDValue getControlRoot();
 
   DebugLoc getCurDebugLoc() const { return CurDebugLoc; }
-  void setCurDebugLoc(DebugLoc dl) { CurDebugLoc = dl; }
 
   unsigned getSDNodeOrder() const { return SDNodeOrder; }
 
-  void CopyValueToVirtualRegister(Value *V, unsigned Reg);
+  void CopyValueToVirtualRegister(const Value *V, unsigned Reg);
 
   /// AssignOrderingToNode - Assign an ordering to the node. The order is gotten
   /// from how the code appeared in the source. The ordering is used by the
   /// scheduler to effectively turn off scheduling.
   void AssignOrderingToNode(const SDNode *Node);
 
-  void visit(Instruction &I);
+  void visit(const Instruction &I);
 
-  void visit(unsigned Opcode, User &I);
-
-  void setCurrentBasicBlock(MachineBasicBlock *MBB) { CurMBB = MBB; }
+  void visit(unsigned Opcode, const User &I);
 
   SDValue getValue(const Value *V);
 
@@ -365,136 +355,146 @@
                             std::set<unsigned> &OutputRegs, 
                             std::set<unsigned> &InputRegs);
 
-  void FindMergedConditions(Value *Cond, MachineBasicBlock *TBB,
+  void FindMergedConditions(const Value *Cond, MachineBasicBlock *TBB,
                             MachineBasicBlock *FBB, MachineBasicBlock *CurBB,
-                            unsigned Opc);
-  void EmitBranchForMergedCondition(Value *Cond, MachineBasicBlock *TBB,
+                            MachineBasicBlock *SwitchBB, unsigned Opc);
+  void EmitBranchForMergedCondition(const Value *Cond, MachineBasicBlock *TBB,
                                     MachineBasicBlock *FBB,
-                                    MachineBasicBlock *CurBB);
+                                    MachineBasicBlock *CurBB,
+                                    MachineBasicBlock *SwitchBB);
   bool ShouldEmitAsBranches(const std::vector<CaseBlock> &Cases);
-  bool isExportableFromCurrentBlock(Value *V, const BasicBlock *FromBB);
-  void CopyToExportRegsIfNeeded(Value *V);
-  void ExportFromCurrentBlock(Value *V);
-  void LowerCallTo(CallSite CS, SDValue Callee, bool IsTailCall,
+  bool isExportableFromCurrentBlock(const Value *V, const BasicBlock *FromBB);
+  void CopyToExportRegsIfNeeded(const Value *V);
+  void ExportFromCurrentBlock(const Value *V);
+  void LowerCallTo(ImmutableCallSite CS, SDValue Callee, bool IsTailCall,
                    MachineBasicBlock *LandingPad = NULL);
 
 private:
   // Terminator instructions.
-  void visitRet(ReturnInst &I);
-  void visitBr(BranchInst &I);
-  void visitSwitch(SwitchInst &I);
-  void visitIndirectBr(IndirectBrInst &I);
-  void visitUnreachable(UnreachableInst &I) { /* noop */ }
+  void visitRet(const ReturnInst &I);
+  void visitBr(const BranchInst &I);
+  void visitSwitch(const SwitchInst &I);
+  void visitIndirectBr(const IndirectBrInst &I);
+  void visitUnreachable(const UnreachableInst &I) { /* noop */ }
 
   // Helpers for visitSwitch
   bool handleSmallSwitchRange(CaseRec& CR,
                               CaseRecVector& WorkList,
-                              Value* SV,
-                              MachineBasicBlock* Default);
+                              const Value* SV,
+                              MachineBasicBlock* Default,
+                              MachineBasicBlock *SwitchBB);
   bool handleJTSwitchCase(CaseRec& CR,
                           CaseRecVector& WorkList,
-                          Value* SV,
-                          MachineBasicBlock* Default);
+                          const Value* SV,
+                          MachineBasicBlock* Default,
+                          MachineBasicBlock *SwitchBB);
   bool handleBTSplitSwitchCase(CaseRec& CR,
                                CaseRecVector& WorkList,
-                               Value* SV,
-                               MachineBasicBlock* Default);
+                               const Value* SV,
+                               MachineBasicBlock* Default,
+                               MachineBasicBlock *SwitchBB);
   bool handleBitTestsSwitchCase(CaseRec& CR,
                                 CaseRecVector& WorkList,
-                                Value* SV,
-                                MachineBasicBlock* Default);  
+                                const Value* SV,
+                                MachineBasicBlock* Default,
+                                MachineBasicBlock *SwitchBB);
 public:
-  void visitSwitchCase(CaseBlock &CB);
-  void visitBitTestHeader(BitTestBlock &B);
+  void visitSwitchCase(CaseBlock &CB,
+                       MachineBasicBlock *SwitchBB);
+  void visitBitTestHeader(BitTestBlock &B, MachineBasicBlock *SwitchBB);
   void visitBitTestCase(MachineBasicBlock* NextMBB,
                         unsigned Reg,
-                        BitTestCase &B);
+                        BitTestCase &B,
+                        MachineBasicBlock *SwitchBB);
   void visitJumpTable(JumpTable &JT);
-  void visitJumpTableHeader(JumpTable &JT, JumpTableHeader &JTH);
+  void visitJumpTableHeader(JumpTable &JT, JumpTableHeader &JTH,
+                            MachineBasicBlock *SwitchBB);
   
 private:
   // These all get lowered before this pass.
-  void visitInvoke(InvokeInst &I);
-  void visitUnwind(UnwindInst &I);
+  void visitInvoke(const InvokeInst &I);
+  void visitUnwind(const UnwindInst &I);
 
-  void visitBinary(User &I, unsigned OpCode);
-  void visitShift(User &I, unsigned Opcode);
-  void visitAdd(User &I)  { visitBinary(I, ISD::ADD); }
-  void visitFAdd(User &I) { visitBinary(I, ISD::FADD); }
-  void visitSub(User &I)  { visitBinary(I, ISD::SUB); }
-  void visitFSub(User &I);
-  void visitMul(User &I)  { visitBinary(I, ISD::MUL); }
-  void visitFMul(User &I) { visitBinary(I, ISD::FMUL); }
-  void visitURem(User &I) { visitBinary(I, ISD::UREM); }
-  void visitSRem(User &I) { visitBinary(I, ISD::SREM); }
-  void visitFRem(User &I) { visitBinary(I, ISD::FREM); }
-  void visitUDiv(User &I) { visitBinary(I, ISD::UDIV); }
-  void visitSDiv(User &I) { visitBinary(I, ISD::SDIV); }
-  void visitFDiv(User &I) { visitBinary(I, ISD::FDIV); }
-  void visitAnd (User &I) { visitBinary(I, ISD::AND); }
-  void visitOr  (User &I) { visitBinary(I, ISD::OR); }
-  void visitXor (User &I) { visitBinary(I, ISD::XOR); }
-  void visitShl (User &I) { visitShift(I, ISD::SHL); }
-  void visitLShr(User &I) { visitShift(I, ISD::SRL); }
-  void visitAShr(User &I) { visitShift(I, ISD::SRA); }
-  void visitICmp(User &I);
-  void visitFCmp(User &I);
+  void visitBinary(const User &I, unsigned OpCode);
+  void visitShift(const User &I, unsigned Opcode);
+  void visitAdd(const User &I)  { visitBinary(I, ISD::ADD); }
+  void visitFAdd(const User &I) { visitBinary(I, ISD::FADD); }
+  void visitSub(const User &I)  { visitBinary(I, ISD::SUB); }
+  void visitFSub(const User &I);
+  void visitMul(const User &I)  { visitBinary(I, ISD::MUL); }
+  void visitFMul(const User &I) { visitBinary(I, ISD::FMUL); }
+  void visitURem(const User &I) { visitBinary(I, ISD::UREM); }
+  void visitSRem(const User &I) { visitBinary(I, ISD::SREM); }
+  void visitFRem(const User &I) { visitBinary(I, ISD::FREM); }
+  void visitUDiv(const User &I) { visitBinary(I, ISD::UDIV); }
+  void visitSDiv(const User &I) { visitBinary(I, ISD::SDIV); }
+  void visitFDiv(const User &I) { visitBinary(I, ISD::FDIV); }
+  void visitAnd (const User &I) { visitBinary(I, ISD::AND); }
+  void visitOr  (const User &I) { visitBinary(I, ISD::OR); }
+  void visitXor (const User &I) { visitBinary(I, ISD::XOR); }
+  void visitShl (const User &I) { visitShift(I, ISD::SHL); }
+  void visitLShr(const User &I) { visitShift(I, ISD::SRL); }
+  void visitAShr(const User &I) { visitShift(I, ISD::SRA); }
+  void visitICmp(const User &I);
+  void visitFCmp(const User &I);
   // Visit the conversion instructions
-  void visitTrunc(User &I);
-  void visitZExt(User &I);
-  void visitSExt(User &I);
-  void visitFPTrunc(User &I);
-  void visitFPExt(User &I);
-  void visitFPToUI(User &I);
-  void visitFPToSI(User &I);
-  void visitUIToFP(User &I);
-  void visitSIToFP(User &I);
-  void visitPtrToInt(User &I);
-  void visitIntToPtr(User &I);
-  void visitBitCast(User &I);
+  void visitTrunc(const User &I);
+  void visitZExt(const User &I);
+  void visitSExt(const User &I);
+  void visitFPTrunc(const User &I);
+  void visitFPExt(const User &I);
+  void visitFPToUI(const User &I);
+  void visitFPToSI(const User &I);
+  void visitUIToFP(const User &I);
+  void visitSIToFP(const User &I);
+  void visitPtrToInt(const User &I);
+  void visitIntToPtr(const User &I);
+  void visitBitCast(const User &I);
 
-  void visitExtractElement(User &I);
-  void visitInsertElement(User &I);
-  void visitShuffleVector(User &I);
+  void visitExtractElement(const User &I);
+  void visitInsertElement(const User &I);
+  void visitShuffleVector(const User &I);
 
-  void visitExtractValue(ExtractValueInst &I);
-  void visitInsertValue(InsertValueInst &I);
+  void visitExtractValue(const ExtractValueInst &I);
+  void visitInsertValue(const InsertValueInst &I);
 
-  void visitGetElementPtr(User &I);
-  void visitSelect(User &I);
+  void visitGetElementPtr(const User &I);
+  void visitSelect(const User &I);
 
-  void visitAlloca(AllocaInst &I);
-  void visitLoad(LoadInst &I);
-  void visitStore(StoreInst &I);
-  void visitPHI(PHINode &I) { } // PHI nodes are handled specially.
-  void visitCall(CallInst &I);
-  bool visitMemCmpCall(CallInst &I);
+  void visitAlloca(const AllocaInst &I);
+  void visitLoad(const LoadInst &I);
+  void visitStore(const StoreInst &I);
+  void visitPHI(const PHINode &I);
+  void visitCall(const CallInst &I);
+  bool visitMemCmpCall(const CallInst &I);
   
-  void visitInlineAsm(CallSite CS);
-  const char *visitIntrinsicCall(CallInst &I, unsigned Intrinsic);
-  void visitTargetIntrinsic(CallInst &I, unsigned Intrinsic);
+  void visitInlineAsm(ImmutableCallSite CS);
+  const char *visitIntrinsicCall(const CallInst &I, unsigned Intrinsic);
+  void visitTargetIntrinsic(const CallInst &I, unsigned Intrinsic);
 
-  void visitPow(CallInst &I);
-  void visitExp2(CallInst &I);
-  void visitExp(CallInst &I);
-  void visitLog(CallInst &I);
-  void visitLog2(CallInst &I);
-  void visitLog10(CallInst &I);
+  void visitPow(const CallInst &I);
+  void visitExp2(const CallInst &I);
+  void visitExp(const CallInst &I);
+  void visitLog(const CallInst &I);
+  void visitLog2(const CallInst &I);
+  void visitLog10(const CallInst &I);
 
-  void visitVAStart(CallInst &I);
-  void visitVAArg(VAArgInst &I);
-  void visitVAEnd(CallInst &I);
-  void visitVACopy(CallInst &I);
+  void visitVAStart(const CallInst &I);
+  void visitVAArg(const VAArgInst &I);
+  void visitVAEnd(const CallInst &I);
+  void visitVACopy(const CallInst &I);
 
-  void visitUserOp1(Instruction &I) {
+  void visitUserOp1(const Instruction &I) {
     llvm_unreachable("UserOp1 should not exist at instruction selection time!");
   }
-  void visitUserOp2(Instruction &I) {
+  void visitUserOp2(const Instruction &I) {
     llvm_unreachable("UserOp2 should not exist at instruction selection time!");
   }
   
-  const char *implVisitBinaryAtomic(CallInst& I, ISD::NodeType Op);
-  const char *implVisitAluOverflow(CallInst &I, ISD::NodeType Op);
+  const char *implVisitBinaryAtomic(const CallInst& I, ISD::NodeType Op);
+  const char *implVisitAluOverflow(const CallInst &I, ISD::NodeType Op);
+
+  void HandlePHINodesInSuccessorBlocks(const BasicBlock *LLVMBB);
 };
 
 } // end namespace llvm
diff --git a/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp b/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp
index 05f9f1f..2cdd1cc 100644
--- a/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp
+++ b/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp
@@ -19,10 +19,7 @@
 #include "llvm/Analysis/AliasAnalysis.h"
 #include "llvm/Analysis/DebugInfo.h"
 #include "llvm/Constants.h"
-#include "llvm/CallingConv.h"
-#include "llvm/DerivedTypes.h"
 #include "llvm/Function.h"
-#include "llvm/GlobalVariable.h"
 #include "llvm/InlineAsm.h"
 #include "llvm/Instructions.h"
 #include "llvm/Intrinsics.h"
@@ -32,19 +29,13 @@
 #include "llvm/CodeGen/GCStrategy.h"
 #include "llvm/CodeGen/GCMetadata.h"
 #include "llvm/CodeGen/MachineFunction.h"
-#include "llvm/CodeGen/MachineFunctionAnalysis.h"
-#include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
-#include "llvm/CodeGen/MachineJumpTableInfo.h"
 #include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/ScheduleHazardRecognizer.h"
 #include "llvm/CodeGen/SchedulerRegistry.h"
 #include "llvm/CodeGen/SelectionDAG.h"
-#include "llvm/CodeGen/DwarfWriter.h"
 #include "llvm/Target/TargetRegisterInfo.h"
-#include "llvm/Target/TargetData.h"
-#include "llvm/Target/TargetFrameInfo.h"
 #include "llvm/Target/TargetIntrinsicInfo.h"
 #include "llvm/Target/TargetInstrInfo.h"
 #include "llvm/Target/TargetLowering.h"
@@ -53,7 +44,6 @@
 #include "llvm/Support/Compiler.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
-#include "llvm/Support/MathExtras.h"
 #include "llvm/Support/Timer.h"
 #include "llvm/Support/raw_ostream.h"
 #include "llvm/ADT/Statistic.h"
@@ -61,6 +51,7 @@
 using namespace llvm;
 
 STATISTIC(NumFastIselFailures, "Number of instructions fast isel failed on");
+STATISTIC(NumDAGIselRetries,"Number of times dag isel has to try another path");
 
 static cl::opt<bool>
 EnableFastISelVerbose("fast-isel-verbose", cl::Hidden,
@@ -69,10 +60,6 @@
 static cl::opt<bool>
 EnableFastISelAbort("fast-isel-abort", cl::Hidden,
           cl::desc("Enable abort calls when \"fast\" instruction fails"));
-static cl::opt<bool>
-SchedLiveInCopies("schedule-livein-copies", cl::Hidden,
-                  cl::desc("Schedule copies of livein registers"),
-                  cl::init(false));
 
 #ifndef NDEBUG
 static cl::opt<bool>
@@ -173,115 +160,15 @@
   return 0;
 }
 
-/// EmitLiveInCopy - Emit a copy for a live in physical register. If the
-/// physical register has only a single copy use, then coalesced the copy
-/// if possible.
-static void EmitLiveInCopy(MachineBasicBlock *MBB,
-                           MachineBasicBlock::iterator &InsertPos,
-                           unsigned VirtReg, unsigned PhysReg,
-                           const TargetRegisterClass *RC,
-                           DenseMap<MachineInstr*, unsigned> &CopyRegMap,
-                           const MachineRegisterInfo &MRI,
-                           const TargetRegisterInfo &TRI,
-                           const TargetInstrInfo &TII) {
-  unsigned NumUses = 0;
-  MachineInstr *UseMI = NULL;
-  for (MachineRegisterInfo::use_iterator UI = MRI.use_begin(VirtReg),
-         UE = MRI.use_end(); UI != UE; ++UI) {
-    UseMI = &*UI;
-    if (++NumUses > 1)
-      break;
-  }
-
-  // If the number of uses is not one, or the use is not a move instruction,
-  // don't coalesce. Also, only coalesce away a virtual register to virtual
-  // register copy.
-  bool Coalesced = false;
-  unsigned SrcReg, DstReg, SrcSubReg, DstSubReg;
-  if (NumUses == 1 &&
-      TII.isMoveInstr(*UseMI, SrcReg, DstReg, SrcSubReg, DstSubReg) &&
-      TargetRegisterInfo::isVirtualRegister(DstReg)) {
-    VirtReg = DstReg;
-    Coalesced = true;
-  }
-
-  // Now find an ideal location to insert the copy.
-  MachineBasicBlock::iterator Pos = InsertPos;
-  while (Pos != MBB->begin()) {
-    MachineInstr *PrevMI = prior(Pos);
-    DenseMap<MachineInstr*, unsigned>::iterator RI = CopyRegMap.find(PrevMI);
-    // copyRegToReg might emit multiple instructions to do a copy.
-    unsigned CopyDstReg = (RI == CopyRegMap.end()) ? 0 : RI->second;
-    if (CopyDstReg && !TRI.regsOverlap(CopyDstReg, PhysReg))
-      // This is what the BB looks like right now:
-      // r1024 = mov r0
-      // ...
-      // r1    = mov r1024
-      //
-      // We want to insert "r1025 = mov r1". Inserting this copy below the
-      // move to r1024 makes it impossible for that move to be coalesced.
-      //
-      // r1025 = mov r1
-      // r1024 = mov r0
-      // ...
-      // r1    = mov 1024
-      // r2    = mov 1025
-      break; // Woot! Found a good location.
-    --Pos;
-  }
-
-  bool Emitted = TII.copyRegToReg(*MBB, Pos, VirtReg, PhysReg, RC, RC);
-  assert(Emitted && "Unable to issue a live-in copy instruction!\n");
-  (void) Emitted;
-
-  CopyRegMap.insert(std::make_pair(prior(Pos), VirtReg));
-  if (Coalesced) {
-    if (&*InsertPos == UseMI) ++InsertPos;
-    MBB->erase(UseMI);
-  }
-}
-
-/// EmitLiveInCopies - If this is the first basic block in the function,
-/// and if it has live ins that need to be copied into vregs, emit the
-/// copies into the block.
-static void EmitLiveInCopies(MachineBasicBlock *EntryMBB,
-                             const MachineRegisterInfo &MRI,
-                             const TargetRegisterInfo &TRI,
-                             const TargetInstrInfo &TII) {
-  if (SchedLiveInCopies) {
-    // Emit the copies at a heuristically-determined location in the block.
-    DenseMap<MachineInstr*, unsigned> CopyRegMap;
-    MachineBasicBlock::iterator InsertPos = EntryMBB->begin();
-    for (MachineRegisterInfo::livein_iterator LI = MRI.livein_begin(),
-           E = MRI.livein_end(); LI != E; ++LI)
-      if (LI->second) {
-        const TargetRegisterClass *RC = MRI.getRegClass(LI->second);
-        EmitLiveInCopy(EntryMBB, InsertPos, LI->second, LI->first,
-                       RC, CopyRegMap, MRI, TRI, TII);
-      }
-  } else {
-    // Emit the copies into the top of the block.
-    for (MachineRegisterInfo::livein_iterator LI = MRI.livein_begin(),
-           E = MRI.livein_end(); LI != E; ++LI)
-      if (LI->second) {
-        const TargetRegisterClass *RC = MRI.getRegClass(LI->second);
-        bool Emitted = TII.copyRegToReg(*EntryMBB, EntryMBB->begin(),
-                                        LI->second, LI->first, RC, RC);
-        assert(Emitted && "Unable to issue a live-in copy instruction!\n");
-        (void) Emitted;
-      }
-  }
-}
-
 //===----------------------------------------------------------------------===//
 // SelectionDAGISel code
 //===----------------------------------------------------------------------===//
 
-SelectionDAGISel::SelectionDAGISel(TargetMachine &tm, CodeGenOpt::Level OL) :
+SelectionDAGISel::SelectionDAGISel(const TargetMachine &tm, CodeGenOpt::Level OL) :
   MachineFunctionPass(&ID), TM(tm), TLI(*tm.getTargetLowering()),
   FuncInfo(new FunctionLoweringInfo(TLI)),
-  CurDAG(new SelectionDAG(TLI, *FuncInfo)),
-  SDB(new SelectionDAGBuilder(*CurDAG, TLI, *FuncInfo, OL)),
+  CurDAG(new SelectionDAG(tm, *FuncInfo)),
+  SDB(new SelectionDAGBuilder(*CurDAG, *FuncInfo, OL)),
   GFI(),
   OptLevel(OL),
   DAGSize(0)
@@ -293,152 +180,70 @@
   delete FuncInfo;
 }
 
-unsigned SelectionDAGISel::MakeReg(EVT VT) {
-  return RegInfo->createVirtualRegister(TLI.getRegClassFor(VT));
-}
-
 void SelectionDAGISel::getAnalysisUsage(AnalysisUsage &AU) const {
   AU.addRequired<AliasAnalysis>();
   AU.addPreserved<AliasAnalysis>();
   AU.addRequired<GCModuleInfo>();
   AU.addPreserved<GCModuleInfo>();
-  AU.addRequired<DwarfWriter>();
-  AU.addPreserved<DwarfWriter>();
   MachineFunctionPass::getAnalysisUsage(AU);
 }
 
 bool SelectionDAGISel::runOnMachineFunction(MachineFunction &mf) {
-  Function &Fn = *mf.getFunction();
-
   // Do some sanity-checking on the command-line options.
   assert((!EnableFastISelVerbose || EnableFastISel) &&
          "-fast-isel-verbose requires -fast-isel");
   assert((!EnableFastISelAbort || EnableFastISel) &&
          "-fast-isel-abort requires -fast-isel");
 
-  // Get alias analysis for load/store combining.
-  AA = &getAnalysis<AliasAnalysis>();
-
-  MF = &mf;
+  const Function &Fn = *mf.getFunction();
   const TargetInstrInfo &TII = *TM.getInstrInfo();
   const TargetRegisterInfo &TRI = *TM.getRegisterInfo();
 
-  if (Fn.hasGC())
-    GFI = &getAnalysis<GCModuleInfo>().getFunctionInfo(Fn);
-  else
-    GFI = 0;
+  MF = &mf;
   RegInfo = &MF->getRegInfo();
+  AA = &getAnalysis<AliasAnalysis>();
+  GFI = Fn.hasGC() ? &getAnalysis<GCModuleInfo>().getFunctionInfo(Fn) : 0;
+
   DEBUG(dbgs() << "\n\n\n=== " << Fn.getName() << "\n");
 
-  MachineModuleInfo *MMI = getAnalysisIfAvailable<MachineModuleInfo>();
-  DwarfWriter *DW = getAnalysisIfAvailable<DwarfWriter>();
-  CurDAG->init(*MF, MMI, DW);
+  CurDAG->init(*MF);
   FuncInfo->set(Fn, *MF, EnableFastISel);
   SDB->init(GFI, *AA);
 
-  for (Function::iterator I = Fn.begin(), E = Fn.end(); I != E; ++I)
-    if (InvokeInst *Invoke = dyn_cast<InvokeInst>(I->getTerminator()))
-      // Mark landing pad.
-      FuncInfo->MBBMap[Invoke->getSuccessor(1)]->setIsLandingPad();
+  SelectAllBasicBlocks(Fn);
 
-  SelectAllBasicBlocks(Fn, *MF, MMI, DW, TII);
+  // Release function-specific state. SDB and CurDAG are already cleared
+  // at this point.
+  FuncInfo->clear();
 
   // If the first basic block in the function has live ins that need to be
   // copied into vregs, emit the copies into the top of the block before
   // emitting the code for the block.
-  EmitLiveInCopies(MF->begin(), *RegInfo, TRI, TII);
-
-  // Add function live-ins to entry block live-in set.
-  for (MachineRegisterInfo::livein_iterator I = RegInfo->livein_begin(),
-         E = RegInfo->livein_end(); I != E; ++I)
-    MF->begin()->addLiveIn(I->first);
-
-#ifndef NDEBUG
-  assert(FuncInfo->CatchInfoFound.size() == FuncInfo->CatchInfoLost.size() &&
-         "Not all catch info was assigned to a landing pad!");
-#endif
-
-  FuncInfo->clear();
+  RegInfo->EmitLiveInCopies(MF->begin(), TRI, TII);
 
   return true;
 }
 
-/// SetDebugLoc - Update MF's and SDB's DebugLocs if debug information is
-/// attached with this instruction.
-static void SetDebugLoc(unsigned MDDbgKind, Instruction *I,
-                        SelectionDAGBuilder *SDB,
-                        FastISel *FastIS, MachineFunction *MF) {
-  if (isa<DbgInfoIntrinsic>(I)) return;
-  
-  if (MDNode *Dbg = I->getMetadata(MDDbgKind)) {
-    DILocation DILoc(Dbg);
-    DebugLoc Loc = ExtractDebugLocation(DILoc, MF->getDebugLocInfo());
-
-    SDB->setCurDebugLoc(Loc);
-
-    if (FastIS)
-      FastIS->setCurDebugLoc(Loc);
-
-    // If the function doesn't have a default debug location yet, set
-    // it. This is kind of a hack.
-    if (MF->getDefaultDebugLoc().isUnknown())
-      MF->setDefaultDebugLoc(Loc);
-  }
-}
-
-/// ResetDebugLoc - Set MF's and SDB's DebugLocs to Unknown.
-static void ResetDebugLoc(SelectionDAGBuilder *SDB, FastISel *FastIS) {
-  SDB->setCurDebugLoc(DebugLoc::getUnknownLoc());
-  if (FastIS)
-    FastIS->setCurDebugLoc(DebugLoc::getUnknownLoc());
-}
-
-void SelectionDAGISel::SelectBasicBlock(BasicBlock *LLVMBB,
-                                        BasicBlock::iterator Begin,
-                                        BasicBlock::iterator End,
-                                        bool &HadTailCall) {
-  SDB->setCurrentBasicBlock(BB);
-  unsigned MDDbgKind = LLVMBB->getContext().getMDKindID("dbg");
-
+MachineBasicBlock *
+SelectionDAGISel::SelectBasicBlock(MachineBasicBlock *BB,
+                                   const BasicBlock *LLVMBB,
+                                   BasicBlock::const_iterator Begin,
+                                   BasicBlock::const_iterator End,
+                                   bool &HadTailCall) {
   // Lower all of the non-terminator instructions. If a call is emitted
-  // as a tail call, cease emitting nodes for this block.
-  for (BasicBlock::iterator I = Begin; I != End && !SDB->HasTailCall; ++I) {
-    SetDebugLoc(MDDbgKind, I, SDB, 0, MF);
-
-    if (!isa<TerminatorInst>(I)) {
-      SDB->visit(*I);
-
-      // Set the current debug location back to "unknown" so that it doesn't
-      // spuriously apply to subsequent instructions.
-      ResetDebugLoc(SDB, 0);
-    }
-  }
-
-  if (!SDB->HasTailCall) {
-    // Ensure that all instructions which are used outside of their defining
-    // blocks are available as virtual registers.  Invoke is handled elsewhere.
-    for (BasicBlock::iterator I = Begin; I != End; ++I)
-      if (!isa<PHINode>(I) && !isa<InvokeInst>(I))
-        SDB->CopyToExportRegsIfNeeded(I);
-
-    // Handle PHI nodes in successor blocks.
-    if (End == LLVMBB->end()) {
-      HandlePHINodesInSuccessorBlocks(LLVMBB);
-
-      // Lower the terminator after the copies are emitted.
-      SetDebugLoc(MDDbgKind, LLVMBB->getTerminator(), SDB, 0, MF);
-      SDB->visit(*LLVMBB->getTerminator());
-      ResetDebugLoc(SDB, 0);
-    }
-  }
+  // as a tail call, cease emitting nodes for this block. Terminators
+  // are handled below.
+  for (BasicBlock::const_iterator I = Begin; I != End && !SDB->HasTailCall; ++I)
+    SDB->visit(*I);
 
   // Make sure the root of the DAG is up-to-date.
   CurDAG->setRoot(SDB->getControlRoot());
 
   // Final step, emit the lowered DAG as machine code.
-  CodeGenAndEmitDAG();
+  BB = CodeGenAndEmitDAG(BB);
   HadTailCall = SDB->HasTailCall;
   SDB->clear();
+  return BB;
 }
 
 namespace {
@@ -446,12 +251,25 @@
 /// nodes from the worklist.
 class SDOPsWorkListRemover : public SelectionDAG::DAGUpdateListener {
   SmallVector<SDNode*, 128> &Worklist;
+  SmallPtrSet<SDNode*, 128> &InWorklist;
 public:
-  SDOPsWorkListRemover(SmallVector<SDNode*, 128> &wl) : Worklist(wl) {}
+  SDOPsWorkListRemover(SmallVector<SDNode*, 128> &wl,
+                       SmallPtrSet<SDNode*, 128> &inwl)
+    : Worklist(wl), InWorklist(inwl) {}
 
+  void RemoveFromWorklist(SDNode *N) {
+    if (!InWorklist.erase(N)) return;
+    
+    SmallVector<SDNode*, 128>::iterator I =
+    std::find(Worklist.begin(), Worklist.end(), N);
+    assert(I != Worklist.end() && "Not in worklist");
+    
+    *I = Worklist.back();
+    Worklist.pop_back();
+  }
+  
   virtual void NodeDeleted(SDNode *N, SDNode *E) {
-    Worklist.erase(std::remove(Worklist.begin(), Worklist.end(), N),
-                   Worklist.end());
+    RemoveFromWorklist(N);
   }
 
   virtual void NodeUpdated(SDNode *N) {
@@ -480,70 +298,79 @@
 /// x+y to (VT)((SmallVT)x+(SmallVT)y) if the casts are free.
 void SelectionDAGISel::ShrinkDemandedOps() {
   SmallVector<SDNode*, 128> Worklist;
+  SmallPtrSet<SDNode*, 128> InWorklist;
 
   // Add all the dag nodes to the worklist.
   Worklist.reserve(CurDAG->allnodes_size());
   for (SelectionDAG::allnodes_iterator I = CurDAG->allnodes_begin(),
-       E = CurDAG->allnodes_end(); I != E; ++I)
+       E = CurDAG->allnodes_end(); I != E; ++I) {
     Worklist.push_back(I);
+    InWorklist.insert(I);
+  }
 
-  APInt Mask;
-  APInt KnownZero;
-  APInt KnownOne;
-
-  TargetLowering::TargetLoweringOpt TLO(*CurDAG, true);
+  TargetLowering::TargetLoweringOpt TLO(*CurDAG, true, true, true);
   while (!Worklist.empty()) {
     SDNode *N = Worklist.pop_back_val();
+    InWorklist.erase(N);
 
     if (N->use_empty() && N != CurDAG->getRoot().getNode()) {
+      // Deleting this node may make its operands dead, add them to the worklist
+      // if they aren't already there.
+      for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i)
+        if (InWorklist.insert(N->getOperand(i).getNode()))
+          Worklist.push_back(N->getOperand(i).getNode());
+      
       CurDAG->DeleteNode(N);
       continue;
     }
 
     // Run ShrinkDemandedOp on scalar binary operations.
-    if (N->getNumValues() == 1 &&
-        N->getValueType(0).isSimple() && N->getValueType(0).isInteger()) {
-      unsigned BitWidth = N->getValueType(0).getScalarType().getSizeInBits();
-      APInt Demanded = APInt::getAllOnesValue(BitWidth);
-      APInt KnownZero, KnownOne;
-      if (TLI.SimplifyDemandedBits(SDValue(N, 0), Demanded,
-                                   KnownZero, KnownOne, TLO) ||
-          (N->getOpcode() == ISD::TRUNCATE &&
-           TrivialTruncElim(SDValue(N, 0), TLO))) {
-        // Revisit the node.
-        Worklist.erase(std::remove(Worklist.begin(), Worklist.end(), N),
-                       Worklist.end());
-        Worklist.push_back(N);
+    if (N->getNumValues() != 1 ||
+        !N->getValueType(0).isSimple() || !N->getValueType(0).isInteger())
+      continue;
+    
+    unsigned BitWidth = N->getValueType(0).getScalarType().getSizeInBits();
+    APInt Demanded = APInt::getAllOnesValue(BitWidth);
+    APInt KnownZero, KnownOne;
+    if (!TLI.SimplifyDemandedBits(SDValue(N, 0), Demanded,
+                                  KnownZero, KnownOne, TLO) &&
+        (N->getOpcode() != ISD::TRUNCATE ||
+         !TrivialTruncElim(SDValue(N, 0), TLO)))
+      continue;
+    
+    // Revisit the node.
+    assert(!InWorklist.count(N) && "Already in worklist");
+    Worklist.push_back(N);
+    InWorklist.insert(N);
 
-        // Replace the old value with the new one.
-        DEBUG(errs() << "\nReplacing "; 
-              TLO.Old.getNode()->dump(CurDAG);
-              errs() << "\nWith: ";
-              TLO.New.getNode()->dump(CurDAG);
-              errs() << '\n');
+    // Replace the old value with the new one.
+    DEBUG(errs() << "\nShrinkDemandedOps replacing "; 
+          TLO.Old.getNode()->dump(CurDAG);
+          errs() << "\nWith: ";
+          TLO.New.getNode()->dump(CurDAG);
+          errs() << '\n');
 
-        Worklist.push_back(TLO.New.getNode());
+    if (InWorklist.insert(TLO.New.getNode()))
+      Worklist.push_back(TLO.New.getNode());
 
-        SDOPsWorkListRemover DeadNodes(Worklist);
-        CurDAG->ReplaceAllUsesOfValueWith(TLO.Old, TLO.New, &DeadNodes);
+    SDOPsWorkListRemover DeadNodes(Worklist, InWorklist);
+    CurDAG->ReplaceAllUsesOfValueWith(TLO.Old, TLO.New, &DeadNodes);
 
-        if (TLO.Old.getNode()->use_empty()) {
-          for (unsigned i = 0, e = TLO.Old.getNode()->getNumOperands();
-               i != e; ++i) {
-            SDNode *OpNode = TLO.Old.getNode()->getOperand(i).getNode(); 
-            if (OpNode->hasOneUse()) {
-              Worklist.erase(std::remove(Worklist.begin(), Worklist.end(),
-                                         OpNode), Worklist.end());
-              Worklist.push_back(OpNode);
-            }
-          }
-
-          Worklist.erase(std::remove(Worklist.begin(), Worklist.end(),
-                                     TLO.Old.getNode()), Worklist.end());
-          CurDAG->DeleteNode(TLO.Old.getNode());
-        }
+    if (!TLO.Old.getNode()->use_empty()) continue;
+        
+    for (unsigned i = 0, e = TLO.Old.getNode()->getNumOperands();
+         i != e; ++i) {
+      SDNode *OpNode = TLO.Old.getNode()->getOperand(i).getNode(); 
+      if (OpNode->hasOneUse()) {
+        // Add OpNode to the end of the list to revisit.
+        DeadNodes.RemoveFromWorklist(OpNode);
+        Worklist.push_back(OpNode);
+        InWorklist.insert(OpNode);
       }
     }
+
+    DeadNodes.RemoveFromWorklist(TLO.Old.getNode());
+    CurDAG->DeleteNode(TLO.Old.getNode());
   }
 }
 
@@ -601,7 +428,7 @@
   } while (!Worklist.empty());
 }
 
-void SelectionDAGISel::CodeGenAndEmitDAG() {
+MachineBasicBlock *SelectionDAGISel::CodeGenAndEmitDAG(MachineBasicBlock *BB) {
   std::string GroupName;
   if (TimePassesIsEnabled)
     GroupName = "Instruction Selection and Scheduling";
@@ -715,13 +542,13 @@
   DEBUG(dbgs() << "Optimized legalized selection DAG:\n");
   DEBUG(CurDAG->dump());
 
-  if (ViewISelDAGs) CurDAG->viewGraph("isel input for " + BlockName);
-
   if (OptLevel != CodeGenOpt::None) {
     ShrinkDemandedOps();
     ComputeLiveOutVRegInfo();
   }
 
+  if (ViewISelDAGs) CurDAG->viewGraph("isel input for " + BlockName);
+
   // Third, instruction select all of the operations to machine code, adding the
   // code to the MachineBasicBlock.
   if (TimePassesIsEnabled) {
@@ -764,8 +591,7 @@
     delete Scheduler;
   }
 
-  DEBUG(dbgs() << "Selected machine code:\n");
-  DEBUG(BB->dump());
+  return BB;
 }
 
 void SelectionDAGISel::DoInstructionSelection() {
@@ -822,39 +648,72 @@
   DEBUG(errs() << "===== Instruction selection ends:\n");
 
   PostprocessISelDAG();
-  
-  // FIXME: This shouldn't be needed, remove it.
-  CurDAG->RemoveDeadNodes();
 }
 
+/// PrepareEHLandingPad - Emit an EH_LABEL, set up live-in registers, and
+/// do other setup for EH landing-pad blocks.
+void SelectionDAGISel::PrepareEHLandingPad(MachineBasicBlock *BB) {
+  // Add a label to mark the beginning of the landing pad.  Deletion of the
+  // landing pad can thus be detected via the MachineModuleInfo.
+  MCSymbol *Label = MF->getMMI().addLandingPad(BB);
 
-void SelectionDAGISel::SelectAllBasicBlocks(Function &Fn,
-                                            MachineFunction &MF,
-                                            MachineModuleInfo *MMI,
-                                            DwarfWriter *DW,
-                                            const TargetInstrInfo &TII) {
+  const TargetInstrDesc &II = TM.getInstrInfo()->get(TargetOpcode::EH_LABEL);
+  BuildMI(BB, SDB->getCurDebugLoc(), II).addSym(Label);
+
+  // Mark exception register as live in.
+  unsigned Reg = TLI.getExceptionAddressRegister();
+  if (Reg) BB->addLiveIn(Reg);
+
+  // Mark exception selector register as live in.
+  Reg = TLI.getExceptionSelectorRegister();
+  if (Reg) BB->addLiveIn(Reg);
+
+  // FIXME: Hack around an exception handling flaw (PR1508): the personality
+  // function and list of typeids logically belong to the invoke (or, if you
+  // like, the basic block containing the invoke), and need to be associated
+  // with it in the dwarf exception handling tables.  Currently however the
+  // information is provided by an intrinsic (eh.selector) that can be moved
+  // to unexpected places by the optimizers: if the unwind edge is critical,
+  // then breaking it can result in the intrinsics being in the successor of
+  // the landing pad, not the landing pad itself.  This results
+  // in exceptions not being caught because no typeids are associated with
+  // the invoke.  This may not be the only way things can go wrong, but it
+  // is the only way we try to work around for the moment.
+  const BasicBlock *LLVMBB = BB->getBasicBlock();
+  const BranchInst *Br = dyn_cast<BranchInst>(LLVMBB->getTerminator());
+
+  if (Br && Br->isUnconditional()) { // Critical edge?
+    BasicBlock::const_iterator I, E;
+    for (I = LLVMBB->begin(), E = --LLVMBB->end(); I != E; ++I)
+      if (isa<EHSelectorInst>(I))
+        break;
+
+    if (I == E)
+      // No catch info found - try to extract some from the successor.
+      CopyCatchInfo(Br->getSuccessor(0), LLVMBB, &MF->getMMI(), *FuncInfo);
+  }
+}
+
+void SelectionDAGISel::SelectAllBasicBlocks(const Function &Fn) {
   // Initialize the Fast-ISel state, if needed.
   FastISel *FastIS = 0;
   if (EnableFastISel)
-    FastIS = TLI.createFastISel(MF, MMI, DW,
-                                FuncInfo->ValueMap,
-                                FuncInfo->MBBMap,
-                                FuncInfo->StaticAllocaMap
+    FastIS = TLI.createFastISel(*MF, FuncInfo->ValueMap, FuncInfo->MBBMap,
+                                FuncInfo->StaticAllocaMap,
+                                FuncInfo->PHINodesToUpdate
 #ifndef NDEBUG
                                 , FuncInfo->CatchInfoLost
 #endif
                                 );
 
-  unsigned MDDbgKind = Fn.getContext().getMDKindID("dbg");
-
   // Iterate over all basic blocks in the function.
-  for (Function::iterator I = Fn.begin(), E = Fn.end(); I != E; ++I) {
-    BasicBlock *LLVMBB = &*I;
-    BB = FuncInfo->MBBMap[LLVMBB];
+  for (Function::const_iterator I = Fn.begin(), E = Fn.end(); I != E; ++I) {
+    const BasicBlock *LLVMBB = &*I;
+    MachineBasicBlock *BB = FuncInfo->MBBMap[LLVMBB];
 
-    BasicBlock::iterator const Begin = LLVMBB->begin();
-    BasicBlock::iterator const End = LLVMBB->end();
-    BasicBlock::iterator BI = Begin;
+    BasicBlock::const_iterator const Begin = LLVMBB->getFirstNonPHI();
+    BasicBlock::const_iterator const End = LLVMBB->end();
+    BasicBlock::const_iterator BI = Begin;
 
     // Lower any arguments needed in this block if this is the entry block.
     bool SuppressFastISel = false;
@@ -865,7 +724,7 @@
       // fast-isel in the entry block.
       if (FastIS) {
         unsigned j = 1;
-        for (Function::arg_iterator I = Fn.arg_begin(), E = Fn.arg_end();
+        for (Function::const_arg_iterator I = Fn.arg_begin(), E = Fn.arg_end();
              I != E; ++I, ++j)
           if (Fn.paramHasAttr(j, Attribute::ByVal)) {
             if (EnableFastISelVerbose || EnableFastISelAbort)
@@ -876,85 +735,25 @@
       }
     }
 
-    if (MMI && BB->isLandingPad()) {
-      // Add a label to mark the beginning of the landing pad.  Deletion of the
-      // landing pad can thus be detected via the MachineModuleInfo.
-      unsigned LabelID = MMI->addLandingPad(BB);
-
-      const TargetInstrDesc &II = TII.get(TargetOpcode::EH_LABEL);
-      BuildMI(BB, SDB->getCurDebugLoc(), II).addImm(LabelID);
-
-      // Mark exception register as live in.
-      unsigned Reg = TLI.getExceptionAddressRegister();
-      if (Reg) BB->addLiveIn(Reg);
-
-      // Mark exception selector register as live in.
-      Reg = TLI.getExceptionSelectorRegister();
-      if (Reg) BB->addLiveIn(Reg);
-
-      // FIXME: Hack around an exception handling flaw (PR1508): the personality
-      // function and list of typeids logically belong to the invoke (or, if you
-      // like, the basic block containing the invoke), and need to be associated
-      // with it in the dwarf exception handling tables.  Currently however the
-      // information is provided by an intrinsic (eh.selector) that can be moved
-      // to unexpected places by the optimizers: if the unwind edge is critical,
-      // then breaking it can result in the intrinsics being in the successor of
-      // the landing pad, not the landing pad itself.  This results
-      // in exceptions not being caught because no typeids are associated with
-      // the invoke.  This may not be the only way things can go wrong, but it
-      // is the only way we try to work around for the moment.
-      BranchInst *Br = dyn_cast<BranchInst>(LLVMBB->getTerminator());
-
-      if (Br && Br->isUnconditional()) { // Critical edge?
-        BasicBlock::iterator I, E;
-        for (I = LLVMBB->begin(), E = --LLVMBB->end(); I != E; ++I)
-          if (isa<EHSelectorInst>(I))
-            break;
-
-        if (I == E)
-          // No catch info found - try to extract some from the successor.
-          CopyCatchInfo(Br->getSuccessor(0), LLVMBB, MMI, *FuncInfo);
-      }
-    }
-
+    // Setup an EH landing-pad block.
+    if (BB->isLandingPad())
+      PrepareEHLandingPad(BB);
+    
     // Before doing SelectionDAG ISel, see if FastISel has been requested.
     if (FastIS && !SuppressFastISel) {
       // Emit code for any incoming arguments. This must happen before
       // beginning FastISel on the entry block.
       if (LLVMBB == &Fn.getEntryBlock()) {
         CurDAG->setRoot(SDB->getControlRoot());
-        CodeGenAndEmitDAG();
+        BB = CodeGenAndEmitDAG(BB);
         SDB->clear();
       }
       FastIS->startNewBlock(BB);
       // Do FastISel on as many instructions as possible.
       for (; BI != End; ++BI) {
-        // Just before the terminator instruction, insert instructions to
-        // feed PHI nodes in successor blocks.
-        if (isa<TerminatorInst>(BI))
-          if (!HandlePHINodesInSuccessorBlocksFast(LLVMBB, FastIS)) {
-            ++NumFastIselFailures;
-            ResetDebugLoc(SDB, FastIS);
-            if (EnableFastISelVerbose || EnableFastISelAbort) {
-              dbgs() << "FastISel miss: ";
-              BI->dump();
-            }
-            assert(!EnableFastISelAbort &&
-                   "FastISel didn't handle a PHI in a successor");
-            break;
-          }
-
-        SetDebugLoc(MDDbgKind, BI, SDB, FastIS, &MF);
-
         // Try to select the instruction with FastISel.
-        if (FastIS->SelectInstruction(BI)) {
-          ResetDebugLoc(SDB, FastIS);
+        if (FastIS->SelectInstruction(BI))
           continue;
-        }
-
-        // Clear out the debug location so that it doesn't carry over to
-        // unrelated instructions.
-        ResetDebugLoc(SDB, FastIS);
 
         // Then handle certain instructions as single-LLVM-Instruction blocks.
         if (isa<CallInst>(BI)) {
@@ -964,14 +763,14 @@
             BI->dump();
           }
 
-          if (!BI->getType()->isVoidTy()) {
+          if (!BI->getType()->isVoidTy() && !BI->use_empty()) {
             unsigned &R = FuncInfo->ValueMap[BI];
             if (!R)
               R = FuncInfo->CreateRegForValue(BI);
           }
 
           bool HadTailCall = false;
-          SelectBasicBlock(LLVMBB, BI, llvm::next(BI), HadTailCall);
+          BB = SelectBasicBlock(BB, LLVMBB, BI, llvm::next(BI), HadTailCall);
 
           // If the call was emitted as a tail call, we're done with the block.
           if (HadTailCall) {
@@ -1007,44 +806,44 @@
     // block.
     if (BI != End) {
       bool HadTailCall;
-      SelectBasicBlock(LLVMBB, BI, End, HadTailCall);
+      BB = SelectBasicBlock(BB, LLVMBB, BI, End, HadTailCall);
     }
 
-    FinishBasicBlock();
+    FinishBasicBlock(BB);
+    FuncInfo->PHINodesToUpdate.clear();
   }
 
   delete FastIS;
 }
 
 void
-SelectionDAGISel::FinishBasicBlock() {
+SelectionDAGISel::FinishBasicBlock(MachineBasicBlock *BB) {
 
   DEBUG(dbgs() << "Target-post-processed machine code:\n");
   DEBUG(BB->dump());
 
   DEBUG(dbgs() << "Total amount of phi nodes to update: "
-               << SDB->PHINodesToUpdate.size() << "\n");
-  DEBUG(for (unsigned i = 0, e = SDB->PHINodesToUpdate.size(); i != e; ++i)
+               << FuncInfo->PHINodesToUpdate.size() << "\n");
+  DEBUG(for (unsigned i = 0, e = FuncInfo->PHINodesToUpdate.size(); i != e; ++i)
           dbgs() << "Node " << i << " : ("
-                 << SDB->PHINodesToUpdate[i].first
-                 << ", " << SDB->PHINodesToUpdate[i].second << ")\n");
+                 << FuncInfo->PHINodesToUpdate[i].first
+                 << ", " << FuncInfo->PHINodesToUpdate[i].second << ")\n");
 
   // Next, now that we know what the last MBB the LLVM BB expanded is, update
   // PHI nodes in successors.
   if (SDB->SwitchCases.empty() &&
       SDB->JTCases.empty() &&
       SDB->BitTestCases.empty()) {
-    for (unsigned i = 0, e = SDB->PHINodesToUpdate.size(); i != e; ++i) {
-      MachineInstr *PHI = SDB->PHINodesToUpdate[i].first;
+    for (unsigned i = 0, e = FuncInfo->PHINodesToUpdate.size(); i != e; ++i) {
+      MachineInstr *PHI = FuncInfo->PHINodesToUpdate[i].first;
       assert(PHI->isPHI() &&
              "This is not a machine PHI node that we are updating!");
       if (!BB->isSuccessor(PHI->getParent()))
         continue;
-      PHI->addOperand(MachineOperand::CreateReg(SDB->PHINodesToUpdate[i].second,
-                                                false));
+      PHI->addOperand(
+        MachineOperand::CreateReg(FuncInfo->PHINodesToUpdate[i].second, false));
       PHI->addOperand(MachineOperand::CreateMBB(BB));
     }
-    SDB->PHINodesToUpdate.clear();
     return;
   }
 
@@ -1053,37 +852,38 @@
     if (!SDB->BitTestCases[i].Emitted) {
       // Set the current basic block to the mbb we wish to insert the code into
       BB = SDB->BitTestCases[i].Parent;
-      SDB->setCurrentBasicBlock(BB);
       // Emit the code
-      SDB->visitBitTestHeader(SDB->BitTestCases[i]);
+      SDB->visitBitTestHeader(SDB->BitTestCases[i], BB);
       CurDAG->setRoot(SDB->getRoot());
-      CodeGenAndEmitDAG();
+      BB = CodeGenAndEmitDAG(BB);
       SDB->clear();
     }
 
     for (unsigned j = 0, ej = SDB->BitTestCases[i].Cases.size(); j != ej; ++j) {
       // Set the current basic block to the mbb we wish to insert the code into
       BB = SDB->BitTestCases[i].Cases[j].ThisBB;
-      SDB->setCurrentBasicBlock(BB);
       // Emit the code
       if (j+1 != ej)
         SDB->visitBitTestCase(SDB->BitTestCases[i].Cases[j+1].ThisBB,
                               SDB->BitTestCases[i].Reg,
-                              SDB->BitTestCases[i].Cases[j]);
+                              SDB->BitTestCases[i].Cases[j],
+                              BB);
       else
         SDB->visitBitTestCase(SDB->BitTestCases[i].Default,
                               SDB->BitTestCases[i].Reg,
-                              SDB->BitTestCases[i].Cases[j]);
+                              SDB->BitTestCases[i].Cases[j],
+                              BB);
 
 
       CurDAG->setRoot(SDB->getRoot());
-      CodeGenAndEmitDAG();
+      BB = CodeGenAndEmitDAG(BB);
       SDB->clear();
     }
 
     // Update PHI Nodes
-    for (unsigned pi = 0, pe = SDB->PHINodesToUpdate.size(); pi != pe; ++pi) {
-      MachineInstr *PHI = SDB->PHINodesToUpdate[pi].first;
+    for (unsigned pi = 0, pe = FuncInfo->PHINodesToUpdate.size();
+         pi != pe; ++pi) {
+      MachineInstr *PHI = FuncInfo->PHINodesToUpdate[pi].first;
       MachineBasicBlock *PHIBB = PHI->getParent();
       assert(PHI->isPHI() &&
              "This is not a machine PHI node that we are updating!");
@@ -1091,10 +891,12 @@
       // from last "case" BB.
       if (PHIBB == SDB->BitTestCases[i].Default) {
         PHI->addOperand(MachineOperand::
-                        CreateReg(SDB->PHINodesToUpdate[pi].second, false));
+                        CreateReg(FuncInfo->PHINodesToUpdate[pi].second,
+                                  false));
         PHI->addOperand(MachineOperand::CreateMBB(SDB->BitTestCases[i].Parent));
         PHI->addOperand(MachineOperand::
-                        CreateReg(SDB->PHINodesToUpdate[pi].second, false));
+                        CreateReg(FuncInfo->PHINodesToUpdate[pi].second,
+                                  false));
         PHI->addOperand(MachineOperand::CreateMBB(SDB->BitTestCases[i].Cases.
                                                   back().ThisBB));
       }
@@ -1104,7 +906,8 @@
         MachineBasicBlock* cBB = SDB->BitTestCases[i].Cases[j].ThisBB;
         if (cBB->isSuccessor(PHIBB)) {
           PHI->addOperand(MachineOperand::
-                          CreateReg(SDB->PHINodesToUpdate[pi].second, false));
+                          CreateReg(FuncInfo->PHINodesToUpdate[pi].second,
+                                    false));
           PHI->addOperand(MachineOperand::CreateMBB(cBB));
         }
       }
@@ -1120,40 +923,42 @@
     if (!SDB->JTCases[i].first.Emitted) {
       // Set the current basic block to the mbb we wish to insert the code into
       BB = SDB->JTCases[i].first.HeaderBB;
-      SDB->setCurrentBasicBlock(BB);
       // Emit the code
-      SDB->visitJumpTableHeader(SDB->JTCases[i].second, SDB->JTCases[i].first);
+      SDB->visitJumpTableHeader(SDB->JTCases[i].second, SDB->JTCases[i].first,
+                                BB);
       CurDAG->setRoot(SDB->getRoot());
-      CodeGenAndEmitDAG();
+      BB = CodeGenAndEmitDAG(BB);
       SDB->clear();
     }
 
     // Set the current basic block to the mbb we wish to insert the code into
     BB = SDB->JTCases[i].second.MBB;
-    SDB->setCurrentBasicBlock(BB);
     // Emit the code
     SDB->visitJumpTable(SDB->JTCases[i].second);
     CurDAG->setRoot(SDB->getRoot());
-    CodeGenAndEmitDAG();
+    BB = CodeGenAndEmitDAG(BB);
     SDB->clear();
 
     // Update PHI Nodes
-    for (unsigned pi = 0, pe = SDB->PHINodesToUpdate.size(); pi != pe; ++pi) {
-      MachineInstr *PHI = SDB->PHINodesToUpdate[pi].first;
+    for (unsigned pi = 0, pe = FuncInfo->PHINodesToUpdate.size();
+         pi != pe; ++pi) {
+      MachineInstr *PHI = FuncInfo->PHINodesToUpdate[pi].first;
       MachineBasicBlock *PHIBB = PHI->getParent();
       assert(PHI->isPHI() &&
              "This is not a machine PHI node that we are updating!");
       // "default" BB. We can go there only from header BB.
       if (PHIBB == SDB->JTCases[i].second.Default) {
         PHI->addOperand
-          (MachineOperand::CreateReg(SDB->PHINodesToUpdate[pi].second, false));
+          (MachineOperand::CreateReg(FuncInfo->PHINodesToUpdate[pi].second,
+                                     false));
         PHI->addOperand
           (MachineOperand::CreateMBB(SDB->JTCases[i].first.HeaderBB));
       }
       // JT BB. Just iterate over successors here
       if (BB->isSuccessor(PHIBB)) {
         PHI->addOperand
-          (MachineOperand::CreateReg(SDB->PHINodesToUpdate[pi].second, false));
+          (MachineOperand::CreateReg(FuncInfo->PHINodesToUpdate[pi].second,
+                                     false));
         PHI->addOperand(MachineOperand::CreateMBB(BB));
       }
     }
@@ -1162,13 +967,13 @@
 
   // If the switch block involved a branch to one of the actual successors, we
   // need to update PHI nodes in that block.
-  for (unsigned i = 0, e = SDB->PHINodesToUpdate.size(); i != e; ++i) {
-    MachineInstr *PHI = SDB->PHINodesToUpdate[i].first;
+  for (unsigned i = 0, e = FuncInfo->PHINodesToUpdate.size(); i != e; ++i) {
+    MachineInstr *PHI = FuncInfo->PHINodesToUpdate[i].first;
     assert(PHI->isPHI() &&
            "This is not a machine PHI node that we are updating!");
     if (BB->isSuccessor(PHI->getParent())) {
-      PHI->addOperand(MachineOperand::CreateReg(SDB->PHINodesToUpdate[i].second,
-                                                false));
+      PHI->addOperand(
+        MachineOperand::CreateReg(FuncInfo->PHINodesToUpdate[i].second, false));
       PHI->addOperand(MachineOperand::CreateMBB(BB));
     }
   }
@@ -1178,12 +983,11 @@
   for (unsigned i = 0, e = SDB->SwitchCases.size(); i != e; ++i) {
     // Set the current basic block to the mbb we wish to insert the code into
     MachineBasicBlock *ThisBB = BB = SDB->SwitchCases[i].ThisBB;
-    SDB->setCurrentBasicBlock(BB);
 
     // Emit the code
-    SDB->visitSwitchCase(SDB->SwitchCases[i]);
+    SDB->visitSwitchCase(SDB->SwitchCases[i], BB);
     CurDAG->setRoot(SDB->getRoot());
-    CodeGenAndEmitDAG();
+    BB = CodeGenAndEmitDAG(BB);
 
     // Handle any PHI nodes in successors of this chunk, as if we were coming
     // from the original BB before switch expansion.  Note that PHI nodes can
@@ -1205,11 +1009,11 @@
              ++Phi) {
           // This value for this PHI node is recorded in PHINodesToUpdate.
           for (unsigned pn = 0; ; ++pn) {
-            assert(pn != SDB->PHINodesToUpdate.size() &&
+            assert(pn != FuncInfo->PHINodesToUpdate.size() &&
                    "Didn't find PHI entry!");
-            if (SDB->PHINodesToUpdate[pn].first == Phi) {
+            if (FuncInfo->PHINodesToUpdate[pn].first == Phi) {
               Phi->addOperand(MachineOperand::
-                              CreateReg(SDB->PHINodesToUpdate[pn].second,
+                              CreateReg(FuncInfo->PHINodesToUpdate[pn].second,
                                         false));
               Phi->addOperand(MachineOperand::CreateMBB(ThisBB));
               break;
@@ -1230,8 +1034,6 @@
     SDB->clear();
   }
   SDB->SwitchCases.clear();
-
-  SDB->PHINodesToUpdate.clear();
 }
 
 
@@ -1330,16 +1132,17 @@
   std::vector<SDValue> InOps;
   std::swap(InOps, Ops);
 
-  Ops.push_back(InOps[0]);  // input chain.
-  Ops.push_back(InOps[1]);  // input asm string.
+  Ops.push_back(InOps[InlineAsm::Op_InputChain]); // 0
+  Ops.push_back(InOps[InlineAsm::Op_AsmString]);  // 1
+  Ops.push_back(InOps[InlineAsm::Op_MDNode]);     // 2, !srcloc
 
-  unsigned i = 2, e = InOps.size();
+  unsigned i = InlineAsm::Op_FirstOperand, e = InOps.size();
   if (InOps[e-1].getValueType() == MVT::Flag)
     --e;  // Don't process a flag operand if it is here.
 
   while (i != e) {
     unsigned Flags = cast<ConstantSDNode>(InOps[i])->getZExtValue();
-    if ((Flags & 7) != 4 /*MEM*/) {
+    if (!InlineAsm::isMemKind(Flags)) {
       // Just skip over this operand, copying the operands verbatim.
       Ops.insert(Ops.end(), InOps.begin()+i,
                  InOps.begin()+i+InlineAsm::getNumOperandRegisters(Flags) + 1);
@@ -1349,14 +1152,14 @@
              "Memory operand with multiple values?");
       // Otherwise, this is a memory operand.  Ask the target to select it.
       std::vector<SDValue> SelOps;
-      if (SelectInlineAsmMemoryOperand(InOps[i+1], 'm', SelOps)) {
-        llvm_report_error("Could not match memory address.  Inline asm"
-                          " failure!");
-      }
+      if (SelectInlineAsmMemoryOperand(InOps[i+1], 'm', SelOps))
+        report_fatal_error("Could not match memory address.  Inline asm"
+                           " failure!");
 
       // Add this to the output node.
-      Ops.push_back(CurDAG->getTargetConstant(4/*MEM*/ | (SelOps.size()<< 3),
-                                              MVT::i32));
+      unsigned NewFlags =
+        InlineAsm::getFlagWord(InlineAsm::Kind_Mem, SelOps.size());
+      Ops.push_back(CurDAG->getTargetConstant(NewFlags, MVT::i32));
       Ops.insert(Ops.end(), SelOps.begin(), SelOps.end());
       i += 2;
     }
@@ -1433,7 +1236,8 @@
 /// IsLegalToFold - Returns true if the specific operand node N of
 /// U can be folded during instruction selection that starts at Root.
 bool SelectionDAGISel::IsLegalToFold(SDValue N, SDNode *U, SDNode *Root,
-                                     bool IgnoreChains) const {
+                                     CodeGenOpt::Level OptLevel,
+                                     bool IgnoreChains) {
   if (OptLevel == CodeGenOpt::None) return false;
 
   // If Root use can somehow reach N through a path that that doesn't contain
@@ -1517,14 +1321,6 @@
   return CurDAG->SelectNodeTo(N, TargetOpcode::IMPLICIT_DEF,N->getValueType(0));
 }
 
-SDNode *SelectionDAGISel::Select_EH_LABEL(SDNode *N) {
-  SDValue Chain = N->getOperand(0);
-  unsigned C = cast<LabelSDNode>(N)->getLabelID();
-  SDValue Tmp = CurDAG->getTargetConstant(C, MVT::i32);
-  return CurDAG->SelectNodeTo(N, TargetOpcode::EH_LABEL,
-                              MVT::Other, Tmp, Chain);
-}
-
 /// GetVBR - decode a vbr encoding whose top bit is set.
 ALWAYS_INLINE static uint64_t
 GetVBR(uint64_t Val, const unsigned char *MatcherTable, unsigned &Idx) {
@@ -1580,8 +1376,9 @@
       assert(ChainVal.getValueType() == MVT::Other && "Not a chain?");
       CurDAG->ReplaceAllUsesOfValueWith(ChainVal, InputChain, &ISU);
       
-      // If the node became dead, delete it.
-      if (ChainNode->use_empty())
+      // If the node became dead and we haven't already seen it, delete it.
+      if (ChainNode->use_empty() &&
+          !std::count(NowDeadNodes.begin(), NowDeadNodes.end(), ChainNode))
         NowDeadNodes.push_back(ChainNode);
     }
   }
@@ -1602,8 +1399,9 @@
       CurDAG->ReplaceAllUsesOfValueWith(SDValue(FRN, FRN->getNumValues()-1),
                                         InputFlag, &ISU);
       
-      // If the node became dead, delete it.
-      if (FRN->use_empty())
+      // If the node became dead and we haven't already seen it, delete it.
+      if (FRN->use_empty() &&
+          !std::count(NowDeadNodes.begin(), NowDeadNodes.end(), FRN))
         NowDeadNodes.push_back(FRN);
     }
   }
@@ -1651,7 +1449,8 @@
     
     if (User->getOpcode() == ISD::CopyToReg ||
         User->getOpcode() == ISD::CopyFromReg ||
-        User->getOpcode() == ISD::INLINEASM) {
+        User->getOpcode() == ISD::INLINEASM ||
+        User->getOpcode() == ISD::EH_LABEL) {
       // If their node ID got reset to -1 then they've already been selected.
       // Treat them like a MachineOpcode.
       if (User->getNodeId() == -1)
@@ -1797,9 +1596,9 @@
   // It is possible we're using MorphNodeTo to replace a node with no
   // normal results with one that has a normal result (or we could be
   // adding a chain) and the input could have flags and chains as well.
-  // In this case we need to shifting the operands down.
+  // In this case we need to shift the operands down.
   // FIXME: This is a horrible hack and broken in obscure cases, no worse
-  // than the old isel though.  We should sink this into MorphNodeTo.
+  // than the old isel though.
   int OldFlagResultNo = -1, OldChainResultNo = -1;
 
   unsigned NTMNumResults = Node->getNumValues();
@@ -1875,7 +1674,9 @@
 ALWAYS_INLINE static bool
 CheckOpcode(const unsigned char *MatcherTable, unsigned &MatcherIndex,
             SDNode *N) {
-  return N->getOpcode() == MatcherTable[MatcherIndex++];
+  uint16_t Opc = MatcherTable[MatcherIndex++];
+  Opc |= (unsigned short)MatcherTable[MatcherIndex++] << 8;
+  return N->getOpcode() == Opc;
 }
 
 ALWAYS_INLINE static bool
@@ -2011,6 +1812,7 @@
   }
 }
 
+namespace {
 
 struct MatchScope {
   /// FailIndex - If this match fails, this is the index to continue with.
@@ -2032,6 +1834,8 @@
   bool HasChainNodesMatched, HasFlagResultNodesMatched;
 };
 
+}
+
 SDNode *SelectionDAGISel::
 SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable,
                  unsigned TableSize) {
@@ -2042,7 +1846,10 @@
   case ISD::EntryToken:       // These nodes remain the same.
   case ISD::BasicBlock:
   case ISD::Register:
+  //case ISD::VALUETYPE:
+  //case ISD::CONDCODE:
   case ISD::HANDLENODE:
+  case ISD::MDNODE_SDNODE:
   case ISD::TargetConstant:
   case ISD::TargetConstantFP:
   case ISD::TargetConstantPool:
@@ -2055,6 +1862,7 @@
   case ISD::TokenFactor:
   case ISD::CopyFromReg:
   case ISD::CopyToReg:
+  case ISD::EH_LABEL:
     NodeToMatch->setNodeId(-1); // Mark selected.
     return 0;
   case ISD::AssertSext:
@@ -2063,7 +1871,6 @@
                                       NodeToMatch->getOperand(0));
     return 0;
   case ISD::INLINEASM: return Select_INLINEASM(NodeToMatch);
-  case ISD::EH_LABEL:  return Select_EH_LABEL(NodeToMatch);
   case ISD::UNDEF:     return Select_UNDEF(NodeToMatch);
   }
   
@@ -2127,7 +1934,8 @@
       if (CaseSize == 0) break;
 
       // Get the opcode, add the index to the table.
-      unsigned Opc = MatcherTable[Idx++];
+      uint16_t Opc = MatcherTable[Idx++];
+      Opc |= (unsigned short)MatcherTable[Idx++] << 8;
       if (Opc >= OpcodeOffset.size())
         OpcodeOffset.resize((Opc+1)*2);
       OpcodeOffset[Opc] = Idx;
@@ -2141,6 +1949,9 @@
   
   while (1) {
     assert(MatcherIndex < TableSize && "Invalid index");
+#ifndef NDEBUG
+    unsigned CurrentOpcodeIndex = MatcherIndex;
+#endif
     BuiltinOpcodes Opcode = (BuiltinOpcodes)MatcherTable[MatcherIndex++];
     switch (Opcode) {
     case OPC_Scope: {
@@ -2163,6 +1974,9 @@
         
         FailIndex = MatcherIndex+NumToSkip;
         
+        unsigned MatcherIndexOfPredicate = MatcherIndex;
+        (void)MatcherIndexOfPredicate; // silence warning.
+        
         // If we can't evaluate this predicate without pushing a scope (e.g. if
         // it is a 'MoveParent') or if the predicate succeeds on this node, we
         // push the scope and evaluate the full predicate chain.
@@ -2172,9 +1986,10 @@
         if (!Result)
           break;
         
-        DEBUG(errs() << "  Skipped scope entry at index " << MatcherIndex
-              << " continuing at " << FailIndex << "\n");
-
+        DEBUG(errs() << "  Skipped scope entry (due to false predicate) at "
+                     << "index " << MatcherIndexOfPredicate
+                     << ", continuing at " << FailIndex << "\n");
+        ++NumDAGIselRetries;
         
         // Otherwise, we know that this case of the Scope is guaranteed to fail,
         // move to the next case.
@@ -2280,8 +2095,11 @@
           CaseSize = GetVBR(CaseSize, MatcherTable, MatcherIndex);
         if (CaseSize == 0) break;
 
+        uint16_t Opc = MatcherTable[MatcherIndex++];
+        Opc |= (unsigned short)MatcherTable[MatcherIndex++] << 8;
+
         // If the opcode matches, then we will execute this case.
-        if (CurNodeOpcode == MatcherTable[MatcherIndex++])
+        if (CurNodeOpcode == Opc)
           break;
       
         // Otherwise, skip over this case.
@@ -2370,7 +2188,8 @@
       if (!IsProfitableToFold(N, NodeStack[NodeStack.size()-2].getNode(),
                               NodeToMatch) ||
           !IsLegalToFold(N, NodeStack[NodeStack.size()-2].getNode(),
-                         NodeToMatch, true/*We validate our own chains*/))
+                         NodeToMatch, OptLevel,
+                         true/*We validate our own chains*/))
         break;
       
       continue;
@@ -2410,6 +2229,35 @@
       continue;
     }
         
+    case OPC_EmitMergeInputChains1_0:    // OPC_EmitMergeInputChains, 1, 0
+    case OPC_EmitMergeInputChains1_1: {  // OPC_EmitMergeInputChains, 1, 1
+      // These are space-optimized forms of OPC_EmitMergeInputChains.
+      assert(InputChain.getNode() == 0 &&
+             "EmitMergeInputChains should be the first chain producing node");
+      assert(ChainNodesMatched.empty() &&
+             "Should only have one EmitMergeInputChains per match");
+      
+      // Read all of the chained nodes.
+      unsigned RecNo = Opcode == OPC_EmitMergeInputChains1_1;
+      assert(RecNo < RecordedNodes.size() && "Invalid CheckSame");
+      ChainNodesMatched.push_back(RecordedNodes[RecNo].getNode());
+        
+      // FIXME: What if other value results of the node have uses not matched
+      // by this pattern?
+      if (ChainNodesMatched.back() != NodeToMatch &&
+          !RecordedNodes[RecNo].hasOneUse()) {
+        ChainNodesMatched.clear();
+        break;
+      }
+      
+      // Merge the input chains if they are not intra-pattern references.
+      InputChain = HandleMergeInputChains(ChainNodesMatched, CurDAG);
+      
+      if (InputChain.getNode() == 0)
+        break;  // Failed to merge.
+      continue;
+    }
+        
     case OPC_EmitMergeInputChains: {
       assert(InputChain.getNode() == 0 &&
              "EmitMergeInputChains should be the first chain producing node");
@@ -2628,14 +2476,10 @@
         assert(ResSlot < RecordedNodes.size() && "Invalid CheckSame");
         SDValue Res = RecordedNodes[ResSlot];
         
-        // FIXME2: Eliminate this horrible hack by fixing the 'Gen' program
-        // after (parallel) on input patterns are removed.  This would also
-        // allow us to stop encoding #results in OPC_CompleteMatch's table
-        // entry.
-        if (NodeToMatch->getNumValues() <= i ||
-            NodeToMatch->getValueType(i) == MVT::Other ||
-            NodeToMatch->getValueType(i) == MVT::Flag)
-          break;
+        assert(i < NodeToMatch->getNumValues() &&
+               NodeToMatch->getValueType(i) != MVT::Other &&
+               NodeToMatch->getValueType(i) != MVT::Flag &&
+               "Invalid number of results to complete!");
         assert((NodeToMatch->getValueType(i) == Res.getValueType() ||
                 NodeToMatch->getValueType(i) == MVT::iPTR ||
                 Res.getValueType() == MVT::iPTR ||
@@ -2666,6 +2510,8 @@
     // If the code reached this point, then the match failed.  See if there is
     // another child to try in the current 'Scope', otherwise pop it until we
     // find a case to check.
+    DEBUG(errs() << "  Match failed at index " << CurrentOpcodeIndex << "\n");
+    ++NumDAGIselRetries;
     while (1) {
       if (MatchScopes.empty()) {
         CannotYetSelect(NodeToMatch);
@@ -2680,13 +2526,12 @@
       NodeStack.append(LastScope.NodeStack.begin(), LastScope.NodeStack.end());
       N = NodeStack.back();
 
-      DEBUG(errs() << "  Match failed at index " << MatcherIndex
-                   << " continuing at " << LastScope.FailIndex << "\n");
-    
       if (LastScope.NumMatchedMemRefs != MatchedMemRefs.size())
         MatchedMemRefs.resize(LastScope.NumMatchedMemRefs);
       MatcherIndex = LastScope.FailIndex;
       
+      DEBUG(errs() << "  Continuing at " << MatcherIndex << "\n");
+    
       InputChain = LastScope.InputChain;
       InputFlag = LastScope.InputFlag;
       if (!LastScope.HasChainNodesMatched)
@@ -2737,7 +2582,7 @@
     else
       Msg << "unknown intrinsic #" << iid;
   }
-  llvm_report_error(Msg.str());
+  report_fatal_error(Msg.str());
 }
 
 char SelectionDAGISel::ID = 0;
diff --git a/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index 8d0d884..f59f851 100644
--- a/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -175,17 +175,25 @@
   Names[RTLIB::FLOOR_F64] = "floor";
   Names[RTLIB::FLOOR_F80] = "floorl";
   Names[RTLIB::FLOOR_PPCF128] = "floorl";
+  Names[RTLIB::COPYSIGN_F32] = "copysignf";
+  Names[RTLIB::COPYSIGN_F64] = "copysign";
+  Names[RTLIB::COPYSIGN_F80] = "copysignl";
+  Names[RTLIB::COPYSIGN_PPCF128] = "copysignl";
   Names[RTLIB::FPEXT_F32_F64] = "__extendsfdf2";
+  Names[RTLIB::FPEXT_F16_F32] = "__gnu_h2f_ieee";
+  Names[RTLIB::FPROUND_F32_F16] = "__gnu_f2h_ieee";
   Names[RTLIB::FPROUND_F64_F32] = "__truncdfsf2";
   Names[RTLIB::FPROUND_F80_F32] = "__truncxfsf2";
   Names[RTLIB::FPROUND_PPCF128_F32] = "__trunctfsf2";
   Names[RTLIB::FPROUND_F80_F64] = "__truncxfdf2";
   Names[RTLIB::FPROUND_PPCF128_F64] = "__trunctfdf2";
-  Names[RTLIB::FPTOSINT_F32_I8] = "__fixsfi8";
-  Names[RTLIB::FPTOSINT_F32_I16] = "__fixsfi16";
+  Names[RTLIB::FPTOSINT_F32_I8] = "__fixsfqi";
+  Names[RTLIB::FPTOSINT_F32_I16] = "__fixsfhi";
   Names[RTLIB::FPTOSINT_F32_I32] = "__fixsfsi";
   Names[RTLIB::FPTOSINT_F32_I64] = "__fixsfdi";
   Names[RTLIB::FPTOSINT_F32_I128] = "__fixsfti";
+  Names[RTLIB::FPTOSINT_F64_I8] = "__fixdfqi";
+  Names[RTLIB::FPTOSINT_F64_I16] = "__fixdfhi";
   Names[RTLIB::FPTOSINT_F64_I32] = "__fixdfsi";
   Names[RTLIB::FPTOSINT_F64_I64] = "__fixdfdi";
   Names[RTLIB::FPTOSINT_F64_I128] = "__fixdfti";
@@ -195,11 +203,13 @@
   Names[RTLIB::FPTOSINT_PPCF128_I32] = "__fixtfsi";
   Names[RTLIB::FPTOSINT_PPCF128_I64] = "__fixtfdi";
   Names[RTLIB::FPTOSINT_PPCF128_I128] = "__fixtfti";
-  Names[RTLIB::FPTOUINT_F32_I8] = "__fixunssfi8";
-  Names[RTLIB::FPTOUINT_F32_I16] = "__fixunssfi16";
+  Names[RTLIB::FPTOUINT_F32_I8] = "__fixunssfqi";
+  Names[RTLIB::FPTOUINT_F32_I16] = "__fixunssfhi";
   Names[RTLIB::FPTOUINT_F32_I32] = "__fixunssfsi";
   Names[RTLIB::FPTOUINT_F32_I64] = "__fixunssfdi";
   Names[RTLIB::FPTOUINT_F32_I128] = "__fixunssfti";
+  Names[RTLIB::FPTOUINT_F64_I8] = "__fixunsdfqi";
+  Names[RTLIB::FPTOUINT_F64_I16] = "__fixunsdfhi";
   Names[RTLIB::FPTOUINT_F64_I32] = "__fixunsdfsi";
   Names[RTLIB::FPTOUINT_F64_I64] = "__fixunsdfdi";
   Names[RTLIB::FPTOUINT_F64_I128] = "__fixunsdfti";
@@ -270,6 +280,7 @@
     if (RetVT == MVT::f64)
       return FPEXT_F32_F64;
   }
+
   return UNKNOWN_LIBCALL;
 }
 
@@ -289,6 +300,7 @@
     if (OpVT == MVT::ppcf128)
       return FPROUND_PPCF128_F64;
   }
+
   return UNKNOWN_LIBCALL;
 }
 
@@ -307,6 +319,10 @@
     if (RetVT == MVT::i128)
       return FPTOSINT_F32_I128;
   } else if (OpVT == MVT::f64) {
+    if (RetVT == MVT::i8)
+      return FPTOSINT_F64_I8;
+    if (RetVT == MVT::i16)
+      return FPTOSINT_F64_I16;
     if (RetVT == MVT::i32)
       return FPTOSINT_F64_I32;
     if (RetVT == MVT::i64)
@@ -346,6 +362,10 @@
     if (RetVT == MVT::i128)
       return FPTOUINT_F32_I128;
   } else if (OpVT == MVT::f64) {
+    if (RetVT == MVT::i8)
+      return FPTOUINT_F64_I8;
+    if (RetVT == MVT::i16)
+      return FPTOUINT_F64_I16;
     if (RetVT == MVT::i32)
       return FPTOUINT_F64_I32;
     if (RetVT == MVT::i64)
@@ -461,14 +481,14 @@
 }
 
 /// NOTE: The constructor takes ownership of TLOF.
-TargetLowering::TargetLowering(TargetMachine &tm,TargetLoweringObjectFile *tlof)
+TargetLowering::TargetLowering(const TargetMachine &tm,
+                               const TargetLoweringObjectFile *tlof)
   : TM(tm), TD(TM.getTargetData()), TLOF(*tlof) {
   // All operations default to being supported.
   memset(OpActions, 0, sizeof(OpActions));
   memset(LoadExtActions, 0, sizeof(LoadExtActions));
   memset(TruncStoreActions, 0, sizeof(TruncStoreActions));
   memset(IndexedModeActions, 0, sizeof(IndexedModeActions));
-  memset(ConvertActions, 0, sizeof(ConvertActions));
   memset(CondCodeActions, 0, sizeof(CondCodeActions));
 
   // Set default actions for various operations.
@@ -702,7 +722,7 @@
       unsigned NElts = VT.getVectorNumElements();
       for (unsigned nVT = i+1; nVT <= MVT::LAST_VECTOR_VALUETYPE; ++nVT) {
         EVT SVT = (MVT::SimpleValueType)nVT;
-        if (isTypeLegal(SVT) && SVT.getVectorElementType() == EltVT &&
+        if (isTypeSynthesizable(SVT) && SVT.getVectorElementType() == EltVT &&
             SVT.getVectorNumElements() > NElts && NElts != 1) {
           TransformToType[i] = SVT;
           ValueTypeActions.setTypeAction(VT, Promote);
@@ -793,20 +813,6 @@
   return 1;
 }
 
-/// getWidenVectorType: given a vector type, returns the type to widen to
-/// (e.g., v7i8 to v8i8). If the vector type is legal, it returns itself.
-/// If there is no vector type that we want to widen to, returns MVT::Other
-/// When and where to widen is target dependent based on the cost of
-/// scalarizing vs using the wider vector type.
-EVT TargetLowering::getWidenVectorType(EVT VT) const {
-  assert(VT.isVector());
-  if (isTypeLegal(VT))
-    return VT;
- 
-  // Default is not to widen until moved to LegalizeTypes
-  return MVT::Other;
-}
-
 /// getByValTypeAlignment - Return the desired alignment for ByVal aggregate
 /// function arguments in the caller parameter area.  This is the actual
 /// alignment, not its logarithm.
@@ -1275,8 +1281,9 @@
     // variable.  The low bit of the shift cannot be an input sign bit unless
     // the shift amount is >= the size of the datatype, which is undefined.
     if (DemandedMask == 1)
-      return TLO.CombineTo(Op, TLO.DAG.getNode(ISD::SRL, dl, Op.getValueType(),
-                                               Op.getOperand(0), Op.getOperand(1)));
+      return TLO.CombineTo(Op,
+                           TLO.DAG.getNode(ISD::SRL, dl, Op.getValueType(),
+                                           Op.getOperand(0), Op.getOperand(1)));
 
     if (ConstantSDNode *SA = dyn_cast<ConstantSDNode>(Op.getOperand(1))) {
       EVT VT = Op.getValueType();
@@ -1461,23 +1468,29 @@
       case ISD::SRL:
         // Shrink SRL by a constant if none of the high bits shifted in are
         // demanded.
-        if (ConstantSDNode *ShAmt = dyn_cast<ConstantSDNode>(In.getOperand(1))){
-          APInt HighBits = APInt::getHighBitsSet(OperandBitWidth,
-                                                 OperandBitWidth - BitWidth);
-          HighBits = HighBits.lshr(ShAmt->getZExtValue());
-          HighBits.trunc(BitWidth);
-          
-          if (ShAmt->getZExtValue() < BitWidth && !(HighBits & NewMask)) {
-            // None of the shifted in bits are needed.  Add a truncate of the
-            // shift input, then shift it.
-            SDValue NewTrunc = TLO.DAG.getNode(ISD::TRUNCATE, dl,
-                                                 Op.getValueType(), 
-                                                 In.getOperand(0));
-            return TLO.CombineTo(Op, TLO.DAG.getNode(ISD::SRL, dl,
-                                                     Op.getValueType(),
-                                                     NewTrunc, 
-                                                     In.getOperand(1)));
-          }
+        if (TLO.LegalTypes() &&
+            !isTypeDesirableForOp(ISD::SRL, Op.getValueType()))
+          // Do not turn (vt1 truncate (vt2 srl)) into (vt1 srl) if vt1 is
+          // undesirable.
+          break;
+        ConstantSDNode *ShAmt = dyn_cast<ConstantSDNode>(In.getOperand(1));
+        if (!ShAmt)
+          break;
+        APInt HighBits = APInt::getHighBitsSet(OperandBitWidth,
+                                               OperandBitWidth - BitWidth);
+        HighBits = HighBits.lshr(ShAmt->getZExtValue());
+        HighBits.trunc(BitWidth);
+
+        if (ShAmt->getZExtValue() < BitWidth && !(HighBits & NewMask)) {
+          // None of the shifted in bits are needed.  Add a truncate of the
+          // shift input, then shift it.
+          SDValue NewTrunc = TLO.DAG.getNode(ISD::TRUNCATE, dl,
+                                             Op.getValueType(), 
+                                             In.getOperand(0));
+          return TLO.CombineTo(Op, TLO.DAG.getNode(ISD::SRL, dl,
+                                                   Op.getValueType(),
+                                                   NewTrunc, 
+                                                   In.getOperand(1)));
         }
         break;
       }
@@ -2241,7 +2254,7 @@
 
 /// isGAPlusOffset - Returns true (and the GlobalValue and the offset) if the
 /// node is a GlobalAddress + offset.
-bool TargetLowering::isGAPlusOffset(SDNode *N, GlobalValue* &GA,
+bool TargetLowering::isGAPlusOffset(SDNode *N, const GlobalValue* &GA,
                                     int64_t &Offset) const {
   if (isa<GlobalAddressSDNode>(N)) {
     GlobalAddressSDNode *GASD = cast<GlobalAddressSDNode>(N);
diff --git a/lib/CodeGen/SelectionDAG/TargetSelectionDAGInfo.cpp b/lib/CodeGen/SelectionDAG/TargetSelectionDAGInfo.cpp
new file mode 100644
index 0000000..d20477f
--- /dev/null
+++ b/lib/CodeGen/SelectionDAG/TargetSelectionDAGInfo.cpp
@@ -0,0 +1,21 @@
+//===-- TargetSelectionDAGInfo.cpp - SelectionDAG Info --------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This implements the TargetSelectionDAGInfo class.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Target/TargetSelectionDAGInfo.h"
+using namespace llvm;
+
+TargetSelectionDAGInfo::TargetSelectionDAGInfo() {
+}
+
+TargetSelectionDAGInfo::~TargetSelectionDAGInfo() {
+}