Vendor import of llvm tags/RELEASE_33/final r183502 (effectively, 3.3

release):
http://llvm.org/svn/llvm-project/llvm/tags/RELEASE_33/final@183502

21 files changed, 317 insertions, 103 deletions

diff --git a/include/llvm/CodeGen/AsmPrinter.h b/include/llvm/CodeGen/AsmPrinter.h
index e0a6e3f..c2fd6ce 100644
--- a/include/llvm/CodeGen/AsmPrinter.h
+++ b/include/llvm/CodeGen/AsmPrinter.h
@@ -25,6 +25,7 @@ namespace llvm {
   class BlockAddress;
   class GCStrategy;
   class Constant;
+  class ConstantArray;
   class GCMetadataPrinter;
   class GlobalValue;
   class GlobalVariable;
@@ -134,6 +135,9 @@ namespace llvm {
     /// getDataLayout - Return information about data layout.
     const DataLayout &getDataLayout() const;
 
+    /// getTargetTriple - Return the target triple string.
+    StringRef getTargetTriple() const;
+
     /// getCurrentSection() - Return the current section we are emitting to.
     const MCSection *getCurrentSection() const;
 
@@ -480,7 +484,7 @@ namespace llvm {
     void EmitJumpTableEntry(const MachineJumpTableInfo *MJTI,
                             const MachineBasicBlock *MBB,
                             unsigned uid) const;
-    void EmitLLVMUsedList(const Constant *List);
+    void EmitLLVMUsedList(const ConstantArray *InitList);
     void EmitXXStructorList(const Constant *List, bool isCtor);
     GCMetadataPrinter *GetOrCreateGCPrinter(GCStrategy *C);
   };
diff --git a/include/llvm/CodeGen/CallingConvLower.h b/include/llvm/CodeGen/CallingConvLower.h
index c035e07..fa9d60f 100644
--- a/include/llvm/CodeGen/CallingConvLower.h
+++ b/include/llvm/CodeGen/CallingConvLower.h
@@ -163,8 +163,56 @@ private:
 
   unsigned StackOffset;
   SmallVector<uint32_t, 16> UsedRegs;
-  unsigned FirstByValReg;
-  bool FirstByValRegValid;
+
+  // ByValInfo and SmallVector<ByValInfo, 4> ByValRegs:
+  //
+  // Vector of ByValInfo instances (ByValRegs) is introduced for byval registers
+  // tracking.
+  // Or, in another words it tracks byval parameters that are stored in
+  // general purpose registers.
+  //
+  // For 4 byte stack alignment,
+  // instance index means byval parameter number in formal
+  // arguments set. Assume, we have some "struct_type" with size = 4 bytes,
+  // then, for function "foo":
+  //
+  // i32 foo(i32 %p, %struct_type* %r, i32 %s, %struct_type* %t)
+  //
+  // ByValRegs[0] describes how "%r" is stored (Begin == r1, End == r2)
+  // ByValRegs[1] describes how "%t" is stored (Begin == r3, End == r4).
+  //
+  // In case of 8 bytes stack alignment,
+  // ByValRegs may also contain information about wasted registers.
+  // In function shown above, r3 would be wasted according to AAPCS rules.
+  // And in that case ByValRegs[1].Waste would be "true".
+  // ByValRegs vector size still would be 2,
+  // while "%t" goes to the stack: it wouldn't be described in ByValRegs.
+  //
+  // Supposed use-case for this collection:
+  // 1. Initially ByValRegs is empty, InRegsParamsProceed is 0.
+  // 2. HandleByVal fillups ByValRegs.
+  // 3. Argument analysis (LowerFormatArguments, for example). After
+  // some byval argument was analyzed, InRegsParamsProceed is increased.
+  struct ByValInfo {
+    ByValInfo(unsigned B, unsigned E, bool IsWaste = false) :
+      Begin(B), End(E), Waste(IsWaste) {}
+    // First register allocated for current parameter.
+    unsigned Begin;
+
+    // First after last register allocated for current parameter.
+    unsigned End;
+
+    // Means that current range of registers doesn't belong to any
+    // parameters. It was wasted due to stack alignment rules.
+    // For more information see:
+    // AAPCS, 5.5 Parameter Passing, Stage C, C.3.
+    bool Waste;
+  };
+  SmallVector<ByValInfo, 4 > ByValRegs;
+
+  // InRegsParamsProceed - shows how many instances of ByValRegs was proceed
+  // during argument analysis.
+  unsigned InRegsParamsProceed;
 
 protected:
   ParmContext CallOrPrologue;
@@ -306,12 +354,45 @@ public:
                    MVT LocVT, CCValAssign::LocInfo LocInfo,
                    int MinSize, int MinAlign, ISD::ArgFlagsTy ArgFlags);
 
-  // First GPR that carries part of a byval aggregate that's split
-  // between registers and memory.
-  unsigned getFirstByValReg() const { return FirstByValRegValid ? FirstByValReg : 0; }
-  void setFirstByValReg(unsigned r) { FirstByValReg = r; FirstByValRegValid = true; }
-  void clearFirstByValReg() { FirstByValReg = 0; FirstByValRegValid = false; }
-  bool isFirstByValRegValid() const { return FirstByValRegValid; }
+  // Returns count of byval arguments that are to be stored (even partly)
+  // in registers.
+  unsigned getInRegsParamsCount() const { return ByValRegs.size(); }
+
+  // Returns count of byval in-regs arguments proceed.
+  unsigned getInRegsParamsProceed() const { return InRegsParamsProceed; }
+
+  // Get information about N-th byval parameter that is stored in registers.
+  // Here "ByValParamIndex" is N.
+  void getInRegsParamInfo(unsigned InRegsParamRecordIndex,
+                          unsigned& BeginReg, unsigned& EndReg) const {
+    assert(InRegsParamRecordIndex < ByValRegs.size() &&
+           "Wrong ByVal parameter index");
+
+    const ByValInfo& info = ByValRegs[InRegsParamRecordIndex];
+    BeginReg = info.Begin;
+    EndReg = info.End;
+  }
+
+  // Add information about parameter that is kept in registers.
+  void addInRegsParamInfo(unsigned RegBegin, unsigned RegEnd) {
+    ByValRegs.push_back(ByValInfo(RegBegin, RegEnd));
+  }
+
+  // Goes either to next byval parameter (excluding "waste" record), or
+  // to the end of collection.
+  // Returns false, if end is reached.
+  bool nextInRegsParam() {
+    unsigned e = ByValRegs.size();
+    if (InRegsParamsProceed < e)
+      ++InRegsParamsProceed;
+    return InRegsParamsProceed < e;
+  }
+
+  // Clear byval registers tracking info.
+  void clearByValRegsInfo() {
+    InRegsParamsProceed = 0;
+    ByValRegs.clear();
+  }
 
   ParmContext getCallOrPrologue() const { return CallOrPrologue; }
 
diff --git a/include/llvm/CodeGen/FastISel.h b/include/llvm/CodeGen/FastISel.h
index 705db7e..471e9bf 100644
--- a/include/llvm/CodeGen/FastISel.h
+++ b/include/llvm/CodeGen/FastISel.h
@@ -123,12 +123,28 @@ public:
   /// index value.
   std::pair<unsigned, bool> getRegForGEPIndex(const Value *V);
 
-  /// TryToFoldLoad - The specified machine instr operand is a vreg, and that
+  /// \brief We're checking to see if we can fold \p LI into \p FoldInst.
+  /// Note that we could have a sequence where multiple LLVM IR instructions
+  /// are folded into the same machineinstr.  For example we could have:
+  ///   A: x = load i32 *P
+  ///   B: y = icmp A, 42
+  ///   C: br y, ...
+  ///
+  /// In this scenario, \p LI is "A", and \p FoldInst is "C".  We know
+  /// about "B" (and any other folded instructions) because it is between
+  /// A and C.
+  ///
+  /// If we succeed folding, return true.
+  ///
+  bool tryToFoldLoad(const LoadInst *LI, const Instruction *FoldInst);
+
+  /// \brief The specified machine instr operand is a vreg, and that
   /// vreg is being provided by the specified load instruction.  If possible,
   /// try to fold the load as an operand to the instruction, returning true if
   /// possible.
-  virtual bool TryToFoldLoad(MachineInstr * /*MI*/, unsigned /*OpNo*/,
-                             const LoadInst * /*LI*/) {
+  /// This method should be implemented by targets.
+  virtual bool tryToFoldLoadIntoMI(MachineInstr * /*MI*/, unsigned /*OpNo*/,
+                                   const LoadInst * /*LI*/) {
     return false;
   }
 
diff --git a/include/llvm/CodeGen/ISDOpcodes.h b/include/llvm/CodeGen/ISDOpcodes.h
index 442729b..0fd211b 100644
--- a/include/llvm/CodeGen/ISDOpcodes.h
+++ b/include/llvm/CodeGen/ISDOpcodes.h
@@ -602,14 +602,6 @@ namespace ISD {
     /// specifier.
     PREFETCH,
 
-    /// OUTCHAIN = MEMBARRIER(INCHAIN, load-load, load-store, store-load,
-    ///                       store-store, device)
-    /// This corresponds to the memory.barrier intrinsic.
-    /// it takes an input chain, 4 operands to specify the type of barrier, an
-    /// operand specifying if the barrier applies to device and uncached memory
-    /// and produces an output chain.
-    MEMBARRIER,
-
     /// OUTCHAIN = ATOMIC_FENCE(INCHAIN, ordering, scope)
     /// This corresponds to the fence instruction. It takes an input chain, and
     /// two integer constants: an AtomicOrdering and a SynchronizationScope.
diff --git a/include/llvm/CodeGen/LiveInterval.h b/include/llvm/CodeGen/LiveInterval.h
index 244be9c..cb09a49 100644
--- a/include/llvm/CodeGen/LiveInterval.h
+++ b/include/llvm/CodeGen/LiveInterval.h
@@ -399,6 +399,15 @@ namespace llvm {
       return r != end() && r->containsRange(Start, End);
     }
 
+    /// True iff this live range is a single segment that lies between the
+    /// specified boundaries, exclusively. Vregs live across a backedge are not
+    /// considered local. The boundaries are expected to lie within an extended
+    /// basic block, so vregs that are not live out should contain no holes.
+    bool isLocal(SlotIndex Start, SlotIndex End) const {
+      return beginIndex() > Start.getBaseIndex() &&
+        endIndex() < End.getBoundaryIndex();
+    }
+
     /// removeRange - Remove the specified range from this interval.  Note that
     /// the range must be a single LiveRange in its entirety.
     void removeRange(SlotIndex Start, SlotIndex End,
diff --git a/include/llvm/CodeGen/LiveRangeEdit.h b/include/llvm/CodeGen/LiveRangeEdit.h
index 8a32a3c..e59276f 100644
--- a/include/llvm/CodeGen/LiveRangeEdit.h
+++ b/include/llvm/CodeGen/LiveRangeEdit.h
@@ -196,8 +196,7 @@ public:
   /// allocator.  These registers should not be split into new intervals
   /// as currently those new intervals are not guaranteed to spill.
   void eliminateDeadDefs(SmallVectorImpl<MachineInstr*> &Dead,
-                         ArrayRef<unsigned> RegsBeingSpilled 
-                          = ArrayRef<unsigned>());
+                         ArrayRef<unsigned> RegsBeingSpilled = None);
 
   /// calculateRegClassAndHint - Recompute register class and hint for each new
   /// register.
diff --git a/include/llvm/CodeGen/MachineBasicBlock.h b/include/llvm/CodeGen/MachineBasicBlock.h
index 492a3ff..0f2f874 100644
--- a/include/llvm/CodeGen/MachineBasicBlock.h
+++ b/include/llvm/CodeGen/MachineBasicBlock.h
@@ -71,7 +71,6 @@ class MachineBasicBlock : public ilist_node<MachineBasicBlock> {
   std::vector<MachineBasicBlock *> Predecessors;
   std::vector<MachineBasicBlock *> Successors;
 
-
   /// Weights - Keep track of the weights to the successors. This vector
   /// has the same order as Successors, or it is empty if we don't use it
   /// (disable optimization).
@@ -96,6 +95,10 @@ class MachineBasicBlock : public ilist_node<MachineBasicBlock> {
   /// target of an indirect branch.
   bool AddressTaken;
 
+  /// \brief since getSymbol is a relatively heavy-weight operation, the symbol
+  /// is only computed once and is cached.
+  mutable MCSymbol *CachedMCSymbol;
+
   // Intrusive list support
   MachineBasicBlock() {}
 
diff --git a/include/llvm/CodeGen/MachineFunction.h b/include/llvm/CodeGen/MachineFunction.h
index 82c4cd6..c886e25 100644
--- a/include/llvm/CodeGen/MachineFunction.h
+++ b/include/llvm/CodeGen/MachineFunction.h
@@ -352,8 +352,8 @@ public:
   // Internal functions used to automatically number MachineBasicBlocks
   //
 
-  /// getNextMBBNumber - Returns the next unique number to be assigned
-  /// to a MachineBasicBlock in this MachineFunction.
+  /// \brief Adds the MBB to the internal numbering. Returns the unique number
+  /// assigned to the MBB.
   ///
   unsigned addToMBBNumbering(MachineBasicBlock *MBB) {
     MBBNumbering.push_back(MBB);
diff --git a/include/llvm/CodeGen/MachineMemOperand.h b/include/llvm/CodeGen/MachineMemOperand.h
index ddb1271..00a55b5 100644
--- a/include/llvm/CodeGen/MachineMemOperand.h
+++ b/include/llvm/CodeGen/MachineMemOperand.h
@@ -34,22 +34,22 @@ struct MachinePointerInfo {
   /// If this is null, then the access is to a pointer in the default address
   /// space.
   const Value *V;
-  
+
   /// Offset - This is an offset from the base Value*.
   int64_t Offset;
-  
+
   explicit MachinePointerInfo(const Value *v = 0, int64_t offset = 0)
     : V(v), Offset(offset) {}
-  
+
   MachinePointerInfo getWithOffset(int64_t O) const {
     if (V == 0) return MachinePointerInfo(0, 0);
     return MachinePointerInfo(V, Offset+O);
   }
-  
+
   /// getAddrSpace - Return the LLVM IR address space number that this pointer
   /// points into.
   unsigned getAddrSpace() const;
-  
+
   /// getConstantPool - Return a MachinePointerInfo record that refers to the
   /// constant pool.
   static MachinePointerInfo getConstantPool();
@@ -57,20 +57,20 @@ struct MachinePointerInfo {
   /// getFixedStack - Return a MachinePointerInfo record that refers to the
   /// the specified FrameIndex.
   static MachinePointerInfo getFixedStack(int FI, int64_t offset = 0);
-  
+
   /// getJumpTable - Return a MachinePointerInfo record that refers to a
   /// jump table entry.
   static MachinePointerInfo getJumpTable();
-  
+
   /// getGOT - Return a MachinePointerInfo record that refers to a
   /// GOT entry.
   static MachinePointerInfo getGOT();
-  
+
   /// getStack - stack pointer relative access.
   static MachinePointerInfo getStack(int64_t Offset);
 };
-  
-  
+
+
 //===----------------------------------------------------------------------===//
 /// MachineMemOperand - A description of a memory reference used in the backend.
 /// Instead of holding a StoreInst or LoadInst, this class holds the address
@@ -99,8 +99,11 @@ public:
     MONonTemporal = 8,
     /// The memory access is invariant.
     MOInvariant = 16,
+    // Target hints allow target passes to annotate memory operations.
+    MOTargetStartBit = 5,
+    MOTargetNumBits = 3,
     // This is the number of bits we need to represent flags.
-    MOMaxBits = 5
+    MOMaxBits = 8
   };
 
   /// MachineMemOperand - Construct an MachineMemOperand object with the
@@ -110,7 +113,7 @@ public:
                     const MDNode *Ranges = 0);
 
   const MachinePointerInfo &getPointerInfo() const { return PtrInfo; }
-  
+
   /// getValue - Return the base address of the memory access. This may either
   /// be a normal LLVM IR Value, or one of the special values used in CodeGen.
   /// Special values are those obtained via
@@ -123,6 +126,9 @@ public:
   /// getFlags - Return the raw flags of the source value, \see MemOperandFlags.
   unsigned int getFlags() const { return Flags & ((1 << MOMaxBits) - 1); }
 
+  /// Bitwise OR the current flags with the given flags.
+  void setFlags(unsigned f) { Flags |= (f & ((1 << MOMaxBits) - 1)); }
+
   /// getOffset - For normal values, this is a byte offset added to the base
   /// address. For PseudoSourceValue::FPRel values, this is the FrameIndex
   /// number.
diff --git a/include/llvm/CodeGen/MachineRegisterInfo.h b/include/llvm/CodeGen/MachineRegisterInfo.h
index 4b43cc1..24ba7bb 100644
--- a/include/llvm/CodeGen/MachineRegisterInfo.h
+++ b/include/llvm/CodeGen/MachineRegisterInfo.h
@@ -157,6 +157,12 @@ public:
   // Strictly for use by MachineInstr.cpp.
   void moveOperands(MachineOperand *Dst, MachineOperand *Src, unsigned NumOps);
 
+  /// Verify the sanity of the use list for Reg.
+  void verifyUseList(unsigned Reg) const;
+
+  /// Verify the use list of all registers.
+  void verifyUseLists() const;
+
   /// reg_begin/reg_end - Provide iteration support to walk over all definitions
   /// and uses of a register within the MachineFunction that corresponds to this
   /// MachineRegisterInfo object.
diff --git a/include/llvm/CodeGen/MachineScheduler.h b/include/llvm/CodeGen/MachineScheduler.h
index 57febe7..769e4b4 100644
--- a/include/llvm/CodeGen/MachineScheduler.h
+++ b/include/llvm/CodeGen/MachineScheduler.h
@@ -274,6 +274,10 @@ public:
     Mutations.push_back(Mutation);
   }
 
+  /// \brief True if an edge can be added from PredSU to SuccSU without creating
+  /// a cycle.
+  bool canAddEdge(SUnit *SuccSU, SUnit *PredSU);
+
   /// \brief Add a DAG edge to the given SU with the given predecessor
   /// dependence data.
   ///
@@ -297,6 +301,10 @@ public:
   /// reorderable instructions.
   virtual void schedule();
 
+  /// Change the position of an instruction within the basic block and update
+  /// live ranges and region boundary iterators.
+  void moveInstruction(MachineInstr *MI, MachineBasicBlock::iterator InsertPos);
+
   /// Get current register pressure for the top scheduled instructions.
   const IntervalPressure &getTopPressure() const { return TopPressure; }
   const RegPressureTracker &getTopRPTracker() const { return TopRPTracker; }
@@ -362,7 +370,6 @@ protected:
 
   void updateScheduledPressure(const std::vector<unsigned> &NewMaxPressure);
 
-  void moveInstruction(MachineInstr *MI, MachineBasicBlock::iterator InsertPos);
   bool checkSchedLimit();
 
   void findRootsAndBiasEdges(SmallVectorImpl<SUnit*> &TopRoots,
diff --git a/include/llvm/CodeGen/MachineTraceMetrics.h b/include/llvm/CodeGen/MachineTraceMetrics.h
index 2775a04..9794707 100644
--- a/include/llvm/CodeGen/MachineTraceMetrics.h
+++ b/include/llvm/CodeGen/MachineTraceMetrics.h
@@ -260,9 +260,12 @@ public:
     /// independent, exposing the maximum instruction-level parallelism.
     ///
     /// Any blocks in Extrablocks are included as if they were part of the
-    /// trace.
-    unsigned getResourceLength(ArrayRef<const MachineBasicBlock*> Extrablocks =
-                               ArrayRef<const MachineBasicBlock*>()) const;
+    /// trace. Likewise, extra resources required by the specified scheduling
+    /// classes are included. For the caller to account for extra machine
+    /// instructions, it must first resolve each instruction's scheduling class.
+    unsigned getResourceLength(
+                ArrayRef<const MachineBasicBlock*> Extrablocks = None,
+                ArrayRef<const MCSchedClassDesc*> ExtraInstrs = None) const;
 
     /// Return the length of the (data dependency) critical path through the
     /// trace.
diff --git a/include/llvm/CodeGen/Passes.h b/include/llvm/CodeGen/Passes.h
index fc8aa75..b02f63e 100644
--- a/include/llvm/CodeGen/Passes.h
+++ b/include/llvm/CodeGen/Passes.h
@@ -35,6 +35,48 @@ namespace llvm {
 
 class PassConfigImpl;
 
+/// Discriminated union of Pass ID types.
+///
+/// The PassConfig API prefers dealing with IDs because they are safer and more
+/// efficient. IDs decouple configuration from instantiation. This way, when a
+/// pass is overriden, it isn't unnecessarily instantiated. It is also unsafe to
+/// refer to a Pass pointer after adding it to a pass manager, which deletes
+/// redundant pass instances.
+///
+/// However, it is convient to directly instantiate target passes with
+/// non-default ctors. These often don't have a registered PassInfo. Rather than
+/// force all target passes to implement the pass registry boilerplate, allow
+/// the PassConfig API to handle either type.
+///
+/// AnalysisID is sadly char*, so PointerIntPair won't work.
+class IdentifyingPassPtr {
+  union {
+    AnalysisID ID;
+    Pass *P;
+  };
+  bool IsInstance;
+public:
+  IdentifyingPassPtr() : P(0), IsInstance(false) {}
+  IdentifyingPassPtr(AnalysisID IDPtr) : ID(IDPtr), IsInstance(false) {}
+  IdentifyingPassPtr(Pass *InstancePtr) : P(InstancePtr), IsInstance(true) {}
+
+  bool isValid() const { return P; }
+  bool isInstance() const { return IsInstance; }
+
+  AnalysisID getID() const {
+    assert(!IsInstance && "Not a Pass ID");
+    return ID;
+  }
+  Pass *getInstance() const {
+    assert(IsInstance && "Not a Pass Instance");
+    return P;
+  }
+};
+
+template <> struct isPodLike<IdentifyingPassPtr> {
+  static const bool value = true;
+};
+
 /// Target-Independent Code Generator Pass Configuration Options.
 ///
 /// This is an ImmutablePass solely for the purpose of exposing CodeGen options
@@ -117,20 +159,22 @@ public:
   /// Allow the target to override a specific pass without overriding the pass
   /// pipeline. When passes are added to the standard pipeline at the
   /// point where StandardID is expected, add TargetID in its place.
-  void substitutePass(AnalysisID StandardID, AnalysisID TargetID);
+  void substitutePass(AnalysisID StandardID, IdentifyingPassPtr TargetID);
 
   /// Insert InsertedPassID pass after TargetPassID pass.
-  void insertPass(AnalysisID TargetPassID, AnalysisID InsertedPassID);
+  void insertPass(AnalysisID TargetPassID, IdentifyingPassPtr InsertedPassID);
 
   /// Allow the target to enable a specific standard pass by default.
   void enablePass(AnalysisID PassID) { substitutePass(PassID, PassID); }
 
   /// Allow the target to disable a specific standard pass by default.
-  void disablePass(AnalysisID PassID) { substitutePass(PassID, 0); }
+  void disablePass(AnalysisID PassID) {
+    substitutePass(PassID, IdentifyingPassPtr());
+  }
 
   /// Return the pass substituted for StandardID by the target.
   /// If no substitution exists, return StandardID.
-  AnalysisID getPassSubstitution(AnalysisID StandardID) const;
+  IdentifyingPassPtr getPassSubstitution(AnalysisID StandardID) const;
 
   /// Return true if the optimized regalloc pipeline is enabled.
   bool getOptimizeRegAlloc() const;
@@ -222,17 +266,6 @@ protected:
     return false;
   }
 
-  /// addFinalizeRegAlloc - This method may be implemented by targets that want
-  /// to run passes within the regalloc pipeline, immediately after the register
-  /// allocation pass itself. These passes run as soon as virtual regisiters
-  /// have been rewritten to physical registers but before and other postRA
-  /// optimization happens. Targets that have marked instructions for bundling
-  /// must have finalized those bundles by the time these passes have run,
-  /// because subsequent passes are not guaranteed to be bundle-aware.
-  virtual bool addFinalizeRegAlloc() {
-    return false;
-  }
-
   /// addPostRegAlloc - This method may be implemented by targets that want to
   /// run passes after register allocation pass pipeline but before
   /// prolog-epilog insertion.  This should return true if -print-machineinstrs
diff --git a/include/llvm/CodeGen/RegAllocPBQP.h b/include/llvm/CodeGen/RegAllocPBQP.h
index b617c14..8b8e3d9 100644
--- a/include/llvm/CodeGen/RegAllocPBQP.h
+++ b/include/llvm/CodeGen/RegAllocPBQP.h
@@ -29,6 +29,7 @@ namespace llvm {
   class MachineFunction;
   class MachineLoopInfo;
   class TargetRegisterInfo;
+  template<class T> class OwningPtr;
 
   /// This class wraps up a PBQP instance representing a register allocation
   /// problem, plus the structures necessary to map back from the PBQP solution
@@ -123,11 +124,9 @@ namespace llvm {
 
     /// Build a PBQP instance to represent the register allocation problem for
     /// the given MachineFunction.
-    virtual std::auto_ptr<PBQPRAProblem> build(
-                                              MachineFunction *mf,
-                                              const LiveIntervals *lis,
-                                              const MachineLoopInfo *loopInfo,
-                                              const RegSet &vregs);
+    virtual PBQPRAProblem *build(MachineFunction *mf, const LiveIntervals *lis,
+                                 const MachineLoopInfo *loopInfo,
+                                 const RegSet &vregs);
   private:
 
     void addSpillCosts(PBQP::Vector &costVec, PBQP::PBQPNum spillCost);
@@ -144,11 +143,9 @@ namespace llvm {
  
     /// Build a PBQP instance to represent the register allocation problem for
     /// the given MachineFunction.
-    virtual std::auto_ptr<PBQPRAProblem> build(
-                                              MachineFunction *mf,
-                                              const LiveIntervals *lis,
-                                              const MachineLoopInfo *loopInfo,
-                                              const RegSet &vregs);   
+    virtual PBQPRAProblem *build(MachineFunction *mf, const LiveIntervals *lis,
+                                 const MachineLoopInfo *loopInfo,
+                                 const RegSet &vregs);   
 
   private:
 
@@ -161,7 +158,7 @@ namespace llvm {
                             PBQP::PBQPNum benefit);
   };
 
-  FunctionPass* createPBQPRegisterAllocator(std::auto_ptr<PBQPBuilder> builder,
+  FunctionPass* createPBQPRegisterAllocator(OwningPtr<PBQPBuilder> &builder,
                                             char *customPassID=0);
 }
 
diff --git a/include/llvm/CodeGen/ScheduleDAG.h b/include/llvm/CodeGen/ScheduleDAG.h
index 8c959da..7cff27e 100644
--- a/include/llvm/CodeGen/ScheduleDAG.h
+++ b/include/llvm/CodeGen/ScheduleDAG.h
@@ -302,6 +302,7 @@ namespace llvm {
     bool isCallOp         : 1;          // Is a function call operand.
     bool isTwoAddress     : 1;          // Is a two-address instruction.
     bool isCommutable     : 1;          // Is a commutable instruction.
+    bool hasPhysRegUses   : 1;          // Has physreg uses.
     bool hasPhysRegDefs   : 1;          // Has physreg defs that are being used.
     bool hasPhysRegClobbers : 1;        // Has any physreg defs, used or not.
     bool isPending        : 1;          // True once pending.
@@ -331,10 +332,10 @@ namespace llvm {
         NodeQueueId(0), NumPreds(0), NumSuccs(0), NumPredsLeft(0),
         NumSuccsLeft(0), WeakPredsLeft(0), WeakSuccsLeft(0), NumRegDefsLeft(0),
         Latency(0), isVRegCycle(false), isCall(false), isCallOp(false),
-        isTwoAddress(false), isCommutable(false), hasPhysRegDefs(false),
-        hasPhysRegClobbers(false), isPending(false), isAvailable(false),
-        isScheduled(false), isScheduleHigh(false), isScheduleLow(false),
-        isCloned(false), SchedulingPref(Sched::None),
+        isTwoAddress(false), isCommutable(false), hasPhysRegUses(false),
+        hasPhysRegDefs(false), hasPhysRegClobbers(false), isPending(false),
+        isAvailable(false), isScheduled(false), isScheduleHigh(false),
+        isScheduleLow(false), isCloned(false), SchedulingPref(Sched::None),
         isDepthCurrent(false), isHeightCurrent(false), Depth(0), Height(0),
         TopReadyCycle(0), BotReadyCycle(0), CopyDstRC(NULL), CopySrcRC(NULL) {}
 
@@ -345,10 +346,10 @@ namespace llvm {
         NodeQueueId(0), NumPreds(0), NumSuccs(0), NumPredsLeft(0),
         NumSuccsLeft(0), WeakPredsLeft(0), WeakSuccsLeft(0), NumRegDefsLeft(0),
         Latency(0), isVRegCycle(false), isCall(false), isCallOp(false),
-        isTwoAddress(false), isCommutable(false), hasPhysRegDefs(false),
-        hasPhysRegClobbers(false), isPending(false), isAvailable(false),
-        isScheduled(false), isScheduleHigh(false), isScheduleLow(false),
-        isCloned(false), SchedulingPref(Sched::None),
+        isTwoAddress(false), isCommutable(false), hasPhysRegUses(false),
+        hasPhysRegDefs(false), hasPhysRegClobbers(false), isPending(false),
+        isAvailable(false), isScheduled(false), isScheduleHigh(false),
+        isScheduleLow(false), isCloned(false), SchedulingPref(Sched::None),
         isDepthCurrent(false), isHeightCurrent(false), Depth(0), Height(0),
         TopReadyCycle(0), BotReadyCycle(0), CopyDstRC(NULL), CopySrcRC(NULL) {}
 
@@ -358,10 +359,10 @@ namespace llvm {
         NodeQueueId(0), NumPreds(0), NumSuccs(0), NumPredsLeft(0),
         NumSuccsLeft(0), WeakPredsLeft(0), WeakSuccsLeft(0), NumRegDefsLeft(0),
         Latency(0), isVRegCycle(false), isCall(false), isCallOp(false),
-        isTwoAddress(false), isCommutable(false), hasPhysRegDefs(false),
-        hasPhysRegClobbers(false), isPending(false), isAvailable(false),
-        isScheduled(false), isScheduleHigh(false), isScheduleLow(false),
-        isCloned(false), SchedulingPref(Sched::None),
+        isTwoAddress(false), isCommutable(false), hasPhysRegUses(false),
+        hasPhysRegDefs(false), hasPhysRegClobbers(false), isPending(false),
+        isAvailable(false), isScheduled(false), isScheduleHigh(false),
+        isScheduleLow(false), isCloned(false), SchedulingPref(Sched::None),
         isDepthCurrent(false), isHeightCurrent(false), Depth(0), Height(0),
         TopReadyCycle(0), BotReadyCycle(0), CopyDstRC(NULL), CopySrcRC(NULL) {}
 
@@ -726,9 +727,8 @@ namespace llvm {
     /// IsReachable - Checks if SU is reachable from TargetSU.
     bool IsReachable(const SUnit *SU, const SUnit *TargetSU);
 
-    /// WillCreateCycle - Returns true if adding an edge from SU to TargetSU
-    /// will create a cycle.
-    bool WillCreateCycle(SUnit *SU, SUnit *TargetSU);
+    /// WillCreateCycle - Return true if addPred(TargetSU, SU) creates a cycle.
+    bool WillCreateCycle(SUnit *TargetSU, SUnit *SU);
 
     /// AddPred - Updates the topological ordering to accommodate an edge
     /// to be added from SUnit X to SUnit Y.
diff --git a/include/llvm/CodeGen/ScheduleDAGInstrs.h b/include/llvm/CodeGen/ScheduleDAGInstrs.h
index 2219520..990cac6 100644
--- a/include/llvm/CodeGen/ScheduleDAGInstrs.h
+++ b/include/llvm/CodeGen/ScheduleDAGInstrs.h
@@ -105,6 +105,10 @@ namespace llvm {
     MachineBasicBlock::iterator RegionEnd;
 
     /// The index in BB of RegionEnd.
+    ///
+    /// This is the instruction number from the top of the current block, not
+    /// the SlotIndex. It is only used by the AntiDepBreaker and should be
+    /// removed once that client is obsolete.
     unsigned EndIndex;
 
     /// After calling BuildSchedGraph, each machine instruction in the current
@@ -146,6 +150,9 @@ namespace llvm {
 
     virtual ~ScheduleDAGInstrs() {}
 
+    /// \brief Expose LiveIntervals for use in DAG mutators and such.
+    LiveIntervals *getLIS() const { return LIS; }
+
     /// \brief Get the machine model for instruction scheduling.
     const TargetSchedModel *getSchedModel() const { return &SchedModel; }
 
diff --git a/include/llvm/CodeGen/SelectionDAG.h b/include/llvm/CodeGen/SelectionDAG.h
index e5adf67..8c064bf 100644
--- a/include/llvm/CodeGen/SelectionDAG.h
+++ b/include/llvm/CodeGen/SelectionDAG.h
@@ -810,31 +810,32 @@ public:
   MachineSDNode *getMachineNode(unsigned Opcode, DebugLoc dl, EVT VT,
                                 SDValue Op1, SDValue Op2);
   MachineSDNode *getMachineNode(unsigned Opcode, DebugLoc dl, EVT VT,
-                         SDValue Op1, SDValue Op2, SDValue Op3);
+                                SDValue Op1, SDValue Op2, SDValue Op3);
   MachineSDNode *getMachineNode(unsigned Opcode, DebugLoc dl, EVT VT,
-                         const SDValue *Ops, unsigned NumOps);
+                                ArrayRef<SDValue> Ops);
   MachineSDNode *getMachineNode(unsigned Opcode, DebugLoc dl, EVT VT1, EVT VT2);
   MachineSDNode *getMachineNode(unsigned Opcode, DebugLoc dl, EVT VT1, EVT VT2,
-                         SDValue Op1);
-  MachineSDNode *getMachineNode(unsigned Opcode, DebugLoc dl, EVT VT1,
-                         EVT VT2, SDValue Op1, SDValue Op2);
-  MachineSDNode *getMachineNode(unsigned Opcode, DebugLoc dl, EVT VT1,
-                         EVT VT2, SDValue Op1, SDValue Op2, SDValue Op3);
+                                SDValue Op1);
+  MachineSDNode *getMachineNode(unsigned Opcode, DebugLoc dl, EVT VT1, EVT VT2,
+                                SDValue Op1, SDValue Op2);
+  MachineSDNode *getMachineNode(unsigned Opcode, DebugLoc dl, EVT VT1, EVT VT2,
+                                SDValue Op1, SDValue Op2, SDValue Op3);
   MachineSDNode *getMachineNode(unsigned Opcode, DebugLoc dl, EVT VT1, EVT VT2,
-                         const SDValue *Ops, unsigned NumOps);
+                                ArrayRef<SDValue> Ops);
   MachineSDNode *getMachineNode(unsigned Opcode, DebugLoc dl, EVT VT1, EVT VT2,
-                         EVT VT3, SDValue Op1, SDValue Op2);
+                                EVT VT3, SDValue Op1, SDValue Op2);
   MachineSDNode *getMachineNode(unsigned Opcode, DebugLoc dl, EVT VT1, EVT VT2,
-                         EVT VT3, SDValue Op1, SDValue Op2, SDValue Op3);
+                                EVT VT3, SDValue Op1, SDValue Op2,
+                                SDValue Op3);
   MachineSDNode *getMachineNode(unsigned Opcode, DebugLoc dl, EVT VT1, EVT VT2,
-                         EVT VT3, const SDValue *Ops, unsigned NumOps);
+                                EVT VT3, ArrayRef<SDValue> Ops);
   MachineSDNode *getMachineNode(unsigned Opcode, DebugLoc dl, EVT VT1, EVT VT2,
-                         EVT VT3, EVT VT4, const SDValue *Ops, unsigned NumOps);
+                                EVT VT3, EVT VT4, ArrayRef<SDValue> Ops);
   MachineSDNode *getMachineNode(unsigned Opcode, DebugLoc dl,
-                         ArrayRef<EVT> ResultTys, const SDValue *Ops,
-                         unsigned NumOps);
+                                ArrayRef<EVT> ResultTys,
+                                ArrayRef<SDValue> Ops);
   MachineSDNode *getMachineNode(unsigned Opcode, DebugLoc dl, SDVTList VTs,
-                         const SDValue *Ops, unsigned NumOps);
+                                ArrayRef<SDValue> Ops);
 
   /// getTargetExtractSubreg - A convenience function for creating
   /// TargetInstrInfo::EXTRACT_SUBREG nodes.
diff --git a/include/llvm/CodeGen/SelectionDAGISel.h b/include/llvm/CodeGen/SelectionDAGISel.h
index 5f503de..a4721db 100644
--- a/include/llvm/CodeGen/SelectionDAGISel.h
+++ b/include/llvm/CodeGen/SelectionDAGISel.h
@@ -259,9 +259,6 @@ private:
   void SelectBasicBlock(BasicBlock::const_iterator Begin,
                         BasicBlock::const_iterator End,
                         bool &HadTailCall);
-
-  bool TryToFoldFastISelLoad(const LoadInst *LI, const Instruction *FoldInst,
-                             FastISel *FastIS);
   void FinishBasicBlock();
 
   void CodeGenAndEmitDAG();
diff --git a/include/llvm/CodeGen/SlotIndexes.h b/include/llvm/CodeGen/SlotIndexes.h
index a277080..26d0433 100644
--- a/include/llvm/CodeGen/SlotIndexes.h
+++ b/include/llvm/CodeGen/SlotIndexes.h
@@ -53,6 +53,20 @@ namespace llvm {
       this->index = index;
     }
 
+#ifdef EXPENSIVE_CHECKS
+    // When EXPENSIVE_CHECKS is defined, "erased" index list entries will
+    // actually be moved to a "graveyard" list, and have their pointers
+    // poisoned, so that dangling SlotIndex access can be reliably detected.
+    void setPoison() {
+      intptr_t tmp = reinterpret_cast<intptr_t>(mi);
+      assert(((tmp & 0x1) == 0x0) && "Pointer already poisoned?");  
+      tmp |= 0x1;
+      mi = reinterpret_cast<MachineInstr*>(tmp);
+    }
+
+    bool isPoisoned() const { return (reinterpret_cast<intptr_t>(mi) & 0x1) == 0x1; }
+#endif // EXPENSIVE_CHECKS
+
   };
 
   template <>
@@ -109,6 +123,10 @@ namespace llvm {
 
     IndexListEntry* listEntry() const {
       assert(isValid() && "Attempt to compare reserved index.");
+#ifdef EXPENSIVE_CHECKS
+      assert(!lie.getPointer()->isPoisoned() &&
+             "Attempt to access deleted list-entry.");
+#endif // EXPENSIVE_CHECKS
       return lie.getPointer();
     }
 
@@ -282,7 +300,6 @@ namespace llvm {
 
   template <> struct isPodLike<SlotIndex> { static const bool value = true; };
 
-
   inline raw_ostream& operator<<(raw_ostream &os, SlotIndex li) {
     li.print(os);
     return os;
@@ -313,6 +330,10 @@ namespace llvm {
     typedef ilist<IndexListEntry> IndexList;
     IndexList indexList;
 
+#ifdef EXPENSIVE_CHECKS
+    IndexList graveyardList;
+#endif // EXPENSIVE_CHECKS
+
     MachineFunction *mf;
 
     typedef DenseMap<const MachineInstr*, SlotIndex> Mi2IndexMap;
@@ -643,6 +664,32 @@ namespace llvm {
       std::sort(idx2MBBMap.begin(), idx2MBBMap.end(), Idx2MBBCompare());
     }
 
+    /// \brief Free the resources that were required to maintain a SlotIndex.
+    ///
+    /// Once an index is no longer needed (for instance because the instruction
+    /// at that index has been moved), the resources required to maintain the
+    /// index can be relinquished to reduce memory use and improve renumbering
+    /// performance. Any remaining SlotIndex objects that point to the same
+    /// index are left 'dangling' (much the same as a dangling pointer to a
+    /// freed object) and should not be accessed, except to destruct them.
+    /// 
+    /// Like dangling pointers, access to dangling SlotIndexes can cause
+    /// painful-to-track-down bugs, especially if the memory for the index
+    /// previously pointed to has been re-used. To detect dangling SlotIndex
+    /// bugs, build with EXPENSIVE_CHECKS=1. This will cause "erased" indexes to
+    /// be retained in a graveyard instead of being freed. Operations on indexes
+    /// in the graveyard will trigger an assertion.
+    void eraseIndex(SlotIndex index) {
+      IndexListEntry *entry = index.listEntry();
+#ifdef EXPENSIVE_CHECKS
+      indexList.remove(entry);
+      graveyardList.push_back(entry);
+      entry->setPoison();
+#else
+      indexList.erase(entry);
+#endif
+    }
+
   };
 
 
diff --git a/include/llvm/CodeGen/TargetLoweringObjectFileImpl.h b/include/llvm/CodeGen/TargetLoweringObjectFileImpl.h
index e7098e4..5b22c9c 100644
--- a/include/llvm/CodeGen/TargetLoweringObjectFileImpl.h
+++ b/include/llvm/CodeGen/TargetLoweringObjectFileImpl.h
@@ -128,6 +128,12 @@ public:
   virtual const MCSection *
   SelectSectionForGlobal(const GlobalValue *GV, SectionKind Kind,
                          Mangler *Mang, const TargetMachine &TM) const;
+
+  /// emitModuleFlags - Emit Obj-C garbage collection and linker options.  Only
+  /// linker option emission is implemented for COFF.
+  virtual void emitModuleFlags(MCStreamer &Streamer,
+                               ArrayRef<Module::ModuleFlagEntry> ModuleFlags,
+                               Mangler *Mang, const TargetMachine &TM) const;
 };
 
 } // end namespace llvm
diff --git a/include/llvm/CodeGen/ValueTypes.td b/include/llvm/CodeGen/ValueTypes.td
index 76df6ac..da26985 100644
--- a/include/llvm/CodeGen/ValueTypes.td
+++ b/include/llvm/CodeGen/ValueTypes.td
@@ -44,13 +44,13 @@ def v4i8   : ValueType<32 , 20>;   //  4 x i8  vector value
 def v8i8   : ValueType<64 , 21>;   //  8 x i8  vector value
 def v16i8  : ValueType<128, 22>;   // 16 x i8  vector value
 def v32i8  : ValueType<256, 23>;   // 32 x i8 vector value
-def v64i8  : ValueType<256, 24>;   // 64 x i8 vector value
+def v64i8  : ValueType<512, 24>;   // 64 x i8 vector value
 def v1i16  : ValueType<16 , 25>;   //  1 x i16 vector value
 def v2i16  : ValueType<32 , 26>;   //  2 x i16 vector value
 def v4i16  : ValueType<64 , 27>;   //  4 x i16 vector value
 def v8i16  : ValueType<128, 28>;   //  8 x i16 vector value
 def v16i16 : ValueType<256, 29>;   // 16 x i16 vector value
-def v32i16 : ValueType<256, 30>;   // 32 x i16 vector value
+def v32i16 : ValueType<512, 30>;   // 32 x i16 vector value
 def v1i32  : ValueType<32 , 31>;   //  1 x i32 vector value
 def v2i32  : ValueType<64 , 32>;   //  2 x i32 vector value
 def v4i32  : ValueType<128, 33>;   //  4 x i32 vector value