diff --git a/tools/llvm-diff/CMakeLists.txt b/tools/llvm-diff/CMakeLists.txt
new file mode 100644
index 0000000..f6d65c9
--- /dev/null
+++ b/tools/llvm-diff/CMakeLists.txt
@@ -0,0 +1,6 @@
+set(LLVM_LINK_COMPONENTS support asmparser bitreader)
+
+add_llvm_tool(llvm-diff
+  llvm-diff.cpp
+  DifferenceEngine.cpp
+  )
diff --git a/tools/llvm-diff/DifferenceEngine.cpp b/tools/llvm-diff/DifferenceEngine.cpp
new file mode 100644
index 0000000..400ab9f
--- /dev/null
+++ b/tools/llvm-diff/DifferenceEngine.cpp
@@ -0,0 +1,600 @@
+//===-- DifferenceEngine.cpp - Structural function/module comparison ------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This header defines the interface to the LLVM difference engine,
+// which structurally compares functions within a module.
+//
+//===----------------------------------------------------------------------===//
+
+#include <utility>
+
+#include <llvm/ADT/DenseMap.h>
+#include <llvm/ADT/DenseSet.h>
+#include <llvm/ADT/SmallVector.h>
+#include <llvm/ADT/StringRef.h>
+#include <llvm/ADT/StringSet.h>
+
+#include <llvm/Module.h>
+#include <llvm/Function.h>
+#include <llvm/Instructions.h>
+#include <llvm/Support/CFG.h>
+
+#include <llvm/Support/raw_ostream.h>
+#include <llvm/Support/type_traits.h>
+#include <llvm/Support/ErrorHandling.h>
+#include <llvm/Support/CallSite.h>
+
+#include "DifferenceEngine.h"
+
+using namespace llvm;
+
+namespace {
+
+/// A priority queue, implemented as a heap.
+template <class T, class Sorter, unsigned InlineCapacity>
+class PriorityQueue {
+  Sorter Precedes;
+  llvm::SmallVector<T, InlineCapacity> Storage;
+
+public:
+  PriorityQueue(const Sorter &Precedes) : Precedes(Precedes) {}
+
+  /// Checks whether the heap is empty.
+  bool empty() const { return Storage.empty(); }
+
+  /// Insert a new value on the heap.
+  void insert(const T &V) {
+    unsigned Index = Storage.size();
+    Storage.push_back(V);
+    if (Index == 0) return;
+
+    T *data = Storage.data();
+    while (true) {
+      unsigned Target = (Index + 1) / 2 - 1;
+      if (!Precedes(data[Index], data[Target])) return;
+      std::swap(data[Index], data[Target]);
+      if (Target == 0) return;
+      Index = Target;
+    }
+  }
+
+  /// Remove the minimum value in the heap.  Only valid on a non-empty heap.
+  T remove_min() {
+    assert(!empty());
+    T tmp = Storage[0];
+    
+    unsigned NewSize = Storage.size() - 1;
+    if (NewSize) {
+      // Move the slot at the end to the beginning.
+      if (isPodLike<T>::value)
+        Storage[0] = Storage[NewSize];
+      else
+        std::swap(Storage[0], Storage[NewSize]);
+
+      // Bubble the root up as necessary.
+      unsigned Index = 0;
+      while (true) {
+        // With a 1-based index, the children would be Index*2 and Index*2+1.
+        unsigned R = (Index + 1) * 2;
+        unsigned L = R - 1;
+
+        // If R is out of bounds, we're done after this in any case.
+        if (R >= NewSize) {
+          // If L is also out of bounds, we're done immediately.
+          if (L >= NewSize) break;
+
+          // Otherwise, test whether we should swap L and Index.
+          if (Precedes(Storage[L], Storage[Index]))
+            std::swap(Storage[L], Storage[Index]);
+          break;
+        }
+
+        // Otherwise, we need to compare with the smaller of L and R.
+        // Prefer R because it's closer to the end of the array.
+        unsigned IndexToTest = (Precedes(Storage[L], Storage[R]) ? L : R);
+
+        // If Index is >= the min of L and R, then heap ordering is restored.
+        if (!Precedes(Storage[IndexToTest], Storage[Index]))
+          break;
+
+        // Otherwise, keep bubbling up.
+        std::swap(Storage[IndexToTest], Storage[Index]);
+        Index = IndexToTest;
+      }
+    }
+    Storage.pop_back();
+
+    return tmp;
+  }
+};
+
+/// A function-scope difference engine.
+class FunctionDifferenceEngine {
+  DifferenceEngine &Engine;
+
+  /// The current mapping from old local values to new local values.
+  DenseMap<Value*, Value*> Values;
+
+  /// The current mapping from old blocks to new blocks.
+  DenseMap<BasicBlock*, BasicBlock*> Blocks;
+
+  DenseSet<std::pair<Value*, Value*> > TentativeValuePairs;
+
+  unsigned getUnprocPredCount(BasicBlock *Block) const {
+    unsigned Count = 0;
+    for (pred_iterator I = pred_begin(Block), E = pred_end(Block); I != E; ++I)
+      if (!Blocks.count(*I)) Count++;
+    return Count;
+  }
+
+  typedef std::pair<BasicBlock*, BasicBlock*> BlockPair;
+
+  /// A type which sorts a priority queue by the number of unprocessed
+  /// predecessor blocks it has remaining.
+  ///
+  /// This is actually really expensive to calculate.
+  struct QueueSorter {
+    const FunctionDifferenceEngine &fde;
+    explicit QueueSorter(const FunctionDifferenceEngine &fde) : fde(fde) {}
+
+    bool operator()(const BlockPair &Old, const BlockPair &New) {
+      return fde.getUnprocPredCount(Old.first)
+           < fde.getUnprocPredCount(New.first);
+    }
+  };
+
+  /// A queue of unified blocks to process.
+  PriorityQueue<BlockPair, QueueSorter, 20> Queue;
+
+  /// Try to unify the given two blocks.  Enqueues them for processing
+  /// if they haven't already been processed.
+  ///
+  /// Returns true if there was a problem unifying them.
+  bool tryUnify(BasicBlock *L, BasicBlock *R) {
+    BasicBlock *&Ref = Blocks[L];
+
+    if (Ref) {
+      if (Ref == R) return false;
+
+      Engine.logf("successor %s cannot be equivalent to %s; "
+                  "it's already equivalent to %s")
+        << L << R << Ref;
+      return true;
+    }
+
+    Ref = R;
+    Queue.insert(BlockPair(L, R));
+    return false;
+  }
+
+  void processQueue() {
+    while (!Queue.empty()) {
+      BlockPair Pair = Queue.remove_min();
+      diff(Pair.first, Pair.second);
+    }
+  }
+
+  void diff(BasicBlock *L, BasicBlock *R) {
+    DifferenceEngine::Context C(Engine, L, R);
+
+    BasicBlock::iterator LI = L->begin(), LE = L->end();
+    BasicBlock::iterator RI = R->begin(), RE = R->end();
+
+    do {
+      assert(LI != LE && RI != RE);
+      Instruction *LeftI = &*LI, *RightI = &*RI;
+
+      // If the instructions differ, start the more sophisticated diff
+      // algorithm here.
+      if (diff(LeftI, RightI, false, true))
+        return runBlockDiff(LI, RI);
+
+      // Otherwise, unify them.
+      if (!LeftI->use_empty())
+        Values[LeftI] = RightI;
+
+      ++LI, ++RI;
+    } while (LI != LE); // This is sufficient: we can't get equality of
+                        // terminators if there are residual instructions.
+  }
+
+  bool matchForBlockDiff(Instruction *L, Instruction *R);
+  void runBlockDiff(BasicBlock::iterator LI, BasicBlock::iterator RI);
+
+  bool diffCallSites(CallSite L, CallSite R, bool Complain) {
+    // FIXME: call attributes
+    if (!equivalentAsOperands(L.getCalledValue(), R.getCalledValue())) {
+      if (Complain) Engine.log("called functions differ");
+      return true;
+    }
+    if (L.arg_size() != R.arg_size()) {
+      if (Complain) Engine.log("argument counts differ");
+      return true;
+    }
+    for (unsigned I = 0, E = L.arg_size(); I != E; ++I)
+      if (!equivalentAsOperands(L.getArgument(I), R.getArgument(I))) {
+        if (Complain)
+          Engine.logf("arguments %s and %s differ")
+            << L.getArgument(I) << R.getArgument(I);
+        return true;
+      }
+    return false;
+  }
+
+  bool diff(Instruction *L, Instruction *R, bool Complain, bool TryUnify) {
+    // FIXME: metadata (if Complain is set)
+
+    // Different opcodes always imply different operations.
+    if (L->getOpcode() != R->getOpcode()) {
+      if (Complain) Engine.log("different instruction types");
+      return true;
+    }
+
+    if (isa<CmpInst>(L)) {
+      if (cast<CmpInst>(L)->getPredicate()
+            != cast<CmpInst>(R)->getPredicate()) {
+        if (Complain) Engine.log("different predicates");
+        return true;
+      }
+    } else if (isa<CallInst>(L)) {
+      return diffCallSites(CallSite(L), CallSite(R), Complain);
+    } else if (isa<PHINode>(L)) {
+      // FIXME: implement.
+
+      // This is really wierd;  type uniquing is broken?
+      if (L->getType() != R->getType()) {
+        if (!L->getType()->isPointerTy() || !R->getType()->isPointerTy()) {
+          if (Complain) Engine.log("different phi types");
+          return true;
+        }
+      }
+      return false;
+
+    // Terminators.
+    } else if (isa<InvokeInst>(L)) {
+      InvokeInst *LI = cast<InvokeInst>(L);
+      InvokeInst *RI = cast<InvokeInst>(R);
+      if (diffCallSites(CallSite(LI), CallSite(RI), Complain))
+        return true;
+
+      if (TryUnify) {
+        tryUnify(LI->getNormalDest(), RI->getNormalDest());
+        tryUnify(LI->getUnwindDest(), RI->getUnwindDest());
+      }
+      return false;
+
+    } else if (isa<BranchInst>(L)) {
+      BranchInst *LI = cast<BranchInst>(L);
+      BranchInst *RI = cast<BranchInst>(R);
+      if (LI->isConditional() != RI->isConditional()) {
+        if (Complain) Engine.log("branch conditionality differs");
+        return true;
+      }
+
+      if (LI->isConditional()) {
+        if (!equivalentAsOperands(LI->getCondition(), RI->getCondition())) {
+          if (Complain) Engine.log("branch conditions differ");
+          return true;
+        }
+        if (TryUnify) tryUnify(LI->getSuccessor(1), RI->getSuccessor(1));
+      }
+      if (TryUnify) tryUnify(LI->getSuccessor(0), RI->getSuccessor(0));
+      return false;
+
+    } else if (isa<SwitchInst>(L)) {
+      SwitchInst *LI = cast<SwitchInst>(L);
+      SwitchInst *RI = cast<SwitchInst>(R);
+      if (!equivalentAsOperands(LI->getCondition(), RI->getCondition())) {
+        if (Complain) Engine.log("switch conditions differ");
+        return true;
+      }
+      if (TryUnify) tryUnify(LI->getDefaultDest(), RI->getDefaultDest());
+
+      bool Difference = false;
+
+      DenseMap<ConstantInt*,BasicBlock*> LCases;
+      for (unsigned I = 1, E = LI->getNumCases(); I != E; ++I)
+        LCases[LI->getCaseValue(I)] = LI->getSuccessor(I);
+      for (unsigned I = 1, E = RI->getNumCases(); I != E; ++I) {
+        ConstantInt *CaseValue = RI->getCaseValue(I);
+        BasicBlock *LCase = LCases[CaseValue];
+        if (LCase) {
+          if (TryUnify) tryUnify(LCase, RI->getSuccessor(I));
+          LCases.erase(CaseValue);
+        } else if (!Difference) {
+          if (Complain)
+            Engine.logf("right switch has extra case %s") << CaseValue;
+          Difference = true;
+        }
+      }
+      if (!Difference)
+        for (DenseMap<ConstantInt*,BasicBlock*>::iterator
+               I = LCases.begin(), E = LCases.end(); I != E; ++I) {
+          if (Complain)
+            Engine.logf("left switch has extra case %s") << I->first;
+          Difference = true;
+        }
+      return Difference;
+    } else if (isa<UnreachableInst>(L)) {
+      return false;
+    }
+
+    if (L->getNumOperands() != R->getNumOperands()) {
+      if (Complain) Engine.log("instructions have different operand counts");
+      return true;
+    }
+
+    for (unsigned I = 0, E = L->getNumOperands(); I != E; ++I) {
+      Value *LO = L->getOperand(I), *RO = R->getOperand(I);
+      if (!equivalentAsOperands(LO, RO)) {
+        if (Complain) Engine.logf("operands %s and %s differ") << LO << RO;
+        return true;
+      }
+    }
+
+    return false;
+  }
+
+  bool equivalentAsOperands(Constant *L, Constant *R) {
+    // Use equality as a preliminary filter.
+    if (L == R)
+      return true;
+
+    if (L->getValueID() != R->getValueID())
+      return false;
+    
+    // Ask the engine about global values.
+    if (isa<GlobalValue>(L))
+      return Engine.equivalentAsOperands(cast<GlobalValue>(L),
+                                         cast<GlobalValue>(R));
+
+    // Compare constant expressions structurally.
+    if (isa<ConstantExpr>(L))
+      return equivalentAsOperands(cast<ConstantExpr>(L),
+                                  cast<ConstantExpr>(R));
+
+    // Nulls of the "same type" don't always actually have the same
+    // type; I don't know why.  Just white-list them.
+    if (isa<ConstantPointerNull>(L))
+      return true;
+
+    // Block addresses only match if we've already encountered the
+    // block.  FIXME: tentative matches?
+    if (isa<BlockAddress>(L))
+      return Blocks[cast<BlockAddress>(L)->getBasicBlock()]
+                 == cast<BlockAddress>(R)->getBasicBlock();
+
+    return false;
+  }
+
+  bool equivalentAsOperands(ConstantExpr *L, ConstantExpr *R) {
+    if (L == R)
+      return true;
+    if (L->getOpcode() != R->getOpcode())
+      return false;
+
+    switch (L->getOpcode()) {
+    case Instruction::ICmp:
+    case Instruction::FCmp:
+      if (L->getPredicate() != R->getPredicate())
+        return false;
+      break;
+
+    case Instruction::GetElementPtr:
+      // FIXME: inbounds?
+      break;
+
+    default:
+      break;
+    }
+
+    if (L->getNumOperands() != R->getNumOperands())
+      return false;
+
+    for (unsigned I = 0, E = L->getNumOperands(); I != E; ++I)
+      if (!equivalentAsOperands(L->getOperand(I), R->getOperand(I)))
+        return false;
+
+    return true;
+  }
+
+  bool equivalentAsOperands(Value *L, Value *R) {
+    // Fall out if the values have different kind.
+    // This possibly shouldn't take priority over oracles.
+    if (L->getValueID() != R->getValueID())
+      return false;
+
+    // Value subtypes:  Argument, Constant, Instruction, BasicBlock,
+    //                  InlineAsm, MDNode, MDString, PseudoSourceValue
+
+    if (isa<Constant>(L))
+      return equivalentAsOperands(cast<Constant>(L), cast<Constant>(R));
+
+    if (isa<Instruction>(L))
+      return Values[L] == R || TentativeValuePairs.count(std::make_pair(L, R));
+
+    if (isa<Argument>(L))
+      return Values[L] == R;
+
+    if (isa<BasicBlock>(L))
+      return Blocks[cast<BasicBlock>(L)] != R;
+
+    // Pretend everything else is identical.
+    return true;
+  }
+
+  // Avoid a gcc warning about accessing 'this' in an initializer.
+  FunctionDifferenceEngine *this_() { return this; }
+
+public:
+  FunctionDifferenceEngine(DifferenceEngine &Engine) :
+    Engine(Engine), Queue(QueueSorter(*this_())) {}
+
+  void diff(Function *L, Function *R) {
+    if (L->arg_size() != R->arg_size())
+      Engine.log("different argument counts");
+
+    // Map the arguments.
+    for (Function::arg_iterator
+           LI = L->arg_begin(), LE = L->arg_end(),
+           RI = R->arg_begin(), RE = R->arg_end();
+         LI != LE && RI != RE; ++LI, ++RI)
+      Values[&*LI] = &*RI;
+
+    tryUnify(&*L->begin(), &*R->begin());
+    processQueue();
+  }
+};
+
+struct DiffEntry {
+  DiffEntry() : Cost(0) {}
+
+  unsigned Cost;
+  llvm::SmallVector<char, 8> Path; // actually of DifferenceEngine::DiffChange
+};
+
+bool FunctionDifferenceEngine::matchForBlockDiff(Instruction *L,
+                                                 Instruction *R) {
+  return !diff(L, R, false, false);
+}
+
+void FunctionDifferenceEngine::runBlockDiff(BasicBlock::iterator LStart,
+                                            BasicBlock::iterator RStart) {
+  BasicBlock::iterator LE = LStart->getParent()->end();
+  BasicBlock::iterator RE = RStart->getParent()->end();
+
+  unsigned NL = std::distance(LStart, LE);
+
+  SmallVector<DiffEntry, 20> Paths1(NL+1);
+  SmallVector<DiffEntry, 20> Paths2(NL+1);
+
+  DiffEntry *Cur = Paths1.data();
+  DiffEntry *Next = Paths2.data();
+
+  assert(TentativeValuePairs.empty());
+
+  // Initialize the first column.
+  for (unsigned I = 0; I != NL+1; ++I) {
+    Cur[I].Cost = I;
+    for (unsigned J = 0; J != I; ++J)
+      Cur[I].Path.push_back(DifferenceEngine::DC_left);
+  }
+
+  for (BasicBlock::iterator RI = RStart; RI != RE; ++RI) {
+    // Initialize the first row.
+    Next[0] = Cur[0];
+    Next[0].Path.push_back(DifferenceEngine::DC_right);
+
+    unsigned Index = 1;
+    for (BasicBlock::iterator LI = LStart; LI != LE; ++LI, ++Index) {
+      if (matchForBlockDiff(&*LI, &*RI)) {
+        Next[Index] = Cur[Index-1];
+        Next[Index].Path.push_back(DifferenceEngine::DC_match);
+        TentativeValuePairs.insert(std::make_pair(&*LI, &*RI));
+      } else if (Next[Index-1].Cost <= Cur[Index].Cost) {
+        Next[Index] = Next[Index-1];
+        Next[Index].Path.push_back(DifferenceEngine::DC_left);
+      } else {
+        Next[Index] = Cur[Index];
+        Next[Index].Path.push_back(DifferenceEngine::DC_right);
+      }
+    }
+
+    std::swap(Cur, Next);
+  }
+
+  SmallVectorImpl<char> &Path = Cur[NL].Path;
+  BasicBlock::iterator LI = LStart, RI = RStart;
+
+  DifferenceEngine::DiffLogBuilder Diff(Engine);
+
+  // Drop trailing matches.
+  while (Path.back() == DifferenceEngine::DC_match)
+    Path.pop_back();
+
+  for (SmallVectorImpl<char>::iterator
+         PI = Path.begin(), PE = Path.end(); PI != PE; ++PI) {
+    switch (static_cast<DifferenceEngine::DiffChange>(*PI)) {
+    case DifferenceEngine::DC_match:
+      assert(LI != LE && RI != RE);
+      {
+        Instruction *L = &*LI, *R = &*RI;
+        DifferenceEngine::Context C(Engine, L, R);
+        diff(L, R, false, true);
+        Diff.addMatch(L, R);
+      }
+      ++LI; ++RI;
+      break;
+
+    case DifferenceEngine::DC_left:
+      assert(LI != LE);
+      Diff.addLeft(&*LI);
+      ++LI;
+      break;
+
+    case DifferenceEngine::DC_right:
+      assert(RI != RE);
+      Diff.addRight(&*RI);
+      ++RI;
+      break;
+    }
+  }
+
+  TentativeValuePairs.clear();
+}
+
+}
+
+void DifferenceEngine::diff(Function *L, Function *R) {
+  Context C(*this, L, R);
+
+  // FIXME: types
+  // FIXME: attributes and CC
+  // FIXME: parameter attributes
+  
+  // If both are declarations, we're done.
+  if (L->empty() && R->empty())
+    return;
+  else if (L->empty())
+    log("left function is declaration, right function is definition");
+  else if (R->empty())
+    log("right function is declaration, left function is definition");
+  else
+    FunctionDifferenceEngine(*this).diff(L, R);
+}
+
+void DifferenceEngine::diff(Module *L, Module *R) {
+  StringSet<> LNames;
+  SmallVector<std::pair<Function*,Function*>, 20> Queue;
+
+  for (Module::iterator I = L->begin(), E = L->end(); I != E; ++I) {
+    Function *LFn = &*I;
+    LNames.insert(LFn->getName());
+
+    if (Function *RFn = R->getFunction(LFn->getName()))
+      Queue.push_back(std::make_pair(LFn, RFn));
+    else
+      logf("function %s exists only in left module") << LFn;
+  }
+
+  for (Module::iterator I = R->begin(), E = R->end(); I != E; ++I) {
+    Function *RFn = &*I;
+    if (!LNames.count(RFn->getName()))
+      logf("function %s exists only in right module") << RFn;
+  }
+
+  for (SmallVectorImpl<std::pair<Function*,Function*> >::iterator
+         I = Queue.begin(), E = Queue.end(); I != E; ++I)
+    diff(I->first, I->second);
+}
+
+bool DifferenceEngine::equivalentAsOperands(GlobalValue *L, GlobalValue *R) {
+  if (globalValueOracle) return (*globalValueOracle)(L, R);
+  return L->getName() == R->getName();
+}
diff --git a/tools/llvm-diff/DifferenceEngine.h b/tools/llvm-diff/DifferenceEngine.h
new file mode 100644
index 0000000..9d0ebec
--- /dev/null
+++ b/tools/llvm-diff/DifferenceEngine.h
@@ -0,0 +1,171 @@
+//===-- DifferenceEngine.h - Module comparator ------------------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This header defines the interface to the LLVM difference engine,
+// which structurally compares functions within a module.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef _LLVM_DIFFERENCE_ENGINE_H_
+#define _LLVM_DIFFERENCE_ENGINE_H_
+
+#include <utility>
+#include <llvm/ADT/SmallVector.h>
+#include <llvm/ADT/StringRef.h>
+
+namespace llvm {
+  class LLVMContext;
+  class Module;
+  class Function;
+  class Twine;
+  class Value;
+  class GlobalValue;
+
+  /// A class for performing structural comparisons of LLVM assembly.
+  class DifferenceEngine {
+  public:
+    /// A temporary-object class for building up log messages.
+    class LogBuilder {
+      DifferenceEngine &Engine;
+
+      /// The use of a stored StringRef here is okay because
+      /// LogBuilder should be used only as a temporary, and as a
+      /// temporary it will be destructed before whatever temporary
+      /// might be initializing this format.
+      StringRef Format;
+
+      SmallVector<Value*, 4> Arguments;
+
+    public:
+      LogBuilder(DifferenceEngine &Engine, StringRef Format)
+        : Engine(Engine), Format(Format) {}
+
+      LogBuilder &operator<<(Value *V) {
+        Arguments.push_back(V);
+        return *this;
+      }
+
+      ~LogBuilder() {
+        Engine.consumer.logf(*this);
+      }
+
+      StringRef getFormat() const { return Format; }
+
+      unsigned getNumArguments() const { return Arguments.size(); }
+      Value *getArgument(unsigned I) const { return Arguments[I]; }
+    };
+
+    enum DiffChange { DC_match, DC_left, DC_right };
+
+    /// A temporary-object class for building up diff messages.
+    class DiffLogBuilder {
+      typedef std::pair<Instruction*,Instruction*> DiffRecord;
+      SmallVector<DiffRecord, 20> Diff;
+
+      DifferenceEngine &Engine;
+
+    public:
+      DiffLogBuilder(DifferenceEngine &Engine) : Engine(Engine) {}
+      ~DiffLogBuilder() { Engine.consumer.logd(*this); }
+
+      void addMatch(Instruction *L, Instruction *R) {
+        Diff.push_back(DiffRecord(L, R));
+      }
+      void addLeft(Instruction *L) { Diff.push_back(DiffRecord(L, 0)); }
+      void addRight(Instruction *R) { Diff.push_back(DiffRecord(0, R)); }
+
+      unsigned getNumLines() const { return Diff.size(); }
+      DiffChange getLineKind(unsigned I) const {
+        return (Diff[I].first ? (Diff[I].second ? DC_match : DC_left)
+                              : DC_right);
+      }
+      Instruction *getLeft(unsigned I) const { return Diff[I].first; }
+      Instruction *getRight(unsigned I) const { return Diff[I].second; }
+    };
+
+    /// The interface for consumers of difference data.
+    struct Consumer {
+      /// Record that a local context has been entered.  Left and
+      /// Right are IR "containers" of some sort which are being
+      /// considered for structural equivalence: global variables,
+      /// functions, blocks, instructions, etc.
+      virtual void enterContext(Value *Left, Value *Right) = 0;
+
+      /// Record that a local context has been exited.
+      virtual void exitContext() = 0;
+
+      /// Record a difference within the current context.
+      virtual void log(StringRef Text) = 0;
+
+      /// Record a formatted difference within the current context.
+      virtual void logf(const LogBuilder &Log) = 0;
+
+      /// Record a line-by-line instruction diff.
+      virtual void logd(const DiffLogBuilder &Log) = 0;
+
+    protected:
+      ~Consumer() {}
+    };
+
+    /// A RAII object for recording the current context.
+    struct Context {
+      Context(DifferenceEngine &Engine, Value *L, Value *R) : Engine(Engine) {
+        Engine.consumer.enterContext(L, R);
+      }
+
+      ~Context() {
+        Engine.consumer.exitContext();
+      }
+
+    private:
+      DifferenceEngine &Engine;
+    };
+
+    /// An oracle for answering whether two values are equivalent as
+    /// operands.
+    struct Oracle {
+      virtual bool operator()(Value *L, Value *R) = 0;
+
+    protected:
+      ~Oracle() {}
+    };
+
+    DifferenceEngine(LLVMContext &context, Consumer &consumer)
+      : context(context), consumer(consumer), globalValueOracle(0) {}
+
+    void diff(Module *L, Module *R);
+    void diff(Function *L, Function *R);
+
+    void log(StringRef text) {
+      consumer.log(text);
+    }
+
+    LogBuilder logf(StringRef text) {
+      return LogBuilder(*this, text);
+    }
+
+    /// Installs an oracle to decide whether two global values are
+    /// equivalent as operands.  Without an oracle, global values are
+    /// considered equivalent as operands precisely when they have the
+    /// same name.
+    void setGlobalValueOracle(Oracle *oracle) {
+      globalValueOracle = oracle;
+    }
+
+    /// Determines whether two global values are equivalent.
+    bool equivalentAsOperands(GlobalValue *L, GlobalValue *R);
+
+  private:
+    LLVMContext &context;
+    Consumer &consumer;
+    Oracle *globalValueOracle;
+  };
+}
+
+#endif
diff --git a/tools/llvm-diff/Makefile b/tools/llvm-diff/Makefile
new file mode 100644
index 0000000..58e49fa
--- /dev/null
+++ b/tools/llvm-diff/Makefile
@@ -0,0 +1,17 @@
+##===- tools/llvm-diff/Makefile ----------------------------*- Makefile -*-===##
+# 
+#                     The LLVM Compiler Infrastructure
+#
+# This file is distributed under the University of Illinois Open Source
+# License. See LICENSE.TXT for details.
+# 
+##===----------------------------------------------------------------------===##
+
+LEVEL = ../..
+TOOLNAME = llvm-diff
+LINK_COMPONENTS := asmparser bitreader
+
+# This tool has no plugins, optimize startup time.
+TOOL_NO_EXPORTS = 1
+
+include $(LEVEL)/Makefile.common
diff --git a/tools/llvm-diff/llvm-diff.cpp b/tools/llvm-diff/llvm-diff.cpp
new file mode 100644
index 0000000..7c2a5d6
--- /dev/null
+++ b/tools/llvm-diff/llvm-diff.cpp
@@ -0,0 +1,301 @@
+#include <string>
+#include <utility>
+
+#include <llvm/ADT/StringRef.h>
+#include <llvm/ADT/SmallVector.h>
+#include <llvm/ADT/DenseMap.h>
+
+// Required to parse .ll files.
+#include <llvm/Support/SourceMgr.h>
+#include <llvm/Assembly/Parser.h>
+
+// Required to parse .bc files.
+#include <llvm/Support/MemoryBuffer.h>
+#include <llvm/Bitcode/ReaderWriter.h>
+
+#include <llvm/Support/raw_ostream.h>
+#include <llvm/LLVMContext.h>
+#include <llvm/Module.h>
+#include <llvm/Type.h>
+#include <llvm/Instructions.h>
+
+#include "DifferenceEngine.h"
+
+using namespace llvm;
+
+/// Reads a module from a file.  If the filename ends in .ll, it is
+/// interpreted as an assembly file;  otherwise, it is interpreted as
+/// bitcode.  On error, messages are written to stderr and null is
+/// returned.
+static Module *ReadModule(LLVMContext &Context, StringRef Name) {
+  // LLVM assembly path.
+  if (Name.endswith(".ll")) {
+    SMDiagnostic Diag;
+    Module *M = ParseAssemblyFile(Name, Diag, Context);
+    if (M) return M;
+
+    Diag.Print("llvmdiff", errs());
+    return 0;
+  }
+
+  // Bitcode path.
+  MemoryBuffer *Buffer = MemoryBuffer::getFile(Name);
+
+  // ParseBitcodeFile takes ownership of the buffer if it succeeds.
+  std::string Error;
+  Module *M = ParseBitcodeFile(Buffer, Context, &Error);
+  if (M) return M;
+
+  errs() << "error parsing " << Name << ": " << Error;
+  delete Buffer;
+  return 0;
+}
+
+static int usage() {
+  errs() << "expected usage:\n";
+  errs() << "  llvm-diff oldmodule.ll newmodule.ll [function list]\n";
+  errs() << "Assembly or bitcode modules may be used interchangeably.\n";
+  errs() << "If no functions are provided, all functions will be compared.\n";
+  return 1;
+}
+
+namespace {
+struct DiffContext {
+  DiffContext(Value *L, Value *R)
+    : L(L), R(R), Differences(false), IsFunction(isa<Function>(L)) {}
+  Value *L;
+  Value *R;
+  bool Differences;
+  bool IsFunction;
+  DenseMap<Value*,unsigned> LNumbering;
+  DenseMap<Value*,unsigned> RNumbering;
+};
+
+void ComputeNumbering(Function *F, DenseMap<Value*,unsigned> &Numbering) {
+  unsigned BBN = 0;
+  unsigned IN = 0;
+
+  // Arguments get the first numbers.
+  for (Function::arg_iterator
+         AI = F->arg_begin(), AE = F->arg_end(); AI != AE; ++AI)
+    if (!AI->hasName())
+      Numbering[&*AI] = IN++;
+
+  // Walk the basic blocks in order.
+  for (Function::iterator FI = F->begin(), FE = F->end(); FI != FE; ++FI) {
+    // Basic blocks have their own 'namespace'.
+    if (!FI->hasName())
+      Numbering[&*FI] = BBN++;
+
+    // Walk the instructions in order.
+    for (BasicBlock::iterator BI = FI->begin(), BE = FI->end(); BI != BE; ++BI)
+      // void instructions don't get numbers.
+      if (!BI->hasName() && !BI->getType()->isVoidTy())
+        Numbering[&*BI] = IN++;
+  }
+
+  assert(!Numbering.empty() && "asked for numbering but numbering was no-op");
+}
+
+class DiffConsumer : public DifferenceEngine::Consumer {
+private:
+  Module *LModule;
+  Module *RModule;
+  SmallVector<DiffContext, 5> contexts;
+  bool Differences;
+  unsigned Indent;
+
+  void printValue(Value *V, bool isL) {
+    if (V->hasName()) {
+      errs() << (isa<GlobalValue>(V) ? '@' : '%') << V->getName();
+      return;
+    }
+    if (V->getType()->isVoidTy()) {
+      if (isa<StoreInst>(V)) {
+        errs() << "store to ";
+        printValue(cast<StoreInst>(V)->getPointerOperand(), isL);
+      } else if (isa<CallInst>(V)) {
+        errs() << "call to ";
+        printValue(cast<CallInst>(V)->getCalledValue(), isL);
+      } else if (isa<InvokeInst>(V)) {
+        errs() << "invoke to ";
+        printValue(cast<InvokeInst>(V)->getCalledValue(), isL);
+      } else {
+        errs() << *V;
+      }
+      return;
+    }
+
+    unsigned N = contexts.size();
+    while (N > 0) {
+      --N;
+      DiffContext &ctxt = contexts[N];
+      if (!ctxt.IsFunction) continue;
+      if (isL) {
+        if (ctxt.LNumbering.empty())
+          ComputeNumbering(cast<Function>(ctxt.L), ctxt.LNumbering);
+        errs() << '%' << ctxt.LNumbering[V];
+        return;
+      } else {
+        if (ctxt.RNumbering.empty())
+          ComputeNumbering(cast<Function>(ctxt.R), ctxt.RNumbering);
+        errs() << '%' << ctxt.RNumbering[V];
+        return;
+      }
+    }
+
+    errs() << "<anonymous>";
+  }
+
+  void header() {
+    if (contexts.empty()) return;
+    for (SmallVectorImpl<DiffContext>::iterator
+           I = contexts.begin(), E = contexts.end(); I != E; ++I) {
+      if (I->Differences) continue;
+      if (isa<Function>(I->L)) {
+        // Extra newline between functions.
+        if (Differences) errs() << "\n";
+
+        Function *L = cast<Function>(I->L);
+        Function *R = cast<Function>(I->R);
+        if (L->getName() != R->getName())
+          errs() << "in function " << L->getName() << " / " << R->getName() << ":\n";
+        else
+          errs() << "in function " << L->getName() << ":\n";
+      } else if (isa<BasicBlock>(I->L)) {
+        BasicBlock *L = cast<BasicBlock>(I->L);
+        BasicBlock *R = cast<BasicBlock>(I->R);
+        errs() << "  in block ";
+        printValue(L, true);
+        errs() << " / ";
+        printValue(R, false);
+        errs() << ":\n";
+      } else if (isa<Instruction>(I->L)) {
+        errs() << "    in instruction ";
+        printValue(I->L, true);
+        errs() << " / ";
+        printValue(I->R, false);
+        errs() << ":\n";
+      }
+
+      I->Differences = true;
+    }
+  }
+
+  void indent() {
+    unsigned N = Indent;
+    while (N--) errs() << ' ';
+  }
+
+public:
+  DiffConsumer(Module *L, Module *R)
+    : LModule(L), RModule(R), Differences(false), Indent(0) {}
+
+  bool hadDifferences() const { return Differences; }
+
+  void enterContext(Value *L, Value *R) {
+    contexts.push_back(DiffContext(L, R));
+    Indent += 2;
+  }
+  void exitContext() {
+    Differences |= contexts.back().Differences;
+    contexts.pop_back();
+    Indent -= 2;
+  }
+
+  void log(StringRef text) {
+    header();
+    indent();
+    errs() << text << "\n";
+  }
+
+  void logf(const DifferenceEngine::LogBuilder &Log) {
+    header();
+    indent();
+
+    // FIXME: we don't know whether these are l-values or r-values (ha!)
+    // Print them in some saner way!
+    errs() << Log.getFormat() << "\n";
+    for (unsigned I = 0, E = Log.getNumArguments(); I != E; ++I)
+      Log.getArgument(I)->dump();
+  }
+
+  void logd(const DifferenceEngine::DiffLogBuilder &Log) {
+    header();
+
+    for (unsigned I = 0, E = Log.getNumLines(); I != E; ++I) {
+      indent();
+      switch (Log.getLineKind(I)) {
+      case DifferenceEngine::DC_match:
+        errs() << "  ";
+        Log.getLeft(I)->dump();
+        //printValue(Log.getLeft(I), true);
+        break;
+      case DifferenceEngine::DC_left:
+        errs() << "< ";
+        Log.getLeft(I)->dump();
+        //printValue(Log.getLeft(I), true);
+        break;
+      case DifferenceEngine::DC_right:
+        errs() << "> ";
+        Log.getRight(I)->dump();
+        //printValue(Log.getRight(I), false);
+        break;
+      }
+      //errs() << "\n";
+    }
+  }
+  
+};
+}
+
+int main(int argc, const char **argv) {
+  if (argc < 3) return usage();
+
+  // Don't make StringRef locals like this at home.
+  StringRef LModuleFile = argv[1];
+  StringRef RModuleFile = argv[2];
+
+  LLVMContext Context;
+  
+  // Load both modules.  Die if that fails.
+  Module *LModule = ReadModule(Context, LModuleFile);
+  Module *RModule = ReadModule(Context, RModuleFile);
+  if (!LModule || !RModule) return 1;
+
+  DiffConsumer Consumer(LModule, RModule);
+  DifferenceEngine Engine(Context, Consumer);
+
+  // If any function names were given, just diff those.
+  const char **FnNames = argv + 3;
+  unsigned NumFnNames = argc - 3;
+  if (NumFnNames) {
+    for (unsigned I = 0; I != NumFnNames; ++I) {
+      StringRef FnName = FnNames[I];
+
+      // Drop leading sigils from the function name.
+      if (FnName.startswith("@")) FnName = FnName.substr(1);
+
+      Function *LFn = LModule->getFunction(FnName);
+      Function *RFn = RModule->getFunction(FnName);
+      if (LFn && RFn)
+        Engine.diff(LFn, RFn);
+      else {
+        if (!LFn && !RFn)
+          errs() << "No function named @" << FnName << " in either module\n";
+        else if (!LFn)
+          errs() << "No function named @" << FnName << " in left module\n";
+        else
+          errs() << "No function named @" << FnName << " in right module\n";
+      }
+    }
+  } else {
+    // Otherwise, diff all functions in the modules.
+    Engine.diff(LModule, RModule);
+  }
+
+  delete LModule;
+  delete RModule;
+
+  return Consumer.hadDifferences();
+}