[llvm-branch-commits] [llvm-branch] r172541 [6/8] - in /llvm/branches/AMDILBackend: ./ autoconf/ bindings/ocaml/executionengine/ bindings/ocaml/llvm/ bindings/ocaml/target/ cmake/ cmake/modules/ cmake/platforms/ docs/ docs/CommandGuide/ docs/_themes/ docs/_themes/llvm-theme/ docs/_themes/llvm-theme/static/ docs/llvm-theme/ docs/llvm-theme/static/ docs/tutorial/ examples/ExceptionDemo/ examples/Fibonacci/ examples/Kaleidoscope/Chapter4/ examples/Kaleidoscope/Chapter5/ examples/Kaleidoscope/Chapter6/ examples/Kaleidoscope/Chapt...

Richard Relph Richard.Relph at amd.com
Tue Jan 15 09:16:26 PST 2013


Modified: llvm/branches/AMDILBackend/lib/Transforms/IPO/DeadArgumentElimination.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/IPO/DeadArgumentElimination.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/IPO/DeadArgumentElimination.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/IPO/DeadArgumentElimination.cpp Tue Jan 15 11:16:16 2013
@@ -21,7 +21,9 @@
 #include "llvm/Transforms/IPO.h"
 #include "llvm/CallingConv.h"
 #include "llvm/Constant.h"
+#include "llvm/DebugInfo.h"
 #include "llvm/DerivedTypes.h"
+#include "llvm/DIBuilder.h"
 #include "llvm/Instructions.h"
 #include "llvm/IntrinsicInst.h"
 #include "llvm/LLVMContext.h"
@@ -30,6 +32,7 @@
 #include "llvm/Support/CallSite.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/raw_ostream.h"
+#include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/Statistic.h"
 #include "llvm/ADT/StringExtras.h"
@@ -121,6 +124,15 @@
 
     typedef SmallVector<RetOrArg, 5> UseVector;
 
+    // Map each LLVM function to corresponding metadata with debug info. If
+    // the function is replaced with another one, we should patch the pointer
+    // to LLVM function in metadata.
+    // As the code generation for module is finished (and DIBuilder is
+    // finalized) we assume that subprogram descriptors won't be changed, and
+    // they are stored in map for short duration anyway.
+    typedef DenseMap<Function*, DISubprogram> FunctionDIMap;
+    FunctionDIMap FunctionDIs;
+
   protected:
     // DAH uses this to specify a different ID.
     explicit DAE(char &ID) : ModulePass(ID) {}
@@ -141,6 +153,7 @@
                        unsigned RetValNum = 0);
     Liveness SurveyUses(const Value *V, UseVector &MaybeLiveUses);
 
+    void CollectFunctionDIs(Module &M);
     void SurveyFunction(const Function &F);
     void MarkValue(const RetOrArg &RA, Liveness L,
                    const UseVector &MaybeLiveUses);
@@ -180,6 +193,33 @@
 ModulePass *llvm::createDeadArgEliminationPass() { return new DAE(); }
 ModulePass *llvm::createDeadArgHackingPass() { return new DAH(); }
 
+/// CollectFunctionDIs - Map each function in the module to its debug info
+/// descriptor.
+void DAE::CollectFunctionDIs(Module &M) {
+  FunctionDIs.clear();
+
+  for (Module::named_metadata_iterator I = M.named_metadata_begin(),
+       E = M.named_metadata_end(); I != E; ++I) {
+    NamedMDNode &NMD = *I;
+    for (unsigned MDIndex = 0, MDNum = NMD.getNumOperands();
+         MDIndex < MDNum; ++MDIndex) {
+      MDNode *Node = NMD.getOperand(MDIndex);
+      if (!DIDescriptor(Node).isCompileUnit())
+        continue;
+      DICompileUnit CU(Node);
+      const DIArray &SPs = CU.getSubprograms();
+      for (unsigned SPIndex = 0, SPNum = SPs.getNumElements();
+           SPIndex < SPNum; ++SPIndex) {
+        DISubprogram SP(SPs.getElement(SPIndex));
+        if (!SP.Verify())
+          continue;
+        if (Function *F = SP.getFunction())
+          FunctionDIs[F] = SP;
+      }
+    }
+  }
+}
+
 /// DeleteDeadVarargs - If this is an function that takes a ... list, and if
 /// llvm.vastart is never called, the varargs list is dead for the function.
 bool DAE::DeleteDeadVarargs(Function &Fn) {
@@ -236,9 +276,11 @@
       SmallVector<AttributeWithIndex, 8> AttributesVec;
       for (unsigned i = 0; PAL.getSlot(i).Index <= NumArgs; ++i)
         AttributesVec.push_back(PAL.getSlot(i));
-      if (Attributes FnAttrs = PAL.getFnAttributes())
-        AttributesVec.push_back(AttributeWithIndex::get(~0, FnAttrs));
-      PAL = AttrListPtr::get(AttributesVec);
+      Attributes FnAttrs = PAL.getFnAttributes();
+      if (FnAttrs.hasAttributes())
+        AttributesVec.push_back(AttributeWithIndex::get(AttrListPtr::FunctionIndex,
+                                                        FnAttrs));
+      PAL = AttrListPtr::get(Fn.getContext(), AttributesVec);
     }
 
     Instruction *New;
@@ -284,6 +326,11 @@
     I2->takeName(I);
   }
 
+  // Patch the pointer to LLVM function in debug info descriptor.
+  FunctionDIMap::iterator DI = FunctionDIs.find(&Fn);
+  if (DI != FunctionDIs.end())
+    DI->second.replaceFunction(NF);
+
   // Finally, nuke the old function.
   Fn.eraseFromParent();
   return true;
@@ -717,13 +764,17 @@
   // here. Currently, this should not be possible, but special handling might be
   // required when new return value attributes are added.
   if (NRetTy->isVoidTy())
-    RAttrs &= ~Attribute::typeIncompatible(NRetTy);
+    RAttrs =
+      Attributes::get(NRetTy->getContext(), AttrBuilder(RAttrs).
+                      removeAttributes(Attributes::typeIncompatible(NRetTy)));
   else
-    assert((RAttrs & Attribute::typeIncompatible(NRetTy)) == 0
-           && "Return attributes no longer compatible?");
-
-  if (RAttrs)
-    AttributesVec.push_back(AttributeWithIndex::get(0, RAttrs));
+    assert(!AttrBuilder(RAttrs).
+             hasAttributes(Attributes::typeIncompatible(NRetTy)) &&
+           "Return attributes no longer compatible?");
+
+  if (RAttrs.hasAttributes())
+    AttributesVec.push_back(AttributeWithIndex::get(AttrListPtr::ReturnIndex,
+                                                    RAttrs));
 
   // Remember which arguments are still alive.
   SmallVector<bool, 10> ArgAlive(FTy->getNumParams(), false);
@@ -740,7 +791,8 @@
 
       // Get the original parameter attributes (skipping the first one, that is
       // for the return value.
-      if (Attributes Attrs = PAL.getParamAttributes(i + 1))
+      Attributes Attrs = PAL.getParamAttributes(i + 1);
+      if (Attrs.hasAttributes())
         AttributesVec.push_back(AttributeWithIndex::get(Params.size(), Attrs));
     } else {
       ++NumArgumentsEliminated;
@@ -749,11 +801,12 @@
     }
   }
 
-  if (FnAttrs != Attribute::None)
-    AttributesVec.push_back(AttributeWithIndex::get(~0, FnAttrs));
+  if (FnAttrs.hasAttributes())
+    AttributesVec.push_back(AttributeWithIndex::get(AttrListPtr::FunctionIndex,
+                                                    FnAttrs));
 
   // Reconstruct the AttributesList based on the vector we constructed.
-  AttrListPtr NewPAL = AttrListPtr::get(AttributesVec);
+  AttrListPtr NewPAL = AttrListPtr::get(F->getContext(), AttributesVec);
 
   // Create the new function type based on the recomputed parameters.
   FunctionType *NFTy = FunctionType::get(NRetTy, Params, FTy->isVarArg());
@@ -786,9 +839,12 @@
     Attributes RAttrs = CallPAL.getRetAttributes();
     Attributes FnAttrs = CallPAL.getFnAttributes();
     // Adjust in case the function was changed to return void.
-    RAttrs &= ~Attribute::typeIncompatible(NF->getReturnType());
-    if (RAttrs)
-      AttributesVec.push_back(AttributeWithIndex::get(0, RAttrs));
+    RAttrs =
+      Attributes::get(NF->getContext(), AttrBuilder(RAttrs).
+           removeAttributes(Attributes::typeIncompatible(NF->getReturnType())));
+    if (RAttrs.hasAttributes())
+      AttributesVec.push_back(AttributeWithIndex::get(AttrListPtr::ReturnIndex,
+                                                      RAttrs));
 
     // Declare these outside of the loops, so we can reuse them for the second
     // loop, which loops the varargs.
@@ -800,22 +856,25 @@
       if (ArgAlive[i]) {
         Args.push_back(*I);
         // Get original parameter attributes, but skip return attributes.
-        if (Attributes Attrs = CallPAL.getParamAttributes(i + 1))
+        Attributes Attrs = CallPAL.getParamAttributes(i + 1);
+        if (Attrs.hasAttributes())
           AttributesVec.push_back(AttributeWithIndex::get(Args.size(), Attrs));
       }
 
     // Push any varargs arguments on the list. Don't forget their attributes.
     for (CallSite::arg_iterator E = CS.arg_end(); I != E; ++I, ++i) {
       Args.push_back(*I);
-      if (Attributes Attrs = CallPAL.getParamAttributes(i + 1))
+      Attributes Attrs = CallPAL.getParamAttributes(i + 1);
+      if (Attrs.hasAttributes())
         AttributesVec.push_back(AttributeWithIndex::get(Args.size(), Attrs));
     }
 
-    if (FnAttrs != Attribute::None)
-      AttributesVec.push_back(AttributeWithIndex::get(~0, FnAttrs));
+    if (FnAttrs.hasAttributes())
+      AttributesVec.push_back(AttributeWithIndex::get(AttrListPtr::FunctionIndex,
+                                                      FnAttrs));
 
     // Reconstruct the AttributesList based on the vector we constructed.
-    AttrListPtr NewCallPAL = AttrListPtr::get(AttributesVec);
+    AttrListPtr NewCallPAL = AttrListPtr::get(F->getContext(), AttributesVec);
 
     Instruction *New;
     if (InvokeInst *II = dyn_cast<InvokeInst>(Call)) {
@@ -952,6 +1011,11 @@
         BB->getInstList().erase(RI);
       }
 
+  // Patch the pointer to LLVM function in debug info descriptor.
+  FunctionDIMap::iterator DI = FunctionDIs.find(F);
+  if (DI != FunctionDIs.end())
+    DI->second.replaceFunction(NF);
+
   // Now that the old function is dead, delete it.
   F->eraseFromParent();
 
@@ -961,6 +1025,9 @@
 bool DAE::runOnModule(Module &M) {
   bool Changed = false;
 
+  // Collect debug info descriptors for functions.
+  CollectFunctionDIs(M);
+
   // First pass: Do a simple check to see if any functions can have their "..."
   // removed.  We can do this if they never call va_start.  This loop cannot be
   // fused with the next loop, because deleting a function invalidates

Modified: llvm/branches/AMDILBackend/lib/Transforms/IPO/ExtractGV.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/IPO/ExtractGV.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/IPO/ExtractGV.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/IPO/ExtractGV.cpp Tue Jan 15 11:16:16 2013
@@ -51,32 +51,75 @@
       // Visit the GlobalVariables.
       for (Module::global_iterator I = M.global_begin(), E = M.global_end();
            I != E; ++I) {
-        if (deleteStuff == (bool)Named.count(I) && !I->isDeclaration()) {
-          I->setInitializer(0);
-        } else {
+        bool Delete =
+          deleteStuff == (bool)Named.count(I) && !I->isDeclaration();
+        if (!Delete) {
           if (I->hasAvailableExternallyLinkage())
             continue;
           if (I->getName() == "llvm.global_ctors")
             continue;
         }
 
-        if (I->hasLocalLinkage())
+        bool Local = I->hasLocalLinkage();
+        if (Local)
           I->setVisibility(GlobalValue::HiddenVisibility);
-        I->setLinkage(GlobalValue::ExternalLinkage);
+
+        if (Local || Delete)
+          I->setLinkage(GlobalValue::ExternalLinkage);
+
+        if (Delete)
+          I->setInitializer(0);
       }
 
       // Visit the Functions.
       for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I) {
-        if (deleteStuff == (bool)Named.count(I) && !I->isDeclaration()) {
-          I->deleteBody();
-        } else {
+        bool Delete =
+          deleteStuff == (bool)Named.count(I) && !I->isDeclaration();
+        if (!Delete) {
           if (I->hasAvailableExternallyLinkage())
             continue;
         }
 
-        if (I->hasLocalLinkage())
+        bool Local = I->hasLocalLinkage();
+        if (Local)
           I->setVisibility(GlobalValue::HiddenVisibility);
-        I->setLinkage(GlobalValue::ExternalLinkage);
+
+        if (Local || Delete)
+          I->setLinkage(GlobalValue::ExternalLinkage);
+
+        if (Delete)
+          I->deleteBody();
+      }
+
+      // Visit the Aliases.
+      for (Module::alias_iterator I = M.alias_begin(), E = M.alias_end();
+           I != E;) {
+        Module::alias_iterator CurI = I;
+        ++I;
+
+        if (CurI->hasLocalLinkage()) {
+          CurI->setVisibility(GlobalValue::HiddenVisibility);
+          CurI->setLinkage(GlobalValue::ExternalLinkage);
+        }
+
+        if (deleteStuff == (bool)Named.count(CurI)) {
+          Type *Ty =  CurI->getType()->getElementType();
+
+          CurI->removeFromParent();
+          llvm::Value *Declaration;
+          if (FunctionType *FTy = dyn_cast<FunctionType>(Ty)) {
+            Declaration = Function::Create(FTy, GlobalValue::ExternalLinkage,
+                                           CurI->getName(), &M);
+
+          } else {
+            Declaration =
+              new GlobalVariable(M, Ty, false, GlobalValue::ExternalLinkage,
+                                 0, CurI->getName());
+
+          }
+          CurI->replaceAllUsesWith(Declaration);
+          delete CurI;
+        }
       }
 
       return true;

Modified: llvm/branches/AMDILBackend/lib/Transforms/IPO/FunctionAttrs.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/IPO/FunctionAttrs.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/IPO/FunctionAttrs.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/IPO/FunctionAttrs.cpp Tue Jan 15 11:16:16 2013
@@ -28,9 +28,9 @@
 #include "llvm/Analysis/CallGraph.h"
 #include "llvm/Analysis/CaptureTracking.h"
 #include "llvm/ADT/SCCIterator.h"
+#include "llvm/ADT/SetVector.h"
 #include "llvm/ADT/SmallSet.h"
 #include "llvm/ADT/Statistic.h"
-#include "llvm/ADT/UniqueVector.h"
 #include "llvm/Support/InstIterator.h"
 using namespace llvm;
 
@@ -212,10 +212,17 @@
     MadeChange = true;
 
     // Clear out any existing attributes.
-    F->removeAttribute(~0, Attribute::ReadOnly | Attribute::ReadNone);
+    AttrBuilder B;
+    B.addAttribute(Attributes::ReadOnly)
+      .addAttribute(Attributes::ReadNone);
+    F->removeAttribute(AttrListPtr::FunctionIndex,
+                       Attributes::get(F->getContext(), B));
 
     // Add in the new attribute.
-    F->addAttribute(~0, ReadsMemory? Attribute::ReadOnly : Attribute::ReadNone);
+    B.clear();
+    B.addAttribute(ReadsMemory ? Attributes::ReadOnly : Attributes::ReadNone);
+    F->addAttribute(AttrListPtr::FunctionIndex,
+                    Attributes::get(F->getContext(), B));
 
     if (ReadsMemory)
       ++NumReadOnly;
@@ -276,8 +283,6 @@
 
     void tooManyUses() { Captured = true; }
 
-    bool shouldExplore(Use *U) { return true; }
-
     bool captured(Use *U) {
       CallSite CS(U->getUser());
       if (!CS.getInstruction()) { Captured = true; return true; }
@@ -352,6 +357,9 @@
 
   ArgumentGraph AG;
 
+  AttrBuilder B;
+  B.addAttribute(Attributes::NoCapture);
+
   // Check each function in turn, determining which pointer arguments are not
   // captured.
   for (CallGraphSCC::iterator I = SCC.begin(), E = SCC.end(); I != E; ++I) {
@@ -373,7 +381,7 @@
       for (Function::arg_iterator A = F->arg_begin(), E = F->arg_end();
            A != E; ++A) {
         if (A->getType()->isPointerTy() && !A->hasNoCaptureAttr()) {
-          A->addAttr(Attribute::NoCapture);
+          A->addAttr(Attributes::get(F->getContext(), B));
           ++NumNoCapture;
           Changed = true;
         }
@@ -388,7 +396,7 @@
         if (!Tracker.Captured) {
           if (Tracker.Uses.empty()) {
             // If it's trivially not captured, mark it nocapture now.
-            A->addAttr(Attribute::NoCapture);
+            A->addAttr(Attributes::get(F->getContext(), B));
             ++NumNoCapture;
             Changed = true;
           } else {
@@ -421,7 +429,9 @@
       // eg. "void f(int* x) { if (...) f(x); }"
       if (ArgumentSCC[0]->Uses.size() == 1 &&
           ArgumentSCC[0]->Uses[0] == ArgumentSCC[0]) {
-        ArgumentSCC[0]->Definition->addAttr(Attribute::NoCapture);
+        ArgumentSCC[0]->
+          Definition->
+          addAttr(Attributes::get(ArgumentSCC[0]->Definition->getContext(), B));
         ++NumNoCapture;
         Changed = true;
       }
@@ -463,7 +473,7 @@
 
     for (unsigned i = 0, e = ArgumentSCC.size(); i != e; ++i) {
       Argument *A = ArgumentSCC[i]->Definition;
-      A->addAttr(Attribute::NoCapture);
+      A->addAttr(Attributes::get(A->getContext(), B));
       ++NumNoCapture;
       Changed = true;
     }
@@ -476,13 +486,13 @@
 /// or a pointer that doesn't alias any other pointer visible to the caller.
 bool FunctionAttrs::IsFunctionMallocLike(Function *F,
                               SmallPtrSet<Function*, 8> &SCCNodes) const {
-  UniqueVector<Value *> FlowsToReturn;
+  SmallSetVector<Value *, 8> FlowsToReturn;
   for (Function::iterator I = F->begin(), E = F->end(); I != E; ++I)
     if (ReturnInst *Ret = dyn_cast<ReturnInst>(I->getTerminator()))
       FlowsToReturn.insert(Ret->getReturnValue());
 
   for (unsigned i = 0; i != FlowsToReturn.size(); ++i) {
-    Value *RetVal = FlowsToReturn[i+1];   // UniqueVector[0] is reserved.
+    Value *RetVal = FlowsToReturn[i];
 
     if (Constant *C = dyn_cast<Constant>(RetVal)) {
       if (!C->isNullValue() && !isa<UndefValue>(C))
@@ -520,7 +530,7 @@
         case Instruction::Call:
         case Instruction::Invoke: {
           CallSite CS(RVI);
-          if (CS.paramHasAttr(0, Attribute::NoAlias))
+          if (CS.paramHasAttr(0, Attributes::NoAlias))
             break;
           if (CS.getCalledFunction() &&
               SCCNodes.count(CS.getCalledFunction()))

Modified: llvm/branches/AMDILBackend/lib/Transforms/IPO/GlobalOpt.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/IPO/GlobalOpt.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/IPO/GlobalOpt.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/IPO/GlobalOpt.cpp Tue Jan 15 11:16:16 2013
@@ -25,7 +25,7 @@
 #include "llvm/Pass.h"
 #include "llvm/Analysis/ConstantFolding.h"
 #include "llvm/Analysis/MemoryBuiltins.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Target/TargetLibraryInfo.h"
 #include "llvm/Support/CallSite.h"
 #include "llvm/Support/Debug.h"
@@ -83,7 +83,7 @@
                                const GlobalStatus &GS);
     bool OptimizeEmptyGlobalCXXDtors(Function *CXAAtExitFn);
 
-    TargetData *TD;
+    DataLayout *TD;
     TargetLibraryInfo *TLI;
   };
 }
@@ -225,6 +225,7 @@
 
         // Don't hack on volatile stores.
         if (SI->isVolatile()) return true;
+
         GS.Ordering = StrongerOrdering(GS.Ordering, SI->getOrdering());
 
         // If this is a direct store to the global (i.e., the global is a scalar
@@ -234,6 +235,14 @@
           if (const GlobalVariable *GV = dyn_cast<GlobalVariable>(
                                                            SI->getOperand(1))) {
             Value *StoredVal = SI->getOperand(0);
+
+            if (Constant *C = dyn_cast<Constant>(StoredVal)) {
+              if (C->isThreadDependent()) {
+                // The stored value changes between threads; don't track it.
+                return true;
+              }
+            }
+
             if (StoredVal == GV->getInitializer()) {
               if (GS.StoredType < GlobalStatus::isInitializerStored)
                 GS.StoredType = GlobalStatus::isInitializerStored;
@@ -346,7 +355,7 @@
 /// Given a value that is stored to a global but never read, determine whether
 /// it's safe to remove the store and the chain of computation that feeds the
 /// store.
-static bool IsSafeComputationToRemove(Value *V) {
+static bool IsSafeComputationToRemove(Value *V, const TargetLibraryInfo *TLI) {
   do {
     if (isa<Constant>(V))
       return true;
@@ -355,7 +364,7 @@
     if (isa<LoadInst>(V) || isa<InvokeInst>(V) || isa<Argument>(V) ||
         isa<GlobalValue>(V))
       return false;
-    if (isAllocationFn(V))
+    if (isAllocationFn(V, TLI))
       return true;
 
     Instruction *I = cast<Instruction>(V);
@@ -376,7 +385,8 @@
 /// of the global and clean up any that obviously don't assign the global a
 /// value that isn't dynamically allocated.
 ///
-static bool CleanupPointerRootUsers(GlobalVariable *GV) {
+static bool CleanupPointerRootUsers(GlobalVariable *GV,
+                                    const TargetLibraryInfo *TLI) {
   // A brief explanation of leak checkers.  The goal is to find bugs where
   // pointers are forgotten, causing an accumulating growth in memory
   // usage over time.  The common strategy for leak checkers is to whitelist the
@@ -432,18 +442,18 @@
         C->destroyConstant();
         // This could have invalidated UI, start over from scratch.
         Dead.clear();
-        CleanupPointerRootUsers(GV);
+        CleanupPointerRootUsers(GV, TLI);
         return true;
       }
     }
   }
 
   for (int i = 0, e = Dead.size(); i != e; ++i) {
-    if (IsSafeComputationToRemove(Dead[i].first)) {
+    if (IsSafeComputationToRemove(Dead[i].first, TLI)) {
       Dead[i].second->eraseFromParent();
       Instruction *I = Dead[i].first;
       do {
-	if (isAllocationFn(I))
+	if (isAllocationFn(I, TLI))
 	  break;
         Instruction *J = dyn_cast<Instruction>(I->getOperand(0));
         if (!J)
@@ -463,7 +473,7 @@
 /// quick scan over the use list to clean up the easy and obvious cruft.  This
 /// returns true if it made a change.
 static bool CleanupConstantGlobalUsers(Value *V, Constant *Init,
-                                       TargetData *TD, TargetLibraryInfo *TLI) {
+                                       DataLayout *TD, TargetLibraryInfo *TLI) {
   bool Changed = false;
   for (Value::use_iterator UI = V->use_begin(), E = V->use_end(); UI != E;) {
     User *U = *UI++;
@@ -655,7 +665,7 @@
 /// behavior of the program in a more fine-grained way.  We have determined that
 /// this transformation is safe already.  We return the first global variable we
 /// insert so that the caller can reprocess it.
-static GlobalVariable *SRAGlobal(GlobalVariable *GV, const TargetData &TD) {
+static GlobalVariable *SRAGlobal(GlobalVariable *GV, const DataLayout &TD) {
   // Make sure this global only has simple uses that we can SRA.
   if (!GlobalUsersSafeToSRA(GV))
     return 0;
@@ -931,7 +941,7 @@
 /// if the loaded value is dynamically null, then we know that they cannot be
 /// reachable with a null optimize away the load.
 static bool OptimizeAwayTrappingUsesOfLoads(GlobalVariable *GV, Constant *LV,
-                                            TargetData *TD,
+                                            DataLayout *TD,
                                             TargetLibraryInfo *TLI) {
   bool Changed = false;
 
@@ -961,7 +971,9 @@
       // If we get here we could have other crazy uses that are transitively
       // loaded.
       assert((isa<PHINode>(GlobalUser) || isa<SelectInst>(GlobalUser) ||
-              isa<ConstantExpr>(GlobalUser) || isa<CmpInst>(GlobalUser)) &&
+              isa<ConstantExpr>(GlobalUser) || isa<CmpInst>(GlobalUser) ||
+              isa<BitCastInst>(GlobalUser) ||
+              isa<GetElementPtrInst>(GlobalUser)) &&
              "Only expect load and stores!");
     }
   }
@@ -975,7 +987,7 @@
   // nor is the global.
   if (AllNonStoreUsesGone) {
     if (isLeakCheckerRoot(GV)) {
-      Changed |= CleanupPointerRootUsers(GV);
+      Changed |= CleanupPointerRootUsers(GV, TLI);
     } else {
       Changed = true;
       CleanupConstantGlobalUsers(GV, 0, TD, TLI);
@@ -993,7 +1005,7 @@
 /// ConstantPropUsersOf - Walk the use list of V, constant folding all of the
 /// instructions that are foldable.
 static void ConstantPropUsersOf(Value *V,
-                                TargetData *TD, TargetLibraryInfo *TLI) {
+                                DataLayout *TD, TargetLibraryInfo *TLI) {
   for (Value::use_iterator UI = V->use_begin(), E = V->use_end(); UI != E; )
     if (Instruction *I = dyn_cast<Instruction>(*UI++))
       if (Constant *NewC = ConstantFoldInstruction(I, TD, TLI)) {
@@ -1016,7 +1028,7 @@
                                                      CallInst *CI,
                                                      Type *AllocTy,
                                                      ConstantInt *NElements,
-                                                     TargetData *TD,
+                                                     DataLayout *TD,
                                                      TargetLibraryInfo *TLI) {
   DEBUG(errs() << "PROMOTING GLOBAL: " << *GV << "  CALL = " << *CI << '\n');
 
@@ -1465,9 +1477,10 @@
 /// PerformHeapAllocSRoA - CI is an allocation of an array of structures.  Break
 /// it up into multiple allocations of arrays of the fields.
 static GlobalVariable *PerformHeapAllocSRoA(GlobalVariable *GV, CallInst *CI,
-                                            Value *NElems, TargetData *TD) {
+                                            Value *NElems, DataLayout *TD,
+                                            const TargetLibraryInfo *TLI) {
   DEBUG(dbgs() << "SROA HEAP ALLOC: " << *GV << "  MALLOC = " << *CI << '\n');
-  Type *MAT = getMallocAllocatedType(CI);
+  Type *MAT = getMallocAllocatedType(CI, TLI);
   StructType *STy = cast<StructType>(MAT);
 
   // There is guaranteed to be at least one use of the malloc (storing
@@ -1656,7 +1669,7 @@
                                                Type *AllocTy,
                                                AtomicOrdering Ordering,
                                                Module::global_iterator &GVI,
-                                               TargetData *TD,
+                                               DataLayout *TD,
                                                TargetLibraryInfo *TLI) {
   if (!TD)
     return false;
@@ -1688,7 +1701,7 @@
   // This eliminates dynamic allocation, avoids an indirection accessing the
   // data, and exposes the resultant global to further GlobalOpt.
   // We cannot optimize the malloc if we cannot determine malloc array size.
-  Value *NElems = getMallocArraySize(CI, TD, true);
+  Value *NElems = getMallocArraySize(CI, TD, TLI, true);
   if (!NElems)
     return false;
 
@@ -1725,7 +1738,7 @@
 
     // If this is a fixed size array, transform the Malloc to be an alloc of
     // structs.  malloc [100 x struct],1 -> malloc struct, 100
-    if (ArrayType *AT = dyn_cast<ArrayType>(getMallocAllocatedType(CI))) {
+    if (ArrayType *AT = dyn_cast<ArrayType>(getMallocAllocatedType(CI, TLI))) {
       Type *IntPtrTy = TD->getIntPtrType(CI->getContext());
       unsigned TypeSize = TD->getStructLayout(AllocSTy)->getSizeInBytes();
       Value *AllocSize = ConstantInt::get(IntPtrTy, TypeSize);
@@ -1742,7 +1755,8 @@
         CI = cast<CallInst>(Malloc);
     }
 
-    GVI = PerformHeapAllocSRoA(GV, CI, getMallocArraySize(CI, TD, true), TD);
+    GVI = PerformHeapAllocSRoA(GV, CI, getMallocArraySize(CI, TD, TLI, true),
+                               TD, TLI);
     return true;
   }
 
@@ -1754,7 +1768,7 @@
 static bool OptimizeOnceStoredGlobal(GlobalVariable *GV, Value *StoredOnceVal,
                                      AtomicOrdering Ordering,
                                      Module::global_iterator &GVI,
-                                     TargetData *TD, TargetLibraryInfo *TLI) {
+                                     DataLayout *TD, TargetLibraryInfo *TLI) {
   // Ignore no-op GEPs and bitcasts.
   StoredOnceVal = StoredOnceVal->stripPointerCasts();
 
@@ -1771,8 +1785,8 @@
       // Optimize away any trapping uses of the loaded value.
       if (OptimizeAwayTrappingUsesOfLoads(GV, SOVC, TD, TLI))
         return true;
-    } else if (CallInst *CI = extractMallocCall(StoredOnceVal)) {
-      Type *MallocType = getMallocAllocatedType(CI);
+    } else if (CallInst *CI = extractMallocCall(StoredOnceVal, TLI)) {
+      Type *MallocType = getMallocAllocatedType(CI, TLI);
       if (MallocType &&
           TryToOptimizeStoreOfMallocToGlobal(GV, CI, MallocType, Ordering, GVI,
                                              TD, TLI))
@@ -1964,7 +1978,7 @@
     bool Changed;
     if (isLeakCheckerRoot(GV)) {
       // Delete any constant stores to the global.
-      Changed = CleanupPointerRootUsers(GV);
+      Changed = CleanupPointerRootUsers(GV, TLI);
     } else {
       // Delete any stores we can find to the global.  We may not be able to
       // make it completely dead though.
@@ -1997,7 +2011,7 @@
     ++NumMarked;
     return true;
   } else if (!GV->getInitializer()->getType()->isSingleValueType()) {
-    if (TargetData *TD = getAnalysisIfAvailable<TargetData>())
+    if (DataLayout *TD = getAnalysisIfAvailable<DataLayout>())
       if (GlobalVariable *FirstNewGV = SRAGlobal(GV, *TD)) {
         GVI = FirstNewGV;  // Don't skip the newly produced globals!
         return true;
@@ -2056,25 +2070,26 @@
   }
 }
 
-static AttrListPtr StripNest(const AttrListPtr &Attrs) {
+static AttrListPtr StripNest(LLVMContext &C, const AttrListPtr &Attrs) {
   for (unsigned i = 0, e = Attrs.getNumSlots(); i != e; ++i) {
-    if ((Attrs.getSlot(i).Attrs & Attribute::Nest) == 0)
+    if (!Attrs.getSlot(i).Attrs.hasAttribute(Attributes::Nest))
       continue;
 
     // There can be only one.
-    return Attrs.removeAttr(Attrs.getSlot(i).Index, Attribute::Nest);
+    return Attrs.removeAttr(C, Attrs.getSlot(i).Index,
+                            Attributes::get(C, Attributes::Nest));
   }
 
   return Attrs;
 }
 
 static void RemoveNestAttribute(Function *F) {
-  F->setAttributes(StripNest(F->getAttributes()));
+  F->setAttributes(StripNest(F->getContext(), F->getAttributes()));
   for (Value::use_iterator UI = F->use_begin(), E = F->use_end(); UI != E;++UI){
     if (isa<BlockAddress>(*UI))
       continue;
     CallSite User(cast<Instruction>(*UI));
-    User.setAttributes(StripNest(User.getAttributes()));
+    User.setAttributes(StripNest(F->getContext(), User.getAttributes()));
   }
 }
 
@@ -2103,7 +2118,7 @@
         Changed = true;
       }
 
-      if (F->getAttributes().hasAttrSomewhere(Attribute::Nest) &&
+      if (F->getAttributes().hasAttrSomewhere(Attributes::Nest) &&
           !F->hasAddressTaken()) {
         // The function is not used by a trampoline intrinsic, so it is safe
         // to remove the 'nest' attribute.
@@ -2251,7 +2266,7 @@
 static inline bool 
 isSimpleEnoughValueToCommit(Constant *C,
                             SmallPtrSet<Constant*, 8> &SimpleConstants,
-                            const TargetData *TD);
+                            const DataLayout *TD);
 
 
 /// isSimpleEnoughValueToCommit - Return true if the specified constant can be
@@ -2264,7 +2279,7 @@
 /// time.
 static bool isSimpleEnoughValueToCommitHelper(Constant *C,
                                    SmallPtrSet<Constant*, 8> &SimpleConstants,
-                                   const TargetData *TD) {
+                                   const DataLayout *TD) {
   // Simple integer, undef, constant aggregate zero, global addresses, etc are
   // all supported.
   if (C->getNumOperands() == 0 || isa<BlockAddress>(C) ||
@@ -2319,7 +2334,7 @@
 static inline bool 
 isSimpleEnoughValueToCommit(Constant *C,
                             SmallPtrSet<Constant*, 8> &SimpleConstants,
-                            const TargetData *TD) {
+                            const DataLayout *TD) {
   // If we already checked this constant, we win.
   if (!SimpleConstants.insert(C)) return true;
   // Check the constant.
@@ -2450,7 +2465,7 @@
 /// Once an evaluation call fails, the evaluation object should not be reused.
 class Evaluator {
 public:
-  Evaluator(const TargetData *TD, const TargetLibraryInfo *TLI)
+  Evaluator(const DataLayout *TD, const TargetLibraryInfo *TLI)
     : TD(TD), TLI(TLI) {
     ValueStack.push_back(new DenseMap<Value*, Constant*>);
   }
@@ -2531,7 +2546,7 @@
   /// simple enough to live in a static initializer of a global.
   SmallPtrSet<Constant*, 8> SimpleConstants;
 
-  const TargetData *TD;
+  const DataLayout *TD;
   const TargetLibraryInfo *TLI;
 };
 
@@ -2869,7 +2884,7 @@
 
 /// EvaluateStaticConstructor - Evaluate static constructors in the function, if
 /// we can.  Return true if we can, false otherwise.
-static bool EvaluateStaticConstructor(Function *F, const TargetData *TD,
+static bool EvaluateStaticConstructor(Function *F, const DataLayout *TD,
                                       const TargetLibraryInfo *TLI) {
   // Call the function.
   Evaluator Eval(TD, TLI);
@@ -3110,7 +3125,7 @@
 bool GlobalOpt::runOnModule(Module &M) {
   bool Changed = false;
 
-  TD = getAnalysisIfAvailable<TargetData>();
+  TD = getAnalysisIfAvailable<DataLayout>();
   TLI = &getAnalysis<TargetLibraryInfo>();
 
   // Try to find the llvm.globalctors list.

Modified: llvm/branches/AMDILBackend/lib/Transforms/IPO/IPO.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/IPO/IPO.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/IPO/IPO.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/IPO/IPO.cpp Tue Jan 15 11:16:16 2013
@@ -1,4 +1,4 @@
-//===-- Scalar.cpp --------------------------------------------------------===//
+//===-- IPO.cpp -----------------------------------------------------------===//
 //
 //                     The LLVM Compiler Infrastructure
 //
@@ -95,7 +95,10 @@
 }
 
 void LLVMAddInternalizePass(LLVMPassManagerRef PM, unsigned AllButMain) {
-  unwrap(PM)->add(createInternalizePass(AllButMain != 0));
+  std::vector<const char *> Export;
+  if (AllButMain)
+    Export.push_back("main");
+  unwrap(PM)->add(createInternalizePass(Export));
 }
 
 void LLVMAddStripDeadPrototypesPass(LLVMPassManagerRef PM) {

Modified: llvm/branches/AMDILBackend/lib/Transforms/IPO/InlineAlways.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/IPO/InlineAlways.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/IPO/InlineAlways.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/IPO/InlineAlways.cpp Tue Jan 15 11:16:16 2013
@@ -23,7 +23,7 @@
 #include "llvm/Support/CallSite.h"
 #include "llvm/Transforms/IPO.h"
 #include "llvm/Transforms/IPO/InlinerPass.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/ADT/SmallPtrSet.h"
 
 using namespace llvm;
@@ -65,7 +65,7 @@
 
 /// \brief Minimal filter to detect invalid constructs for inlining.
 static bool isInlineViable(Function &F) {
-  bool ReturnsTwice = F.hasFnAttr(Attribute::ReturnsTwice);
+  bool ReturnsTwice =F.getFnAttributes().hasAttribute(Attributes::ReturnsTwice);
   for (Function::iterator BI = F.begin(), BE = F.end(); BI != BE; ++BI) {
     // Disallow inlining of functions which contain an indirect branch.
     if (isa<IndirectBrInst>(BI->getTerminator()))
@@ -114,7 +114,7 @@
   if (Callee->isDeclaration()) return InlineCost::getNever();
 
   // Return never for anything not marked as always inline.
-  if (!Callee->hasFnAttr(Attribute::AlwaysInline))
+  if (!Callee->getFnAttributes().hasAttribute(Attributes::AlwaysInline))
     return InlineCost::getNever();
 
   // Do some minimal analysis to preclude non-viable functions.

Modified: llvm/branches/AMDILBackend/lib/Transforms/IPO/InlineSimple.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/IPO/InlineSimple.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/IPO/InlineSimple.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/IPO/InlineSimple.cpp Tue Jan 15 11:16:16 2013
@@ -22,7 +22,7 @@
 #include "llvm/Support/CallSite.h"
 #include "llvm/Transforms/IPO.h"
 #include "llvm/Transforms/IPO/InlinerPass.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 
 using namespace llvm;
 
@@ -62,7 +62,7 @@
 // doInitialization - Initializes the vector of functions that have been
 // annotated with the noinline attribute.
 bool SimpleInliner::doInitialization(CallGraph &CG) {
-  CA.setTargetData(getAnalysisIfAvailable<TargetData>());
+  CA.setDataLayout(getAnalysisIfAvailable<DataLayout>());
   return false;
 }
 

Modified: llvm/branches/AMDILBackend/lib/Transforms/IPO/Inliner.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/IPO/Inliner.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/IPO/Inliner.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/IPO/Inliner.cpp Tue Jan 15 11:16:16 2013
@@ -19,7 +19,8 @@
 #include "llvm/IntrinsicInst.h"
 #include "llvm/Analysis/CallGraph.h"
 #include "llvm/Analysis/InlineCost.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
+#include "llvm/Target/TargetLibraryInfo.h"
 #include "llvm/Transforms/IPO/InlinerPass.h"
 #include "llvm/Transforms/Utils/Cloning.h"
 #include "llvm/Transforms/Utils/Local.h"
@@ -92,11 +93,11 @@
 
   // If the inlined function had a higher stack protection level than the
   // calling function, then bump up the caller's stack protection level.
-  if (Callee->hasFnAttr(Attribute::StackProtectReq))
-    Caller->addFnAttr(Attribute::StackProtectReq);
-  else if (Callee->hasFnAttr(Attribute::StackProtect) &&
-           !Caller->hasFnAttr(Attribute::StackProtectReq))
-    Caller->addFnAttr(Attribute::StackProtect);
+  if (Callee->getFnAttributes().hasAttribute(Attributes::StackProtectReq))
+    Caller->addFnAttr(Attributes::StackProtectReq);
+  else if (Callee->getFnAttributes().hasAttribute(Attributes::StackProtect) &&
+           !Caller->getFnAttributes().hasAttribute(Attributes::StackProtectReq))
+    Caller->addFnAttr(Attributes::StackProtect);
 
   // Look at all of the allocas that we inlined through this call site.  If we
   // have already inlined other allocas through other calls into this function,
@@ -208,14 +209,15 @@
   // would decrease the threshold.
   Function *Caller = CS.getCaller();
   bool OptSize = Caller && !Caller->isDeclaration() &&
-    Caller->hasFnAttr(Attribute::OptimizeForSize);
-  if (!(InlineLimit.getNumOccurrences() > 0) && OptSize && OptSizeThreshold < thres)
+    Caller->getFnAttributes().hasAttribute(Attributes::OptimizeForSize);
+  if (!(InlineLimit.getNumOccurrences() > 0) && OptSize &&
+      OptSizeThreshold < thres)
     thres = OptSizeThreshold;
 
   // Listen to the inlinehint attribute when it would increase the threshold.
   Function *Callee = CS.getCalledFunction();
   bool InlineHint = Callee && !Callee->isDeclaration() &&
-    Callee->hasFnAttr(Attribute::InlineHint);
+    Callee->getFnAttributes().hasAttribute(Attributes::InlineHint);
   if (InlineHint && HintThreshold > thres)
     thres = HintThreshold;
 
@@ -338,7 +340,8 @@
 
 bool Inliner::runOnSCC(CallGraphSCC &SCC) {
   CallGraph &CG = getAnalysis<CallGraph>();
-  const TargetData *TD = getAnalysisIfAvailable<TargetData>();
+  const DataLayout *TD = getAnalysisIfAvailable<DataLayout>();
+  const TargetLibraryInfo *TLI = getAnalysisIfAvailable<TargetLibraryInfo>();
 
   SmallPtrSet<Function*, 8> SCCFunctions;
   DEBUG(dbgs() << "Inliner visiting SCC:");
@@ -417,7 +420,7 @@
       // just delete the call instead of trying to inline it, regardless of
       // size.  This happens because IPSCCP propagates the result out of the
       // call and then we're left with the dead call.
-      if (isInstructionTriviallyDead(CS.getInstruction())) {
+      if (isInstructionTriviallyDead(CS.getInstruction(), TLI)) {
         DEBUG(dbgs() << "    -> Deleting dead call: "
                      << *CS.getInstruction() << "\n");
         // Update the call graph by deleting the edge from Callee to Caller.
@@ -530,7 +533,8 @@
     // Handle the case when this function is called and we only want to care
     // about always-inline functions. This is a bit of a hack to share code
     // between here and the InlineAlways pass.
-    if (AlwaysInlineOnly && !F->hasFnAttr(Attribute::AlwaysInline))
+    if (AlwaysInlineOnly &&
+        !F->getFnAttributes().hasAttribute(Attributes::AlwaysInline))
       continue;
 
     // If the only remaining users of the function are dead constants, remove

Modified: llvm/branches/AMDILBackend/lib/Transforms/IPO/Internalize.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/IPO/Internalize.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/IPO/Internalize.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/IPO/Internalize.cpp Tue Jan 15 11:16:16 2013
@@ -7,9 +7,9 @@
 //
 //===----------------------------------------------------------------------===//
 //
-// This pass loops over all of the functions in the input module, looking for a
-// main function.  If a main function is found, all other functions and all
-// global variables with initializers are marked as internal.
+// This pass loops over all of the functions and variables in the input module.
+// If the function or variable is not in the list of external names given to
+// the pass it is marked as internal.
 //
 //===----------------------------------------------------------------------===//
 
@@ -45,12 +45,9 @@
 namespace {
   class InternalizePass : public ModulePass {
     std::set<std::string> ExternalNames;
-    /// If no api symbols were specified and a main function is defined,
-    /// assume the main function is the only API
-    bool AllButMain;
   public:
     static char ID; // Pass identification, replacement for typeid
-    explicit InternalizePass(bool AllButMain = true);
+    explicit InternalizePass();
     explicit InternalizePass(const std::vector <const char *>& exportList);
     void LoadFile(const char *Filename);
     virtual bool runOnModule(Module &M);
@@ -66,8 +63,8 @@
 INITIALIZE_PASS(InternalizePass, "internalize",
                 "Internalize Global Symbols", false, false)
 
-InternalizePass::InternalizePass(bool AllButMain)
-  : ModulePass(ID), AllButMain(AllButMain){
+InternalizePass::InternalizePass()
+  : ModulePass(ID) {
   initializeInternalizePassPass(*PassRegistry::getPassRegistry());
   if (!APIFile.empty())           // If a filename is specified, use it.
     LoadFile(APIFile.c_str());
@@ -76,7 +73,7 @@
 }
 
 InternalizePass::InternalizePass(const std::vector<const char *>&exportList)
-  : ModulePass(ID), AllButMain(false){
+  : ModulePass(ID){
   initializeInternalizePassPass(*PassRegistry::getPassRegistry());
   for(std::vector<const char *>::const_iterator itr = exportList.begin();
         itr != exportList.end(); itr++) {
@@ -103,23 +100,6 @@
 bool InternalizePass::runOnModule(Module &M) {
   CallGraph *CG = getAnalysisIfAvailable<CallGraph>();
   CallGraphNode *ExternalNode = CG ? CG->getExternalCallingNode() : 0;
-  
-  if (ExternalNames.empty()) {
-    // Return if we're not in 'all but main' mode and have no external api
-    if (!AllButMain)
-      return false;
-    // If no list or file of symbols was specified, check to see if there is a
-    // "main" symbol defined in the module.  If so, use it, otherwise do not
-    // internalize the module, it must be a library or something.
-    //
-    Function *MainFunc = M.getFunction("main");
-    if (MainFunc == 0 || MainFunc->isDeclaration())
-      return false;  // No main found, must be a library...
-
-    // Preserve main, internalize all else.
-    ExternalNames.insert(MainFunc->getName());
-  }
-
   bool Changed = false;
 
   // Never internalize functions which code-gen might insert.
@@ -189,8 +169,8 @@
   return Changed;
 }
 
-ModulePass *llvm::createInternalizePass(bool AllButMain) {
-  return new InternalizePass(AllButMain);
+ModulePass *llvm::createInternalizePass() {
+  return new InternalizePass();
 }
 
 ModulePass *llvm::createInternalizePass(const std::vector <const char *> &el) {

Modified: llvm/branches/AMDILBackend/lib/Transforms/IPO/MergeFunctions.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/IPO/MergeFunctions.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/IPO/MergeFunctions.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/IPO/MergeFunctions.cpp Tue Jan 15 11:16:16 2013
@@ -63,7 +63,7 @@
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/ValueHandle.h"
 #include "llvm/Support/raw_ostream.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include <vector>
 using namespace llvm;
 
@@ -92,19 +92,19 @@
 namespace {
 
 /// ComparableFunction - A struct that pairs together functions with a
-/// TargetData so that we can keep them together as elements in the DenseSet.
+/// DataLayout so that we can keep them together as elements in the DenseSet.
 class ComparableFunction {
 public:
   static const ComparableFunction EmptyKey;
   static const ComparableFunction TombstoneKey;
-  static TargetData * const LookupOnly;
+  static DataLayout * const LookupOnly;
 
-  ComparableFunction(Function *Func, TargetData *TD)
+  ComparableFunction(Function *Func, DataLayout *TD)
     : Func(Func), Hash(profileFunction(Func)), TD(TD) {}
 
   Function *getFunc() const { return Func; }
   unsigned getHash() const { return Hash; }
-  TargetData *getTD() const { return TD; }
+  DataLayout *getTD() const { return TD; }
 
   // Drops AssertingVH reference to the function. Outside of debug mode, this
   // does nothing.
@@ -120,13 +120,13 @@
 
   AssertingVH<Function> Func;
   unsigned Hash;
-  TargetData *TD;
+  DataLayout *TD;
 };
 
 const ComparableFunction ComparableFunction::EmptyKey = ComparableFunction(0);
 const ComparableFunction ComparableFunction::TombstoneKey =
     ComparableFunction(1);
-TargetData *const ComparableFunction::LookupOnly = (TargetData*)(-1);
+DataLayout *const ComparableFunction::LookupOnly = (DataLayout*)(-1);
 
 }
 
@@ -150,12 +150,12 @@
 namespace {
 
 /// FunctionComparator - Compares two functions to determine whether or not
-/// they will generate machine code with the same behaviour. TargetData is
+/// they will generate machine code with the same behaviour. DataLayout is
 /// used if available. The comparator always fails conservatively (erring on the
 /// side of claiming that two functions are different).
 class FunctionComparator {
 public:
-  FunctionComparator(const TargetData *TD, const Function *F1,
+  FunctionComparator(const DataLayout *TD, const Function *F1,
                      const Function *F2)
     : F1(F1), F2(F2), TD(TD) {}
 
@@ -190,7 +190,7 @@
   // The two functions undergoing comparison.
   const Function *F1, *F2;
 
-  const TargetData *TD;
+  const DataLayout *TD;
 
   DenseMap<const Value *, const Value *> id_map;
   DenseSet<const Value *> seen_values;
@@ -591,8 +591,8 @@
   /// to modify it.
   FnSetType FnSet;
 
-  /// TargetData for more accurate GEP comparisons. May be NULL.
-  TargetData *TD;
+  /// DataLayout for more accurate GEP comparisons. May be NULL.
+  DataLayout *TD;
 
   /// Whether or not the target supports global aliases.
   bool HasGlobalAliases;
@@ -609,7 +609,7 @@
 
 bool MergeFunctions::runOnModule(Module &M) {
   bool Changed = false;
-  TD = getAnalysisIfAvailable<TargetData>();
+  TD = getAnalysisIfAvailable<DataLayout>();
 
   for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I) {
     if (!I->isDeclaration() && !I->hasAvailableExternallyLinkage())

Modified: llvm/branches/AMDILBackend/lib/Transforms/IPO/PassManagerBuilder.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/IPO/PassManagerBuilder.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/IPO/PassManagerBuilder.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/IPO/PassManagerBuilder.cpp Tue Jan 15 11:16:16 2013
@@ -33,13 +33,21 @@
 using namespace llvm;
 
 static cl::opt<bool>
-RunVectorization("vectorize", cl::desc("Run vectorization passes"));
+RunLoopVectorization("vectorize-loops",
+                     cl::desc("Run the Loop vectorization passes"));
+
+static cl::opt<bool>
+RunBBVectorization("vectorize", cl::desc("Run the BB vectorization passes"));
 
 static cl::opt<bool>
 UseGVNAfterVectorization("use-gvn-after-vectorization",
   cl::init(false), cl::Hidden,
   cl::desc("Run GVN instead of Early CSE after vectorization passes"));
 
+static cl::opt<bool> UseNewSROA("use-new-sroa",
+  cl::init(true), cl::Hidden,
+  cl::desc("Enable the new, experimental SROA pass"));
+
 PassManagerBuilder::PassManagerBuilder() {
     OptLevel = 2;
     SizeLevel = 0;
@@ -48,7 +56,8 @@
     DisableSimplifyLibCalls = false;
     DisableUnitAtATime = false;
     DisableUnrollLoops = false;
-    Vectorize = RunVectorization;
+    Vectorize = RunBBVectorization;
+    LoopVectorize = RunLoopVectorization;
 }
 
 PassManagerBuilder::~PassManagerBuilder() {
@@ -100,7 +109,10 @@
   addInitialAliasAnalysisPasses(FPM);
 
   FPM.add(createCFGSimplificationPass());
-  FPM.add(createScalarReplAggregatesPass());
+  if (UseNewSROA)
+    FPM.add(createSROAPass());
+  else
+    FPM.add(createScalarReplAggregatesPass());
   FPM.add(createEarlyCSEPass());
   FPM.add(createLowerExpectIntrinsicPass());
 }
@@ -112,6 +124,14 @@
       MPM.add(Inliner);
       Inliner = 0;
     }
+
+    // FIXME: This is a HACK! The inliner pass above implicitly creates a CGSCC
+    // pass manager, but we don't want to add extensions into that pass manager.
+    // To prevent this we must insert a no-op module pass to reset the pass
+    // manager to get the same behavior as EP_OptimizerLast in non-O0 builds.
+    if (!GlobalExtensions->empty() || !Extensions.empty())
+      MPM.add(createBarrierNoopPass());
+
     addExtensionsToPM(EP_EnabledOnOptLevel0, MPM);
     return;
   }
@@ -147,7 +167,10 @@
 
   // Start of function pass.
   // Break up aggregate allocas, using SSAUpdater.
-  MPM.add(createScalarReplAggregatesPass(-1, false));
+  if (UseNewSROA)
+    MPM.add(createSROAPass(/*RequiresDomTree*/ false));
+  else
+    MPM.add(createScalarReplAggregatesPass(-1, false));
   MPM.add(createEarlyCSEPass());              // Catch trivial redundancies
   if (!DisableSimplifyLibCalls)
     MPM.add(createSimplifyLibCallsPass());    // Library Call Optimizations
@@ -166,6 +189,12 @@
   MPM.add(createIndVarSimplifyPass());        // Canonicalize indvars
   MPM.add(createLoopIdiomPass());             // Recognize idioms like memset.
   MPM.add(createLoopDeletionPass());          // Delete dead loops
+
+  if (LoopVectorize) {
+    MPM.add(createLoopVectorizePass());
+    MPM.add(createLICMPass());
+  }
+
   if (!DisableUnrollLoops)
     MPM.add(createLoopUnrollPass());          // Unroll small loops
   addExtensionsToPM(EP_LoopOptimizerEnd, MPM);
@@ -201,13 +230,12 @@
     // FIXME: We shouldn't bother with this anymore.
     MPM.add(createStripDeadPrototypesPass()); // Get rid of dead prototypes
 
-    // GlobalOpt already deletes dead functions and globals, at -O3 try a
+    // GlobalOpt already deletes dead functions and globals, at -O2 try a
     // late pass of GlobalDCE.  It is capable of deleting dead cycles.
-    if (OptLevel > 2)
+    if (OptLevel > 1) {
       MPM.add(createGlobalDCEPass());         // Remove dead fns and globals.
-
-    if (OptLevel > 1)
       MPM.add(createConstantMergePass());     // Merge dup global constants
+    }
   }
   addExtensionsToPM(EP_OptimizerLast, MPM);
 }
@@ -222,8 +250,11 @@
   // Now that composite has been compiled, scan through the module, looking
   // for a main function.  If main is defined, mark all other functions
   // internal.
-  if (Internalize)
-    PM.add(createInternalizePass(true));
+  if (Internalize) {
+    std::vector<const char*> E;
+    E.push_back("main");
+    PM.add(createInternalizePass(E));
+  }
 
   // Propagate constants at call sites into the functions they call.  This
   // opens opportunities for globalopt (and inlining) by substituting function
@@ -265,7 +296,10 @@
   PM.add(createInstructionCombiningPass());
   PM.add(createJumpThreadingPass());
   // Break up allocas
-  PM.add(createScalarReplAggregatesPass());
+  if (UseNewSROA)
+    PM.add(createSROAPass());
+  else
+    PM.add(createScalarReplAggregatesPass());
 
   // Run a few AA driven optimizations here and now, to cleanup the code.
   PM.add(createFunctionAttrsPass()); // Add nocapture.

Modified: llvm/branches/AMDILBackend/lib/Transforms/IPO/PruneEH.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/IPO/PruneEH.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/IPO/PruneEH.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/IPO/PruneEH.cpp Tue Jan 15 11:16:16 2013
@@ -137,16 +137,18 @@
   // If the SCC doesn't unwind or doesn't throw, note this fact.
   if (!SCCMightUnwind || !SCCMightReturn)
     for (CallGraphSCC::iterator I = SCC.begin(), E = SCC.end(); I != E; ++I) {
-      Attributes NewAttributes = Attribute::None;
+      AttrBuilder NewAttributes;
 
       if (!SCCMightUnwind)
-        NewAttributes |= Attribute::NoUnwind;
+        NewAttributes.addAttribute(Attributes::NoUnwind);
       if (!SCCMightReturn)
-        NewAttributes |= Attribute::NoReturn;
+        NewAttributes.addAttribute(Attributes::NoReturn);
 
       Function *F = (*I)->getFunction();
       const AttrListPtr &PAL = F->getAttributes();
-      const AttrListPtr &NPAL = PAL.addAttr(~0, NewAttributes);
+      const AttrListPtr &NPAL = PAL.addAttr(F->getContext(), ~0,
+                                            Attributes::get(F->getContext(),
+                                                            NewAttributes));
       if (PAL != NPAL) {
         MadeChange = true;
         F->setAttributes(NPAL);

Modified: llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombine.h
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombine.h?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombine.h (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombine.h Tue Jan 15 11:16:16 2013
@@ -18,10 +18,11 @@
 #include "llvm/Analysis/ValueTracking.h"
 #include "llvm/Support/InstVisitor.h"
 #include "llvm/Support/TargetFolder.h"
+#include "llvm/Transforms/Utils/SimplifyLibCalls.h"
 
 namespace llvm {
   class CallSite;
-  class TargetData;
+  class DataLayout;
   class TargetLibraryInfo;
   class DbgDeclareInst;
   class MemIntrinsic;
@@ -71,9 +72,10 @@
 class LLVM_LIBRARY_VISIBILITY InstCombiner
                              : public FunctionPass,
                                public InstVisitor<InstCombiner, Instruction*> {
-  TargetData *TD;
+  DataLayout *TD;
   TargetLibraryInfo *TLI;
   bool MadeIRChange;
+  LibCallSimplifier *Simplifier;
 public:
   /// Worklist - All of the instructions that need to be simplified.
   InstCombineWorklist Worklist;
@@ -95,7 +97,7 @@
 
   virtual void getAnalysisUsage(AnalysisUsage &AU) const;
 
-  TargetData *getTargetData() const { return TD; }
+  DataLayout *getDataLayout() const { return TD; }
 
   TargetLibraryInfo *getTargetLibraryInfo() const { return TLI; }
 
@@ -218,7 +220,7 @@
                           Type *Ty);
 
   Instruction *visitCallSite(CallSite CS);
-  Instruction *tryOptimizeCall(CallInst *CI, const TargetData *TD);
+  Instruction *tryOptimizeCall(CallInst *CI, const DataLayout *TD);
   bool transformConstExprCastCall(CallSite CS);
   Instruction *transformCallThroughTrampoline(CallSite CS,
                                               IntrinsicInst *Tramp);
@@ -365,6 +367,10 @@
 
 
   Value *EvaluateInDifferentType(Value *V, Type *Ty, bool isSigned);
+
+  /// Descale - Return a value X such that Val = X * Scale, or null if none.  If
+  /// the multiplication is known not to overflow then NoSignedWrap is set.
+  Value *Descale(Value *Val, APInt Scale, bool &NoSignedWrap);
 };
 
       

Modified: llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineAddSub.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineAddSub.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineAddSub.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineAddSub.cpp Tue Jan 15 11:16:16 2013
@@ -13,7 +13,7 @@
 
 #include "InstCombine.h"
 #include "llvm/Analysis/InstructionSimplify.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Support/GetElementPtrTypeIterator.h"
 #include "llvm/Support/PatternMatch.h"
 using namespace llvm;

Modified: llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineCalls.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineCalls.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineCalls.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineCalls.cpp Tue Jan 15 11:16:16 2013
@@ -13,7 +13,7 @@
 
 #include "InstCombine.h"
 #include "llvm/Support/CallSite.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Analysis/MemoryBuiltins.h"
 #include "llvm/Transforms/Utils/BuildLibCalls.h"
 #include "llvm/Transforms/Utils/Local.h"
@@ -29,6 +29,26 @@
   return Ty;
 }
 
+/// reduceToSingleValueType - Given an aggregate type which ultimately holds a
+/// single scalar element, like {{{type}}} or [1 x type], return type.
+static Type *reduceToSingleValueType(Type *T) {
+  while (!T->isSingleValueType()) {
+    if (StructType *STy = dyn_cast<StructType>(T)) {
+      if (STy->getNumElements() == 1)
+        T = STy->getElementType(0);
+      else
+        break;
+    } else if (ArrayType *ATy = dyn_cast<ArrayType>(T)) {
+      if (ATy->getNumElements() == 1)
+        T = ATy->getElementType();
+      else
+        break;
+    } else
+      break;
+  }
+
+  return T;
+}
 
 Instruction *InstCombiner::SimplifyMemTransfer(MemIntrinsic *MI) {
   unsigned DstAlign = getKnownAlignment(MI->getArgOperand(0), TD);
@@ -51,8 +71,8 @@
   // if the size is something we can handle with a single primitive load/store.
   // A single load+store correctly handles overlapping memory in the memmove
   // case.
-  unsigned Size = MemOpLength->getZExtValue();
-  if (Size == 0) return MI;  // Delete this mem transfer.
+  uint64_t Size = MemOpLength->getLimitedValue();
+  assert(Size && "0-sized memory transfering should be removed already.");
 
   if (Size > 8 || (Size&(Size-1)))
     return 0;  // If not 1/2/4/8 bytes, exit.
@@ -74,35 +94,37 @@
   // dest address will be promotable.  See if we can find a better type than the
   // integer datatype.
   Value *StrippedDest = MI->getArgOperand(0)->stripPointerCasts();
+  MDNode *CopyMD = 0;
   if (StrippedDest != MI->getArgOperand(0)) {
     Type *SrcETy = cast<PointerType>(StrippedDest->getType())
                                     ->getElementType();
     if (TD && SrcETy->isSized() && TD->getTypeStoreSize(SrcETy) == Size) {
       // The SrcETy might be something like {{{double}}} or [1 x double].  Rip
       // down through these levels if so.
-      while (!SrcETy->isSingleValueType()) {
-        if (StructType *STy = dyn_cast<StructType>(SrcETy)) {
-          if (STy->getNumElements() == 1)
-            SrcETy = STy->getElementType(0);
-          else
-            break;
-        } else if (ArrayType *ATy = dyn_cast<ArrayType>(SrcETy)) {
-          if (ATy->getNumElements() == 1)
-            SrcETy = ATy->getElementType();
-          else
-            break;
-        } else
-          break;
-      }
+      SrcETy = reduceToSingleValueType(SrcETy);
 
       if (SrcETy->isSingleValueType()) {
         NewSrcPtrTy = PointerType::get(SrcETy, SrcAddrSp);
         NewDstPtrTy = PointerType::get(SrcETy, DstAddrSp);
+
+        // If the memcpy has metadata describing the members, see if we can
+        // get the TBAA tag describing our copy.
+        if (MDNode *M = MI->getMetadata(LLVMContext::MD_tbaa_struct)) {
+          if (M->getNumOperands() == 3 &&
+              M->getOperand(0) &&
+              isa<ConstantInt>(M->getOperand(0)) &&
+              cast<ConstantInt>(M->getOperand(0))->isNullValue() &&
+              M->getOperand(1) &&
+              isa<ConstantInt>(M->getOperand(1)) &&
+              cast<ConstantInt>(M->getOperand(1))->getValue() == Size &&
+              M->getOperand(2) &&
+              isa<MDNode>(M->getOperand(2)))
+            CopyMD = cast<MDNode>(M->getOperand(2));
+        }
       }
     }
   }
 
-
   // If the memcpy/memmove provides better alignment info than we can
   // infer, use it.
   SrcAlign = std::max(SrcAlign, CopyAlign);
@@ -112,8 +134,12 @@
   Value *Dest = Builder->CreateBitCast(MI->getArgOperand(0), NewDstPtrTy);
   LoadInst *L = Builder->CreateLoad(Src, MI->isVolatile());
   L->setAlignment(SrcAlign);
+  if (CopyMD)
+    L->setMetadata(LLVMContext::MD_tbaa, CopyMD);
   StoreInst *S = Builder->CreateStore(L, Dest, MI->isVolatile());
   S->setAlignment(DstAlign);
+  if (CopyMD)
+    S->setMetadata(LLVMContext::MD_tbaa, CopyMD);
 
   // Set the size of the copy to 0, it will be deleted on the next iteration.
   MI->setArgOperand(2, Constant::getNullValue(MemOpLength->getType()));
@@ -133,11 +159,9 @@
   ConstantInt *FillC = dyn_cast<ConstantInt>(MI->getValue());
   if (!LenC || !FillC || !FillC->getType()->isIntegerTy(8))
     return 0;
-  uint64_t Len = LenC->getZExtValue();
+  uint64_t Len = LenC->getLimitedValue();
   Alignment = MI->getAlignment();
-
-  // If the length is zero, this is a no-op
-  if (Len == 0) return MI; // memset(d,c,0,a) -> noop
+  assert(Len && "0-sized memory setting should be removed already.");
 
   // memset(s,c,n) -> store s, c (for n=1,2,4,8)
   if (Len <= 8 && isPowerOf2_32((uint32_t)Len)) {
@@ -170,7 +194,7 @@
 /// the heavy lifting.
 ///
 Instruction *InstCombiner::visitCallInst(CallInst &CI) {
-  if (isFreeCall(&CI))
+  if (isFreeCall(&CI, TLI))
     return visitFree(CI);
 
   // If the caller function is nounwind, mark the call as nounwind, even if the
@@ -245,7 +269,7 @@
   default: break;
   case Intrinsic::objectsize: {
     uint64_t Size;
-    if (getObjectSize(II->getArgOperand(0), Size, TD))
+    if (getObjectSize(II->getArgOperand(0), Size, TD, TLI))
       return ReplaceInstUsesWith(CI, ConstantInt::get(CI.getType(), Size));
     return 0;
   }
@@ -733,7 +757,7 @@
 /// passed through the varargs area, we can eliminate the use of the cast.
 static bool isSafeToEliminateVarargsCast(const CallSite CS,
                                          const CastInst * const CI,
-                                         const TargetData * const TD,
+                                         const DataLayout * const TD,
                                          const int ix) {
   if (!CI->isLosslessCast())
     return false;
@@ -754,49 +778,17 @@
   return true;
 }
 
-namespace {
-class InstCombineFortifiedLibCalls : public SimplifyFortifiedLibCalls {
-  InstCombiner *IC;
-protected:
-  void replaceCall(Value *With) {
-    NewInstruction = IC->ReplaceInstUsesWith(*CI, With);
-  }
-  bool isFoldable(unsigned SizeCIOp, unsigned SizeArgOp, bool isString) const {
-    if (CI->getArgOperand(SizeCIOp) == CI->getArgOperand(SizeArgOp))
-      return true;
-    if (ConstantInt *SizeCI =
-                           dyn_cast<ConstantInt>(CI->getArgOperand(SizeCIOp))) {
-      if (SizeCI->isAllOnesValue())
-        return true;
-      if (isString) {
-        uint64_t Len = GetStringLength(CI->getArgOperand(SizeArgOp));
-        // If the length is 0 we don't know how long it is and so we can't
-        // remove the check.
-        if (Len == 0) return false;
-        return SizeCI->getZExtValue() >= Len;
-      }
-      if (ConstantInt *Arg = dyn_cast<ConstantInt>(
-                                                  CI->getArgOperand(SizeArgOp)))
-        return SizeCI->getZExtValue() >= Arg->getZExtValue();
-    }
-    return false;
-  }
-public:
-  InstCombineFortifiedLibCalls(InstCombiner *IC) : IC(IC), NewInstruction(0) { }
-  Instruction *NewInstruction;
-};
-} // end anonymous namespace
-
 // Try to fold some different type of calls here.
 // Currently we're only working with the checking functions, memcpy_chk,
 // mempcpy_chk, memmove_chk, memset_chk, strcpy_chk, stpcpy_chk, strncpy_chk,
 // strcat_chk and strncat_chk.
-Instruction *InstCombiner::tryOptimizeCall(CallInst *CI, const TargetData *TD) {
+Instruction *InstCombiner::tryOptimizeCall(CallInst *CI, const DataLayout *TD) {
   if (CI->getCalledFunction() == 0) return 0;
 
-  InstCombineFortifiedLibCalls Simplifier(this);
-  Simplifier.fold(CI, TD, TLI);
-  return Simplifier.NewInstruction;
+  if (Value *With = Simplifier->optimizeCall(CI))
+    return ReplaceInstUsesWith(*CI, With);
+
+  return 0;
 }
 
 static IntrinsicInst *FindInitTrampolineFromAlloca(Value *TrampMem) {
@@ -879,7 +871,7 @@
 // visitCallSite - Improvements for call and invoke instructions.
 //
 Instruction *InstCombiner::visitCallSite(CallSite CS) {
-  if (isAllocLikeFn(CS.getInstruction()))
+  if (isAllocLikeFn(CS.getInstruction(), TLI))
     return visitAllocSite(*CS.getInstruction());
 
   bool Changed = false;
@@ -963,7 +955,7 @@
     Changed = true;
   }
 
-  // Try to optimize the call if possible, we require TargetData for most of
+  // Try to optimize the call if possible, we require DataLayout for most of
   // this.  None of these calls are seen as possibly dead so go ahead and
   // delete the instruction now.
   if (CallInst *CI = dyn_cast<CallInst>(CS.getInstruction())) {
@@ -1015,8 +1007,8 @@
       return false;   // Cannot transform this return value.
 
     if (!CallerPAL.isEmpty() && !Caller->use_empty()) {
-      Attributes RAttrs = CallerPAL.getRetAttributes();
-      if (RAttrs & Attribute::typeIncompatible(NewRetTy))
+      AttrBuilder RAttrs = CallerPAL.getRetAttributes();
+      if (RAttrs.hasAttributes(Attributes::typeIncompatible(NewRetTy)))
         return false;   // Attribute not compatible with transformed value.
     }
 
@@ -1046,12 +1038,13 @@
       return false;   // Cannot transform this parameter value.
 
     Attributes Attrs = CallerPAL.getParamAttributes(i + 1);
-    if (Attrs & Attribute::typeIncompatible(ParamTy))
+    if (AttrBuilder(Attrs).
+          hasAttributes(Attributes::typeIncompatible(ParamTy)))
       return false;   // Attribute not compatible with transformed value.
 
     // If the parameter is passed as a byval argument, then we have to have a
     // sized type and the sized type has to have the same size as the old type.
-    if (ParamTy != ActTy && (Attrs & Attribute::ByVal)) {
+    if (ParamTy != ActTy && Attrs.hasAttribute(Attributes::ByVal)) {
       PointerType *ParamPTy = dyn_cast<PointerType>(ParamTy);
       if (ParamPTy == 0 || !ParamPTy->getElementType()->isSized() || TD == 0)
         return false;
@@ -1103,7 +1096,7 @@
       if (CallerPAL.getSlot(i - 1).Index <= FT->getNumParams())
         break;
       Attributes PAttrs = CallerPAL.getSlot(i - 1).Attrs;
-      if (PAttrs & Attribute::VarArgsIncompatible)
+      if (PAttrs.hasIncompatibleWithVarArgsAttrs())
         return false;
     }
 
@@ -1116,15 +1109,17 @@
   attrVec.reserve(NumCommonArgs);
 
   // Get any return attributes.
-  Attributes RAttrs = CallerPAL.getRetAttributes();
+  AttrBuilder RAttrs = CallerPAL.getRetAttributes();
 
   // If the return value is not being used, the type may not be compatible
   // with the existing attributes.  Wipe out any problematic attributes.
-  RAttrs &= ~Attribute::typeIncompatible(NewRetTy);
+  RAttrs.removeAttributes(Attributes::typeIncompatible(NewRetTy));
 
   // Add the new return attributes.
-  if (RAttrs)
-    attrVec.push_back(AttributeWithIndex::get(0, RAttrs));
+  if (RAttrs.hasAttributes())
+    attrVec.push_back(
+      AttributeWithIndex::get(AttrListPtr::ReturnIndex,
+                              Attributes::get(FT->getContext(), RAttrs)));
 
   AI = CS.arg_begin();
   for (unsigned i = 0; i != NumCommonArgs; ++i, ++AI) {
@@ -1138,7 +1133,8 @@
     }
 
     // Add any parameter attributes.
-    if (Attributes PAttrs = CallerPAL.getParamAttributes(i + 1))
+    Attributes PAttrs = CallerPAL.getParamAttributes(i + 1);
+    if (PAttrs.hasAttributes())
       attrVec.push_back(AttributeWithIndex::get(i + 1, PAttrs));
   }
 
@@ -1166,19 +1162,23 @@
         }
 
         // Add any parameter attributes.
-        if (Attributes PAttrs = CallerPAL.getParamAttributes(i + 1))
+        Attributes PAttrs = CallerPAL.getParamAttributes(i + 1);
+        if (PAttrs.hasAttributes())
           attrVec.push_back(AttributeWithIndex::get(i + 1, PAttrs));
       }
     }
   }
 
-  if (Attributes FnAttrs =  CallerPAL.getFnAttributes())
-    attrVec.push_back(AttributeWithIndex::get(~0, FnAttrs));
+  Attributes FnAttrs = CallerPAL.getFnAttributes();
+  if (FnAttrs.hasAttributes())
+    attrVec.push_back(AttributeWithIndex::get(AttrListPtr::FunctionIndex,
+                                              FnAttrs));
 
   if (NewRetTy->isVoidTy())
     Caller->setName("");   // Void type should not have a name.
 
-  const AttrListPtr &NewCallerPAL = AttrListPtr::get(attrVec);
+  const AttrListPtr &NewCallerPAL = AttrListPtr::get(Callee->getContext(),
+                                                     attrVec);
 
   Instruction *NC;
   if (InvokeInst *II = dyn_cast<InvokeInst>(Caller)) {
@@ -1242,8 +1242,9 @@
 
   // If the call already has the 'nest' attribute somewhere then give up -
   // otherwise 'nest' would occur twice after splicing in the chain.
-  if (Attrs.hasAttrSomewhere(Attribute::Nest))
-    return 0;
+  for (unsigned I = 0, E = Attrs.getNumAttrs(); I != E; ++I)
+    if (Attrs.getAttributesAtIndex(I).hasAttribute(Attributes::Nest))
+      return 0;
 
   assert(Tramp &&
          "transformCallThroughTrampoline called with incorrect CallSite.");
@@ -1256,12 +1257,12 @@
   if (!NestAttrs.isEmpty()) {
     unsigned NestIdx = 1;
     Type *NestTy = 0;
-    Attributes NestAttr = Attribute::None;
+    Attributes NestAttr;
 
     // Look for a parameter marked with the 'nest' attribute.
     for (FunctionType::param_iterator I = NestFTy->param_begin(),
          E = NestFTy->param_end(); I != E; ++NestIdx, ++I)
-      if (NestAttrs.paramHasAttr(NestIdx, Attribute::Nest)) {
+      if (NestAttrs.getParamAttributes(NestIdx).hasAttribute(Attributes::Nest)){
         // Record the parameter type and any other attributes.
         NestTy = *I;
         NestAttr = NestAttrs.getParamAttributes(NestIdx);
@@ -1280,8 +1281,10 @@
       // mean appending it.  Likewise for attributes.
 
       // Add any result attributes.
-      if (Attributes Attr = Attrs.getRetAttributes())
-        NewAttrs.push_back(AttributeWithIndex::get(0, Attr));
+      Attributes Attr = Attrs.getRetAttributes();
+      if (Attr.hasAttributes())
+        NewAttrs.push_back(AttributeWithIndex::get(AttrListPtr::ReturnIndex,
+                                                   Attr));
 
       {
         unsigned Idx = 1;
@@ -1301,7 +1304,8 @@
 
           // Add the original argument and attributes.
           NewArgs.push_back(*I);
-          if (Attributes Attr = Attrs.getParamAttributes(Idx))
+          Attr = Attrs.getParamAttributes(Idx);
+          if (Attr.hasAttributes())
             NewAttrs.push_back
               (AttributeWithIndex::get(Idx + (Idx >= NestIdx), Attr));
 
@@ -1310,8 +1314,10 @@
       }
 
       // Add any function attributes.
-      if (Attributes Attr = Attrs.getFnAttributes())
-        NewAttrs.push_back(AttributeWithIndex::get(~0, Attr));
+      Attr = Attrs.getFnAttributes();
+      if (Attr.hasAttributes())
+        NewAttrs.push_back(AttributeWithIndex::get(AttrListPtr::FunctionIndex,
+                                                   Attr));
 
       // The trampoline may have been bitcast to a bogus type (FTy).
       // Handle this by synthesizing a new function type, equal to FTy
@@ -1350,7 +1356,7 @@
         NestF->getType() == PointerType::getUnqual(NewFTy) ?
         NestF : ConstantExpr::getBitCast(NestF,
                                          PointerType::getUnqual(NewFTy));
-      const AttrListPtr &NewPAL = AttrListPtr::get(NewAttrs);
+      const AttrListPtr &NewPAL = AttrListPtr::get(FTy->getContext(), NewAttrs);
 
       Instruction *NewCaller;
       if (InvokeInst *II = dyn_cast<InvokeInst>(Caller)) {

Modified: llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineCasts.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineCasts.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineCasts.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineCasts.cpp Tue Jan 15 11:16:16 2013
@@ -13,7 +13,7 @@
 
 #include "InstCombine.h"
 #include "llvm/Analysis/ConstantFolding.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Target/TargetLibraryInfo.h"
 #include "llvm/Support/PatternMatch.h"
 using namespace llvm;
@@ -78,7 +78,7 @@
 /// try to eliminate the cast by moving the type information into the alloc.
 Instruction *InstCombiner::PromoteCastOfAllocation(BitCastInst &CI,
                                                    AllocaInst &AI) {
-  // This requires TargetData to get the alloca alignment and size information.
+  // This requires DataLayout to get the alloca alignment and size information.
   if (!TD) return 0;
 
   PointerType *PTy = cast<PointerType>(CI.getType());
@@ -229,7 +229,7 @@
   const CastInst *CI, ///< The first cast instruction
   unsigned opcode,       ///< The opcode of the second cast instruction
   Type *DstTy,     ///< The target type for the second cast instruction
-  TargetData *TD         ///< The target data for pointer size
+  DataLayout *TD         ///< The target data for pointer size
 ) {
 
   Type *SrcTy = CI->getOperand(0)->getType();   // A from above
@@ -238,17 +238,20 @@
   // Get the opcodes of the two Cast instructions
   Instruction::CastOps firstOp = Instruction::CastOps(CI->getOpcode());
   Instruction::CastOps secondOp = Instruction::CastOps(opcode);
-
+  Type *SrcIntPtrTy = TD && SrcTy->isPtrOrPtrVectorTy() ?
+    TD->getIntPtrType(SrcTy) : 0;
+  Type *MidIntPtrTy = TD && MidTy->isPtrOrPtrVectorTy() ?
+    TD->getIntPtrType(MidTy) : 0;
+  Type *DstIntPtrTy = TD && DstTy->isPtrOrPtrVectorTy() ?
+    TD->getIntPtrType(DstTy) : 0;
   unsigned Res = CastInst::isEliminableCastPair(firstOp, secondOp, SrcTy, MidTy,
-                                                DstTy,
-                                  TD ? TD->getIntPtrType(CI->getContext()) : 0);
-  
+                                                DstTy, SrcIntPtrTy, MidIntPtrTy,
+                                                DstIntPtrTy);
+
   // We don't want to form an inttoptr or ptrtoint that converts to an integer
   // type that differs from the pointer size.
-  if ((Res == Instruction::IntToPtr &&
-          (!TD || SrcTy != TD->getIntPtrType(CI->getContext()))) ||
-      (Res == Instruction::PtrToInt &&
-          (!TD || DstTy != TD->getIntPtrType(CI->getContext()))))
+  if ((Res == Instruction::IntToPtr && SrcTy != DstIntPtrTy) ||
+      (Res == Instruction::PtrToInt && DstTy != SrcIntPtrTy))
     Res = 0;
   
   return Instruction::CastOps(Res);

Modified: llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineCompares.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineCompares.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineCompares.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineCompares.cpp Tue Jan 15 11:16:16 2013
@@ -16,7 +16,8 @@
 #include "llvm/Analysis/ConstantFolding.h"
 #include "llvm/Analysis/InstructionSimplify.h"
 #include "llvm/Analysis/MemoryBuiltins.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
+#include "llvm/Target/TargetLibraryInfo.h"
 #include "llvm/Support/ConstantRange.h"
 #include "llvm/Support/GetElementPtrTypeIterator.h"
 #include "llvm/Support/PatternMatch.h"
@@ -473,7 +474,7 @@
 /// If we can't emit an optimized form for this expression, this returns null.
 ///
 static Value *EvaluateGEPOffsetExpression(User *GEP, InstCombiner &IC) {
-  TargetData &TD = *IC.getTargetData();
+  DataLayout &TD = *IC.getDataLayout();
   gep_type_iterator GTI = gep_type_begin(GEP);
 
   // Check to see if this gep only has a single variable index.  If so, and if
@@ -2355,8 +2356,25 @@
         // Try not to increase register pressure.
         BO0->hasOneUse() && BO1->hasOneUse()) {
       // Determine Y and Z in the form icmp (X+Y), (X+Z).
-      Value *Y = (A == C || A == D) ? B : A;
-      Value *Z = (C == A || C == B) ? D : C;
+      Value *Y, *Z;
+      if (A == C) {
+        // C + B == C + D  ->  B == D
+        Y = B;
+        Z = D;
+      } else if (A == D) {
+        // D + B == C + D  ->  B == C
+        Y = B;
+        Z = C;
+      } else if (B == C) {
+        // A + C == C + D  ->  A == D
+        Y = A;
+        Z = D;
+      } else {
+        assert(B == D);
+        // A + D == C + D  ->  A == C
+        Y = A;
+        Z = C;
+      }
       return new ICmpInst(Pred, Y, Z);
     }
 
@@ -2894,10 +2912,6 @@
         if (!RHSF)
           break;
 
-        // We can't convert a PPC double double.
-        if (RHSF->getType()->isPPC_FP128Ty())
-          break;
-
         const fltSemantics *Sem;
         // FIXME: This shouldn't be here.
         if (LHSExt->getSrcTy()->isHalfTy())
@@ -2910,6 +2924,8 @@
           Sem = &APFloat::IEEEquad;
         else if (LHSExt->getSrcTy()->isX86_FP80Ty())
           Sem = &APFloat::x87DoubleExtended;
+        else if (LHSExt->getSrcTy()->isPPC_FP128Ty())
+          Sem = &APFloat::PPCDoubleDouble;
         else
           break;
 
@@ -2985,6 +3001,44 @@
                 return Res;
         }
         break;
+      case Instruction::Call: {
+        CallInst *CI = cast<CallInst>(LHSI);
+        LibFunc::Func Func;
+        // Various optimization for fabs compared with zero.
+        if (RHSC->isNullValue() && CI->getCalledFunction() &&
+            TLI->getLibFunc(CI->getCalledFunction()->getName(), Func) &&
+            TLI->has(Func)) {
+          if (Func == LibFunc::fabs || Func == LibFunc::fabsf ||
+              Func == LibFunc::fabsl) {
+            switch (I.getPredicate()) {
+            default: break;
+            // fabs(x) < 0 --> false
+            case FCmpInst::FCMP_OLT:
+              return ReplaceInstUsesWith(I, Builder->getFalse());
+            // fabs(x) > 0 --> x != 0
+            case FCmpInst::FCMP_OGT:
+              return new FCmpInst(FCmpInst::FCMP_ONE, CI->getArgOperand(0),
+                                  RHSC);
+            // fabs(x) <= 0 --> x == 0
+            case FCmpInst::FCMP_OLE:
+              return new FCmpInst(FCmpInst::FCMP_OEQ, CI->getArgOperand(0),
+                                  RHSC);
+            // fabs(x) >= 0 --> !isnan(x)
+            case FCmpInst::FCMP_OGE:
+              return new FCmpInst(FCmpInst::FCMP_ORD, CI->getArgOperand(0),
+                                  RHSC);
+            // fabs(x) == 0 --> x == 0
+            // fabs(x) != 0 --> x != 0
+            case FCmpInst::FCMP_OEQ:
+            case FCmpInst::FCMP_UEQ:
+            case FCmpInst::FCMP_ONE:
+            case FCmpInst::FCMP_UNE:
+              return new FCmpInst(I.getPredicate(), CI->getArgOperand(0),
+                                  RHSC);
+            }
+          }
+        }
+      }
       }
   }
 

Modified: llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp Tue Jan 15 11:16:16 2013
@@ -14,13 +14,161 @@
 #include "InstCombine.h"
 #include "llvm/IntrinsicInst.h"
 #include "llvm/Analysis/Loads.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
 #include "llvm/Transforms/Utils/Local.h"
 #include "llvm/ADT/Statistic.h"
 using namespace llvm;
 
-STATISTIC(NumDeadStore, "Number of dead stores eliminated");
+STATISTIC(NumDeadStore,    "Number of dead stores eliminated");
+STATISTIC(NumGlobalCopies, "Number of allocas copied from constant global");
+
+/// pointsToConstantGlobal - Return true if V (possibly indirectly) points to
+/// some part of a constant global variable.  This intentionally only accepts
+/// constant expressions because we can't rewrite arbitrary instructions.
+static bool pointsToConstantGlobal(Value *V) {
+  if (GlobalVariable *GV = dyn_cast<GlobalVariable>(V))
+    return GV->isConstant();
+  if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V))
+    if (CE->getOpcode() == Instruction::BitCast ||
+        CE->getOpcode() == Instruction::GetElementPtr)
+      return pointsToConstantGlobal(CE->getOperand(0));
+  return false;
+}
+
+/// isOnlyCopiedFromConstantGlobal - Recursively walk the uses of a (derived)
+/// pointer to an alloca.  Ignore any reads of the pointer, return false if we
+/// see any stores or other unknown uses.  If we see pointer arithmetic, keep
+/// track of whether it moves the pointer (with IsOffset) but otherwise traverse
+/// the uses.  If we see a memcpy/memmove that targets an unoffseted pointer to
+/// the alloca, and if the source pointer is a pointer to a constant global, we
+/// can optimize this.
+static bool
+isOnlyCopiedFromConstantGlobal(Value *V, MemTransferInst *&TheCopy,
+                               SmallVectorImpl<Instruction *> &ToDelete,
+                               bool IsOffset = false) {
+  // We track lifetime intrinsics as we encounter them.  If we decide to go
+  // ahead and replace the value with the global, this lets the caller quickly
+  // eliminate the markers.
+
+  for (Value::use_iterator UI = V->use_begin(), E = V->use_end(); UI!=E; ++UI) {
+    User *U = cast<Instruction>(*UI);
+
+    if (LoadInst *LI = dyn_cast<LoadInst>(U)) {
+      // Ignore non-volatile loads, they are always ok.
+      if (!LI->isSimple()) return false;
+      continue;
+    }
+
+    if (BitCastInst *BCI = dyn_cast<BitCastInst>(U)) {
+      // If uses of the bitcast are ok, we are ok.
+      if (!isOnlyCopiedFromConstantGlobal(BCI, TheCopy, ToDelete, IsOffset))
+        return false;
+      continue;
+    }
+    if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(U)) {
+      // If the GEP has all zero indices, it doesn't offset the pointer.  If it
+      // doesn't, it does.
+      if (!isOnlyCopiedFromConstantGlobal(GEP, TheCopy, ToDelete,
+                                          IsOffset || !GEP->hasAllZeroIndices()))
+        return false;
+      continue;
+    }
+
+    if (CallSite CS = U) {
+      // If this is the function being called then we treat it like a load and
+      // ignore it.
+      if (CS.isCallee(UI))
+        continue;
+
+      // If this is a readonly/readnone call site, then we know it is just a
+      // load (but one that potentially returns the value itself), so we can
+      // ignore it if we know that the value isn't captured.
+      unsigned ArgNo = CS.getArgumentNo(UI);
+      if (CS.onlyReadsMemory() &&
+          (CS.getInstruction()->use_empty() || CS.doesNotCapture(ArgNo)))
+        continue;
+
+      // If this is being passed as a byval argument, the caller is making a
+      // copy, so it is only a read of the alloca.
+      if (CS.isByValArgument(ArgNo))
+        continue;
+    }
+
+    // Lifetime intrinsics can be handled by the caller.
+    if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(U)) {
+      if (II->getIntrinsicID() == Intrinsic::lifetime_start ||
+          II->getIntrinsicID() == Intrinsic::lifetime_end) {
+        assert(II->use_empty() && "Lifetime markers have no result to use!");
+        ToDelete.push_back(II);
+        continue;
+      }
+    }
+
+    // If this is isn't our memcpy/memmove, reject it as something we can't
+    // handle.
+    MemTransferInst *MI = dyn_cast<MemTransferInst>(U);
+    if (MI == 0)
+      return false;
+
+    // If the transfer is using the alloca as a source of the transfer, then
+    // ignore it since it is a load (unless the transfer is volatile).
+    if (UI.getOperandNo() == 1) {
+      if (MI->isVolatile()) return false;
+      continue;
+    }
+
+    // If we already have seen a copy, reject the second one.
+    if (TheCopy) return false;
+
+    // If the pointer has been offset from the start of the alloca, we can't
+    // safely handle this.
+    if (IsOffset) return false;
+
+    // If the memintrinsic isn't using the alloca as the dest, reject it.
+    if (UI.getOperandNo() != 0) return false;
+
+    // If the source of the memcpy/move is not a constant global, reject it.
+    if (!pointsToConstantGlobal(MI->getSource()))
+      return false;
+
+    // Otherwise, the transform is safe.  Remember the copy instruction.
+    TheCopy = MI;
+  }
+  return true;
+}
+
+/// isOnlyCopiedFromConstantGlobal - Return true if the specified alloca is only
+/// modified by a copy from a constant global.  If we can prove this, we can
+/// replace any uses of the alloca with uses of the global directly.
+static MemTransferInst *
+isOnlyCopiedFromConstantGlobal(AllocaInst *AI,
+                               SmallVectorImpl<Instruction *> &ToDelete) {
+  MemTransferInst *TheCopy = 0;
+  if (isOnlyCopiedFromConstantGlobal(AI, TheCopy, ToDelete))
+    return TheCopy;
+  return 0;
+}
+
+/// getPointeeAlignment - Compute the minimum alignment of the value pointed
+/// to by the given pointer.
+static unsigned getPointeeAlignment(Value *V, const DataLayout &TD) {
+  if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V))
+    if (CE->getOpcode() == Instruction::BitCast ||
+        (CE->getOpcode() == Instruction::GetElementPtr &&
+         cast<GEPOperator>(CE)->hasAllZeroIndices()))
+      return getPointeeAlignment(CE->getOperand(0), TD);
+
+  if (GlobalVariable *GV = dyn_cast<GlobalVariable>(V))
+    if (!GV->isDeclaration())
+      return TD.getPreferredAlignment(GV);
+
+  if (PointerType *PT = dyn_cast<PointerType>(V->getType()))
+    if (PT->getElementType()->isSized())
+      return TD.getABITypeAlignment(PT->getElementType());
+
+  return 0;
+}
 
 Instruction *InstCombiner::visitAllocaInst(AllocaInst &AI) {
   // Ensure that the alloca array size argument has type intptr_t, so that
@@ -99,12 +247,16 @@
           return &AI;
         }
 
+        // If the alignment of the entry block alloca is 0 (unspecified),
+        // assign it the preferred alignment.
+        if (EntryAI->getAlignment() == 0)
+          EntryAI->setAlignment(
+            TD->getPrefTypeAlignment(EntryAI->getAllocatedType()));
         // Replace this zero-sized alloca with the one at the start of the entry
         // block after ensuring that the address will be aligned enough for both
         // types.
-        unsigned MaxAlign =
-          std::max(TD->getPrefTypeAlignment(EntryAI->getAllocatedType()),
-                   TD->getPrefTypeAlignment(AI.getAllocatedType()));
+        unsigned MaxAlign = std::max(EntryAI->getAlignment(),
+                                     AI.getAlignment());
         EntryAI->setAlignment(MaxAlign);
         if (AI.getType() != EntryAI->getType())
           return new BitCastInst(EntryAI, AI.getType());
@@ -113,6 +265,31 @@
     }
   }
 
+  if (TD) {
+    // Check to see if this allocation is only modified by a memcpy/memmove from
+    // a constant global whose alignment is equal to or exceeds that of the
+    // allocation.  If this is the case, we can change all users to use
+    // the constant global instead.  This is commonly produced by the CFE by
+    // constructs like "void foo() { int A[] = {1,2,3,4,5,6,7,8,9...}; }" if 'A'
+    // is only subsequently read.
+    SmallVector<Instruction *, 4> ToDelete;
+    if (MemTransferInst *Copy = isOnlyCopiedFromConstantGlobal(&AI, ToDelete)) {
+      if (AI.getAlignment() <= getPointeeAlignment(Copy->getSource(), *TD)) {
+        DEBUG(dbgs() << "Found alloca equal to global: " << AI << '\n');
+        DEBUG(dbgs() << "  memcpy = " << *Copy << '\n');
+        for (unsigned i = 0, e = ToDelete.size(); i != e; ++i)
+          EraseInstFromFunction(*ToDelete[i]);
+        Constant *TheSrc = cast<Constant>(Copy->getSource());
+        Instruction *NewI
+          = ReplaceInstUsesWith(AI, ConstantExpr::getBitCast(TheSrc,
+                                                             AI.getType()));
+        EraseInstFromFunction(*Copy);
+        ++NumGlobalCopies;
+        return NewI;
+      }
+    }
+  }
+
   // At last, use the generic allocation site handler to aggressively remove
   // unused allocas.
   return visitAllocSite(AI);
@@ -121,7 +298,7 @@
 
 /// InstCombineLoadCast - Fold 'load (cast P)' -> cast (load P)' when possible.
 static Instruction *InstCombineLoadCast(InstCombiner &IC, LoadInst &LI,
-                                        const TargetData *TD) {
+                                        const DataLayout *TD) {
   User *CI = cast<User>(LI.getOperand(0));
   Value *CastOp = CI->getOperand(0);
 
@@ -151,14 +328,14 @@
             SrcPTy = SrcTy->getElementType();
           }
 
-      if (IC.getTargetData() &&
+      if (IC.getDataLayout() &&
           (SrcPTy->isIntegerTy() || SrcPTy->isPointerTy() || 
             SrcPTy->isVectorTy()) &&
           // Do not allow turning this into a load of an integer, which is then
           // casted to a pointer, this pessimizes pointer analysis a lot.
           (SrcPTy->isPointerTy() == LI.getType()->isPointerTy()) &&
-          IC.getTargetData()->getTypeSizeInBits(SrcPTy) ==
-               IC.getTargetData()->getTypeSizeInBits(DestPTy)) {
+          IC.getDataLayout()->getTypeSizeInBits(SrcPTy) ==
+               IC.getDataLayout()->getTypeSizeInBits(DestPTy)) {
 
         // Okay, we are casting from one integer or pointer type to another of
         // the same size.  Instead of casting the pointer before the load, cast
@@ -336,11 +513,11 @@
   
   // If the pointers point into different address spaces or if they point to
   // values with different sizes, we can't do the transformation.
-  if (!IC.getTargetData() ||
+  if (!IC.getDataLayout() ||
       SrcTy->getAddressSpace() != 
         cast<PointerType>(CI->getType())->getAddressSpace() ||
-      IC.getTargetData()->getTypeSizeInBits(SrcPTy) !=
-      IC.getTargetData()->getTypeSizeInBits(DestPTy))
+      IC.getDataLayout()->getTypeSizeInBits(SrcPTy) !=
+      IC.getDataLayout()->getTypeSizeInBits(DestPTy))
     return 0;
 
   // Okay, we are casting from one integer or pointer type to another of

Modified: llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp Tue Jan 15 11:16:16 2013
@@ -37,7 +37,7 @@
   if (match(V, m_LShr(m_OneUse(m_Shl(m_Value(PowerOf2), m_Value(A))),
                       m_Value(B))) &&
       // The "1" can be any value known to be a power of 2.
-      isPowerOfTwo(PowerOf2, IC.getTargetData())) {
+      isPowerOfTwo(PowerOf2, IC.getDataLayout())) {
     A = IC.Builder->CreateSub(A, B);
     return IC.Builder->CreateShl(PowerOf2, A);
   }
@@ -46,7 +46,7 @@
   // inexact.  Similarly for <<.
   if (BinaryOperator *I = dyn_cast<BinaryOperator>(V))
     if (I->isLogicalShift() &&
-        isPowerOfTwo(I->getOperand(0), IC.getTargetData())) {
+        isPowerOfTwo(I->getOperand(0), IC.getDataLayout())) {
       // We know that this is an exact/nuw shift and that the input is a
       // non-zero context as well.
       if (Value *V2 = simplifyValueKnownNonZero(I->getOperand(0), IC)) {
@@ -462,12 +462,23 @@
     }
   }
 
+  // (x lshr C1) udiv C2 --> x udiv (C2 << C1)
+  if (ConstantInt *C2 = dyn_cast<ConstantInt>(Op1)) {
+    Value *X;
+    ConstantInt *C1;
+    if (match(Op0, m_LShr(m_Value(X), m_ConstantInt(C1)))) {
+      APInt NC = C2->getValue().shl(C1->getLimitedValue(C1->getBitWidth()-1));
+      return BinaryOperator::CreateUDiv(X, Builder->getInt(NC));
+    }
+  }
+
   // X udiv (C1 << N), where C1 is "1<<C2"  -->  X >> (N+C2)
   { const APInt *CI; Value *N;
     if (match(Op1, m_Shl(m_Power2(CI), m_Value(N))) ||
         match(Op1, m_ZExt(m_Shl(m_Power2(CI), m_Value(N))))) {
       if (*CI != 1)
-        N = Builder->CreateAdd(N, ConstantInt::get(I.getType(),CI->logBase2()));
+        N = Builder->CreateAdd(N,
+                               ConstantInt::get(N->getType(), CI->logBase2()));
       if (ZExtInst *Z = dyn_cast<ZExtInst>(Op1))
         N = Builder->CreateZExt(N, Z->getDestTy());
       if (I.isExact())

Modified: llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombinePHI.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombinePHI.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombinePHI.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombinePHI.cpp Tue Jan 15 11:16:16 2013
@@ -13,7 +13,7 @@
 
 #include "InstCombine.h"
 #include "llvm/Analysis/InstructionSimplify.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/ADT/SmallPtrSet.h"
 #include "llvm/ADT/STLExtras.h"
 using namespace llvm;

Modified: llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineSelect.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineSelect.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineSelect.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineSelect.cpp Tue Jan 15 11:16:16 2013
@@ -287,7 +287,7 @@
 /// SimplifyWithOpReplaced - See if V simplifies when its operand Op is
 /// replaced with RepOp.
 static Value *SimplifyWithOpReplaced(Value *V, Value *Op, Value *RepOp,
-                                     const TargetData *TD,
+                                     const DataLayout *TD,
                                      const TargetLibraryInfo *TLI) {
   // Trivial replacement.
   if (V == Op)
@@ -333,6 +333,10 @@
 
     // All operands were constants, fold it.
     if (ConstOps.size() == I->getNumOperands()) {
+      if (CmpInst *C = dyn_cast<CmpInst>(I))
+        return ConstantFoldCompareInstOperands(C->getPredicate(), ConstOps[0],
+                                               ConstOps[1], TD, TLI);
+
       if (LoadInst *LI = dyn_cast<LoadInst>(I))
         if (!LI->isVolatile())
           return ConstantFoldLoadFromConstPtr(ConstOps[0], TD);
@@ -903,7 +907,7 @@
     return &SI;
   }
 
-  if (VectorType* VecTy = dyn_cast<VectorType>(SI.getType())) {
+  if (VectorType *VecTy = dyn_cast<VectorType>(SI.getType())) {
     unsigned VWidth = VecTy->getNumElements();
     APInt UndefElts(VWidth, 0);
     APInt AllOnesEltMask(APInt::getAllOnesValue(VWidth));
@@ -912,6 +916,28 @@
         return ReplaceInstUsesWith(SI, V);
       return &SI;
     }
+
+    if (ConstantVector *CV = dyn_cast<ConstantVector>(CondVal)) {
+      // Form a shufflevector instruction.
+      SmallVector<Constant *, 8> Mask(VWidth);
+      Type *Int32Ty = Type::getInt32Ty(CV->getContext());
+      for (unsigned i = 0; i != VWidth; ++i) {
+        Constant *Elem = cast<Constant>(CV->getOperand(i));
+        if (ConstantInt *E = dyn_cast<ConstantInt>(Elem))
+          Mask[i] = ConstantInt::get(Int32Ty, i + (E->isZero() ? VWidth : 0));
+        else if (isa<UndefValue>(Elem))
+          Mask[i] = UndefValue::get(Int32Ty);
+        else
+          return 0;
+      }
+      Constant *MaskVal = ConstantVector::get(Mask);
+      Value *V = Builder->CreateShuffleVector(TrueVal, FalseVal, MaskVal);
+      return ReplaceInstUsesWith(SI, V);
+    }
+
+    if (isa<ConstantAggregateZero>(CondVal)) {
+      return ReplaceInstUsesWith(SI, FalseVal);
+    }
   }
 
   return 0;

Modified: llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineShifts.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineShifts.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineShifts.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineShifts.cpp Tue Jan 15 11:16:16 2013
@@ -190,7 +190,7 @@
       V = IC.Builder->CreateLShr(C, NumBits);
     // If we got a constantexpr back, try to simplify it with TD info.
     if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V))
-      V = ConstantFoldConstantExpression(CE, IC.getTargetData(),
+      V = ConstantFoldConstantExpression(CE, IC.getDataLayout(),
                                          IC.getTargetLibraryInfo());
     return V;
   }

Modified: llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp Tue Jan 15 11:16:16 2013
@@ -14,7 +14,7 @@
 
 
 #include "InstCombine.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/IntrinsicInst.h"
 
 using namespace llvm;

Modified: llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineVectorOps.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineVectorOps.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineVectorOps.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineVectorOps.cpp Tue Jan 15 11:16:16 2013
@@ -636,8 +636,11 @@
 
       // If LHS's width is changed, shift the mask value accordingly.
       // If newRHS == NULL, i.e. LHSOp0 == RHSOp0, we want to remap any
-      // references to RHSOp0 to LHSOp0, so we don't need to shift the mask.
-      if (eltMask >= 0 && newRHS != NULL)
+      // references from RHSOp0 to LHSOp0, so we don't need to shift the mask.
+      // If newRHS == newLHS, we want to remap any references from newRHS to
+      // newLHS so that we can properly identify splats that may occur due to
+      // obfuscation accross the two vectors.
+      if (eltMask >= 0 && newRHS != NULL && newLHS != newRHS)
         eltMask += newLHSWidth;
     }
 

Modified: llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineWorklist.h
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineWorklist.h?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineWorklist.h (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstCombineWorklist.h Tue Jan 15 11:16:16 2013
@@ -26,8 +26,8 @@
   SmallVector<Instruction*, 256> Worklist;
   DenseMap<Instruction*, unsigned> WorklistMap;
   
-  void operator=(const InstCombineWorklist&RHS);   // DO NOT IMPLEMENT
-  InstCombineWorklist(const InstCombineWorklist&); // DO NOT IMPLEMENT
+  void operator=(const InstCombineWorklist&RHS) LLVM_DELETED_FUNCTION;
+  InstCombineWorklist(const InstCombineWorklist&) LLVM_DELETED_FUNCTION;
 public:
   InstCombineWorklist() {}
   

Modified: llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstructionCombining.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstructionCombining.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstructionCombining.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/InstCombine/InstructionCombining.cpp Tue Jan 15 11:16:16 2013
@@ -40,7 +40,7 @@
 #include "llvm/Analysis/ConstantFolding.h"
 #include "llvm/Analysis/InstructionSimplify.h"
 #include "llvm/Analysis/MemoryBuiltins.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Target/TargetLibraryInfo.h"
 #include "llvm/Transforms/Utils/Local.h"
 #include "llvm/Support/CFG.h"
@@ -88,7 +88,7 @@
 
 
 Value *InstCombiner::EmitGEPOffset(User *GEP) {
-  return llvm::EmitGEPOffset(Builder, *getTargetData(), GEP);
+  return llvm::EmitGEPOffset(Builder, *getDataLayout(), GEP);
 }
 
 /// ShouldChangeType - Return true if it is desirable to convert a computation
@@ -805,6 +805,244 @@
   return true;
 }
 
+/// Descale - Return a value X such that Val = X * Scale, or null if none.  If
+/// the multiplication is known not to overflow then NoSignedWrap is set.
+Value *InstCombiner::Descale(Value *Val, APInt Scale, bool &NoSignedWrap) {
+  assert(isa<IntegerType>(Val->getType()) && "Can only descale integers!");
+  assert(cast<IntegerType>(Val->getType())->getBitWidth() ==
+         Scale.getBitWidth() && "Scale not compatible with value!");
+
+  // If Val is zero or Scale is one then Val = Val * Scale.
+  if (match(Val, m_Zero()) || Scale == 1) {
+    NoSignedWrap = true;
+    return Val;
+  }
+
+  // If Scale is zero then it does not divide Val.
+  if (Scale.isMinValue())
+    return 0;
+
+  // Look through chains of multiplications, searching for a constant that is
+  // divisible by Scale.  For example, descaling X*(Y*(Z*4)) by a factor of 4
+  // will find the constant factor 4 and produce X*(Y*Z).  Descaling X*(Y*8) by
+  // a factor of 4 will produce X*(Y*2).  The principle of operation is to bore
+  // down from Val:
+  //
+  //     Val = M1 * X          ||   Analysis starts here and works down
+  //      M1 = M2 * Y          ||   Doesn't descend into terms with more
+  //      M2 =  Z * 4          \/   than one use
+  //
+  // Then to modify a term at the bottom:
+  //
+  //     Val = M1 * X
+  //      M1 =  Z * Y          ||   Replaced M2 with Z
+  //
+  // Then to work back up correcting nsw flags.
+
+  // Op - the term we are currently analyzing.  Starts at Val then drills down.
+  // Replaced with its descaled value before exiting from the drill down loop.
+  Value *Op = Val;
+
+  // Parent - initially null, but after drilling down notes where Op came from.
+  // In the example above, Parent is (Val, 0) when Op is M1, because M1 is the
+  // 0'th operand of Val.
+  std::pair<Instruction*, unsigned> Parent;
+
+  // RequireNoSignedWrap - Set if the transform requires a descaling at deeper
+  // levels that doesn't overflow.
+  bool RequireNoSignedWrap = false;
+
+  // logScale - log base 2 of the scale.  Negative if not a power of 2.
+  int32_t logScale = Scale.exactLogBase2();
+
+  for (;; Op = Parent.first->getOperand(Parent.second)) { // Drill down
+
+    if (ConstantInt *CI = dyn_cast<ConstantInt>(Op)) {
+      // If Op is a constant divisible by Scale then descale to the quotient.
+      APInt Quotient(Scale), Remainder(Scale); // Init ensures right bitwidth.
+      APInt::sdivrem(CI->getValue(), Scale, Quotient, Remainder);
+      if (!Remainder.isMinValue())
+        // Not divisible by Scale.
+        return 0;
+      // Replace with the quotient in the parent.
+      Op = ConstantInt::get(CI->getType(), Quotient);
+      NoSignedWrap = true;
+      break;
+    }
+
+    if (BinaryOperator *BO = dyn_cast<BinaryOperator>(Op)) {
+
+      if (BO->getOpcode() == Instruction::Mul) {
+        // Multiplication.
+        NoSignedWrap = BO->hasNoSignedWrap();
+        if (RequireNoSignedWrap && !NoSignedWrap)
+          return 0;
+
+        // There are three cases for multiplication: multiplication by exactly
+        // the scale, multiplication by a constant different to the scale, and
+        // multiplication by something else.
+        Value *LHS = BO->getOperand(0);
+        Value *RHS = BO->getOperand(1);
+
+        if (ConstantInt *CI = dyn_cast<ConstantInt>(RHS)) {
+          // Multiplication by a constant.
+          if (CI->getValue() == Scale) {
+            // Multiplication by exactly the scale, replace the multiplication
+            // by its left-hand side in the parent.
+            Op = LHS;
+            break;
+          }
+
+          // Otherwise drill down into the constant.
+          if (!Op->hasOneUse())
+            return 0;
+
+          Parent = std::make_pair(BO, 1);
+          continue;
+        }
+
+        // Multiplication by something else. Drill down into the left-hand side
+        // since that's where the reassociate pass puts the good stuff.
+        if (!Op->hasOneUse())
+          return 0;
+
+        Parent = std::make_pair(BO, 0);
+        continue;
+      }
+
+      if (logScale > 0 && BO->getOpcode() == Instruction::Shl &&
+          isa<ConstantInt>(BO->getOperand(1))) {
+        // Multiplication by a power of 2.
+        NoSignedWrap = BO->hasNoSignedWrap();
+        if (RequireNoSignedWrap && !NoSignedWrap)
+          return 0;
+
+        Value *LHS = BO->getOperand(0);
+        int32_t Amt = cast<ConstantInt>(BO->getOperand(1))->
+          getLimitedValue(Scale.getBitWidth());
+        // Op = LHS << Amt.
+
+        if (Amt == logScale) {
+          // Multiplication by exactly the scale, replace the multiplication
+          // by its left-hand side in the parent.
+          Op = LHS;
+          break;
+        }
+        if (Amt < logScale || !Op->hasOneUse())
+          return 0;
+
+        // Multiplication by more than the scale.  Reduce the multiplying amount
+        // by the scale in the parent.
+        Parent = std::make_pair(BO, 1);
+        Op = ConstantInt::get(BO->getType(), Amt - logScale);
+        break;
+      }
+    }
+
+    if (!Op->hasOneUse())
+      return 0;
+
+    if (CastInst *Cast = dyn_cast<CastInst>(Op)) {
+      if (Cast->getOpcode() == Instruction::SExt) {
+        // Op is sign-extended from a smaller type, descale in the smaller type.
+        unsigned SmallSize = Cast->getSrcTy()->getPrimitiveSizeInBits();
+        APInt SmallScale = Scale.trunc(SmallSize);
+        // Suppose Op = sext X, and we descale X as Y * SmallScale.  We want to
+        // descale Op as (sext Y) * Scale.  In order to have
+        //   sext (Y * SmallScale) = (sext Y) * Scale
+        // some conditions need to hold however: SmallScale must sign-extend to
+        // Scale and the multiplication Y * SmallScale should not overflow.
+        if (SmallScale.sext(Scale.getBitWidth()) != Scale)
+          // SmallScale does not sign-extend to Scale.
+          return 0;
+        assert(SmallScale.exactLogBase2() == logScale);
+        // Require that Y * SmallScale must not overflow.
+        RequireNoSignedWrap = true;
+
+        // Drill down through the cast.
+        Parent = std::make_pair(Cast, 0);
+        Scale = SmallScale;
+        continue;
+      }
+
+      if (Cast->getOpcode() == Instruction::Trunc) {
+        // Op is truncated from a larger type, descale in the larger type.
+        // Suppose Op = trunc X, and we descale X as Y * sext Scale.  Then
+        //   trunc (Y * sext Scale) = (trunc Y) * Scale
+        // always holds.  However (trunc Y) * Scale may overflow even if
+        // trunc (Y * sext Scale) does not, so nsw flags need to be cleared
+        // from this point up in the expression (see later).
+        if (RequireNoSignedWrap)
+          return 0;
+
+        // Drill down through the cast.
+        unsigned LargeSize = Cast->getSrcTy()->getPrimitiveSizeInBits();
+        Parent = std::make_pair(Cast, 0);
+        Scale = Scale.sext(LargeSize);
+        if (logScale + 1 == (int32_t)Cast->getType()->getPrimitiveSizeInBits())
+          logScale = -1;
+        assert(Scale.exactLogBase2() == logScale);
+        continue;
+      }
+    }
+
+    // Unsupported expression, bail out.
+    return 0;
+  }
+
+  // We know that we can successfully descale, so from here on we can safely
+  // modify the IR.  Op holds the descaled version of the deepest term in the
+  // expression.  NoSignedWrap is 'true' if multiplying Op by Scale is known
+  // not to overflow.
+
+  if (!Parent.first)
+    // The expression only had one term.
+    return Op;
+
+  // Rewrite the parent using the descaled version of its operand.
+  assert(Parent.first->hasOneUse() && "Drilled down when more than one use!");
+  assert(Op != Parent.first->getOperand(Parent.second) &&
+         "Descaling was a no-op?");
+  Parent.first->setOperand(Parent.second, Op);
+  Worklist.Add(Parent.first);
+
+  // Now work back up the expression correcting nsw flags.  The logic is based
+  // on the following observation: if X * Y is known not to overflow as a signed
+  // multiplication, and Y is replaced by a value Z with smaller absolute value,
+  // then X * Z will not overflow as a signed multiplication either.  As we work
+  // our way up, having NoSignedWrap 'true' means that the descaled value at the
+  // current level has strictly smaller absolute value than the original.
+  Instruction *Ancestor = Parent.first;
+  do {
+    if (BinaryOperator *BO = dyn_cast<BinaryOperator>(Ancestor)) {
+      // If the multiplication wasn't nsw then we can't say anything about the
+      // value of the descaled multiplication, and we have to clear nsw flags
+      // from this point on up.
+      bool OpNoSignedWrap = BO->hasNoSignedWrap();
+      NoSignedWrap &= OpNoSignedWrap;
+      if (NoSignedWrap != OpNoSignedWrap) {
+        BO->setHasNoSignedWrap(NoSignedWrap);
+        Worklist.Add(Ancestor);
+      }
+    } else if (Ancestor->getOpcode() == Instruction::Trunc) {
+      // The fact that the descaled input to the trunc has smaller absolute
+      // value than the original input doesn't tell us anything useful about
+      // the absolute values of the truncations.
+      NoSignedWrap = false;
+    }
+    assert((Ancestor->getOpcode() != Instruction::SExt || NoSignedWrap) &&
+           "Failed to keep proper track of nsw flags while drilling down?");
+
+    if (Ancestor == Val)
+      // Got to the top, all done!
+      return Val;
+
+    // Move up one level in the expression.
+    assert(Ancestor->hasOneUse() && "Drilled down when more than one use!");
+    Ancestor = Ancestor->use_back();
+  } while (1);
+}
+
 Instruction *InstCombiner::visitGetElementPtrInst(GetElementPtrInst &GEP) {
   SmallVector<Value*, 8> Ops(GEP.op_begin(), GEP.op_end());
 
@@ -817,7 +1055,7 @@
   // by multiples of a zero size type with zero.
   if (TD) {
     bool MadeChange = false;
-    Type *IntPtrTy = TD->getIntPtrType(GEP.getContext());
+    Type *IntPtrTy = TD->getIntPtrType(GEP.getPointerOperandType());
 
     gep_type_iterator GTI = gep_type_begin(GEP);
     for (User::op_iterator I = GEP.op_begin() + 1, E = GEP.op_end();
@@ -836,7 +1074,7 @@
         }
 
       Type *IndexTy = (*I)->getType();
-      if (IndexTy != IntPtrTy && !IndexTy->isVectorTy()) {
+      if (IndexTy != IntPtrTy) {
         // If we are using a wider index than needed for this platform, shrink
         // it to what we need.  If narrower, sign-extend it to what we need.
         // This explicit cast can make subsequent optimizations more obvious.
@@ -855,7 +1093,7 @@
     if (!shouldMergeGEPs(*cast<GEPOperator>(&GEP), *Src))
       return 0;
 
-    // Note that if our source is a gep chain itself that we wait for that
+    // Note that if our source is a gep chain itself then we wait for that
     // chain to be resolved before we perform this transformation.  This
     // avoids us creating a TON of code in some cases.
     if (GEPOperator *SrcGEP =
@@ -987,63 +1225,74 @@
       }
 
       // Transform things like:
+      // %V = mul i64 %N, 4
+      // %t = getelementptr i8* bitcast (i32* %arr to i8*), i32 %V
+      // into:  %t1 = getelementptr i32* %arr, i32 %N; bitcast
+      if (TD && ResElTy->isSized() && SrcElTy->isSized()) {
+        // Check that changing the type amounts to dividing the index by a scale
+        // factor.
+        uint64_t ResSize = TD->getTypeAllocSize(ResElTy);
+        uint64_t SrcSize = TD->getTypeAllocSize(SrcElTy);
+        if (ResSize && SrcSize % ResSize == 0) {
+          Value *Idx = GEP.getOperand(1);
+          unsigned BitWidth = Idx->getType()->getPrimitiveSizeInBits();
+          uint64_t Scale = SrcSize / ResSize;
+
+          // Earlier transforms ensure that the index has type IntPtrType, which
+          // considerably simplifies the logic by eliminating implicit casts.
+          assert(Idx->getType() == TD->getIntPtrType(GEP.getContext()) &&
+                 "Index not cast to pointer width?");
+
+          bool NSW;
+          if (Value *NewIdx = Descale(Idx, APInt(BitWidth, Scale), NSW)) {
+            // Successfully decomposed Idx as NewIdx * Scale, form a new GEP.
+            // If the multiplication NewIdx * Scale may overflow then the new
+            // GEP may not be "inbounds".
+            Value *NewGEP = GEP.isInBounds() && NSW ?
+              Builder->CreateInBoundsGEP(StrippedPtr, NewIdx, GEP.getName()) :
+              Builder->CreateGEP(StrippedPtr, NewIdx, GEP.getName());
+            // The NewGEP must be pointer typed, so must the old one -> BitCast
+            return new BitCastInst(NewGEP, GEP.getType());
+          }
+        }
+      }
+
+      // Similarly, transform things like:
       // getelementptr i8* bitcast ([100 x double]* X to i8*), i32 %tmp
       //   (where tmp = 8*tmp2) into:
       // getelementptr [100 x double]* %arr, i32 0, i32 %tmp2; bitcast
-
-      if (TD && SrcElTy->isArrayTy() && ResElTy->isIntegerTy(8)) {
+      if (TD && ResElTy->isSized() && SrcElTy->isSized() &&
+          SrcElTy->isArrayTy()) {
+        // Check that changing to the array element type amounts to dividing the
+        // index by a scale factor.
+        uint64_t ResSize = TD->getTypeAllocSize(ResElTy);
         uint64_t ArrayEltSize =
-            TD->getTypeAllocSize(cast<ArrayType>(SrcElTy)->getElementType());
-
-        // Check to see if "tmp" is a scale by a multiple of ArrayEltSize.  We
-        // allow either a mul, shift, or constant here.
-        Value *NewIdx = 0;
-        ConstantInt *Scale = 0;
-        if (ArrayEltSize == 1) {
-          NewIdx = GEP.getOperand(1);
-          Scale = ConstantInt::get(cast<IntegerType>(NewIdx->getType()), 1);
-        } else if (ConstantInt *CI = dyn_cast<ConstantInt>(GEP.getOperand(1))) {
-          NewIdx = ConstantInt::get(CI->getType(), 1);
-          Scale = CI;
-        } else if (Instruction *Inst =dyn_cast<Instruction>(GEP.getOperand(1))){
-          if (Inst->getOpcode() == Instruction::Shl &&
-              isa<ConstantInt>(Inst->getOperand(1))) {
-            ConstantInt *ShAmt = cast<ConstantInt>(Inst->getOperand(1));
-            uint32_t ShAmtVal = ShAmt->getLimitedValue(64);
-            Scale = ConstantInt::get(cast<IntegerType>(Inst->getType()),
-                                     1ULL << ShAmtVal);
-            NewIdx = Inst->getOperand(0);
-          } else if (Inst->getOpcode() == Instruction::Mul &&
-                     isa<ConstantInt>(Inst->getOperand(1))) {
-            Scale = cast<ConstantInt>(Inst->getOperand(1));
-            NewIdx = Inst->getOperand(0);
+          TD->getTypeAllocSize(cast<ArrayType>(SrcElTy)->getElementType());
+        if (ResSize && ArrayEltSize % ResSize == 0) {
+          Value *Idx = GEP.getOperand(1);
+          unsigned BitWidth = Idx->getType()->getPrimitiveSizeInBits();
+          uint64_t Scale = ArrayEltSize / ResSize;
+
+          // Earlier transforms ensure that the index has type IntPtrType, which
+          // considerably simplifies the logic by eliminating implicit casts.
+          assert(Idx->getType() == TD->getIntPtrType(GEP.getContext()) &&
+                 "Index not cast to pointer width?");
+
+          bool NSW;
+          if (Value *NewIdx = Descale(Idx, APInt(BitWidth, Scale), NSW)) {
+            // Successfully decomposed Idx as NewIdx * Scale, form a new GEP.
+            // If the multiplication NewIdx * Scale may overflow then the new
+            // GEP may not be "inbounds".
+            Value *Off[2];
+            Off[0] = Constant::getNullValue(Type::getInt32Ty(GEP.getContext()));
+            Off[1] = NewIdx;
+            Value *NewGEP = GEP.isInBounds() && NSW ?
+              Builder->CreateInBoundsGEP(StrippedPtr, Off, GEP.getName()) :
+              Builder->CreateGEP(StrippedPtr, Off, GEP.getName());
+            // The NewGEP must be pointer typed, so must the old one -> BitCast
+            return new BitCastInst(NewGEP, GEP.getType());
           }
         }
-
-        // If the index will be to exactly the right offset with the scale taken
-        // out, perform the transformation. Note, we don't know whether Scale is
-        // signed or not. We'll use unsigned version of division/modulo
-        // operation after making sure Scale doesn't have the sign bit set.
-        if (ArrayEltSize && Scale && Scale->getSExtValue() >= 0LL &&
-            Scale->getZExtValue() % ArrayEltSize == 0) {
-          Scale = ConstantInt::get(Scale->getType(),
-                                   Scale->getZExtValue() / ArrayEltSize);
-          if (Scale->getZExtValue() != 1) {
-            Constant *C = ConstantExpr::getIntegerCast(Scale, NewIdx->getType(),
-                                                       false /*ZExt*/);
-            NewIdx = Builder->CreateMul(NewIdx, C, "idxscale");
-          }
-
-          // Insert the new GEP instruction.
-          Value *Idx[2];
-          Idx[0] = Constant::getNullValue(Type::getInt32Ty(GEP.getContext()));
-          Idx[1] = NewIdx;
-          Value *NewGEP = GEP.isInBounds() ?
-            Builder->CreateInBoundsGEP(StrippedPtr, Idx, GEP.getName()):
-            Builder->CreateGEP(StrippedPtr, Idx, GEP.getName());
-          // The NewGEP must be pointer typed, so must the old one -> BitCast
-          return new BitCastInst(NewGEP, GEP.getType());
-        }
       }
     }
   }
@@ -1068,7 +1317,7 @@
         // If the bitcast is of an allocation, and the allocation will be
         // converted to match the type of the cast, don't touch this.
         if (isa<AllocaInst>(BCI->getOperand(0)) ||
-            isAllocationFn(BCI->getOperand(0))) {
+            isAllocationFn(BCI->getOperand(0), TLI)) {
           // See if the bitcast simplifies, if so, don't nuke this GEP yet.
           if (Instruction *I = visitBitCast(*BCI)) {
             if (I != BCI) {
@@ -1107,7 +1356,8 @@
 
 
 static bool
-isAllocSiteRemovable(Instruction *AI, SmallVectorImpl<WeakVH> &Users) {
+isAllocSiteRemovable(Instruction *AI, SmallVectorImpl<WeakVH> &Users,
+                     const TargetLibraryInfo *TLI) {
   SmallVector<Instruction*, 4> Worklist;
   Worklist.push_back(AI);
 
@@ -1163,7 +1413,7 @@
           }
         }
 
-        if (isFreeCall(I)) {
+        if (isFreeCall(I, TLI)) {
           Users.push_back(I);
           continue;
         }
@@ -1188,7 +1438,7 @@
   // to null and free calls, delete the calls and replace the comparisons with
   // true or false as appropriate.
   SmallVector<WeakVH, 64> Users;
-  if (isAllocSiteRemovable(&MI, Users)) {
+  if (isAllocSiteRemovable(&MI, Users, TLI)) {
     for (unsigned i = 0, e = Users.size(); i != e; ++i) {
       Instruction *I = cast_or_null<Instruction>(&*Users[i]);
       if (!I) continue;
@@ -1853,7 +2103,7 @@
 static bool AddReachableCodeToWorklist(BasicBlock *BB,
                                        SmallPtrSet<BasicBlock*, 64> &Visited,
                                        InstCombiner &IC,
-                                       const TargetData *TD,
+                                       const DataLayout *TD,
                                        const TargetLibraryInfo *TLI) {
   bool MadeIRChange = false;
   SmallVector<BasicBlock*, 256> Worklist;
@@ -1872,7 +2122,7 @@
       Instruction *Inst = BBI++;
 
       // DCE instruction if trivially dead.
-      if (isInstructionTriviallyDead(Inst)) {
+      if (isInstructionTriviallyDead(Inst, TLI)) {
         ++NumDeadInst;
         DEBUG(errs() << "IC: DCE: " << *Inst << '\n');
         Inst->eraseFromParent();
@@ -2002,7 +2252,7 @@
     if (I == 0) continue;  // skip null values.
 
     // Check to see if we can DCE the instruction.
-    if (isInstructionTriviallyDead(I)) {
+    if (isInstructionTriviallyDead(I, TLI)) {
       DEBUG(errs() << "IC: DCE: " << *I << '\n');
       EraseInstFromFunction(*I);
       ++NumDeadInst;
@@ -2102,7 +2352,7 @@
 
         // If the instruction was modified, it's possible that it is now dead.
         // if so, remove it.
-        if (isInstructionTriviallyDead(I)) {
+        if (isInstructionTriviallyDead(I, TLI)) {
           EraseInstFromFunction(*I);
         } else {
           Worklist.Add(I);
@@ -2117,9 +2367,27 @@
   return MadeIRChange;
 }
 
+namespace {
+class InstCombinerLibCallSimplifier : public LibCallSimplifier {
+  InstCombiner *IC;
+public:
+  InstCombinerLibCallSimplifier(const DataLayout *TD,
+                                const TargetLibraryInfo *TLI,
+                                InstCombiner *IC)
+    : LibCallSimplifier(TD, TLI) {
+    this->IC = IC;
+  }
+
+  /// replaceAllUsesWith - override so that instruction replacement
+  /// can be defined in terms of the instruction combiner framework.
+  virtual void replaceAllUsesWith(Instruction *I, Value *With) const {
+    IC->ReplaceInstUsesWith(*I, With);
+  }
+};
+}
 
 bool InstCombiner::runOnFunction(Function &F) {
-  TD = getAnalysisIfAvailable<TargetData>();
+  TD = getAnalysisIfAvailable<DataLayout>();
   TLI = &getAnalysis<TargetLibraryInfo>();
 
   /// Builder - This is an IRBuilder that automatically inserts new
@@ -2129,6 +2397,9 @@
                InstCombineIRInserter(Worklist));
   Builder = &TheBuilder;
 
+  InstCombinerLibCallSimplifier TheSimplifier(TD, TLI, this);
+  Simplifier = &TheSimplifier;
+
   bool EverMadeChange = false;
 
   // Lower dbg.declare intrinsics otherwise their value may be clobbered

Modified: llvm/branches/AMDILBackend/lib/Transforms/Instrumentation/AddressSanitizer.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Instrumentation/AddressSanitizer.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Instrumentation/AddressSanitizer.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Instrumentation/AddressSanitizer.cpp Tue Jan 15 11:16:16 2013
@@ -15,7 +15,7 @@
 
 #define DEBUG_TYPE "asan"
 
-#include "FunctionBlackList.h"
+#include "BlackList.h"
 #include "llvm/Function.h"
 #include "llvm/IRBuilder.h"
 #include "llvm/InlineAsm.h"
@@ -35,7 +35,7 @@
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/raw_ostream.h"
 #include "llvm/Support/system_error.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Target/TargetMachine.h"
 #include "llvm/Transforms/Instrumentation.h"
 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
@@ -61,6 +61,8 @@
 static const char *kAsanReportErrorTemplate = "__asan_report_";
 static const char *kAsanRegisterGlobalsName = "__asan_register_globals";
 static const char *kAsanUnregisterGlobalsName = "__asan_unregister_globals";
+static const char *kAsanPoisonGlobalsName = "__asan_before_dynamic_init";
+static const char *kAsanUnpoisonGlobalsName = "__asan_after_dynamic_init";
 static const char *kAsanInitName = "__asan_init";
 static const char *kAsanHandleNoReturnName = "__asan_handle_no_return";
 static const char *kAsanMappingOffsetName = "__asan_mapping_offset";
@@ -86,8 +88,8 @@
 static cl::opt<bool> ClInstrumentAtomics("asan-instrument-atomics",
        cl::desc("instrument atomic instructions (rmw, cmpxchg)"),
        cl::Hidden, cl::init(true));
-static cl::opt<bool> ClMergeCallbacks("asan-merge-callbacks",
-       cl::desc("merge __asan_report_ callbacks to create fewer BBs"),
+static cl::opt<bool> ClAlwaysSlowPath("asan-always-slow-path",
+       cl::desc("use instrumentation with slow path for all accesses"),
        cl::Hidden, cl::init(false));
 // This flag limits the number of instructions to be instrumented
 // in any given BB. Normally, this should be set to unlimited (INT_MAX),
@@ -106,6 +108,8 @@
 // This flag may need to be replaced with -f[no]asan-globals.
 static cl::opt<bool> ClGlobals("asan-globals",
        cl::desc("Handle global objects"), cl::Hidden, cl::init(true));
+static cl::opt<bool> ClInitializers("asan-initialization-order",
+       cl::desc("Handle C++ initializer order"), cl::Hidden, cl::init(false));
 static cl::opt<bool> ClMemIntrin("asan-memintrin",
        cl::desc("Handle memset/memcpy/memmove"), cl::Hidden, cl::init(true));
 // This flag may need to be replaced with -fasan-blacklist.
@@ -144,54 +148,33 @@
                                cl::Hidden, cl::init(-1));
 
 namespace {
-
-/// When the crash callbacks are merged, they receive some amount of arguments
-/// that are merged in a PHI node. This struct represents arguments from one
-/// call site.
-struct CrashArg {
-  Value *Arg1;
-  Value *Arg2;
-};
-
-/// An object of this type is created while instrumenting every function.
-struct AsanFunctionContext {
-  AsanFunctionContext(Function &Function) : F(Function), CrashBlock() { }
-
-  Function &F;
-  // These are initially zero. If we require at least one call to
-  // __asan_report_{read,write}{1,2,4,8,16}, an appropriate BB is created.
-  BasicBlock *CrashBlock[2][kNumberOfAccessSizes];
-  typedef  SmallVector<CrashArg, 8> CrashArgsVec;
-  CrashArgsVec CrashArgs[2][kNumberOfAccessSizes];
-};
-
 /// AddressSanitizer: instrument the code in module to find memory bugs.
-struct AddressSanitizer : public ModulePass {
+struct AddressSanitizer : public FunctionPass {
   AddressSanitizer();
   virtual const char *getPassName() const;
-  void instrumentMop(AsanFunctionContext &AFC, Instruction *I);
-  void instrumentAddress(AsanFunctionContext &AFC,
-                         Instruction *OrigIns, IRBuilder<> &IRB,
+  void instrumentMop(Instruction *I);
+  void instrumentAddress(Instruction *OrigIns, IRBuilder<> &IRB,
                          Value *Addr, uint32_t TypeSize, bool IsWrite);
   Value *createSlowPathCmp(IRBuilder<> &IRB, Value *AddrLong,
                            Value *ShadowValue, uint32_t TypeSize);
-  Instruction *generateCrashCode(BasicBlock *BB, Value *Addr, Value *PC,
+  Instruction *generateCrashCode(Instruction *InsertBefore, Value *Addr,
                                  bool IsWrite, size_t AccessSizeIndex);
-  bool instrumentMemIntrinsic(AsanFunctionContext &AFC, MemIntrinsic *MI);
-  void instrumentMemIntrinsicParam(AsanFunctionContext &AFC,
-                                   Instruction *OrigIns, Value *Addr,
+  bool instrumentMemIntrinsic(MemIntrinsic *MI);
+  void instrumentMemIntrinsicParam(Instruction *OrigIns, Value *Addr,
                                    Value *Size,
                                    Instruction *InsertBefore, bool IsWrite);
   Value *memToShadow(Value *Shadow, IRBuilder<> &IRB);
-  bool handleFunction(Module &M, Function &F);
+  bool runOnFunction(Function &F);
+  void createInitializerPoisonCalls(Module &M,
+                                    Value *FirstAddr, Value *LastAddr);
   bool maybeInsertAsanInitAtFunctionEntry(Function &F);
-  bool poisonStackInFunction(Module &M, Function &F);
-  virtual bool runOnModule(Module &M);
+  bool poisonStackInFunction(Function &F);
+  virtual bool doInitialization(Module &M);
+  virtual bool doFinalization(Module &M);
   bool insertGlobalRedzones(Module &M);
   static char ID;  // Pass identification, replacement for typeid
 
  private:
-
   uint64_t getAllocaSizeInBytes(AllocaInst *AI) {
     Type *Ty = AI->getAllocatedType();
     uint64_t SizeInBytes = TD->getTypeAllocSize(Ty);
@@ -207,12 +190,15 @@
   }
 
   Function *checkInterfaceFunction(Constant *FuncOrBitcast);
+  bool ShouldInstrumentGlobal(GlobalVariable *G);
   void PoisonStack(const ArrayRef<AllocaInst*> &AllocaVec, IRBuilder<> IRB,
                    Value *ShadowBase, bool DoPoison);
   bool LooksLikeCodeInBug11395(Instruction *I);
+  void FindDynamicInitializers(Module &M);
+  bool HasDynamicInitializer(GlobalVariable *G);
 
   LLVMContext *C;
-  TargetData *TD;
+  DataLayout *TD;
   uint64_t MappingOffset;
   int MappingScale;
   size_t RedzoneSize;
@@ -221,11 +207,15 @@
   Type *IntptrPtrTy;
   Function *AsanCtorFunction;
   Function *AsanInitFunction;
+  Function *AsanStackMallocFunc, *AsanStackFreeFunc;
+  Function *AsanHandleNoReturnFunc;
   Instruction *CtorInsertBefore;
-  OwningPtr<FunctionBlackList> BL;
+  OwningPtr<BlackList> BL;
   // This array is indexed by AccessIsWrite and log2(AccessSize).
   Function *AsanErrorCallback[2][kNumberOfAccessSizes];
   InlineAsm *EmptyAsm;
+  SmallSet<GlobalValue*, 32> DynamicallyInitializedGlobals;
+  SmallSet<GlobalValue*, 32> GlobalsCreatedByAsan;
 };
 
 }  // namespace
@@ -234,8 +224,8 @@
 INITIALIZE_PASS(AddressSanitizer, "asan",
     "AddressSanitizer: detects use-after-free and out-of-bounds bugs.",
     false, false)
-AddressSanitizer::AddressSanitizer() : ModulePass(ID) { }
-ModulePass *llvm::createAddressSanitizerPass() {
+AddressSanitizer::AddressSanitizer() : FunctionPass(ID) { }
+FunctionPass *llvm::createAddressSanitizerPass() {
   return new AddressSanitizer();
 }
 
@@ -256,38 +246,6 @@
                             GlobalValue::PrivateLinkage, StrConst, "");
 }
 
-// Split the basic block and insert an if-then code.
-// Before:
-//   Head
-//   Cmp
-//   Tail
-// After:
-//   Head
-//   if (Cmp)
-//     ThenBlock
-//   Tail
-//
-// If ThenBlock is zero, a new block is created and its terminator is returned.
-// Otherwize 0 is returned.
-static BranchInst *splitBlockAndInsertIfThen(Value *Cmp,
-                                             BasicBlock *ThenBlock = 0) {
-  Instruction *SplitBefore = cast<Instruction>(Cmp)->getNextNode();
-  BasicBlock *Head = SplitBefore->getParent();
-  BasicBlock *Tail = Head->splitBasicBlock(SplitBefore);
-  TerminatorInst *HeadOldTerm = Head->getTerminator();
-  BranchInst *CheckTerm = 0;
-  if (!ThenBlock) {
-    LLVMContext &C = Head->getParent()->getParent()->getContext();
-    ThenBlock = BasicBlock::Create(C, "", Head->getParent(), Tail);
-    CheckTerm = BranchInst::Create(Tail, ThenBlock);
-  }
-  BranchInst *HeadNewTerm =
-    BranchInst::Create(/*ifTrue*/ThenBlock, /*ifFalse*/Tail, Cmp);
-  ReplaceInstWithInst(HeadOldTerm, HeadNewTerm);
-
-  return CheckTerm;
-}
-
 Value *AddressSanitizer::memToShadow(Value *Shadow, IRBuilder<> &IRB) {
   // Shadow >> scale
   Shadow = IRB.CreateLShr(Shadow, MappingScale);
@@ -299,12 +257,12 @@
 }
 
 void AddressSanitizer::instrumentMemIntrinsicParam(
-    AsanFunctionContext &AFC, Instruction *OrigIns,
+    Instruction *OrigIns,
     Value *Addr, Value *Size, Instruction *InsertBefore, bool IsWrite) {
   // Check the first byte.
   {
     IRBuilder<> IRB(InsertBefore);
-    instrumentAddress(AFC, OrigIns, IRB, Addr, 8, IsWrite);
+    instrumentAddress(OrigIns, IRB, Addr, 8, IsWrite);
   }
   // Check the last byte.
   {
@@ -314,13 +272,12 @@
     SizeMinusOne = IRB.CreateIntCast(SizeMinusOne, IntptrTy, false);
     Value *AddrLong = IRB.CreatePointerCast(Addr, IntptrTy);
     Value *AddrPlusSizeMinisOne = IRB.CreateAdd(AddrLong, SizeMinusOne);
-    instrumentAddress(AFC, OrigIns, IRB, AddrPlusSizeMinisOne, 8, IsWrite);
+    instrumentAddress(OrigIns, IRB, AddrPlusSizeMinisOne, 8, IsWrite);
   }
 }
 
 // Instrument memset/memmove/memcpy
-bool AddressSanitizer::instrumentMemIntrinsic(AsanFunctionContext &AFC,
-                                              MemIntrinsic *MI) {
+bool AddressSanitizer::instrumentMemIntrinsic(MemIntrinsic *MI) {
   Value *Dst = MI->getDest();
   MemTransferInst *MemTran = dyn_cast<MemTransferInst>(MI);
   Value *Src = MemTran ? MemTran->getSource() : 0;
@@ -336,12 +293,12 @@
 
     Value *Cmp = IRB.CreateICmpNE(Length,
                                   Constant::getNullValue(Length->getType()));
-    InsertBefore = splitBlockAndInsertIfThen(Cmp);
+    InsertBefore = SplitBlockAndInsertIfThen(cast<Instruction>(Cmp), false);
   }
 
-  instrumentMemIntrinsicParam(AFC, MI, Dst, Length, InsertBefore, true);
+  instrumentMemIntrinsicParam(MI, Dst, Length, InsertBefore, true);
   if (Src)
-    instrumentMemIntrinsicParam(AFC, MI, Src, Length, InsertBefore, false);
+    instrumentMemIntrinsicParam(MI, Src, Length, InsertBefore, false);
   return true;
 }
 
@@ -371,14 +328,50 @@
   return NULL;
 }
 
-void AddressSanitizer::instrumentMop(AsanFunctionContext &AFC, Instruction *I) {
-  bool IsWrite;
+void AddressSanitizer::FindDynamicInitializers(Module& M) {
+  // Clang generates metadata identifying all dynamically initialized globals.
+  NamedMDNode *DynamicGlobals =
+      M.getNamedMetadata("llvm.asan.dynamically_initialized_globals");
+  if (!DynamicGlobals)
+    return;
+  for (int i = 0, n = DynamicGlobals->getNumOperands(); i < n; ++i) {
+    MDNode *MDN = DynamicGlobals->getOperand(i);
+    assert(MDN->getNumOperands() == 1);
+    Value *VG = MDN->getOperand(0);
+    // The optimizer may optimize away a global entirely, in which case we
+    // cannot instrument access to it.
+    if (!VG)
+      continue;
+
+    GlobalVariable *G = cast<GlobalVariable>(VG);
+    DynamicallyInitializedGlobals.insert(G);
+  }
+}
+// Returns true if a global variable is initialized dynamically in this TU.
+bool AddressSanitizer::HasDynamicInitializer(GlobalVariable *G) {
+  return DynamicallyInitializedGlobals.count(G);
+}
+
+void AddressSanitizer::instrumentMop(Instruction *I) {
+  bool IsWrite = false;
   Value *Addr = isInterestingMemoryAccess(I, &IsWrite);
   assert(Addr);
-  if (ClOpt && ClOptGlobals && isa<GlobalVariable>(Addr)) {
-    // We are accessing a global scalar variable. Nothing to catch here.
-    return;
+  if (ClOpt && ClOptGlobals) {
+    if (GlobalVariable *G = dyn_cast<GlobalVariable>(Addr)) {
+      // If initialization order checking is disabled, a simple access to a
+      // dynamically initialized global is always valid.
+      if (!ClInitializers)
+        return;
+      // If a global variable does not have dynamic initialization we don't
+      // have to instrument it.  However, if a global has external linkage, we
+      // assume it has dynamic initialization, as it may have an initializer
+      // in a different TU.
+      if (G->getLinkage() != GlobalVariable::ExternalLinkage &&
+          !HasDynamicInitializer(G))
+        return;
+    }
   }
+
   Type *OrigPtrTy = Addr->getType();
   Type *OrigTy = cast<PointerType>(OrigPtrTy)->getElementType();
 
@@ -392,7 +385,7 @@
   }
 
   IRBuilder<> IRB(I);
-  instrumentAddress(AFC, I, IRB, Addr, TypeSize, IsWrite);
+  instrumentAddress(I, IRB, Addr, TypeSize, IsWrite);
 }
 
 // Validate the result of Module::getOrInsertFunction called for an interface
@@ -407,15 +400,11 @@
 }
 
 Instruction *AddressSanitizer::generateCrashCode(
-    BasicBlock *BB, Value *Addr, Value *PC,
+    Instruction *InsertBefore, Value *Addr,
     bool IsWrite, size_t AccessSizeIndex) {
-  IRBuilder<> IRB(BB->getFirstNonPHI());
-  CallInst *Call;
-  if (PC)
-    Call = IRB.CreateCall2(AsanErrorCallback[IsWrite][AccessSizeIndex],
-                           Addr, PC);
-  else
-    Call = IRB.CreateCall(AsanErrorCallback[IsWrite][AccessSizeIndex], Addr);
+  IRBuilder<> IRB(InsertBefore);
+  CallInst *Call = IRB.CreateCall(AsanErrorCallback[IsWrite][AccessSizeIndex],
+                                  Addr);
   // We don't do Call->setDoesNotReturn() because the BB already has
   // UnreachableInst at the end.
   // This EmptyAsm is required to avoid callback merge.
@@ -436,13 +425,12 @@
         LastAccessedByte, ConstantInt::get(IntptrTy, TypeSize / 8 - 1));
   // (uint8_t) ((Addr & (Granularity-1)) + size - 1)
   LastAccessedByte = IRB.CreateIntCast(
-      LastAccessedByte, IRB.getInt8Ty(), false);
+      LastAccessedByte, ShadowValue->getType(), false);
   // ((uint8_t) ((Addr & (Granularity-1)) + size - 1)) >= ShadowValue
   return IRB.CreateICmpSGE(LastAccessedByte, ShadowValue);
 }
 
-void AddressSanitizer::instrumentAddress(AsanFunctionContext &AFC,
-                                         Instruction *OrigIns,
+void AddressSanitizer::instrumentAddress(Instruction *OrigIns,
                                          IRBuilder<> &IRB, Value *Addr,
                                          uint32_t TypeSize, bool IsWrite) {
   Value *AddrLong = IRB.CreatePointerCast(Addr, IntptrTy);
@@ -456,112 +444,133 @@
       IRB.CreateIntToPtr(ShadowPtr, ShadowPtrTy));
 
   Value *Cmp = IRB.CreateICmpNE(ShadowValue, CmpVal);
-
-  BasicBlock *CrashBlock = 0;
-  if (ClMergeCallbacks) {
-    size_t AccessSizeIndex = TypeSizeToSizeIndex(TypeSize);
-    BasicBlock **Cached = &AFC.CrashBlock[IsWrite][AccessSizeIndex];
-    if (!*Cached) {
-      std::string BBName("crash_bb-");
-      BBName += (IsWrite ? "w-" : "r-") + itostr(1 << AccessSizeIndex);
-      BasicBlock *BB = BasicBlock::Create(*C, BBName, &AFC.F);
-      new UnreachableInst(*C, BB);
-      *Cached = BB;
-    }
-    CrashBlock = *Cached;
-    // We need to pass the PC as the second parameter to __asan_report_*.
-    // There are few problems:
-    //  - Some architectures (e.g. x86_32) don't have a cheap way to get the PC.
-    //  - LLVM doesn't have the appropriate intrinsic.
-    // For now, put a random number into the PC, just to allow experiments.
-    Value *PC = ConstantInt::get(IntptrTy, rand());
-    CrashArg Arg = {AddrLong, PC};
-    AFC.CrashArgs[IsWrite][AccessSizeIndex].push_back(Arg);
-  } else {
-    CrashBlock = BasicBlock::Create(*C, "crash_bb", &AFC.F);
-    new UnreachableInst(*C, CrashBlock);
-    size_t AccessSizeIndex = TypeSizeToSizeIndex(TypeSize);
-    Instruction *Crash =
-        generateCrashCode(CrashBlock, AddrLong, 0, IsWrite, AccessSizeIndex);
-    Crash->setDebugLoc(OrigIns->getDebugLoc());
-  }
-
+  size_t AccessSizeIndex = TypeSizeToSizeIndex(TypeSize);
   size_t Granularity = 1 << MappingScale;
-  if (TypeSize < 8 * Granularity) {
-    BranchInst *CheckTerm = splitBlockAndInsertIfThen(Cmp);
-    assert(CheckTerm->isUnconditional());
+  TerminatorInst *CrashTerm = 0;
+
+  if (ClAlwaysSlowPath || (TypeSize < 8 * Granularity)) {
+    TerminatorInst *CheckTerm =
+        SplitBlockAndInsertIfThen(cast<Instruction>(Cmp), false);
+    assert(dyn_cast<BranchInst>(CheckTerm)->isUnconditional());
     BasicBlock *NextBB = CheckTerm->getSuccessor(0);
     IRB.SetInsertPoint(CheckTerm);
     Value *Cmp2 = createSlowPathCmp(IRB, AddrLong, ShadowValue, TypeSize);
+    BasicBlock *CrashBlock =
+        BasicBlock::Create(*C, "", NextBB->getParent(), NextBB);
+    CrashTerm = new UnreachableInst(*C, CrashBlock);
     BranchInst *NewTerm = BranchInst::Create(CrashBlock, NextBB, Cmp2);
     ReplaceInstWithInst(CheckTerm, NewTerm);
   } else {
-    splitBlockAndInsertIfThen(Cmp, CrashBlock);
+    CrashTerm = SplitBlockAndInsertIfThen(cast<Instruction>(Cmp), true);
+  }
+
+  Instruction *Crash =
+      generateCrashCode(CrashTerm, AddrLong, IsWrite, AccessSizeIndex);
+  Crash->setDebugLoc(OrigIns->getDebugLoc());
+}
+
+void AddressSanitizer::createInitializerPoisonCalls(Module &M,
+                                                    Value *FirstAddr,
+                                                    Value *LastAddr) {
+  // We do all of our poisoning and unpoisoning within _GLOBAL__I_a.
+  Function *GlobalInit = M.getFunction("_GLOBAL__I_a");
+  // If that function is not present, this TU contains no globals, or they have
+  // all been optimized away
+  if (!GlobalInit)
+    return;
+
+  // Set up the arguments to our poison/unpoison functions.
+  IRBuilder<> IRB(GlobalInit->begin()->getFirstInsertionPt());
+
+  // Declare our poisoning and unpoisoning functions.
+  Function *AsanPoisonGlobals = checkInterfaceFunction(M.getOrInsertFunction(
+      kAsanPoisonGlobalsName, IRB.getVoidTy(), IntptrTy, IntptrTy, NULL));
+  AsanPoisonGlobals->setLinkage(Function::ExternalLinkage);
+  Function *AsanUnpoisonGlobals = checkInterfaceFunction(M.getOrInsertFunction(
+      kAsanUnpoisonGlobalsName, IRB.getVoidTy(), NULL));
+  AsanUnpoisonGlobals->setLinkage(Function::ExternalLinkage);
+
+  // Add a call to poison all external globals before the given function starts.
+  IRB.CreateCall2(AsanPoisonGlobals, FirstAddr, LastAddr);
+
+  // Add calls to unpoison all globals before each return instruction.
+  for (Function::iterator I = GlobalInit->begin(), E = GlobalInit->end();
+      I != E; ++I) {
+    if (ReturnInst *RI = dyn_cast<ReturnInst>(I->getTerminator())) {
+      CallInst::Create(AsanUnpoisonGlobals, "", RI);
+    }
   }
 }
 
+bool AddressSanitizer::ShouldInstrumentGlobal(GlobalVariable *G) {
+  Type *Ty = cast<PointerType>(G->getType())->getElementType();
+  DEBUG(dbgs() << "GLOBAL: " << *G << "\n");
+
+  if (BL->isIn(*G)) return false;
+  if (!Ty->isSized()) return false;
+  if (!G->hasInitializer()) return false;
+  if (GlobalsCreatedByAsan.count(G)) return false;  // Our own global.
+  // Touch only those globals that will not be defined in other modules.
+  // Don't handle ODR type linkages since other modules may be built w/o asan.
+  if (G->getLinkage() != GlobalVariable::ExternalLinkage &&
+      G->getLinkage() != GlobalVariable::PrivateLinkage &&
+      G->getLinkage() != GlobalVariable::InternalLinkage)
+    return false;
+  // Two problems with thread-locals:
+  //   - The address of the main thread's copy can't be computed at link-time.
+  //   - Need to poison all copies, not just the main thread's one.
+  if (G->isThreadLocal())
+    return false;
+  // For now, just ignore this Alloca if the alignment is large.
+  if (G->getAlignment() > RedzoneSize) return false;
+
+  // Ignore all the globals with the names starting with "\01L_OBJC_".
+  // Many of those are put into the .cstring section. The linker compresses
+  // that section by removing the spare \0s after the string terminator, so
+  // our redzones get broken.
+  if ((G->getName().find("\01L_OBJC_") == 0) ||
+      (G->getName().find("\01l_OBJC_") == 0)) {
+    DEBUG(dbgs() << "Ignoring \\01L_OBJC_* global: " << *G);
+    return false;
+  }
+
+  if (G->hasSection()) {
+    StringRef Section(G->getSection());
+    // Ignore the globals from the __OBJC section. The ObjC runtime assumes
+    // those conform to /usr/lib/objc/runtime.h, so we can't add redzones to
+    // them.
+    if ((Section.find("__OBJC,") == 0) ||
+        (Section.find("__DATA, __objc_") == 0)) {
+      DEBUG(dbgs() << "Ignoring ObjC runtime global: " << *G);
+      return false;
+    }
+    // See http://code.google.com/p/address-sanitizer/issues/detail?id=32
+    // Constant CFString instances are compiled in the following way:
+    //  -- the string buffer is emitted into
+    //     __TEXT,__cstring,cstring_literals
+    //  -- the constant NSConstantString structure referencing that buffer
+    //     is placed into __DATA,__cfstring
+    // Therefore there's no point in placing redzones into __DATA,__cfstring.
+    // Moreover, it causes the linker to crash on OS X 10.7
+    if (Section.find("__DATA,__cfstring") == 0) {
+      DEBUG(dbgs() << "Ignoring CFString: " << *G);
+      return false;
+    }
+  }
+
+  return true;
+}
+
 // This function replaces all global variables with new variables that have
 // trailing redzones. It also creates a function that poisons
 // redzones and inserts this function into llvm.global_ctors.
 bool AddressSanitizer::insertGlobalRedzones(Module &M) {
   SmallVector<GlobalVariable *, 16> GlobalsToChange;
 
-  for (Module::GlobalListType::iterator G = M.getGlobalList().begin(),
-       E = M.getGlobalList().end(); G != E; ++G) {
-    Type *Ty = cast<PointerType>(G->getType())->getElementType();
-    DEBUG(dbgs() << "GLOBAL: " << *G);
-
-    if (!Ty->isSized()) continue;
-    if (!G->hasInitializer()) continue;
-    // Touch only those globals that will not be defined in other modules.
-    // Don't handle ODR type linkages since other modules may be built w/o asan.
-    if (G->getLinkage() != GlobalVariable::ExternalLinkage &&
-        G->getLinkage() != GlobalVariable::PrivateLinkage &&
-        G->getLinkage() != GlobalVariable::InternalLinkage)
-      continue;
-    // Two problems with thread-locals:
-    //   - The address of the main thread's copy can't be computed at link-time.
-    //   - Need to poison all copies, not just the main thread's one.
-    if (G->isThreadLocal())
-      continue;
-    // For now, just ignore this Alloca if the alignment is large.
-    if (G->getAlignment() > RedzoneSize) continue;
-
-    // Ignore all the globals with the names starting with "\01L_OBJC_".
-    // Many of those are put into the .cstring section. The linker compresses
-    // that section by removing the spare \0s after the string terminator, so
-    // our redzones get broken.
-    if ((G->getName().find("\01L_OBJC_") == 0) ||
-        (G->getName().find("\01l_OBJC_") == 0)) {
-      DEBUG(dbgs() << "Ignoring \\01L_OBJC_* global: " << *G);
-      continue;
-    }
-
-    if (G->hasSection()) {
-      StringRef Section(G->getSection());
-      // Ignore the globals from the __OBJC section. The ObjC runtime assumes
-      // those conform to /usr/lib/objc/runtime.h, so we can't add redzones to
-      // them.
-      if ((Section.find("__OBJC,") == 0) ||
-          (Section.find("__DATA, __objc_") == 0)) {
-        DEBUG(dbgs() << "Ignoring ObjC runtime global: " << *G);
-        continue;
-      }
-      // See http://code.google.com/p/address-sanitizer/issues/detail?id=32
-      // Constant CFString instances are compiled in the following way:
-      //  -- the string buffer is emitted into
-      //     __TEXT,__cstring,cstring_literals
-      //  -- the constant NSConstantString structure referencing that buffer
-      //     is placed into __DATA,__cfstring
-      // Therefore there's no point in placing redzones into __DATA,__cfstring.
-      // Moreover, it causes the linker to crash on OS X 10.7
-      if (Section.find("__DATA,__cfstring") == 0) {
-        DEBUG(dbgs() << "Ignoring CFString: " << *G);
-        continue;
-      }
-    }
-
-    GlobalsToChange.push_back(G);
+  for (Module::GlobalListType::iterator G = M.global_begin(),
+       E = M.global_end(); G != E; ++G) {
+    if (ShouldInstrumentGlobal(G))
+      GlobalsToChange.push_back(G);
   }
 
   size_t n = GlobalsToChange.size();
@@ -572,13 +581,22 @@
   //   size_t size;
   //   size_t size_with_redzone;
   //   const char *name;
+  //   size_t has_dynamic_init;
   // We initialize an array of such structures and pass it to a run-time call.
   StructType *GlobalStructTy = StructType::get(IntptrTy, IntptrTy,
-                                               IntptrTy, IntptrTy, NULL);
-  SmallVector<Constant *, 16> Initializers(n);
+                                               IntptrTy, IntptrTy,
+                                               IntptrTy, NULL);
+  SmallVector<Constant *, 16> Initializers(n), DynamicInit;
 
   IRBuilder<> IRB(CtorInsertBefore);
 
+  if (ClInitializers)
+    FindDynamicInitializers(M);
+
+  // The addresses of the first and last dynamically initialized globals in
+  // this TU.  Used in initialization order checking.
+  Value *FirstDynamic = 0, *LastDynamic = 0;
+
   for (size_t i = 0; i < n; i++) {
     GlobalVariable *G = GlobalsToChange[i];
     PointerType *PtrTy = cast<PointerType>(G->getType());
@@ -587,6 +605,10 @@
     uint64_t RightRedzoneSize = RedzoneSize +
         (RedzoneSize - (SizeInBytes % RedzoneSize));
     Type *RightRedZoneTy = ArrayType::get(IRB.getInt8Ty(), RightRedzoneSize);
+    // Determine whether this global should be poisoned in initialization.
+    bool GlobalHasDynamicInitializer = HasDynamicInitializer(G);
+    // Don't check initialization order if this global is blacklisted.
+    GlobalHasDynamicInitializer &= !BL->isInInit(*G);
 
     StructType *NewTy = StructType::get(Ty, RightRedZoneTy, NULL);
     Constant *NewInitializer = ConstantStruct::get(
@@ -621,8 +643,17 @@
         ConstantInt::get(IntptrTy, SizeInBytes),
         ConstantInt::get(IntptrTy, SizeInBytes + RightRedzoneSize),
         ConstantExpr::getPointerCast(Name, IntptrTy),
+        ConstantInt::get(IntptrTy, GlobalHasDynamicInitializer),
         NULL);
-    DEBUG(dbgs() << "NEW GLOBAL:\n" << *NewGlobal);
+
+    // Populate the first and last globals declared in this TU.
+    if (ClInitializers && GlobalHasDynamicInitializer) {
+      LastDynamic = ConstantExpr::getPointerCast(NewGlobal, IntptrTy);
+      if (FirstDynamic == 0)
+        FirstDynamic = LastDynamic;
+    }
+
+    DEBUG(dbgs() << "NEW GLOBAL: " << *NewGlobal << "\n");
   }
 
   ArrayType *ArrayOfGlobalStructTy = ArrayType::get(GlobalStructTy, n);
@@ -630,8 +661,13 @@
       M, ArrayOfGlobalStructTy, false, GlobalVariable::PrivateLinkage,
       ConstantArray::get(ArrayOfGlobalStructTy, Initializers), "");
 
+  // Create calls for poisoning before initializers run and unpoisoning after.
+  if (ClInitializers && FirstDynamic && LastDynamic)
+    createInitializerPoisonCalls(M, FirstDynamic, LastDynamic);
+
   Function *AsanRegisterGlobals = checkInterfaceFunction(M.getOrInsertFunction(
-      kAsanRegisterGlobalsName, IRB.getVoidTy(), IntptrTy, IntptrTy, NULL));
+      kAsanRegisterGlobalsName, IRB.getVoidTy(),
+      IntptrTy, IntptrTy, NULL));
   AsanRegisterGlobals->setLinkage(Function::ExternalLinkage);
 
   IRB.CreateCall2(AsanRegisterGlobals,
@@ -661,12 +697,13 @@
 }
 
 // virtual
-bool AddressSanitizer::runOnModule(Module &M) {
+bool AddressSanitizer::doInitialization(Module &M) {
   // Initialize the private fields. No one has accessed them before.
-  TD = getAnalysisIfAvailable<TargetData>();
+  TD = getAnalysisIfAvailable<DataLayout>();
+
   if (!TD)
     return false;
-  BL.reset(new FunctionBlackList(ClBlackListFile));
+  BL.reset(new BlackList(ClBlackListFile));
 
   C = &(M.getContext());
   LongSize = TD->getPointerSizeInBits();
@@ -694,22 +731,27 @@
       std::string FunctionName = std::string(kAsanReportErrorTemplate) +
           (AccessIsWrite ? "store" : "load") + itostr(1 << AccessSizeIndex);
       // If we are merging crash callbacks, they have two parameters.
-      if (ClMergeCallbacks)
-        AsanErrorCallback[AccessIsWrite][AccessSizeIndex] = cast<Function>(
-          M.getOrInsertFunction(FunctionName, IRB.getVoidTy(), IntptrTy,
-                                IntptrTy, NULL));
-      else
-        AsanErrorCallback[AccessIsWrite][AccessSizeIndex] = cast<Function>(
-          M.getOrInsertFunction(FunctionName, IRB.getVoidTy(), IntptrTy, NULL));
+      AsanErrorCallback[AccessIsWrite][AccessSizeIndex] =
+          checkInterfaceFunction(M.getOrInsertFunction(
+              FunctionName, IRB.getVoidTy(), IntptrTy, NULL));
     }
   }
+
+  AsanStackMallocFunc = checkInterfaceFunction(M.getOrInsertFunction(
+      kAsanStackMallocName, IntptrTy, IntptrTy, IntptrTy, NULL));
+  AsanStackFreeFunc = checkInterfaceFunction(M.getOrInsertFunction(
+      kAsanStackFreeName, IRB.getVoidTy(),
+      IntptrTy, IntptrTy, IntptrTy, NULL));
+  AsanHandleNoReturnFunc = checkInterfaceFunction(M.getOrInsertFunction(
+      kAsanHandleNoReturnName, IRB.getVoidTy(), NULL));
+
   // We insert an empty inline asm after __asan_report* to avoid callback merge.
   EmptyAsm = InlineAsm::get(FunctionType::get(IRB.getVoidTy(), false),
                             StringRef(""), StringRef(""),
                             /*hasSideEffects=*/true);
 
   llvm::Triple targetTriple(M.getTargetTriple());
-  bool isAndroid = targetTriple.getEnvironment() == llvm::Triple::ANDROIDEABI;
+  bool isAndroid = targetTriple.getEnvironment() == llvm::Triple::Android;
 
   MappingOffset = isAndroid ? kDefaultShadowOffsetAndroid :
     (LongSize == 32 ? kDefaultShadowOffset32 : kDefaultShadowOffset64);
@@ -729,10 +771,6 @@
   // For scales 6 and 7, the redzone has to be 64 and 128 bytes respectively.
   RedzoneSize = std::max(32, (int)(1 << MappingScale));
 
-  bool Res = false;
-
-  if (ClGlobals)
-    Res |= insertGlobalRedzones(M);
 
   if (ClMappingOffsetLog >= 0) {
     // Tell the run-time the current values of mapping offset and scale.
@@ -752,17 +790,20 @@
     IRB.CreateLoad(asan_mapping_scale, true);
   }
 
-
-  for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
-    if (F->isDeclaration()) continue;
-    Res |= handleFunction(M, *F);
-  }
-
   appendToGlobalCtors(M, AsanCtorFunction, kAsanCtorAndCtorPriority);
 
-  return Res;
+  return true;
+}
+
+bool AddressSanitizer::doFinalization(Module &M) {
+  // We transform the globals at the very end so that the optimization analysis
+  // works on the original globals.
+  if (ClGlobals)
+    return insertGlobalRedzones(M);
+  return false;
 }
 
+
 bool AddressSanitizer::maybeInsertAsanInitAtFunctionEntry(Function &F) {
   // For each NSObject descendant having a +load method, this method is invoked
   // by the ObjC runtime before any of the static constructors is called.
@@ -779,19 +820,22 @@
   return false;
 }
 
-bool AddressSanitizer::handleFunction(Module &M, Function &F) {
+bool AddressSanitizer::runOnFunction(Function &F) {
   if (BL->isIn(F)) return false;
   if (&F == AsanCtorFunction) return false;
+  DEBUG(dbgs() << "ASAN instrumenting:\n" << F << "\n");
 
   // If needed, insert __asan_init before checking for AddressSafety attr.
   maybeInsertAsanInitAtFunctionEntry(F);
 
-  if (!F.hasFnAttr(Attribute::AddressSafety)) return false;
+  if (!F.getFnAttributes().hasAttribute(Attributes::AddressSafety))
+    return false;
 
   if (!ClDebugFunc.empty() && ClDebugFunc != F.getName())
     return false;
-  // We want to instrument every address only once per basic block
-  // (unless there are calls between uses).
+
+  // We want to instrument every address only once per basic block (unless there
+  // are calls between uses).
   SmallSet<Value*, 16> TempsToInstrument;
   SmallVector<Instruction*, 16> ToInstrument;
   SmallVector<Instruction*, 8> NoReturnCalls;
@@ -829,8 +873,6 @@
     }
   }
 
-  AsanFunctionContext AFC(F);
-
   // Instrument.
   int NumInstrumented = 0;
   for (size_t i = 0, n = ToInstrument.size(); i != n; i++) {
@@ -838,52 +880,23 @@
     if (ClDebugMin < 0 || ClDebugMax < 0 ||
         (NumInstrumented >= ClDebugMin && NumInstrumented <= ClDebugMax)) {
       if (isInterestingMemoryAccess(Inst, &IsWrite))
-        instrumentMop(AFC, Inst);
+        instrumentMop(Inst);
       else
-        instrumentMemIntrinsic(AFC, cast<MemIntrinsic>(Inst));
+        instrumentMemIntrinsic(cast<MemIntrinsic>(Inst));
     }
     NumInstrumented++;
   }
 
-  // Create PHI nodes and crash callbacks if we are merging crash callbacks.
-  if (NumInstrumented) {
-    for (size_t IsWrite = 0; IsWrite <= 1; IsWrite++) {
-      for (size_t AccessSizeIndex = 0; AccessSizeIndex < kNumberOfAccessSizes;
-           AccessSizeIndex++) {
-        BasicBlock *BB = AFC.CrashBlock[IsWrite][AccessSizeIndex];
-        if (!BB) continue;
-        assert(ClMergeCallbacks);
-        AsanFunctionContext::CrashArgsVec &Args =
-            AFC.CrashArgs[IsWrite][AccessSizeIndex];
-        IRBuilder<> IRB(BB->getFirstNonPHI());
-        size_t n = Args.size();
-        PHINode *PN1 = IRB.CreatePHI(IntptrTy, n);
-        PHINode *PN2 = IRB.CreatePHI(IntptrTy, n);
-        // We need to match crash parameters and the predecessors.
-        for (pred_iterator PI = pred_begin(BB), PE = pred_end(BB);
-             PI != PE; ++PI) {
-          n--;
-          PN1->addIncoming(Args[n].Arg1, *PI);
-          PN2->addIncoming(Args[n].Arg2, *PI);
-        }
-        assert(n == 0);
-        generateCrashCode(BB, PN1, PN2, IsWrite, AccessSizeIndex);
-      }
-    }
-  }
-
-  DEBUG(dbgs() << F);
-
-  bool ChangedStack = poisonStackInFunction(M, F);
+  bool ChangedStack = poisonStackInFunction(F);
 
   // We must unpoison the stack before every NoReturn call (throw, _exit, etc).
   // See e.g. http://code.google.com/p/address-sanitizer/issues/detail?id=37
   for (size_t i = 0, n = NoReturnCalls.size(); i != n; i++) {
     Instruction *CI = NoReturnCalls[i];
     IRBuilder<> IRB(CI);
-    IRB.CreateCall(M.getOrInsertFunction(kAsanHandleNoReturnName,
-                                         IRB.getVoidTy(), NULL));
+    IRB.CreateCall(AsanHandleNoReturnFunc);
   }
+  DEBUG(dbgs() << "ASAN done instrumenting:\n" << F << "\n");
 
   return NumInstrumented > 0 || ChangedStack || !NoReturnCalls.empty();
 }
@@ -996,7 +1009,7 @@
 // compiler hoists the load of the shadow value somewhere too high.
 // This causes asan to report a non-existing bug on 453.povray.
 // It sounds like an LLVM bug.
-bool AddressSanitizer::poisonStackInFunction(Module &M, Function &F) {
+bool AddressSanitizer::poisonStackInFunction(Function &F) {
   if (!ClStack) return false;
   SmallVector<AllocaInst*, 16> AllocaVec;
   SmallVector<Instruction*, 8> RetVec;
@@ -1046,8 +1059,6 @@
   Value *LocalStackBase = OrigStackBase;
 
   if (DoStackMalloc) {
-    Value *AsanStackMallocFunc = M.getOrInsertFunction(
-        kAsanStackMallocName, IntptrTy, IntptrTy, IntptrTy, NULL);
     LocalStackBase = IRB.CreateCall2(AsanStackMallocFunc,
         ConstantInt::get(IntptrTy, LocalStackSize), OrigStackBase);
   }
@@ -1082,22 +1093,16 @@
   Value *BasePlus1 = IRB.CreateAdd(LocalStackBase,
                                    ConstantInt::get(IntptrTy, LongSize/8));
   BasePlus1 = IRB.CreateIntToPtr(BasePlus1, IntptrPtrTy);
-  Value *Description = IRB.CreatePointerCast(
-      createPrivateGlobalForString(M, StackDescription.str()),
-      IntptrTy);
+  GlobalVariable *StackDescriptionGlobal =
+      createPrivateGlobalForString(*F.getParent(), StackDescription.str());
+  GlobalsCreatedByAsan.insert(StackDescriptionGlobal);
+  Value *Description = IRB.CreatePointerCast(StackDescriptionGlobal, IntptrTy);
   IRB.CreateStore(Description, BasePlus1);
 
   // Poison the stack redzones at the entry.
   Value *ShadowBase = memToShadow(LocalStackBase, IRB);
   PoisonStack(ArrayRef<AllocaInst*>(AllocaVec), IRB, ShadowBase, true);
 
-  Value *AsanStackFreeFunc = NULL;
-  if (DoStackMalloc) {
-    AsanStackFreeFunc = M.getOrInsertFunction(
-        kAsanStackFreeName, IRB.getVoidTy(),
-        IntptrTy, IntptrTy, IntptrTy, NULL);
-  }
-
   // Unpoison the stack before all ret instructions.
   for (size_t i = 0, n = RetVec.size(); i < n; i++) {
     Instruction *Ret = RetVec[i];
@@ -1116,6 +1121,10 @@
     }
   }
 
+  // We are done. Remove the old unused alloca instructions.
+  for (size_t i = 0, n = AllocaVec.size(); i < n; i++)
+    AllocaVec[i]->eraseFromParent();
+
   if (ClDebugStack) {
     DEBUG(dbgs() << F);
   }

Modified: llvm/branches/AMDILBackend/lib/Transforms/Instrumentation/BoundsChecking.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Instrumentation/BoundsChecking.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Instrumentation/BoundsChecking.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Instrumentation/BoundsChecking.cpp Tue Jan 15 11:16:16 2013
@@ -23,7 +23,8 @@
 #include "llvm/Support/InstIterator.h"
 #include "llvm/Support/TargetFolder.h"
 #include "llvm/Support/raw_ostream.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
+#include "llvm/Target/TargetLibraryInfo.h"
 #include "llvm/Transforms/Instrumentation.h"
 using namespace llvm;
 
@@ -47,11 +48,13 @@
     virtual bool runOnFunction(Function &F);
 
     virtual void getAnalysisUsage(AnalysisUsage &AU) const {
-      AU.addRequired<TargetData>();
+      AU.addRequired<DataLayout>();
+      AU.addRequired<TargetLibraryInfo>();
     }
 
   private:
-    const TargetData *TD;
+    const DataLayout *TD;
+    const TargetLibraryInfo *TLI;
     ObjectSizeOffsetEvaluator *ObjSizeEval;
     BuilderTy *Builder;
     Instruction *Inst;
@@ -140,7 +143,7 @@
   Value *Offset = SizeOffset.second;
   ConstantInt *SizeCI = dyn_cast<ConstantInt>(Size);
 
-  IntegerType *IntTy = TD->getIntPtrType(Inst->getContext());
+  Type *IntTy = TD->getIntPtrType(Ptr->getType());
   Value *NeededSizeVal = ConstantInt::get(IntTy, NeededSize);
 
   // three checks are required to ensure safety:
@@ -165,12 +168,13 @@
 }
 
 bool BoundsChecking::runOnFunction(Function &F) {
-  TD = &getAnalysis<TargetData>();
+  TD = &getAnalysis<DataLayout>();
+  TLI = &getAnalysis<TargetLibraryInfo>();
 
   TrapBB = 0;
   BuilderTy TheBuilder(F.getContext(), TargetFolder(TD));
   Builder = &TheBuilder;
-  ObjectSizeOffsetEvaluator TheObjSizeEval(TD, F.getContext());
+  ObjectSizeOffsetEvaluator TheObjSizeEval(TD, TLI, F.getContext());
   ObjSizeEval = &TheObjSizeEval;
 
   // check HANDLE_MEMORY_INST in include/llvm/Instruction.def for memory

Modified: llvm/branches/AMDILBackend/lib/Transforms/Instrumentation/CMakeLists.txt
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Instrumentation/CMakeLists.txt?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Instrumentation/CMakeLists.txt (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Instrumentation/CMakeLists.txt Tue Jan 15 11:16:16 2013
@@ -1,8 +1,8 @@
 add_llvm_library(LLVMInstrumentation
   AddressSanitizer.cpp
+  BlackList.cpp
   BoundsChecking.cpp
   EdgeProfiling.cpp
-  FunctionBlackList.cpp
   GCOVProfiling.cpp
   Instrumentation.cpp
   OptimalEdgeProfiling.cpp

Removed: llvm/branches/AMDILBackend/lib/Transforms/Instrumentation/FunctionBlackList.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Instrumentation/FunctionBlackList.cpp?rev=172540&view=auto
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Instrumentation/FunctionBlackList.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Instrumentation/FunctionBlackList.cpp (removed)
@@ -1,79 +0,0 @@
-//===-- FunctionBlackList.cpp - blacklist of functions --------------------===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This is a utility class for instrumentation passes (like AddressSanitizer 
-// or ThreadSanitizer) to avoid instrumenting some functions based on
-// user-supplied blacklist.
-//
-//===----------------------------------------------------------------------===//
-
-#include "FunctionBlackList.h"
-#include "llvm/ADT/OwningPtr.h"
-#include "llvm/ADT/SmallVector.h"
-#include "llvm/ADT/StringExtras.h"
-#include "llvm/Function.h"
-#include "llvm/Support/MemoryBuffer.h"
-#include "llvm/Support/Regex.h"
-#include "llvm/Support/raw_ostream.h"
-#include "llvm/Support/system_error.h"
-
-namespace llvm {
-
-FunctionBlackList::FunctionBlackList(const std::string &Path) {
-  Functions = NULL;
-  const char *kFunPrefix = "fun:";
-  if (!Path.size()) return;
-  std::string Fun;
-
-  OwningPtr<MemoryBuffer> File;
-  if (error_code EC = MemoryBuffer::getFile(Path.c_str(), File)) {
-    report_fatal_error("Can't open blacklist file " + Path + ": " +
-                       EC.message());
-  }
-  MemoryBuffer *Buff = File.take();
-  const char *Data = Buff->getBufferStart();
-  size_t DataLen = Buff->getBufferSize();
-  SmallVector<StringRef, 16> Lines;
-  SplitString(StringRef(Data, DataLen), Lines, "\n\r");
-  for (size_t i = 0, numLines = Lines.size(); i < numLines; i++) {
-    if (Lines[i].startswith(kFunPrefix)) {
-      std::string ThisFunc = Lines[i].substr(strlen(kFunPrefix));
-      std::string ThisFuncRE;
-      // add ThisFunc replacing * with .*
-      for (size_t j = 0, n = ThisFunc.size(); j < n; j++) {
-        if (ThisFunc[j] == '*')
-          ThisFuncRE += '.';
-        ThisFuncRE += ThisFunc[j];
-      }
-      // Check that the regexp is valid.
-      Regex CheckRE(ThisFuncRE);
-      std::string Error;
-      if (!CheckRE.isValid(Error))
-        report_fatal_error("malformed blacklist regex: " + ThisFunc +
-                           ": " + Error);
-      // Append to the final regexp.
-      if (Fun.size())
-        Fun += "|";
-      Fun += ThisFuncRE;
-    }
-  }
-  if (Fun.size()) {
-    Functions = new Regex(Fun);
-  }
-}
-
-bool FunctionBlackList::isIn(const Function &F) {
-  if (Functions) {
-    bool Res = Functions->match(F.getName());
-    return Res;
-  }
-  return false;
-}
-
-}  // namespace llvm

Removed: llvm/branches/AMDILBackend/lib/Transforms/Instrumentation/FunctionBlackList.h
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Instrumentation/FunctionBlackList.h?rev=172540&view=auto
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Instrumentation/FunctionBlackList.h (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Instrumentation/FunctionBlackList.h (removed)
@@ -1,37 +0,0 @@
-//===-- FunctionBlackList.cpp - blacklist of functions ----------*- C++ -*-===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//===----------------------------------------------------------------------===//
-//
-// This is a utility class for instrumentation passes (like AddressSanitizer
-// or ThreadSanitizer) to avoid instrumenting some functions based on
-// user-supplied blacklist.
-//
-//===----------------------------------------------------------------------===//
-//
-
-#include <string>
-
-namespace llvm {
-class Function;
-class Regex;
-
-// Blacklisted functions are not instrumented.
-// The blacklist file contains one or more lines like this:
-// ---
-// fun:FunctionWildCard
-// ---
-// This is similar to the "ignore" feature of ThreadSanitizer.
-// http://code.google.com/p/data-race-test/wiki/ThreadSanitizerIgnores
-class FunctionBlackList {
- public:
-  FunctionBlackList(const std::string &Path);
-  bool isIn(const Function &F);
- private:
-  Regex *Functions;
-};
-
-}  // namespace llvm

Modified: llvm/branches/AMDILBackend/lib/Transforms/Instrumentation/GCOVProfiling.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Instrumentation/GCOVProfiling.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Instrumentation/GCOVProfiling.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Instrumentation/GCOVProfiling.cpp Tue Jan 15 11:16:16 2013
@@ -88,11 +88,11 @@
 
     // Add the function to write out all our counters to the global destructor
     // list.
-    void insertCounterWriteout(SmallVector<std::pair<GlobalVariable *,
-                                                     MDNode *>, 8> &);
+    void insertCounterWriteout(ArrayRef<std::pair<GlobalVariable*, MDNode*> >);
     void insertIndirectCounterIncrement();
+    void insertFlush(ArrayRef<std::pair<GlobalVariable*, MDNode*> >);
 
-    std::string mangleName(DICompileUnit CU, std::string NewStem);
+    std::string mangleName(DICompileUnit CU, const char *NewStem);
 
     bool EmitNotes;
     bool EmitData;
@@ -329,7 +329,7 @@
   };
 }
 
-std::string GCOVProfiler::mangleName(DICompileUnit CU, std::string NewStem) {
+std::string GCOVProfiler::mangleName(DICompileUnit CU, const char *NewStem) {
   if (NamedMDNode *GCov = M->getNamedMetadata("llvm.gcov")) {
     for (int i = 0, e = GCov->getNumOperands(); i != e; ++i) {
       MDNode *N = GCov->getOperand(i);
@@ -519,6 +519,7 @@
     }
 
     insertCounterWriteout(CountersBySP);
+    insertFlush(CountersBySP);
   }
 
   if (InsertIndCounterIncrCode)
@@ -630,14 +631,15 @@
 }
 
 void GCOVProfiler::insertCounterWriteout(
-    SmallVector<std::pair<GlobalVariable *, MDNode *>, 8> &CountersBySP) {
-  FunctionType *WriteoutFTy =
-      FunctionType::get(Type::getVoidTy(*Ctx), false);
-  Function *WriteoutF = Function::Create(WriteoutFTy,
-                                         GlobalValue::InternalLinkage,
-                                         "__llvm_gcov_writeout", M);
+    ArrayRef<std::pair<GlobalVariable *, MDNode *> > CountersBySP) {
+  FunctionType *WriteoutFTy = FunctionType::get(Type::getVoidTy(*Ctx), false);
+  Function *WriteoutF = M->getFunction("__llvm_gcov_writeout");
+  if (!WriteoutF)
+    WriteoutF = Function::Create(WriteoutFTy, GlobalValue::InternalLinkage,
+                                 "__llvm_gcov_writeout", M);
   WriteoutF->setUnnamedAddr(true);
-  BasicBlock *BB = BasicBlock::Create(*Ctx, "", WriteoutF);
+
+  BasicBlock *BB = BasicBlock::Create(*Ctx, "entry", WriteoutF);
   IRBuilder<> Builder(BB);
 
   Constant *StartFile = getStartFileFunc();
@@ -648,11 +650,11 @@
   NamedMDNode *CU_Nodes = M->getNamedMetadata("llvm.dbg.cu");
   if (CU_Nodes) {
     for (unsigned i = 0, e = CU_Nodes->getNumOperands(); i != e; ++i) {
-      DICompileUnit compile_unit(CU_Nodes->getOperand(i));
-      std::string FilenameGcda = mangleName(compile_unit, "gcda");
+      DICompileUnit CU(CU_Nodes->getOperand(i));
+      std::string FilenameGcda = mangleName(CU, "gcda");
       Builder.CreateCall(StartFile,
                          Builder.CreateGlobalStringPtr(FilenameGcda));
-      for (SmallVector<std::pair<GlobalVariable *, MDNode *>, 8>::iterator
+      for (ArrayRef<std::pair<GlobalVariable *, MDNode *> >::iterator
              I = CountersBySP.begin(), E = CountersBySP.end();
            I != E; ++I) {
         DISubprogram SP(I->second);
@@ -680,7 +682,7 @@
                                  "__llvm_gcov_init", M);
   F->setUnnamedAddr(true);
   F->setLinkage(GlobalValue::InternalLinkage);
-  F->addFnAttr(Attribute::NoInline);
+  F->addFnAttr(Attributes::NoInline);
 
   BB = BasicBlock::Create(*Ctx, "entry", F);
   Builder.SetInsertPoint(BB);
@@ -699,7 +701,7 @@
     cast<Function>(GCOVProfiler::getIncrementIndirectCounterFunc());
   Fn->setUnnamedAddr(true);
   Fn->setLinkage(GlobalValue::InternalLinkage);
-  Fn->addFnAttr(Attribute::NoInline);
+  Fn->addFnAttr(Attributes::NoInline);
 
   Type *Int32Ty = Type::getInt32Ty(*Ctx);
   Type *Int64Ty = Type::getInt64Ty(*Ctx);
@@ -745,3 +747,42 @@
   Builder.SetInsertPoint(Exit);
   Builder.CreateRetVoid();
 }
+
+void GCOVProfiler::
+insertFlush(ArrayRef<std::pair<GlobalVariable*, MDNode*> > CountersBySP) {
+  FunctionType *FTy = FunctionType::get(Type::getVoidTy(*Ctx), false);
+  Function *FlushF = M->getFunction("__gcov_flush");
+  if (!FlushF)
+    FlushF = Function::Create(FTy, GlobalValue::InternalLinkage,
+                              "__gcov_flush", M);
+  else
+    FlushF->setLinkage(GlobalValue::InternalLinkage);
+  FlushF->setUnnamedAddr(true);
+
+  BasicBlock *Entry = BasicBlock::Create(*Ctx, "entry", FlushF);
+
+  // Write out the current counters.
+  Constant *WriteoutF = M->getFunction("__llvm_gcov_writeout");
+  assert(WriteoutF && "Need to create the writeout function first!");
+
+  IRBuilder<> Builder(Entry);
+  Builder.CreateCall(WriteoutF);
+
+  // Zero out the counters.
+  for (ArrayRef<std::pair<GlobalVariable *, MDNode *> >::iterator
+         I = CountersBySP.begin(), E = CountersBySP.end();
+       I != E; ++I) {
+    GlobalVariable *GV = I->first;
+    Constant *Null = Constant::getNullValue(GV->getType()->getElementType());
+    Builder.CreateStore(Null, GV);
+  }
+
+  Type *RetTy = FlushF->getReturnType();
+  if (RetTy == Type::getVoidTy(*Ctx))
+    Builder.CreateRetVoid();
+  else if (RetTy->isIntegerTy())
+    // Used if __gcov_flush was implicitly declared.
+    Builder.CreateRet(ConstantInt::get(RetTy, 0));
+  else
+    report_fatal_error("invalid return type for __gcov_flush");
+}

Modified: llvm/branches/AMDILBackend/lib/Transforms/Instrumentation/MaximumSpanningTree.h
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Instrumentation/MaximumSpanningTree.h?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Instrumentation/MaximumSpanningTree.h (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Instrumentation/MaximumSpanningTree.h Tue Jan 15 11:16:16 2013
@@ -26,30 +26,6 @@
   /// The type parameter T determines the type of the nodes of the graph.
   template <typename T>
   class MaximumSpanningTree {
-
-    // A comparing class for comparing weighted edges.
-    template <typename CT>
-    struct EdgeWeightCompare {
-      bool operator()(typename MaximumSpanningTree<CT>::EdgeWeight X, 
-                      typename MaximumSpanningTree<CT>::EdgeWeight Y) const {
-        if (X.second > Y.second) return true;
-        if (X.second < Y.second) return false;
-        if (const BasicBlock *BBX = dyn_cast<BasicBlock>(X.first.first)) {
-          if (const BasicBlock *BBY = dyn_cast<BasicBlock>(Y.first.first)) {
-            if (BBX->size() > BBY->size()) return true;
-            if (BBX->size() < BBY->size()) return false;
-          }
-        }
-        if (const BasicBlock *BBX = dyn_cast<BasicBlock>(X.first.second)) {
-          if (const BasicBlock *BBY = dyn_cast<BasicBlock>(Y.first.second)) {
-            if (BBX->size() > BBY->size()) return true;
-            if (BBX->size() < BBY->size()) return false;
-          }
-        }
-        return false;
-      }
-    };
-
   public:
     typedef std::pair<const T*, const T*> Edge;
     typedef std::pair<Edge, double> EdgeWeight;
@@ -59,6 +35,33 @@
 
     MaxSpanTree MST;
 
+  private:
+    // A comparing class for comparing weighted edges.
+    struct EdgeWeightCompare {
+      static bool getBlockSize(const T *X) {
+        const BasicBlock *BB = dyn_cast_or_null<BasicBlock>(X);
+        return BB ? BB->size() : 0;
+      }
+
+      bool operator()(EdgeWeight X, EdgeWeight Y) const {
+        if (X.second > Y.second) return true;
+        if (X.second < Y.second) return false;
+
+        // Equal edge weights: break ties by comparing block sizes.
+        size_t XSizeA = getBlockSize(X.first.first);
+        size_t YSizeA = getBlockSize(Y.first.first);
+        if (XSizeA > YSizeA) return true;
+        if (XSizeA < YSizeA) return false;
+
+        size_t XSizeB = getBlockSize(X.first.second);
+        size_t YSizeB = getBlockSize(Y.first.second);
+        if (XSizeB > YSizeB) return true;
+        if (XSizeB < YSizeB) return false;
+
+        return false;
+      }
+    };
+
   public:
     static char ID; // Class identification, replacement for typeinfo
 
@@ -66,7 +69,7 @@
     /// spanning tree.
     MaximumSpanningTree(EdgeWeights &EdgeVector) {
 
-      std::stable_sort(EdgeVector.begin(), EdgeVector.end(), EdgeWeightCompare<T>());
+      std::stable_sort(EdgeVector.begin(), EdgeVector.end(), EdgeWeightCompare());
 
       // Create spanning tree, Forest contains a special data structure
       // that makes checking if two nodes are already in a common (sub-)tree

Modified: llvm/branches/AMDILBackend/lib/Transforms/Instrumentation/ThreadSanitizer.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Instrumentation/ThreadSanitizer.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Instrumentation/ThreadSanitizer.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Instrumentation/ThreadSanitizer.cpp Tue Jan 15 11:16:16 2013
@@ -21,7 +21,7 @@
 
 #define DEBUG_TYPE "tsan"
 
-#include "FunctionBlackList.h"
+#include "BlackList.h"
 #include "llvm/Function.h"
 #include "llvm/IRBuilder.h"
 #include "llvm/Intrinsics.h"
@@ -38,7 +38,7 @@
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/MathExtras.h"
 #include "llvm/Support/raw_ostream.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Transforms/Instrumentation.h"
 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
 #include "llvm/Transforms/Utils/ModuleUtils.h"
@@ -47,10 +47,19 @@
 
 static cl::opt<std::string>  ClBlackListFile("tsan-blacklist",
        cl::desc("Blacklist file"), cl::Hidden);
+static cl::opt<bool>  ClInstrumentMemoryAccesses(
+    "tsan-instrument-memory-accesses", cl::init(true),
+    cl::desc("Instrument memory accesses"), cl::Hidden);
+static cl::opt<bool>  ClInstrumentFuncEntryExit(
+    "tsan-instrument-func-entry-exit", cl::init(true),
+    cl::desc("Instrument function entry and exit"), cl::Hidden);
+static cl::opt<bool>  ClInstrumentAtomics(
+    "tsan-instrument-atomics", cl::init(true),
+    cl::desc("Instrument atomics"), cl::Hidden);
 
 STATISTIC(NumInstrumentedReads, "Number of instrumented reads");
 STATISTIC(NumInstrumentedWrites, "Number of instrumented writes");
-STATISTIC(NumOmittedReadsBeforeWrite, 
+STATISTIC(NumOmittedReadsBeforeWrite,
           "Number of reads ignored due to following writes");
 STATISTIC(NumAccessesWithBadSize, "Number of accesses with bad size");
 STATISTIC(NumInstrumentedVtableWrites, "Number of vtable ptr writes");
@@ -76,8 +85,8 @@
   bool addrPointsToConstantData(Value *Addr);
   int getMemoryAccessFuncIndex(Value *Addr);
 
-  TargetData *TD;
-  OwningPtr<FunctionBlackList> BL;
+  DataLayout *TD;
+  OwningPtr<BlackList> BL;
   IntegerType *OrdTy;
   // Callbacks to run-time library are computed in doInitialization.
   Function *TsanFuncEntry;
@@ -88,6 +97,10 @@
   Function *TsanWrite[kNumberOfAccessSizes];
   Function *TsanAtomicLoad[kNumberOfAccessSizes];
   Function *TsanAtomicStore[kNumberOfAccessSizes];
+  Function *TsanAtomicRMW[AtomicRMWInst::LAST_BINOP + 1][kNumberOfAccessSizes];
+  Function *TsanAtomicCAS[kNumberOfAccessSizes];
+  Function *TsanAtomicThreadFence;
+  Function *TsanAtomicSignalFence;
   Function *TsanVptrUpdate;
 };
 }  // namespace
@@ -118,10 +131,10 @@
 }
 
 bool ThreadSanitizer::doInitialization(Module &M) {
-  TD = getAnalysisIfAvailable<TargetData>();
+  TD = getAnalysisIfAvailable<DataLayout>();
   if (!TD)
     return false;
-  BL.reset(new FunctionBlackList(ClBlackListFile));
+  BL.reset(new BlackList(ClBlackListFile));
 
   // Always insert a call to __tsan_init into the module's CTORs.
   IRBuilder<> IRB(M.getContext());
@@ -158,10 +171,42 @@
     TsanAtomicStore[i] = checkInterfaceFunction(M.getOrInsertFunction(
         AtomicStoreName, IRB.getVoidTy(), PtrTy, Ty, OrdTy,
         NULL));
+
+    for (int op = AtomicRMWInst::FIRST_BINOP;
+        op <= AtomicRMWInst::LAST_BINOP; ++op) {
+      TsanAtomicRMW[op][i] = NULL;
+      const char *NamePart = NULL;
+      if (op == AtomicRMWInst::Xchg)
+        NamePart = "_exchange";
+      else if (op == AtomicRMWInst::Add)
+        NamePart = "_fetch_add";
+      else if (op == AtomicRMWInst::Sub)
+        NamePart = "_fetch_sub";
+      else if (op == AtomicRMWInst::And)
+        NamePart = "_fetch_and";
+      else if (op == AtomicRMWInst::Or)
+        NamePart = "_fetch_or";
+      else if (op == AtomicRMWInst::Xor)
+        NamePart = "_fetch_xor";
+      else
+        continue;
+      SmallString<32> RMWName("__tsan_atomic" + itostr(BitSize) + NamePart);
+      TsanAtomicRMW[op][i] = checkInterfaceFunction(M.getOrInsertFunction(
+          RMWName, Ty, PtrTy, Ty, OrdTy, NULL));
+    }
+
+    SmallString<32> AtomicCASName("__tsan_atomic" + itostr(BitSize) +
+                                  "_compare_exchange_val");
+    TsanAtomicCAS[i] = checkInterfaceFunction(M.getOrInsertFunction(
+        AtomicCASName, Ty, PtrTy, Ty, Ty, OrdTy, NULL));
   }
   TsanVptrUpdate = checkInterfaceFunction(M.getOrInsertFunction(
       "__tsan_vptr_update", IRB.getVoidTy(), IRB.getInt8PtrTy(),
       IRB.getInt8PtrTy(), NULL));
+  TsanAtomicThreadFence = checkInterfaceFunction(M.getOrInsertFunction(
+      "__tsan_atomic_thread_fence", IRB.getVoidTy(), OrdTy, NULL));
+  TsanAtomicSignalFence = checkInterfaceFunction(M.getOrInsertFunction(
+      "__tsan_atomic_signal_fence", IRB.getVoidTy(), OrdTy, NULL));
   return true;
 }
 
@@ -186,7 +231,7 @@
       NumOmittedReadsFromConstantGlobals++;
       return true;
     }
-  } else if(LoadInst *L = dyn_cast<LoadInst>(Addr)) {
+  } else if (LoadInst *L = dyn_cast<LoadInst>(Addr)) {
     if (isVtableAccess(L)) {
       // Reads from a vtable pointer can not race with any writes.
       NumOmittedReadsFromVtable++;
@@ -244,8 +289,8 @@
     return true;
   if (isa<AtomicCmpXchgInst>(I))
     return true;
-  if (FenceInst *FI = dyn_cast<FenceInst>(I))
-    return FI->getSynchScope() == CrossThread;
+  if (isa<FenceInst>(I))
+    return true;
   return false;
 }
 
@@ -284,17 +329,19 @@
   // (e.g. variables that do not escape, etc).
 
   // Instrument memory accesses.
-  for (size_t i = 0, n = AllLoadsAndStores.size(); i < n; ++i) {
-    Res |= instrumentLoadOrStore(AllLoadsAndStores[i]);
-  }
+  if (ClInstrumentMemoryAccesses)
+    for (size_t i = 0, n = AllLoadsAndStores.size(); i < n; ++i) {
+      Res |= instrumentLoadOrStore(AllLoadsAndStores[i]);
+    }
 
   // Instrument atomic memory accesses.
-  for (size_t i = 0, n = AtomicAccesses.size(); i < n; ++i) {
-    Res |= instrumentAtomic(AtomicAccesses[i]);
-  }
+  if (ClInstrumentAtomics)
+    for (size_t i = 0, n = AtomicAccesses.size(); i < n; ++i) {
+      Res |= instrumentAtomic(AtomicAccesses[i]);
+    }
 
   // Instrument function entry/exit points if there were instrumented accesses.
-  if (Res || HasCalls) {
+  if ((Res || HasCalls) && ClInstrumentFuncEntryExit) {
     IRBuilder<> IRB(F.getEntryBlock().getFirstNonPHI());
     Value *ReturnAddress = IRB.CreateCall(
         Intrinsic::getDeclaration(F.getParent(), Intrinsic::returnaddress),
@@ -343,12 +390,12 @@
   switch (ord) {
     case NotAtomic:              assert(false);
     case Unordered:              // Fall-through.
-    case Monotonic:              v = 1 << 0; break;
- // case Consume:                v = 1 << 1; break;  // Not specified yet.
-    case Acquire:                v = 1 << 2; break;
-    case Release:                v = 1 << 3; break;
-    case AcquireRelease:         v = 1 << 4; break;
-    case SequentiallyConsistent: v = 1 << 5; break;
+    case Monotonic:              v = 0; break;
+ // case Consume:                v = 1; break;  // Not specified yet.
+    case Acquire:                v = 2; break;
+    case Release:                v = 3; break;
+    case AcquireRelease:         v = 4; break;
+    case SequentiallyConsistent: v = 5; break;
   }
   return IRB->getInt32(v);
 }
@@ -385,12 +432,44 @@
     CallInst *C = CallInst::Create(TsanAtomicStore[Idx],
                                    ArrayRef<Value*>(Args));
     ReplaceInstWithInst(I, C);
-  } else if (isa<AtomicRMWInst>(I)) {
-    // FIXME: Not yet supported.
-  } else if (isa<AtomicCmpXchgInst>(I)) {
-    // FIXME: Not yet supported.
-  } else if (isa<FenceInst>(I)) {
-    // FIXME: Not yet supported.
+  } else if (AtomicRMWInst *RMWI = dyn_cast<AtomicRMWInst>(I)) {
+    Value *Addr = RMWI->getPointerOperand();
+    int Idx = getMemoryAccessFuncIndex(Addr);
+    if (Idx < 0)
+      return false;
+    Function *F = TsanAtomicRMW[RMWI->getOperation()][Idx];
+    if (F == NULL)
+      return false;
+    const size_t ByteSize = 1 << Idx;
+    const size_t BitSize = ByteSize * 8;
+    Type *Ty = Type::getIntNTy(IRB.getContext(), BitSize);
+    Type *PtrTy = Ty->getPointerTo();
+    Value *Args[] = {IRB.CreatePointerCast(Addr, PtrTy),
+                     IRB.CreateIntCast(RMWI->getValOperand(), Ty, false),
+                     createOrdering(&IRB, RMWI->getOrdering())};
+    CallInst *C = CallInst::Create(F, ArrayRef<Value*>(Args));
+    ReplaceInstWithInst(I, C);
+  } else if (AtomicCmpXchgInst *CASI = dyn_cast<AtomicCmpXchgInst>(I)) {
+    Value *Addr = CASI->getPointerOperand();
+    int Idx = getMemoryAccessFuncIndex(Addr);
+    if (Idx < 0)
+      return false;
+    const size_t ByteSize = 1 << Idx;
+    const size_t BitSize = ByteSize * 8;
+    Type *Ty = Type::getIntNTy(IRB.getContext(), BitSize);
+    Type *PtrTy = Ty->getPointerTo();
+    Value *Args[] = {IRB.CreatePointerCast(Addr, PtrTy),
+                     IRB.CreateIntCast(CASI->getCompareOperand(), Ty, false),
+                     IRB.CreateIntCast(CASI->getNewValOperand(), Ty, false),
+                     createOrdering(&IRB, CASI->getOrdering())};
+    CallInst *C = CallInst::Create(TsanAtomicCAS[Idx], ArrayRef<Value*>(Args));
+    ReplaceInstWithInst(I, C);
+  } else if (FenceInst *FI = dyn_cast<FenceInst>(I)) {
+    Value *Args[] = {createOrdering(&IRB, FI->getOrdering())};
+    Function *F = FI->getSynchScope() == SingleThread ?
+        TsanAtomicSignalFence : TsanAtomicThreadFence;
+    CallInst *C = CallInst::Create(F, ArrayRef<Value*>(Args));
+    ReplaceInstWithInst(I, C);
   }
   return true;
 }

Modified: llvm/branches/AMDILBackend/lib/Transforms/Scalar/CMakeLists.txt
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Scalar/CMakeLists.txt?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Scalar/CMakeLists.txt (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Scalar/CMakeLists.txt Tue Jan 15 11:16:16 2013
@@ -25,6 +25,7 @@
   Reassociate.cpp
   Reg2Mem.cpp
   SCCP.cpp
+  SROA.cpp
   Scalar.cpp
   ScalarReplAggregates.cpp
   SimplifyCFGPass.cpp

Modified: llvm/branches/AMDILBackend/lib/Transforms/Scalar/CodeGenPrepare.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Scalar/CodeGenPrepare.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Scalar/CodeGenPrepare.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Scalar/CodeGenPrepare.cpp Tue Jan 15 11:16:16 2013
@@ -27,6 +27,7 @@
 #include "llvm/ADT/SmallSet.h"
 #include "llvm/ADT/Statistic.h"
 #include "llvm/Analysis/Dominators.h"
+#include "llvm/Analysis/DominatorInternals.h"
 #include "llvm/Analysis/InstructionSimplify.h"
 #include "llvm/Analysis/ProfileInfo.h"
 #include "llvm/Assembly/Writer.h"
@@ -37,12 +38,13 @@
 #include "llvm/Support/PatternMatch.h"
 #include "llvm/Support/ValueHandle.h"
 #include "llvm/Support/raw_ostream.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Target/TargetLibraryInfo.h"
 #include "llvm/Target/TargetLowering.h"
 #include "llvm/Transforms/Utils/AddrModeMatcher.h"
 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
 #include "llvm/Transforms/Utils/BuildLibCalls.h"
+#include "llvm/Transforms/Utils/BypassSlowDivision.h"
 #include "llvm/Transforms/Utils/Local.h"
 using namespace llvm;
 using namespace llvm::PatternMatch;
@@ -66,11 +68,6 @@
   "disable-cgp-branch-opts", cl::Hidden, cl::init(false),
   cl::desc("Disable branch optimizations in CodeGenPrepare"));
 
-// FIXME: Remove this abomination once all of the tests pass without it!
-static cl::opt<bool> DisableDeleteDeadBlocks(
-  "disable-cgp-delete-dead-blocks", cl::Hidden, cl::init(false),
-  cl::desc("Disable deleting dead blocks in CodeGenPrepare"));
-
 static cl::opt<bool> DisableSelectToBranch(
   "disable-cgp-select2branch", cl::Hidden, cl::init(false),
   cl::desc("Disable select to branch conversion."));
@@ -116,6 +113,7 @@
     }
 
   private:
+    bool EliminateFallThrough(Function &F);
     bool EliminateMostlyEmptyBlocks(Function &F);
     bool CanMergeBlocks(const BasicBlock *BB, const BasicBlock *DestBB) const;
     void EliminateMostlyEmptyBlock(BasicBlock *BB);
@@ -150,9 +148,18 @@
   TLInfo = &getAnalysis<TargetLibraryInfo>();
   DT = getAnalysisIfAvailable<DominatorTree>();
   PFI = getAnalysisIfAvailable<ProfileInfo>();
-  OptSize = F.hasFnAttr(Attribute::OptimizeForSize);
+  OptSize = F.getFnAttributes().hasAttribute(Attributes::OptimizeForSize);
+
+  /// This optimization identifies DIV instructions that can be
+  /// profitably bypassed and carried out with a shorter, faster divide.
+  if (TLI && TLI->isSlowDivBypassed()) {
+    const DenseMap<unsigned int, unsigned int> &BypassWidths =
+       TLI->getBypassSlowDivWidths();
+    for (Function::iterator I = F.begin(); I != F.end(); I++)
+      EverMadeChange |= bypassSlowDivision(F, I, BypassWidths);
+  }
 
-  // First pass, eliminate blocks that contain only PHI nodes and an
+  // Eliminate blocks that contain only PHI nodes and an
   // unconditional branch.
   EverMadeChange |= EliminateMostlyEmptyBlocks(F);
 
@@ -164,7 +171,7 @@
   bool MadeChange = true;
   while (MadeChange) {
     MadeChange = false;
-    for (Function::iterator I = F.begin(), E = F.end(); I != E; ) {
+    for (Function::iterator I = F.begin(); I != F.end(); ) {
       BasicBlock *BB = I++;
       MadeChange |= OptimizeBlock(*BB);
     }
@@ -187,10 +194,14 @@
           WorkList.insert(*II);
     }
 
-    if (!DisableDeleteDeadBlocks)
-      for (SmallPtrSet<BasicBlock*, 8>::iterator
-             I = WorkList.begin(), E = WorkList.end(); I != E; ++I)
-        DeleteDeadBlock(*I);
+    for (SmallPtrSet<BasicBlock*, 8>::iterator
+           I = WorkList.begin(), E = WorkList.end(); I != E; ++I)
+      DeleteDeadBlock(*I);
+
+    // Merge pairs of basic blocks with unconditional branches, connected by
+    // a single edge.
+    if (EverMadeChange || MadeChange)
+      MadeChange |= EliminateFallThrough(F);
 
     if (MadeChange)
       ModifiedDT = true;
@@ -203,6 +214,40 @@
   return EverMadeChange;
 }
 
+/// EliminateFallThrough - Merge basic blocks which are connected
+/// by a single edge, where one of the basic blocks has a single successor
+/// pointing to the other basic block, which has a single predecessor.
+bool CodeGenPrepare::EliminateFallThrough(Function &F) {
+  bool Changed = false;
+  // Scan all of the blocks in the function, except for the entry block.
+  for (Function::iterator I = ++F.begin(), E = F.end(); I != E; ) {
+    BasicBlock *BB = I++;
+    // If the destination block has a single pred, then this is a trivial
+    // edge, just collapse it.
+    BasicBlock *SinglePred = BB->getSinglePredecessor();
+
+    // Don't merge if BB's address is taken.
+    if (!SinglePred || SinglePred == BB || BB->hasAddressTaken()) continue;
+
+    BranchInst *Term = dyn_cast<BranchInst>(SinglePred->getTerminator());
+    if (Term && !Term->isConditional()) {
+      Changed = true;
+      DEBUG(dbgs() << "To merge:\n"<< *SinglePred << "\n\n\n");
+      // Remember if SinglePred was the entry block of the function.
+      // If so, we will need to move BB back to the entry position.
+      bool isEntry = SinglePred == &SinglePred->getParent()->getEntryBlock();
+      MergeBasicBlockIntoOnlyPred(BB, this);
+
+      if (isEntry && BB != &BB->getParent()->getEntryBlock())
+        BB->moveBefore(&BB->getParent()->getEntryBlock());
+
+      // We have erased a block. Update the iterator.
+      I = BB;
+    }
+  }
+  return Changed;
+}
+
 /// EliminateMostlyEmptyBlocks - eliminate blocks that contain only PHI nodes,
 /// debug info directives, and an unconditional branch.  Passes before isel
 /// (e.g. LSR/loopsimplify) often split edges in ways that are non-optimal for
@@ -577,7 +622,7 @@
     // happens.
     WeakVH IterHandle(CurInstIterator);
 
-    replaceAndRecursivelySimplify(CI, RetVal, TLI ? TLI->getTargetData() : 0,
+    replaceAndRecursivelySimplify(CI, RetVal, TLI ? TLI->getDataLayout() : 0,
                                   TLInfo, ModifiedDT ? 0 : DT);
 
     // If the iterator instruction was recursively deleted, start over at the
@@ -601,8 +646,8 @@
   // From here on out we're working with named functions.
   if (CI->getCalledFunction() == 0) return false;
 
-  // We'll need TargetData from here on out.
-  const TargetData *TD = TLI ? TLI->getTargetData() : 0;
+  // We'll need DataLayout from here on out.
+  const DataLayout *TD = TLI ? TLI->getDataLayout() : 0;
   if (!TD) return false;
 
   // Lower all default uses of _chk calls.  This is very similar
@@ -616,6 +661,7 @@
 /// DupRetToEnableTailCallOpts - Look for opportunities to duplicate return
 /// instructions to the predecessor to enable tail call optimizations. The
 /// case it is currently looking for is:
+/// @code
 /// bb0:
 ///   %tmp0 = tail call i32 @f0()
 ///   br label %return
@@ -628,9 +674,11 @@
 /// return:
 ///   %retval = phi i32 [ %tmp0, %bb0 ], [ %tmp1, %bb1 ], [ %tmp2, %bb2 ]
 ///   ret i32 %retval
+/// @endcode
 ///
 /// =>
 ///
+/// @code
 /// bb0:
 ///   %tmp0 = tail call i32 @f0()
 ///   ret i32 %tmp0
@@ -640,7 +688,7 @@
 /// bb2:
 ///   %tmp2 = tail call i32 @f2()
 ///   ret i32 %tmp2
-///
+/// @endcode
 bool CodeGenPrepare::DupRetToEnableTailCallOpts(ReturnInst *RI) {
   if (!TLI)
     return false;
@@ -666,7 +714,8 @@
   // See llvm::isInTailCallPosition().
   const Function *F = BB->getParent();
   Attributes CallerRetAttr = F->getAttributes().getRetAttributes();
-  if ((CallerRetAttr & Attribute::ZExt) || (CallerRetAttr & Attribute::SExt))
+  if (CallerRetAttr.hasAttribute(Attributes::ZExt) ||
+      CallerRetAttr.hasAttribute(Attributes::SExt))
     return false;
 
   // Make sure there are no instructions between the PHI and return, or that the
@@ -724,7 +773,10 @@
     // Conservatively require the attributes of the call to match those of the
     // return. Ignore noalias because it doesn't affect the call sequence.
     Attributes CalleeRetAttr = CS.getAttributes().getRetAttributes();
-    if ((CalleeRetAttr ^ CallerRetAttr) & ~Attribute::NoAlias)
+    if (AttrBuilder(CalleeRetAttr).
+          removeAttribute(Attributes::NoAlias) !=
+        AttrBuilder(CallerRetAttr).
+          removeAttribute(Attributes::NoAlias))
       continue;
 
     // Make sure the call instruction is followed by an unconditional branch to
@@ -741,7 +793,7 @@
   }
 
   // If we eliminated all predecessors of the block, delete the block now.
-  if (Changed && pred_begin(BB) == pred_end(BB))
+  if (Changed && !BB->hasAddressTaken() && pred_begin(BB) == pred_end(BB))
     BB->eraseFromParent();
 
   return Changed;
@@ -881,7 +933,7 @@
     DEBUG(dbgs() << "CGP: SINKING nonlocal addrmode: " << AddrMode << " for "
                  << *MemoryInst);
     Type *IntPtrTy =
-          TLI->getTargetData()->getIntPtrType(AccessTy->getContext());
+          TLI->getDataLayout()->getIntPtrType(AccessTy->getContext());
 
     Value *Result = 0;
 
@@ -955,7 +1007,7 @@
     WeakVH IterHandle(CurInstIterator);
     BasicBlock *BB = CurInstIterator->getParent();
 
-    RecursivelyDeleteTriviallyDeadInstructions(Repl);
+    RecursivelyDeleteTriviallyDeadInstructions(Repl, TLInfo);
 
     if (IterHandle != CurInstIterator) {
       // If the iterator instruction was recursively deleted, start over at the
@@ -1141,17 +1193,32 @@
 }
 
 
+/// If we have a SelectInst that will likely profit from branch prediction,
+/// turn it into a branch.
 bool CodeGenPrepare::OptimizeSelectInst(SelectInst *SI) {
-  // If we have a SelectInst that will likely profit from branch prediction,
-  // turn it into a branch.
-  if (DisableSelectToBranch || OptSize || !TLI ||
-      !TLI->isPredictableSelectExpensive())
-    return false;
+  bool VectorCond = !SI->getCondition()->getType()->isIntegerTy(1);
 
-  if (!SI->getCondition()->getType()->isIntegerTy(1) ||
-      !isFormingBranchFromSelectProfitable(SI))
+  // Can we convert the 'select' to CF ?
+  if (DisableSelectToBranch || OptSize || !TLI || VectorCond)
     return false;
 
+  TargetLowering::SelectSupportKind SelectKind;
+  if (VectorCond)
+    SelectKind = TargetLowering::VectorMaskSelect;
+  else if (SI->getType()->isVectorTy())
+    SelectKind = TargetLowering::ScalarCondVectorVal;
+  else
+    SelectKind = TargetLowering::ScalarValSelect;
+
+  // Do we have efficient codegen support for this kind of 'selects' ?
+  if (TLI->isSelectSupported(SelectKind)) {
+    // We have efficient codegen support for the select instruction.
+    // Check if it is profitable to keep this 'select'.
+    if (!TLI->isPredictableSelectExpensive() ||
+        !isFormingBranchFromSelectProfitable(SI))
+      return false;
+  }
+
   ModifiedDT = true;
 
   // First, we split the block containing the select into 2 blocks.
@@ -1269,7 +1336,7 @@
   bool MadeChange = false;
 
   CurInstIterator = BB.begin();
-  for (BasicBlock::iterator E = BB.end(); CurInstIterator != E; )
+  while (CurInstIterator != BB.end())
     MadeChange |= OptimizeInst(CurInstIterator++);
 
   return MadeChange;

Modified: llvm/branches/AMDILBackend/lib/Transforms/Scalar/ConstantProp.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Scalar/ConstantProp.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Scalar/ConstantProp.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Scalar/ConstantProp.cpp Tue Jan 15 11:16:16 2013
@@ -24,7 +24,7 @@
 #include "llvm/Constant.h"
 #include "llvm/Instruction.h"
 #include "llvm/Pass.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Target/TargetLibraryInfo.h"
 #include "llvm/Support/InstIterator.h"
 #include "llvm/ADT/Statistic.h"
@@ -67,7 +67,7 @@
       WorkList.insert(&*i);
   }
   bool Changed = false;
-  TargetData *TD = getAnalysisIfAvailable<TargetData>();
+  DataLayout *TD = getAnalysisIfAvailable<DataLayout>();
   TargetLibraryInfo *TLI = &getAnalysis<TargetLibraryInfo>();
 
   while (!WorkList.empty()) {

Modified: llvm/branches/AMDILBackend/lib/Transforms/Scalar/CorrelatedValuePropagation.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Scalar/CorrelatedValuePropagation.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Scalar/CorrelatedValuePropagation.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Scalar/CorrelatedValuePropagation.cpp Tue Jan 15 11:16:16 2013
@@ -235,6 +235,11 @@
       // This case never fires - remove it.
       CI.getCaseSuccessor()->removePredecessor(BB);
       SI->removeCase(CI); // Does not invalidate the iterator.
+
+      // The condition can be modified by removePredecessor's PHI simplification
+      // logic.
+      Cond = SI->getCondition();
+
       ++NumDeadCases;
       Changed = true;
     } else if (State == LazyValueInfo::True) {

Modified: llvm/branches/AMDILBackend/lib/Transforms/Scalar/DCE.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Scalar/DCE.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Scalar/DCE.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Scalar/DCE.cpp Tue Jan 15 11:16:16 2013
@@ -22,6 +22,7 @@
 #include "llvm/Instruction.h"
 #include "llvm/Pass.h"
 #include "llvm/Support/InstIterator.h"
+#include "llvm/Target/TargetLibraryInfo.h"
 #include "llvm/ADT/Statistic.h"
 using namespace llvm;
 
@@ -38,10 +39,11 @@
       initializeDeadInstEliminationPass(*PassRegistry::getPassRegistry());
     }
     virtual bool runOnBasicBlock(BasicBlock &BB) {
+      TargetLibraryInfo *TLI = getAnalysisIfAvailable<TargetLibraryInfo>();
       bool Changed = false;
       for (BasicBlock::iterator DI = BB.begin(); DI != BB.end(); ) {
         Instruction *Inst = DI++;
-        if (isInstructionTriviallyDead(Inst)) {
+        if (isInstructionTriviallyDead(Inst, TLI)) {
           Inst->eraseFromParent();
           Changed = true;
           ++DIEEliminated;
@@ -87,6 +89,8 @@
 INITIALIZE_PASS(DCE, "dce", "Dead Code Elimination", false, false)
 
 bool DCE::runOnFunction(Function &F) {
+  TargetLibraryInfo *TLI = getAnalysisIfAvailable<TargetLibraryInfo>();
+
   // Start out with all of the instructions in the worklist...
   std::vector<Instruction*> WorkList;
   for (inst_iterator i = inst_begin(F), e = inst_end(F); i != e; ++i)
@@ -101,7 +105,7 @@
     Instruction *I = WorkList.back();
     WorkList.pop_back();
 
-    if (isInstructionTriviallyDead(I)) {       // If the instruction is dead.
+    if (isInstructionTriviallyDead(I, TLI)) { // If the instruction is dead.
       // Loop over all of the values that the instruction uses, if there are
       // instructions being used, add them to the worklist, because they might
       // go dead after this one is removed.
@@ -114,13 +118,8 @@
       I->eraseFromParent();
 
       // Remove the instruction from the worklist if it still exists in it.
-      for (std::vector<Instruction*>::iterator WI = WorkList.begin();
-           WI != WorkList.end(); ) {
-        if (*WI == I)
-          WI = WorkList.erase(WI);
-        else
-          ++WI;
-      }
+      WorkList.erase(std::remove(WorkList.begin(), WorkList.end(), I),
+                     WorkList.end());
 
       MadeChange = true;
       ++DCEEliminated;

Modified: llvm/branches/AMDILBackend/lib/Transforms/Scalar/DeadStoreElimination.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Scalar/DeadStoreElimination.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Scalar/DeadStoreElimination.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Scalar/DeadStoreElimination.cpp Tue Jan 15 11:16:16 2013
@@ -29,7 +29,8 @@
 #include "llvm/Analysis/MemoryBuiltins.h"
 #include "llvm/Analysis/MemoryDependenceAnalysis.h"
 #include "llvm/Analysis/ValueTracking.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
+#include "llvm/Target/TargetLibraryInfo.h"
 #include "llvm/Transforms/Utils/Local.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/ADT/SetVector.h"
@@ -45,6 +46,7 @@
     AliasAnalysis *AA;
     MemoryDependenceAnalysis *MD;
     DominatorTree *DT;
+    const TargetLibraryInfo *TLI;
 
     static char ID; // Pass identification, replacement for typeid
     DSE() : FunctionPass(ID), AA(0), MD(0), DT(0) {
@@ -55,6 +57,7 @@
       AA = &getAnalysis<AliasAnalysis>();
       MD = &getAnalysis<MemoryDependenceAnalysis>();
       DT = &getAnalysis<DominatorTree>();
+      TLI = AA->getTargetLibraryInfo();
 
       bool Changed = false;
       for (Function::iterator I = F.begin(), E = F.end(); I != E; ++I)
@@ -106,6 +109,7 @@
 ///
 static void DeleteDeadInstruction(Instruction *I,
                                   MemoryDependenceAnalysis &MD,
+                                  const TargetLibraryInfo *TLI,
                                   SmallSetVector<Value*, 16> *ValueSet = 0) {
   SmallVector<Instruction*, 32> NowDeadInsts;
 
@@ -130,7 +134,7 @@
       if (!Op->use_empty()) continue;
 
       if (Instruction *OpI = dyn_cast<Instruction>(Op))
-        if (isInstructionTriviallyDead(OpI))
+        if (isInstructionTriviallyDead(OpI, TLI))
           NowDeadInsts.push_back(OpI);
     }
 
@@ -143,7 +147,7 @@
 
 /// hasMemoryWrite - Does this instruction write some memory?  This only returns
 /// true for things that we can analyze with other helpers below.
-static bool hasMemoryWrite(Instruction *I) {
+static bool hasMemoryWrite(Instruction *I, const TargetLibraryInfo *TLI) {
   if (isa<StoreInst>(I))
     return true;
   if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
@@ -158,6 +162,26 @@
       return true;
     }
   }
+  if (CallSite CS = I) {
+    if (Function *F = CS.getCalledFunction()) {
+      if (TLI && TLI->has(LibFunc::strcpy) &&
+          F->getName() == TLI->getName(LibFunc::strcpy)) {
+        return true;
+      }
+      if (TLI && TLI->has(LibFunc::strncpy) &&
+          F->getName() == TLI->getName(LibFunc::strncpy)) {
+        return true;
+      }
+      if (TLI && TLI->has(LibFunc::strcat) &&
+          F->getName() == TLI->getName(LibFunc::strcat)) {
+        return true;
+      }
+      if (TLI && TLI->has(LibFunc::strncat) &&
+          F->getName() == TLI->getName(LibFunc::strncat)) {
+        return true;
+      }
+    }
+  }
   return false;
 }
 
@@ -175,7 +199,7 @@
     // If we don't have target data around, an unknown size in Location means
     // that we should use the size of the pointee type.  This isn't valid for
     // memset/memcpy, which writes more than an i8.
-    if (Loc.Size == AliasAnalysis::UnknownSize && AA.getTargetData() == 0)
+    if (Loc.Size == AliasAnalysis::UnknownSize && AA.getDataLayout() == 0)
       return AliasAnalysis::Location();
     return Loc;
   }
@@ -189,7 +213,7 @@
     // If we don't have target data around, an unknown size in Location means
     // that we should use the size of the pointee type.  This isn't valid for
     // init.trampoline, which writes more than an i8.
-    if (AA.getTargetData() == 0) return AliasAnalysis::Location();
+    if (AA.getDataLayout() == 0) return AliasAnalysis::Location();
 
     // FIXME: We don't know the size of the trampoline, so we can't really
     // handle it here.
@@ -205,7 +229,8 @@
 /// instruction if any.
 static AliasAnalysis::Location
 getLocForRead(Instruction *Inst, AliasAnalysis &AA) {
-  assert(hasMemoryWrite(Inst) && "Unknown instruction case");
+  assert(hasMemoryWrite(Inst, AA.getTargetLibraryInfo()) &&
+         "Unknown instruction case");
 
   // The only instructions that both read and write are the mem transfer
   // instructions (memcpy/memmove).
@@ -222,23 +247,29 @@
   if (StoreInst *SI = dyn_cast<StoreInst>(I))
     return SI->isUnordered();
 
-  IntrinsicInst *II = cast<IntrinsicInst>(I);
-  switch (II->getIntrinsicID()) {
-  default: llvm_unreachable("doesn't pass 'hasMemoryWrite' predicate");
-  case Intrinsic::lifetime_end:
-    // Never remove dead lifetime_end's, e.g. because it is followed by a
-    // free.
-    return false;
-  case Intrinsic::init_trampoline:
-    // Always safe to remove init_trampoline.
-    return true;
+  if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
+    switch (II->getIntrinsicID()) {
+    default: llvm_unreachable("doesn't pass 'hasMemoryWrite' predicate");
+    case Intrinsic::lifetime_end:
+      // Never remove dead lifetime_end's, e.g. because it is followed by a
+      // free.
+      return false;
+    case Intrinsic::init_trampoline:
+      // Always safe to remove init_trampoline.
+      return true;
 
-  case Intrinsic::memset:
-  case Intrinsic::memmove:
-  case Intrinsic::memcpy:
-    // Don't remove volatile memory intrinsics.
-    return !cast<MemIntrinsic>(II)->isVolatile();
+    case Intrinsic::memset:
+    case Intrinsic::memmove:
+    case Intrinsic::memcpy:
+      // Don't remove volatile memory intrinsics.
+      return !cast<MemIntrinsic>(II)->isVolatile();
+    }
   }
+
+  if (CallSite CS = I)
+    return CS.getInstruction()->use_empty();
+
+  return false;
 }
 
 
@@ -249,14 +280,19 @@
   if (isa<StoreInst>(I))
     return false;
 
-  IntrinsicInst *II = cast<IntrinsicInst>(I);
-  switch (II->getIntrinsicID()) {
-    default: return false;
-    case Intrinsic::memset:
-    case Intrinsic::memcpy:
-      // Do shorten memory intrinsics.
-      return true;
+  if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
+    switch (II->getIntrinsicID()) {
+      default: return false;
+      case Intrinsic::memset:
+      case Intrinsic::memcpy:
+        // Do shorten memory intrinsics.
+        return true;
+    }
   }
+
+  // Don't shorten libcalls calls for now.
+
+  return false;
 }
 
 /// getStoredPointerOperand - Return the pointer that is being written to.
@@ -266,17 +302,23 @@
   if (MemIntrinsic *MI = dyn_cast<MemIntrinsic>(I))
     return MI->getDest();
 
-  IntrinsicInst *II = cast<IntrinsicInst>(I);
-  switch (II->getIntrinsicID()) {
-  default: llvm_unreachable("Unexpected intrinsic!");
-  case Intrinsic::init_trampoline:
-    return II->getArgOperand(0);
+  if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
+    switch (II->getIntrinsicID()) {
+    default: llvm_unreachable("Unexpected intrinsic!");
+    case Intrinsic::init_trampoline:
+      return II->getArgOperand(0);
+    }
   }
+
+  CallSite CS = I;
+  // All the supported functions so far happen to have dest as their first
+  // argument.
+  return CS.getArgument(0);
 }
 
 static uint64_t getPointerSize(const Value *V, AliasAnalysis &AA) {
   uint64_t Size;
-  if (getObjectSize(V, Size, AA.getTargetData()))
+  if (getObjectSize(V, Size, AA.getDataLayout(), AA.getTargetLibraryInfo()))
     return Size;
   return AliasAnalysis::UnknownSize;
 }
@@ -309,10 +351,10 @@
     // comparison.
     if (Later.Size == AliasAnalysis::UnknownSize ||
         Earlier.Size == AliasAnalysis::UnknownSize) {
-      // If we have no TargetData information around, then the size of the store
+      // If we have no DataLayout information around, then the size of the store
       // is inferrable from the pointee type.  If they are the same type, then
       // we know that the store is safe.
-      if (AA.getTargetData() == 0 &&
+      if (AA.getDataLayout() == 0 &&
           Later.Ptr->getType() == Earlier.Ptr->getType())
         return OverwriteComplete;
 
@@ -328,13 +370,13 @@
   // larger than the earlier one.
   if (Later.Size == AliasAnalysis::UnknownSize ||
       Earlier.Size == AliasAnalysis::UnknownSize ||
-      AA.getTargetData() == 0)
+      AA.getDataLayout() == 0)
     return OverwriteUnknown;
 
   // Check to see if the later store is to the entire object (either a global,
   // an alloca, or a byval argument).  If so, then it clearly overwrites any
   // other store to the same object.
-  const TargetData &TD = *AA.getTargetData();
+  const DataLayout &TD = *AA.getDataLayout();
 
   const Value *UO1 = GetUnderlyingObject(P1, &TD),
               *UO2 = GetUnderlyingObject(P2, &TD);
@@ -378,7 +420,7 @@
   //
   // We have to be careful here as *Off is signed while *.Size is unsigned.
   if (EarlierOff >= LaterOff &&
-      Later.Size > Earlier.Size &&
+      Later.Size >= Earlier.Size &&
       uint64_t(EarlierOff - LaterOff) + Earlier.Size <= Later.Size)
     return OverwriteComplete;
 
@@ -454,13 +496,13 @@
     Instruction *Inst = BBI++;
 
     // Handle 'free' calls specially.
-    if (CallInst *F = isFreeCall(Inst)) {
+    if (CallInst *F = isFreeCall(Inst, TLI)) {
       MadeChange |= HandleFree(F);
       continue;
     }
 
     // If we find something that writes memory, get its memory dependence.
-    if (!hasMemoryWrite(Inst))
+    if (!hasMemoryWrite(Inst, TLI))
       continue;
 
     MemDepResult InstDep = MD->getDependency(Inst);
@@ -483,7 +525,7 @@
           // in case we need it.
           WeakVH NextInst(BBI);
 
-          DeleteDeadInstruction(SI, *MD);
+          DeleteDeadInstruction(SI, *MD, TLI);
 
           if (NextInst == 0)  // Next instruction deleted.
             BBI = BB.begin();
@@ -530,7 +572,7 @@
                 << *DepWrite << "\n  KILLER: " << *Inst << '\n');
 
           // Delete the store and now-dead instructions that feed it.
-          DeleteDeadInstruction(DepWrite, *MD);
+          DeleteDeadInstruction(DepWrite, *MD, TLI);
           ++NumFastStores;
           MadeChange = true;
 
@@ -627,7 +669,7 @@
     MemDepResult Dep = MD->getPointerDependencyFrom(Loc, false, InstPt, BB);
     while (Dep.isDef() || Dep.isClobber()) {
       Instruction *Dependency = Dep.getInst();
-      if (!hasMemoryWrite(Dependency) || !isRemovable(Dependency))
+      if (!hasMemoryWrite(Dependency, TLI) || !isRemovable(Dependency))
         break;
 
       Value *DepPointer =
@@ -640,7 +682,7 @@
       Instruction *Next = llvm::next(BasicBlock::iterator(Dependency));
 
       // DCE instructions only used to calculate that store
-      DeleteDeadInstruction(Dependency, *MD);
+      DeleteDeadInstruction(Dependency, *MD, TLI);
       ++NumFastStores;
       MadeChange = true;
 
@@ -659,6 +701,22 @@
   return MadeChange;
 }
 
+namespace {
+  struct CouldRef {
+    typedef Value *argument_type;
+    const CallSite CS;
+    AliasAnalysis *AA;
+
+    bool operator()(Value *I) {
+      // See if the call site touches the value.
+      AliasAnalysis::ModRefResult A =
+        AA->getModRefInfo(CS, I, getPointerSize(I, *AA));
+
+      return A == AliasAnalysis::ModRef || A == AliasAnalysis::Ref;
+    }
+  };
+}
+
 /// handleEndBlock - Remove dead stores to stack-allocated locations in the
 /// function end block.  Ex:
 /// %A = alloca i32
@@ -680,7 +738,7 @@
 
     // Okay, so these are dead heap objects, but if the pointer never escapes
     // then it's leaked by this function anyways.
-    else if (isAllocLikeFn(I) && !PointerMayBeCaptured(I, true, true))
+    else if (isAllocLikeFn(I, TLI) && !PointerMayBeCaptured(I, true, true))
       DeadStackObjects.insert(I);
   }
 
@@ -696,7 +754,7 @@
     --BBI;
 
     // If we find a store, check to see if it points into a dead stack value.
-    if (hasMemoryWrite(BBI) && isRemovable(BBI)) {
+    if (hasMemoryWrite(BBI, TLI) && isRemovable(BBI)) {
       // See through pointer-to-pointer bitcasts
       SmallVector<Value *, 4> Pointers;
       GetUnderlyingObjects(getStoredPointerOperand(BBI), Pointers);
@@ -724,7 +782,7 @@
               dbgs() << '\n');
 
         // DCE instructions only used to calculate that store.
-        DeleteDeadInstruction(Dead, *MD, &DeadStackObjects);
+        DeleteDeadInstruction(Dead, *MD, TLI, &DeadStackObjects);
         ++NumFastStores;
         MadeChange = true;
         continue;
@@ -732,9 +790,9 @@
     }
 
     // Remove any dead non-memory-mutating instructions.
-    if (isInstructionTriviallyDead(BBI)) {
+    if (isInstructionTriviallyDead(BBI, TLI)) {
       Instruction *Inst = BBI++;
-      DeleteDeadInstruction(Inst, *MD, &DeadStackObjects);
+      DeleteDeadInstruction(Inst, *MD, TLI, &DeadStackObjects);
       ++NumFastOther;
       MadeChange = true;
       continue;
@@ -750,7 +808,7 @@
     if (CallSite CS = cast<Value>(BBI)) {
       // Remove allocation function calls from the list of dead stack objects; 
       // there can't be any references before the definition.
-      if (isAllocLikeFn(BBI))
+      if (isAllocLikeFn(BBI, TLI))
         DeadStackObjects.remove(BBI);
 
       // If this call does not access memory, it can't be loading any of our
@@ -760,20 +818,8 @@
 
       // If the call might load from any of our allocas, then any store above
       // the call is live.
-      SmallVector<Value*, 8> LiveAllocas;
-      for (SmallSetVector<Value*, 16>::iterator I = DeadStackObjects.begin(),
-           E = DeadStackObjects.end(); I != E; ++I) {
-        // See if the call site touches it.
-        AliasAnalysis::ModRefResult A =
-          AA->getModRefInfo(CS, *I, getPointerSize(*I, *AA));
-
-        if (A == AliasAnalysis::ModRef || A == AliasAnalysis::Ref)
-          LiveAllocas.push_back(*I);
-      }
-
-      for (SmallVector<Value*, 8>::iterator I = LiveAllocas.begin(),
-           E = LiveAllocas.end(); I != E; ++I)
-        DeadStackObjects.remove(*I);
+      CouldRef Pred = { CS, AA };
+      DeadStackObjects.remove_if(Pred);
 
       // If all of the allocas were clobbered by the call then we're not going
       // to find anything else to process.
@@ -816,6 +862,20 @@
   return MadeChange;
 }
 
+namespace {
+  struct CouldAlias {
+    typedef Value *argument_type;
+    const AliasAnalysis::Location &LoadedLoc;
+    AliasAnalysis *AA;
+
+    bool operator()(Value *I) {
+      // See if the loaded location could alias the stack location.
+      AliasAnalysis::Location StackLoc(I, getPointerSize(I, *AA));
+      return !AA->isNoAlias(StackLoc, LoadedLoc);
+    }
+  };
+}
+
 /// RemoveAccessedObjects - Check to see if the specified location may alias any
 /// of the stack objects in the DeadStackObjects set.  If so, they become live
 /// because the location is being loaded.
@@ -834,16 +894,7 @@
     return;
   }
 
-  SmallVector<Value*, 16> NowLive;
-  for (SmallSetVector<Value*, 16>::iterator I = DeadStackObjects.begin(),
-       E = DeadStackObjects.end(); I != E; ++I) {
-    // See if the loaded location could alias the stack location.
-    AliasAnalysis::Location StackLoc(*I, getPointerSize(*I, *AA));
-    if (!AA->isNoAlias(StackLoc, LoadedLoc))
-      NowLive.push_back(*I);
-  }
-
-  for (SmallVector<Value*, 16>::iterator I = NowLive.begin(), E = NowLive.end();
-       I != E; ++I)
-    DeadStackObjects.remove(*I);
+  // Remove objects that could alias LoadedLoc.
+  CouldAlias Pred = { LoadedLoc, AA };
+  DeadStackObjects.remove_if(Pred);
 }

Modified: llvm/branches/AMDILBackend/lib/Transforms/Scalar/EarlyCSE.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Scalar/EarlyCSE.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Scalar/EarlyCSE.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Scalar/EarlyCSE.cpp Tue Jan 15 11:16:16 2013
@@ -18,11 +18,12 @@
 #include "llvm/Pass.h"
 #include "llvm/Analysis/Dominators.h"
 #include "llvm/Analysis/InstructionSimplify.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Target/TargetLibraryInfo.h"
 #include "llvm/Transforms/Utils/Local.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/RecyclingAllocator.h"
+#include "llvm/ADT/Hashing.h"
 #include "llvm/ADT/ScopedHashTable.h"
 #include "llvm/ADT/Statistic.h"
 #include <deque>
@@ -90,35 +91,56 @@
 
 unsigned DenseMapInfo<SimpleValue>::getHashValue(SimpleValue Val) {
   Instruction *Inst = Val.Inst;
-
   // Hash in all of the operands as pointers.
-  unsigned Res = 0;
-  for (unsigned i = 0, e = Inst->getNumOperands(); i != e; ++i)
-    Res ^= getHash(Inst->getOperand(i)) << (i & 0xF);
+  if (BinaryOperator* BinOp = dyn_cast<BinaryOperator>(Inst)) {
+    Value *LHS = BinOp->getOperand(0);
+    Value *RHS = BinOp->getOperand(1);
+    if (BinOp->isCommutative() && BinOp->getOperand(0) > BinOp->getOperand(1))
+      std::swap(LHS, RHS);
+
+    if (isa<OverflowingBinaryOperator>(BinOp)) {
+      // Hash the overflow behavior
+      unsigned Overflow =
+        BinOp->hasNoSignedWrap()   * OverflowingBinaryOperator::NoSignedWrap |
+        BinOp->hasNoUnsignedWrap() * OverflowingBinaryOperator::NoUnsignedWrap;
+      return hash_combine(BinOp->getOpcode(), Overflow, LHS, RHS);
+    }
 
-  if (CastInst *CI = dyn_cast<CastInst>(Inst))
-    Res ^= getHash(CI->getType());
-  else if (CmpInst *CI = dyn_cast<CmpInst>(Inst))
-    Res ^= CI->getPredicate();
-  else if (const ExtractValueInst *EVI = dyn_cast<ExtractValueInst>(Inst)) {
-    for (ExtractValueInst::idx_iterator I = EVI->idx_begin(),
-         E = EVI->idx_end(); I != E; ++I)
-      Res ^= *I;
-  } else if (const InsertValueInst *IVI = dyn_cast<InsertValueInst>(Inst)) {
-    for (InsertValueInst::idx_iterator I = IVI->idx_begin(),
-         E = IVI->idx_end(); I != E; ++I)
-      Res ^= *I;
-  } else {
-    // nothing extra to hash in.
-    assert((isa<CallInst>(Inst) ||
-            isa<BinaryOperator>(Inst) || isa<GetElementPtrInst>(Inst) ||
-            isa<SelectInst>(Inst) || isa<ExtractElementInst>(Inst) ||
-            isa<InsertElementInst>(Inst) || isa<ShuffleVectorInst>(Inst)) &&
-           "Invalid/unknown instruction");
+    return hash_combine(BinOp->getOpcode(), LHS, RHS);
   }
 
+  if (CmpInst *CI = dyn_cast<CmpInst>(Inst)) {
+    Value *LHS = CI->getOperand(0);
+    Value *RHS = CI->getOperand(1);
+    CmpInst::Predicate Pred = CI->getPredicate();
+    if (Inst->getOperand(0) > Inst->getOperand(1)) {
+      std::swap(LHS, RHS);
+      Pred = CI->getSwappedPredicate();
+    }
+    return hash_combine(Inst->getOpcode(), Pred, LHS, RHS);
+  }
+
+  if (CastInst *CI = dyn_cast<CastInst>(Inst))
+    return hash_combine(CI->getOpcode(), CI->getType(), CI->getOperand(0));
+
+  if (const ExtractValueInst *EVI = dyn_cast<ExtractValueInst>(Inst))
+    return hash_combine(EVI->getOpcode(), EVI->getOperand(0),
+                        hash_combine_range(EVI->idx_begin(), EVI->idx_end()));
+
+  if (const InsertValueInst *IVI = dyn_cast<InsertValueInst>(Inst))
+    return hash_combine(IVI->getOpcode(), IVI->getOperand(0),
+                        IVI->getOperand(1),
+                        hash_combine_range(IVI->idx_begin(), IVI->idx_end()));
+
+  assert((isa<CallInst>(Inst) || isa<BinaryOperator>(Inst) ||
+          isa<GetElementPtrInst>(Inst) || isa<SelectInst>(Inst) ||
+          isa<ExtractElementInst>(Inst) || isa<InsertElementInst>(Inst) ||
+          isa<ShuffleVectorInst>(Inst)) && "Invalid/unknown instruction");
+
   // Mix in the opcode.
-  return (Res << 1) ^ Inst->getOpcode();
+  return hash_combine(Inst->getOpcode(),
+                      hash_combine_range(Inst->value_op_begin(),
+                                         Inst->value_op_end()));
 }
 
 bool DenseMapInfo<SimpleValue>::isEqual(SimpleValue LHS, SimpleValue RHS) {
@@ -128,7 +150,41 @@
     return LHSI == RHSI;
 
   if (LHSI->getOpcode() != RHSI->getOpcode()) return false;
-  return LHSI->isIdenticalTo(RHSI);
+  if (LHSI->isIdenticalTo(RHSI)) return true;
+
+  // If we're not strictly identical, we still might be a commutable instruction
+  if (BinaryOperator *LHSBinOp = dyn_cast<BinaryOperator>(LHSI)) {
+    if (!LHSBinOp->isCommutative())
+      return false;
+
+    assert(isa<BinaryOperator>(RHSI)
+           && "same opcode, but different instruction type?");
+    BinaryOperator *RHSBinOp = cast<BinaryOperator>(RHSI);
+
+    // Check overflow attributes
+    if (isa<OverflowingBinaryOperator>(LHSBinOp)) {
+      assert(isa<OverflowingBinaryOperator>(RHSBinOp)
+             && "same opcode, but different operator type?");
+      if (LHSBinOp->hasNoUnsignedWrap() != RHSBinOp->hasNoUnsignedWrap() ||
+          LHSBinOp->hasNoSignedWrap() != RHSBinOp->hasNoSignedWrap())
+        return false;
+    }
+
+    // Commuted equality
+    return LHSBinOp->getOperand(0) == RHSBinOp->getOperand(1) &&
+      LHSBinOp->getOperand(1) == RHSBinOp->getOperand(0);
+  }
+  if (CmpInst *LHSCmp = dyn_cast<CmpInst>(LHSI)) {
+    assert(isa<CmpInst>(RHSI)
+           && "same opcode, but different instruction type?");
+    CmpInst *RHSCmp = cast<CmpInst>(RHSI);
+    // Commuted equality
+    return LHSCmp->getOperand(0) == RHSCmp->getOperand(1) &&
+      LHSCmp->getOperand(1) == RHSCmp->getOperand(0) &&
+      LHSCmp->getSwappedPredicate() == RHSCmp->getPredicate();
+  }
+
+  return false;
 }
 
 //===----------------------------------------------------------------------===//
@@ -216,7 +272,7 @@
 /// cases.
 class EarlyCSE : public FunctionPass {
 public:
-  const TargetData *TD;
+  const DataLayout *TD;
   const TargetLibraryInfo *TLI;
   DominatorTree *DT;
   typedef RecyclingAllocator<BumpPtrAllocator,
@@ -274,7 +330,8 @@
         CallScope(*availableCalls) {}
 
    private:
-    NodeScope(const NodeScope&); // DO NOT IMPLEMENT
+    NodeScope(const NodeScope&) LLVM_DELETED_FUNCTION;
+    void operator=(const NodeScope&) LLVM_DELETED_FUNCTION;
 
     ScopedHTType::ScopeTy Scope;
     LoadHTType::ScopeTy LoadScope;
@@ -313,7 +370,8 @@
     void process() { Processed = true; }
 
    private:
-    StackNode(const StackNode&); // DO NOT IMPLEMENT
+    StackNode(const StackNode&) LLVM_DELETED_FUNCTION;
+    void operator=(const StackNode&) LLVM_DELETED_FUNCTION;
 
     // Members.
     unsigned CurrentGeneration;
@@ -374,7 +432,7 @@
     Instruction *Inst = I++;
 
     // Dead instructions should just be removed.
-    if (isInstructionTriviallyDead(Inst)) {
+    if (isInstructionTriviallyDead(Inst, TLI)) {
       DEBUG(dbgs() << "EarlyCSE DCE: " << *Inst << '\n');
       Inst->eraseFromParent();
       Changed = true;
@@ -506,7 +564,7 @@
 bool EarlyCSE::runOnFunction(Function &F) {
   std::deque<StackNode *> nodesToProcess;
 
-  TD = getAnalysisIfAvailable<TargetData>();
+  TD = getAnalysisIfAvailable<DataLayout>();
   TLI = &getAnalysis<TargetLibraryInfo>();
   DT = &getAnalysis<DominatorTree>();
 

Modified: llvm/branches/AMDILBackend/lib/Transforms/Scalar/GVN.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Scalar/GVN.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Scalar/GVN.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Scalar/GVN.cpp Tue Jan 15 11:16:16 2013
@@ -41,7 +41,7 @@
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/PatternMatch.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Target/TargetLibraryInfo.h"
 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
 #include "llvm/Transforms/Utils/SSAUpdater.h"
@@ -271,16 +271,16 @@
   valueNumbering.insert(std::make_pair(V, num));
 }
 
-uint32_t ValueTable::lookup_or_add_call(CallInst* C) {
+uint32_t ValueTable::lookup_or_add_call(CallInst *C) {
   if (AA->doesNotAccessMemory(C)) {
     Expression exp = create_expression(C);
-    uint32_t& e = expressionNumbering[exp];
+    uint32_t &e = expressionNumbering[exp];
     if (!e) e = nextValueNumber++;
     valueNumbering[C] = e;
     return e;
   } else if (AA->onlyReadsMemory(C)) {
     Expression exp = create_expression(C);
-    uint32_t& e = expressionNumbering[exp];
+    uint32_t &e = expressionNumbering[exp];
     if (!e) {
       e = nextValueNumber++;
       valueNumbering[C] = e;
@@ -413,7 +413,7 @@
     case Instruction::LShr:
     case Instruction::AShr:
     case Instruction::And:
-    case Instruction::Or :
+    case Instruction::Or:
     case Instruction::Xor:
     case Instruction::ICmp:
     case Instruction::FCmp:
@@ -503,7 +503,7 @@
     bool NoLoads;
     MemoryDependenceAnalysis *MD;
     DominatorTree *DT;
-    const TargetData *TD;
+    const DataLayout *TD;
     const TargetLibraryInfo *TLI;
 
     ValueTable VN;
@@ -535,7 +535,7 @@
       InstrsToErase.push_back(I);
     }
 
-    const TargetData *getTargetData() const { return TD; }
+    const DataLayout *getDataLayout() const { return TD; }
     DominatorTree &getDominatorTree() const { return *DT; }
     AliasAnalysis *getAliasAnalysis() const { return VN.getAliasAnalysis(); }
     MemoryDependenceAnalysis &getMemDep() const { return *MD; }
@@ -613,8 +613,8 @@
     void verifyRemoved(const Instruction *I) const;
     bool splitCriticalEdges();
     unsigned replaceAllDominatedUsesWith(Value *From, Value *To,
-                                         const BasicBlock *Root);
-    bool propagateEquality(Value *LHS, Value *RHS, const BasicBlock *Root);
+                                         const BasicBlockEdge &Root);
+    bool propagateEquality(Value *LHS, Value *RHS, const BasicBlockEdge &Root);
   };
 
   char GVN::ID = 0;
@@ -632,6 +632,7 @@
 INITIALIZE_AG_DEPENDENCY(AliasAnalysis)
 INITIALIZE_PASS_END(GVN, "gvn", "Global Value Numbering", false, false)
 
+#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
 void GVN::dump(DenseMap<uint32_t, Value*>& d) {
   errs() << "{\n";
   for (DenseMap<uint32_t, Value*>::iterator I = d.begin(),
@@ -641,6 +642,7 @@
   }
   errs() << "}\n";
 }
+#endif
 
 /// IsValueFullyAvailableInBlock - Return true if we can prove that the value
 /// we're analyzing is fully available in the specified block.  As we go, keep
@@ -728,7 +730,7 @@
 /// CoerceAvailableValueToLoadType will succeed.
 static bool CanCoerceMustAliasedValueToLoad(Value *StoredVal,
                                             Type *LoadTy,
-                                            const TargetData &TD) {
+                                            const DataLayout &TD) {
   // If the loaded or stored value is an first class array or struct, don't try
   // to transform them.  We need to be able to bitcast to integer.
   if (LoadTy->isStructTy() || LoadTy->isArrayTy() ||
@@ -744,7 +746,6 @@
   return true;
 }
 
-
 /// CoerceAvailableValueToLoadType - If we saw a store of a value to memory, and
 /// then a load from a must-aliased pointer of a different type, try to coerce
 /// the stored value.  LoadedTy is the type of the load we want to replace and
@@ -754,7 +755,7 @@
 static Value *CoerceAvailableValueToLoadType(Value *StoredVal,
                                              Type *LoadedTy,
                                              Instruction *InsertPt,
-                                             const TargetData &TD) {
+                                             const DataLayout &TD) {
   if (!CanCoerceMustAliasedValueToLoad(StoredVal, LoadedTy, TD))
     return 0;
 
@@ -767,24 +768,25 @@
   // If the store and reload are the same size, we can always reuse it.
   if (StoreSize == LoadSize) {
     // Pointer to Pointer -> use bitcast.
-    if (StoredValTy->isPointerTy() && LoadedTy->isPointerTy())
+    if (StoredValTy->getScalarType()->isPointerTy() &&
+        LoadedTy->getScalarType()->isPointerTy())
       return new BitCastInst(StoredVal, LoadedTy, "", InsertPt);
 
     // Convert source pointers to integers, which can be bitcast.
-    if (StoredValTy->isPointerTy()) {
-      StoredValTy = TD.getIntPtrType(StoredValTy->getContext());
+    if (StoredValTy->getScalarType()->isPointerTy()) {
+      StoredValTy = TD.getIntPtrType(StoredValTy);
       StoredVal = new PtrToIntInst(StoredVal, StoredValTy, "", InsertPt);
     }
 
     Type *TypeToCastTo = LoadedTy;
-    if (TypeToCastTo->isPointerTy())
-      TypeToCastTo = TD.getIntPtrType(StoredValTy->getContext());
+    if (TypeToCastTo->getScalarType()->isPointerTy())
+      TypeToCastTo = TD.getIntPtrType(TypeToCastTo);
 
     if (StoredValTy != TypeToCastTo)
       StoredVal = new BitCastInst(StoredVal, TypeToCastTo, "", InsertPt);
 
     // Cast to pointer if the load needs a pointer type.
-    if (LoadedTy->isPointerTy())
+    if (LoadedTy->getScalarType()->isPointerTy())
       StoredVal = new IntToPtrInst(StoredVal, LoadedTy, "", InsertPt);
 
     return StoredVal;
@@ -796,8 +798,8 @@
   assert(StoreSize >= LoadSize && "CanCoerceMustAliasedValueToLoad fail");
 
   // Convert source pointers to integers, which can be manipulated.
-  if (StoredValTy->isPointerTy()) {
-    StoredValTy = TD.getIntPtrType(StoredValTy->getContext());
+  if (StoredValTy->getScalarType()->isPointerTy()) {
+    StoredValTy = TD.getIntPtrType(StoredValTy);
     StoredVal = new PtrToIntInst(StoredVal, StoredValTy, "", InsertPt);
   }
 
@@ -822,7 +824,7 @@
     return StoredVal;
 
   // If the result is a pointer, inttoptr.
-  if (LoadedTy->isPointerTy())
+  if (LoadedTy->getScalarType()->isPointerTy())
     return new IntToPtrInst(StoredVal, LoadedTy, "inttoptr", InsertPt);
 
   // Otherwise, bitcast.
@@ -840,7 +842,7 @@
 static int AnalyzeLoadFromClobberingWrite(Type *LoadTy, Value *LoadPtr,
                                           Value *WritePtr,
                                           uint64_t WriteSizeInBits,
-                                          const TargetData &TD) {
+                                          const DataLayout &TD) {
   // If the loaded or stored value is a first class array or struct, don't try
   // to transform them.  We need to be able to bitcast to integer.
   if (LoadTy->isStructTy() || LoadTy->isArrayTy())
@@ -913,7 +915,7 @@
 /// memdep query of a load that ends up being a clobbering store.
 static int AnalyzeLoadFromClobberingStore(Type *LoadTy, Value *LoadPtr,
                                           StoreInst *DepSI,
-                                          const TargetData &TD) {
+                                          const DataLayout &TD) {
   // Cannot handle reading from store of first-class aggregate yet.
   if (DepSI->getValueOperand()->getType()->isStructTy() ||
       DepSI->getValueOperand()->getType()->isArrayTy())
@@ -929,7 +931,7 @@
 /// memdep query of a load that ends up being clobbered by another load.  See if
 /// the other load can feed into the second load.
 static int AnalyzeLoadFromClobberingLoad(Type *LoadTy, Value *LoadPtr,
-                                         LoadInst *DepLI, const TargetData &TD){
+                                         LoadInst *DepLI, const DataLayout &TD){
   // Cannot handle reading from store of first-class aggregate yet.
   if (DepLI->getType()->isStructTy() || DepLI->getType()->isArrayTy())
     return -1;
@@ -957,7 +959,7 @@
 
 static int AnalyzeLoadFromClobberingMemInst(Type *LoadTy, Value *LoadPtr,
                                             MemIntrinsic *MI,
-                                            const TargetData &TD) {
+                                            const DataLayout &TD) {
   // If the mem operation is a non-constant size, we can't handle it.
   ConstantInt *SizeCst = dyn_cast<ConstantInt>(MI->getLength());
   if (SizeCst == 0) return -1;
@@ -1007,7 +1009,7 @@
 /// before we give up.
 static Value *GetStoreValueForLoad(Value *SrcVal, unsigned Offset,
                                    Type *LoadTy,
-                                   Instruction *InsertPt, const TargetData &TD){
+                                   Instruction *InsertPt, const DataLayout &TD){
   LLVMContext &Ctx = SrcVal->getType()->getContext();
 
   uint64_t StoreSize = (TD.getTypeSizeInBits(SrcVal->getType()) + 7) / 8;
@@ -1017,8 +1019,9 @@
 
   // Compute which bits of the stored value are being used by the load.  Convert
   // to an integer type to start with.
-  if (SrcVal->getType()->isPointerTy())
-    SrcVal = Builder.CreatePtrToInt(SrcVal, TD.getIntPtrType(Ctx));
+  if (SrcVal->getType()->getScalarType()->isPointerTy())
+    SrcVal = Builder.CreatePtrToInt(SrcVal,
+        TD.getIntPtrType(SrcVal->getType()));
   if (!SrcVal->getType()->isIntegerTy())
     SrcVal = Builder.CreateBitCast(SrcVal, IntegerType::get(Ctx, StoreSize*8));
 
@@ -1046,7 +1049,7 @@
 static Value *GetLoadValueForLoad(LoadInst *SrcVal, unsigned Offset,
                                   Type *LoadTy, Instruction *InsertPt,
                                   GVN &gvn) {
-  const TargetData &TD = *gvn.getTargetData();
+  const DataLayout &TD = *gvn.getDataLayout();
   // If Offset+LoadTy exceeds the size of SrcVal, then we must be wanting to
   // widen SrcVal out to a larger load.
   unsigned SrcValSize = TD.getTypeStoreSize(SrcVal->getType());
@@ -1105,7 +1108,7 @@
 /// memdep query of a load that ends up being a clobbering mem intrinsic.
 static Value *GetMemInstValueForLoad(MemIntrinsic *SrcInst, unsigned Offset,
                                      Type *LoadTy, Instruction *InsertPt,
-                                     const TargetData &TD){
+                                     const DataLayout &TD){
   LLVMContext &Ctx = LoadTy->getContext();
   uint64_t LoadSize = TD.getTypeSizeInBits(LoadTy)/8;
 
@@ -1229,7 +1232,7 @@
     if (isSimpleValue()) {
       Res = getSimpleValue();
       if (Res->getType() != LoadTy) {
-        const TargetData *TD = gvn.getTargetData();
+        const DataLayout *TD = gvn.getDataLayout();
         assert(TD && "Need target data to handle type mismatch case");
         Res = GetStoreValueForLoad(Res, Offset, LoadTy, BB->getTerminator(),
                                    *TD);
@@ -1251,7 +1254,7 @@
                      << *Res << '\n' << "\n\n\n");
       }
     } else {
-      const TargetData *TD = gvn.getTargetData();
+      const DataLayout *TD = gvn.getDataLayout();
       assert(TD && "Need target data to handle type mismatch case");
       Res = GetMemInstValueForLoad(getMemIntrinValue(), Offset,
                                    LoadTy, BB->getTerminator(), *TD);
@@ -1299,7 +1302,7 @@
   Value *V = SSAUpdate.GetValueInMiddleOfBlock(LI->getParent());
 
   // If new PHI nodes were created, notify alias analysis.
-  if (V->getType()->isPointerTy()) {
+  if (V->getType()->getScalarType()->isPointerTy()) {
     AliasAnalysis *AA = gvn.getAliasAnalysis();
 
     for (unsigned i = 0, e = NewPHIs.size(); i != e; ++i)
@@ -1436,7 +1439,7 @@
     Instruction *DepInst = DepInfo.getInst();
 
     // Loading the allocation -> undef.
-    if (isa<AllocaInst>(DepInst) || isMallocLikeFn(DepInst) ||
+    if (isa<AllocaInst>(DepInst) || isMallocLikeFn(DepInst, TLI) ||
         // Loading immediately after lifetime begin -> undef.
         isLifetimeStart(DepInst)) {
       ValuesPerBlock.push_back(AvailableValueInBlock::get(DepBB,
@@ -1496,7 +1499,7 @@
 
     if (isa<PHINode>(V))
       V->takeName(LI);
-    if (V->getType()->isPointerTy())
+    if (V->getType()->getScalarType()->isPointerTy())
       MD->invalidateCachedPointerInfo(V);
     markInstructionForDeletion(LI);
     ++NumGVNLoad;
@@ -1728,7 +1731,7 @@
   LI->replaceAllUsesWith(V);
   if (isa<PHINode>(V))
     V->takeName(LI);
-  if (V->getType()->isPointerTy())
+  if (V->getType()->getScalarType()->isPointerTy())
     MD->invalidateCachedPointerInfo(V);
   markInstructionForDeletion(LI);
   ++NumPRELoad;
@@ -1855,7 +1858,7 @@
 
       // Replace the load!
       L->replaceAllUsesWith(AvailVal);
-      if (AvailVal->getType()->isPointerTy())
+      if (AvailVal->getType()->getScalarType()->isPointerTy())
         MD->invalidateCachedPointerInfo(AvailVal);
       markInstructionForDeletion(L);
       ++NumGVNLoad;
@@ -1912,7 +1915,7 @@
 
     // Remove it!
     L->replaceAllUsesWith(StoredVal);
-    if (StoredVal->getType()->isPointerTy())
+    if (StoredVal->getType()->getScalarType()->isPointerTy())
       MD->invalidateCachedPointerInfo(StoredVal);
     markInstructionForDeletion(L);
     ++NumGVNLoad;
@@ -1941,7 +1944,7 @@
 
     // Remove it!
     patchAndReplaceAllUsesWith(AvailableVal, L);
-    if (DepLI->getType()->isPointerTy())
+    if (DepLI->getType()->getScalarType()->isPointerTy())
       MD->invalidateCachedPointerInfo(DepLI);
     markInstructionForDeletion(L);
     ++NumGVNLoad;
@@ -1951,7 +1954,7 @@
   // If this load really doesn't depend on anything, then we must be loading an
   // undef value.  This can happen when loading for a fresh allocation with no
   // intervening stores, for example.
-  if (isa<AllocaInst>(DepInst) || isMallocLikeFn(DepInst)) {
+  if (isa<AllocaInst>(DepInst) || isMallocLikeFn(DepInst, TLI)) {
     L->replaceAllUsesWith(UndefValue::get(L->getType()));
     markInstructionForDeletion(L);
     ++NumGVNLoad;
@@ -2004,22 +2007,13 @@
 /// use is dominated by the given basic block.  Returns the number of uses that
 /// were replaced.
 unsigned GVN::replaceAllDominatedUsesWith(Value *From, Value *To,
-                                          const BasicBlock *Root) {
+                                          const BasicBlockEdge &Root) {
   unsigned Count = 0;
   for (Value::use_iterator UI = From->use_begin(), UE = From->use_end();
        UI != UE; ) {
     Use &U = (UI++).getUse();
 
-    // If From occurs as a phi node operand then the use implicitly lives in the
-    // corresponding incoming block.  Otherwise it is the block containing the
-    // user that must be dominated by Root.
-    BasicBlock *UsingBlock;
-    if (PHINode *PN = dyn_cast<PHINode>(U.getUser()))
-      UsingBlock = PN->getIncomingBlock(U);
-    else
-      UsingBlock = cast<Instruction>(U.getUser())->getParent();
-
-    if (DT->dominates(Root, UsingBlock)) {
+    if (DT->dominates(Root, U)) {
       U.set(To);
       ++Count;
     }
@@ -2027,13 +2021,34 @@
   return Count;
 }
 
+/// isOnlyReachableViaThisEdge - There is an edge from 'Src' to 'Dst'.  Return
+/// true if every path from the entry block to 'Dst' passes via this edge.  In
+/// particular 'Dst' must not be reachable via another edge from 'Src'.
+static bool isOnlyReachableViaThisEdge(const BasicBlockEdge &E,
+                                       DominatorTree *DT) {
+  // While in theory it is interesting to consider the case in which Dst has
+  // more than one predecessor, because Dst might be part of a loop which is
+  // only reachable from Src, in practice it is pointless since at the time
+  // GVN runs all such loops have preheaders, which means that Dst will have
+  // been changed to have only one predecessor, namely Src.
+  const BasicBlock *Pred = E.getEnd()->getSinglePredecessor();
+  const BasicBlock *Src = E.getStart();
+  assert((!Pred || Pred == Src) && "No edge between these basic blocks!");
+  (void)Src;
+  return Pred != 0;
+}
+
 /// propagateEquality - The given values are known to be equal in every block
 /// dominated by 'Root'.  Exploit this, for example by replacing 'LHS' with
 /// 'RHS' everywhere in the scope.  Returns whether a change was made.
-bool GVN::propagateEquality(Value *LHS, Value *RHS, const BasicBlock *Root) {
+bool GVN::propagateEquality(Value *LHS, Value *RHS,
+                            const BasicBlockEdge &Root) {
   SmallVector<std::pair<Value*, Value*>, 4> Worklist;
   Worklist.push_back(std::make_pair(LHS, RHS));
   bool Changed = false;
+  // For speed, compute a conservative fast approximation to
+  // DT->dominates(Root, Root.getEnd());
+  bool RootDominatesEnd = isOnlyReachableViaThisEdge(Root, DT);
 
   while (!Worklist.empty()) {
     std::pair<Value*, Value*> Item = Worklist.pop_back_val();
@@ -2065,9 +2080,6 @@
         LVN = RVN;
       }
     }
-    assert((!isa<Instruction>(RHS) ||
-            DT->properlyDominates(cast<Instruction>(RHS)->getParent(), Root)) &&
-           "Instruction doesn't dominate scope!");
 
     // If value numbering later sees that an instruction in the scope is equal
     // to 'LHS' then ensure it will be turned into 'RHS'.  In order to preserve
@@ -2076,8 +2088,10 @@
     // if RHS is an instruction (if an instruction in the scope is morphed into
     // LHS then it will be turned into RHS by the next GVN iteration anyway, so
     // using the leader table is about compiling faster, not optimizing better).
-    if (!isa<Instruction>(RHS))
-      addToLeaderTable(LVN, RHS, Root);
+    // The leader table only tracks basic blocks, not edges. Only add to if we
+    // have the simple case where the edge dominates the end.
+    if (RootDominatesEnd && !isa<Instruction>(RHS))
+      addToLeaderTable(LVN, RHS, Root.getEnd());
 
     // Replace all occurrences of 'LHS' with 'RHS' everywhere in the scope.  As
     // LHS always has at least one use that is not dominated by Root, this will
@@ -2136,7 +2150,7 @@
       // If the number we were assigned was brand new then there is no point in
       // looking for an instruction realizing it: there cannot be one!
       if (Num < NextNum) {
-        Value *NotCmp = findLeader(Root, Num);
+        Value *NotCmp = findLeader(Root.getEnd(), Num);
         if (NotCmp && isa<Instruction>(NotCmp)) {
           unsigned NumReplacements =
             replaceAllDominatedUsesWith(NotCmp, NotVal, Root);
@@ -2146,7 +2160,10 @@
       }
       // Ensure that any instruction in scope that gets the "A < B" value number
       // is replaced with false.
-      addToLeaderTable(Num, NotVal, Root);
+      // The leader table only tracks basic blocks, not edges. Only add to if we
+      // have the simple case where the edge dominates the end.
+      if (RootDominatesEnd)
+        addToLeaderTable(Num, NotVal, Root.getEnd());
 
       continue;
     }
@@ -2155,22 +2172,6 @@
   return Changed;
 }
 
-/// isOnlyReachableViaThisEdge - There is an edge from 'Src' to 'Dst'.  Return
-/// true if every path from the entry block to 'Dst' passes via this edge.  In
-/// particular 'Dst' must not be reachable via another edge from 'Src'.
-static bool isOnlyReachableViaThisEdge(BasicBlock *Src, BasicBlock *Dst,
-                                       DominatorTree *DT) {
-  // While in theory it is interesting to consider the case in which Dst has
-  // more than one predecessor, because Dst might be part of a loop which is
-  // only reachable from Src, in practice it is pointless since at the time
-  // GVN runs all such loops have preheaders, which means that Dst will have
-  // been changed to have only one predecessor, namely Src.
-  BasicBlock *Pred = Dst->getSinglePredecessor();
-  assert((!Pred || Pred == Src) && "No edge between these basic blocks!");
-  (void)Src;
-  return Pred != 0;
-}
-
 /// processInstruction - When calculating availability, handle an instruction
 /// by inserting it into the appropriate sets
 bool GVN::processInstruction(Instruction *I) {
@@ -2184,7 +2185,7 @@
   // "%z = and i32 %x, %y" becomes "%z = and i32 %x, %x" which we now simplify.
   if (Value *V = SimplifyInstruction(I, TD, TLI, DT)) {
     I->replaceAllUsesWith(V);
-    if (MD && V->getType()->isPointerTy())
+    if (MD && V->getType()->getScalarType()->isPointerTy())
       MD->invalidateCachedPointerInfo(V);
     markInstructionForDeletion(I);
     ++NumGVNSimpl;
@@ -2210,18 +2211,20 @@
 
     BasicBlock *TrueSucc = BI->getSuccessor(0);
     BasicBlock *FalseSucc = BI->getSuccessor(1);
+    // Avoid multiple edges early.
+    if (TrueSucc == FalseSucc)
+      return false;
+
     BasicBlock *Parent = BI->getParent();
     bool Changed = false;
 
-    if (isOnlyReachableViaThisEdge(Parent, TrueSucc, DT))
-      Changed |= propagateEquality(BranchCond,
-                                   ConstantInt::getTrue(TrueSucc->getContext()),
-                                   TrueSucc);
-
-    if (isOnlyReachableViaThisEdge(Parent, FalseSucc, DT))
-      Changed |= propagateEquality(BranchCond,
-                                   ConstantInt::getFalse(FalseSucc->getContext()),
-                                   FalseSucc);
+    Value *TrueVal = ConstantInt::getTrue(TrueSucc->getContext());
+    BasicBlockEdge TrueE(Parent, TrueSucc);
+    Changed |= propagateEquality(BranchCond, TrueVal, TrueE);
+
+    Value *FalseVal = ConstantInt::getFalse(FalseSucc->getContext());
+    BasicBlockEdge FalseE(Parent, FalseSucc);
+    Changed |= propagateEquality(BranchCond, FalseVal, FalseE);
 
     return Changed;
   }
@@ -2231,11 +2234,20 @@
     Value *SwitchCond = SI->getCondition();
     BasicBlock *Parent = SI->getParent();
     bool Changed = false;
+
+    // Remember how many outgoing edges there are to every successor.
+    SmallDenseMap<BasicBlock *, unsigned, 16> SwitchEdges;
+    for (unsigned i = 0, n = SI->getNumSuccessors(); i != n; ++i)
+      ++SwitchEdges[SI->getSuccessor(i)];
+
     for (SwitchInst::CaseIt i = SI->case_begin(), e = SI->case_end();
          i != e; ++i) {
       BasicBlock *Dst = i.getCaseSuccessor();
-      if (isOnlyReachableViaThisEdge(Parent, Dst, DT))
-        Changed |= propagateEquality(SwitchCond, i.getCaseValue(), Dst);
+      // If there is only a single edge, propagate the case value into it.
+      if (SwitchEdges.lookup(Dst) == 1) {
+        BasicBlockEdge E(Parent, Dst);
+        Changed |= propagateEquality(SwitchCond, i.getCaseValue(), E);
+      }
     }
     return Changed;
   }
@@ -2273,7 +2285,7 @@
 
   // Remove it!
   patchAndReplaceAllUsesWith(repl, I);
-  if (MD && repl->getType()->isPointerTy())
+  if (MD && repl->getType()->getScalarType()->isPointerTy())
     MD->invalidateCachedPointerInfo(repl);
   markInstructionForDeletion(I);
   return true;
@@ -2284,7 +2296,7 @@
   if (!NoLoads)
     MD = &getAnalysis<MemoryDependenceAnalysis>();
   DT = &getAnalysis<DominatorTree>();
-  TD = getAnalysisIfAvailable<TargetData>();
+  TD = getAnalysisIfAvailable<DataLayout>();
   TLI = &getAnalysis<TargetLibraryInfo>();
   VN.setAliasAnalysis(&getAnalysis<AliasAnalysis>());
   VN.setMemDep(MD);
@@ -2521,7 +2533,7 @@
       addToLeaderTable(ValNo, Phi, CurrentBlock);
       Phi->setDebugLoc(CurInst->getDebugLoc());
       CurInst->replaceAllUsesWith(Phi);
-      if (Phi->getType()->isPointerTy()) {
+      if (Phi->getType()->getScalarType()->isPointerTy()) {
         // Because we have added a PHI-use of the pointer value, it has now
         // "escaped" from alias analysis' perspective.  We need to inform
         // AA of this.

Modified: llvm/branches/AMDILBackend/lib/Transforms/Scalar/GlobalMerge.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Scalar/GlobalMerge.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Scalar/GlobalMerge.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Scalar/GlobalMerge.cpp Tue Jan 15 11:16:16 2013
@@ -62,7 +62,7 @@
 #include "llvm/Intrinsics.h"
 #include "llvm/Module.h"
 #include "llvm/Pass.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Target/TargetLowering.h"
 #include "llvm/Target/TargetLoweringObjectFile.h"
 #include "llvm/ADT/Statistic.h"
@@ -98,9 +98,9 @@
     }
 
     struct GlobalCmp {
-      const TargetData *TD;
+      const DataLayout *TD;
 
-      GlobalCmp(const TargetData *td) : TD(td) { }
+      GlobalCmp(const DataLayout *td) : TD(td) { }
 
       bool operator()(const GlobalVariable *GV1, const GlobalVariable *GV2) {
         Type *Ty1 = cast<PointerType>(GV1->getType())->getElementType();
@@ -119,7 +119,7 @@
 
 bool GlobalMerge::doMerge(SmallVectorImpl<GlobalVariable*> &Globals,
                              Module &M, bool isConst) const {
-  const TargetData *TD = TLI->getTargetData();
+  const DataLayout *TD = TLI->getDataLayout();
 
   // FIXME: Infer the maximum possible offset depending on the actual users
   // (these max offsets are different for the users inside Thumb or ARM
@@ -170,7 +170,7 @@
 
 bool GlobalMerge::doInitialization(Module &M) {
   SmallVector<GlobalVariable*, 16> Globals, ConstGlobals, BSSGlobals;
-  const TargetData *TD = TLI->getTargetData();
+  const DataLayout *TD = TLI->getDataLayout();
   unsigned MaxOffset = TLI->getMaximalGlobalOffset();
   bool Changed = false;
 

Modified: llvm/branches/AMDILBackend/lib/Transforms/Scalar/IndVarSimplify.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Scalar/IndVarSimplify.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Scalar/IndVarSimplify.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Scalar/IndVarSimplify.cpp Tue Jan 15 11:16:16 2013
@@ -43,7 +43,8 @@
 #include "llvm/Transforms/Utils/Local.h"
 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
 #include "llvm/Transforms/Utils/SimplifyIndVar.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
+#include "llvm/Target/TargetLibraryInfo.h"
 #include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/Statistic.h"
@@ -67,7 +68,8 @@
     LoopInfo        *LI;
     ScalarEvolution *SE;
     DominatorTree   *DT;
-    TargetData      *TD;
+    DataLayout      *TD;
+    TargetLibraryInfo *TLI;
 
     SmallVector<WeakVH, 16> DeadInsts;
     bool Changed;
@@ -218,8 +220,6 @@
 /// ConvertToSInt - Convert APF to an integer, if possible.
 static bool ConvertToSInt(const APFloat &APF, int64_t &IntVal) {
   bool isExact = false;
-  if (&APF.getSemantics() == &APFloat::PPCDoubleDouble)
-    return false;
   // See if we can convert this to an int64_t
   uint64_t UIntVal;
   if (APF.convertToInteger(&UIntVal, 64, true, APFloat::rmTowardZero,
@@ -414,11 +414,11 @@
   // new comparison.
   NewCompare->takeName(Compare);
   Compare->replaceAllUsesWith(NewCompare);
-  RecursivelyDeleteTriviallyDeadInstructions(Compare);
+  RecursivelyDeleteTriviallyDeadInstructions(Compare, TLI);
 
   // Delete the old floating point increment.
   Incr->replaceAllUsesWith(UndefValue::get(Incr->getType()));
-  RecursivelyDeleteTriviallyDeadInstructions(Incr);
+  RecursivelyDeleteTriviallyDeadInstructions(Incr, TLI);
 
   // If the FP induction variable still has uses, this is because something else
   // in the loop uses its value.  In order to canonicalize the induction
@@ -431,7 +431,7 @@
     Value *Conv = new SIToFPInst(NewPHI, PN->getType(), "indvar.conv",
                                  PN->getParent()->getFirstInsertionPt());
     PN->replaceAllUsesWith(Conv);
-    RecursivelyDeleteTriviallyDeadInstructions(PN);
+    RecursivelyDeleteTriviallyDeadInstructions(PN, TLI);
   }
   Changed = true;
 }
@@ -549,15 +549,17 @@
 
         PN->setIncomingValue(i, ExitVal);
 
-        // If this instruction is dead now, delete it.
-        RecursivelyDeleteTriviallyDeadInstructions(Inst);
+        // If this instruction is dead now, delete it. Don't do it now to avoid
+        // invalidating iterators.
+        if (isInstructionTriviallyDead(Inst, TLI))
+          DeadInsts.push_back(Inst);
 
         if (NumPreds == 1) {
           // Completely replace a single-pred PHI. This is safe, because the
           // NewVal won't be variant in the loop, so we don't need an LCSSA phi
           // node anymore.
           PN->replaceAllUsesWith(ExitVal);
-          RecursivelyDeleteTriviallyDeadInstructions(PN);
+          PN->eraseFromParent();
         }
       }
       if (NumPreds != 1) {
@@ -595,13 +597,13 @@
 
   class WideIVVisitor : public IVVisitor {
     ScalarEvolution *SE;
-    const TargetData *TD;
+    const DataLayout *TD;
 
   public:
     WideIVInfo WI;
 
     WideIVVisitor(PHINode *NarrowIV, ScalarEvolution *SCEV,
-                  const TargetData *TData) :
+                  const DataLayout *TData) :
       SE(SCEV), TD(TData) { WI.NarrowIV = NarrowIV; }
 
     // Implement the interface used by simplifyUsersOfIV.
@@ -1259,8 +1261,13 @@
   if (!Phi)
     return true;
 
+  // Do LFTR if PHI node is defined in the loop, but is *not* a counter.
+  int Idx = Phi->getBasicBlockIndex(L->getLoopLatch());
+  if (Idx < 0)
+    return true;
+
   // Do LFTR if the exit condition's IV is *not* a simple counter.
-  Value *IncV = Phi->getIncomingValueForBlock(L->getLoopLatch());
+  Value *IncV = Phi->getIncomingValue(Idx);
   return Phi != getLoopPhiForCounter(IncV, L, DT);
 }
 
@@ -1339,7 +1346,7 @@
 /// could at least handle constant BECounts.
 static PHINode *
 FindLoopCounter(Loop *L, const SCEV *BECount,
-                ScalarEvolution *SE, DominatorTree *DT, const TargetData *TD) {
+                ScalarEvolution *SE, DominatorTree *DT, const DataLayout *TD) {
   uint64_t BCWidth = SE->getTypeSizeInBits(BECount->getType());
 
   Value *Cond =
@@ -1696,7 +1703,8 @@
   LI = &getAnalysis<LoopInfo>();
   SE = &getAnalysis<ScalarEvolution>();
   DT = &getAnalysis<DominatorTree>();
-  TD = getAnalysisIfAvailable<TargetData>();
+  TD = getAnalysisIfAvailable<DataLayout>();
+  TLI = getAnalysisIfAvailable<TargetLibraryInfo>();
 
   DeadInsts.clear();
   Changed = false;
@@ -1763,7 +1771,7 @@
   while (!DeadInsts.empty())
     if (Instruction *Inst =
           dyn_cast_or_null<Instruction>(&*DeadInsts.pop_back_val()))
-      RecursivelyDeleteTriviallyDeadInstructions(Inst);
+      RecursivelyDeleteTriviallyDeadInstructions(Inst, TLI);
 
   // The Rewriter may not be used from this point on.
 
@@ -1772,7 +1780,7 @@
   SinkUnusedInvariants(L);
 
   // Clean up dead instructions.
-  Changed |= DeleteDeadPHIs(L->getHeader());
+  Changed |= DeleteDeadPHIs(L->getHeader(), TLI);
   // Check a post-condition.
   assert(L->isLCSSAForm(*DT) &&
          "Indvars did not leave the loop in lcssa form!");

Modified: llvm/branches/AMDILBackend/lib/Transforms/Scalar/JumpThreading.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Scalar/JumpThreading.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Scalar/JumpThreading.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Scalar/JumpThreading.cpp Tue Jan 15 11:16:16 2013
@@ -23,7 +23,7 @@
 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
 #include "llvm/Transforms/Utils/Local.h"
 #include "llvm/Transforms/Utils/SSAUpdater.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Target/TargetLibraryInfo.h"
 #include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/DenseSet.h"
@@ -75,7 +75,7 @@
   /// revectored to the false side of the second if.
   ///
   class JumpThreading : public FunctionPass {
-    TargetData *TD;
+    DataLayout *TD;
     TargetLibraryInfo *TLI;
     LazyValueInfo *LVI;
 #ifdef NDEBUG
@@ -147,7 +147,7 @@
 ///
 bool JumpThreading::runOnFunction(Function &F) {
   DEBUG(dbgs() << "Jump threading on function '" << F.getName() << "'\n");
-  TD = getAnalysisIfAvailable<TargetData>();
+  TD = getAnalysisIfAvailable<DataLayout>();
   TLI = &getAnalysis<TargetLibraryInfo>();
   LVI = &getAnalysis<LazyValueInfo>();
 
@@ -1455,7 +1455,7 @@
   // At this point, the IR is fully up to date and consistent.  Do a quick scan
   // over the new instructions and zap any that are constants or dead.  This
   // frequently happens because of phi translation.
-  SimplifyInstructionsInBlock(NewBB, TD);
+  SimplifyInstructionsInBlock(NewBB, TD, TLI);
 
   // Threaded an edge!
   ++NumThreads;

Modified: llvm/branches/AMDILBackend/lib/Transforms/Scalar/LICM.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Scalar/LICM.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Scalar/LICM.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Scalar/LICM.cpp Tue Jan 15 11:16:16 2013
@@ -46,7 +46,7 @@
 #include "llvm/Analysis/ValueTracking.h"
 #include "llvm/Transforms/Utils/Local.h"
 #include "llvm/Transforms/Utils/SSAUpdater.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Target/TargetLibraryInfo.h"
 #include "llvm/Support/CFG.h"
 #include "llvm/Support/CommandLine.h"
@@ -100,7 +100,7 @@
     LoopInfo      *LI;       // Current LoopInfo
     DominatorTree *DT;       // Dominator Tree for the current Loop.
 
-    TargetData *TD;          // TargetData for constant folding.
+    DataLayout *TD;          // DataLayout for constant folding.
     TargetLibraryInfo *TLI;  // TargetLibraryInfo for constant folding.
 
     // State that is updated as we process loops.
@@ -108,6 +108,9 @@
     BasicBlock *Preheader;   // The preheader block of the current loop...
     Loop *CurLoop;           // The current loop we are working on...
     AliasSetTracker *CurAST; // AliasSet information for the current loop...
+    bool MayThrow;           // The current loop contains an instruction which
+                             // may throw, thus preventing code motion of
+                             // instructions with side effects.
     DenseMap<Loop*, AliasSetTracker*> LoopToAliasSetMap;
 
     /// cloneBasicBlockAnalysis - Simple Analysis hook. Clone alias set info.
@@ -204,7 +207,7 @@
   AA = &getAnalysis<AliasAnalysis>();
   DT = &getAnalysis<DominatorTree>();
 
-  TD = getAnalysisIfAvailable<TargetData>();
+  TD = getAnalysisIfAvailable<DataLayout>();
   TLI = &getAnalysis<TargetLibraryInfo>();
 
   CurAST = new AliasSetTracker(*AA);
@@ -240,6 +243,15 @@
       CurAST->add(*BB);                 // Incorporate the specified basic block
   }
 
+  MayThrow = false;
+  // TODO: We've already searched for instructions which may throw in subloops.
+  // We may want to reuse this information.
+  for (Loop::block_iterator BB = L->block_begin(), BBE = L->block_end();
+       (BB != BBE) && !MayThrow ; ++BB)
+    for (BasicBlock::iterator I = (*BB)->begin(), E = (*BB)->end();
+         (I != E) && !MayThrow; ++I)
+      MayThrow |= I->mayThrow();
+
   // We want to visit all of the instructions in this loop... that are not parts
   // of our subloops (they have already had their invariants hoisted out of
   // their loop, into this loop, so there is no need to process the BODIES of
@@ -307,7 +319,7 @@
 
     // If the instruction is dead, we would try to sink it because it isn't used
     // in the loop, instead, just delete it.
-    if (isInstructionTriviallyDead(&I)) {
+    if (isInstructionTriviallyDead(&I, TLI)) {
       DEBUG(dbgs() << "LICM deleting dead inst: " << I << '\n');
       ++II;
       CurAST->deleteValue(&I);
@@ -418,17 +430,22 @@
       if (!FoundMod) return true;
     }
 
-    // FIXME: This should use mod/ref information to see if we can hoist or sink
-    // the call.
+    // FIXME: This should use mod/ref information to see if we can hoist or
+    // sink the call.
 
     return false;
   }
 
-  // Otherwise these instructions are hoistable/sinkable
-  return isa<BinaryOperator>(I) || isa<CastInst>(I) ||
-         isa<SelectInst>(I) || isa<GetElementPtrInst>(I) || isa<CmpInst>(I) ||
-         isa<InsertElementInst>(I) || isa<ExtractElementInst>(I) ||
-         isa<ShuffleVectorInst>(I);
+  // Only these instructions are hoistable/sinkable.
+  bool HoistableKind = (isa<BinaryOperator>(I) || isa<CastInst>(I) ||
+                            isa<SelectInst>(I) || isa<GetElementPtrInst>(I) ||
+                            isa<CmpInst>(I)    || isa<InsertElementInst>(I) ||
+                            isa<ExtractElementInst>(I) ||
+                            isa<ShuffleVectorInst>(I));
+  if (!HoistableKind)
+      return false;
+
+  return isSafeToExecuteUnconditionally(I);
 }
 
 /// isNotUsedInLoop - Return true if the only users of this instruction are
@@ -604,6 +621,12 @@
 }
 
 bool LICM::isGuaranteedToExecute(Instruction &Inst) {
+
+  // Somewhere in this loop there is an instruction which may throw and make us
+  // exit the loop.
+  if (MayThrow)
+    return false;
+
   // Otherwise we have to check to make sure that the instruction dominates all
   // of the exit blocks.  If it doesn't, then there is a path out of the loop
   // which does not execute this instruction, so we can't hoist it.

Modified: llvm/branches/AMDILBackend/lib/Transforms/Scalar/LoopIdiomRecognize.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Scalar/LoopIdiomRecognize.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Scalar/LoopIdiomRecognize.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Scalar/LoopIdiomRecognize.cpp Tue Jan 15 11:16:16 2013
@@ -54,7 +54,7 @@
 #include "llvm/Analysis/ValueTracking.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/raw_ostream.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Target/TargetLibraryInfo.h"
 #include "llvm/Transforms/Utils/Local.h"
 using namespace llvm;
@@ -65,7 +65,7 @@
 namespace {
   class LoopIdiomRecognize : public LoopPass {
     Loop *CurLoop;
-    const TargetData *TD;
+    const DataLayout *TD;
     DominatorTree *DT;
     ScalarEvolution *SE;
     TargetLibraryInfo *TLI;
@@ -132,7 +132,8 @@
 /// and zero out all the operands of this instruction.  If any of them become
 /// dead, delete them and the computation tree that feeds them.
 ///
-static void deleteDeadInstruction(Instruction *I, ScalarEvolution &SE) {
+static void deleteDeadInstruction(Instruction *I, ScalarEvolution &SE,
+                                  const TargetLibraryInfo *TLI) {
   SmallVector<Instruction*, 32> NowDeadInsts;
 
   NowDeadInsts.push_back(I);
@@ -153,7 +154,7 @@
       if (!Op->use_empty()) continue;
 
       if (Instruction *OpI = dyn_cast<Instruction>(Op))
-        if (isInstructionTriviallyDead(OpI))
+        if (isInstructionTriviallyDead(OpI, TLI))
           NowDeadInsts.push_back(OpI);
     }
 
@@ -164,15 +165,21 @@
 
 /// deleteIfDeadInstruction - If the specified value is a dead instruction,
 /// delete it and any recursively used instructions.
-static void deleteIfDeadInstruction(Value *V, ScalarEvolution &SE) {
+static void deleteIfDeadInstruction(Value *V, ScalarEvolution &SE,
+                                    const TargetLibraryInfo *TLI) {
   if (Instruction *I = dyn_cast<Instruction>(V))
-    if (isInstructionTriviallyDead(I))
-      deleteDeadInstruction(I, SE);
+    if (isInstructionTriviallyDead(I, TLI))
+      deleteDeadInstruction(I, SE, TLI);
 }
 
 bool LoopIdiomRecognize::runOnLoop(Loop *L, LPPassManager &LPM) {
   CurLoop = L;
 
+  // If the loop could not be converted to canonical form, it must have an
+  // indirectbr in it, just give up.
+  if (!L->getLoopPreheader())
+    return false;
+
   // Disable loop idiom recognition if the function's name is a common idiom.
   StringRef Name = L->getHeader()->getParent()->getName();
   if (Name == "memset" || Name == "memcpy")
@@ -192,7 +199,7 @@
       return false;
 
   // We require target data for now.
-  TD = getAnalysisIfAvailable<TargetData>();
+  TD = getAnalysisIfAvailable<DataLayout>();
   if (TD == 0) return false;
 
   DT = &getAnalysis<DominatorTree>();
@@ -401,7 +408,7 @@
 ///
 /// Note that we don't ever attempt to use memset_pattern8 or 4, because these
 /// just replicate their input array and then pass on to memset_pattern16.
-static Constant *getMemSetPatternValue(Value *V, const TargetData &TD) {
+static Constant *getMemSetPatternValue(Value *V, const DataLayout &TD) {
   // If the value isn't a constant, we can't promote it to being in a constant
   // array.  We could theoretically do a store to an alloca or something, but
   // that doesn't seem worthwhile.
@@ -490,7 +497,7 @@
                             StoreSize, getAnalysis<AliasAnalysis>(), TheStore)){
     Expander.clear();
     // If we generated new code for the base pointer, clean up.
-    deleteIfDeadInstruction(BasePtr, *SE);
+    deleteIfDeadInstruction(BasePtr, *SE, TLI);
     return false;
   }
 
@@ -538,7 +545,7 @@
 
   // Okay, the memset has been formed.  Zap the original store and anything that
   // feeds into it.
-  deleteDeadInstruction(TheStore, *SE);
+  deleteDeadInstruction(TheStore, *SE, TLI);
   ++NumMemSet;
   return true;
 }
@@ -579,7 +586,7 @@
                             getAnalysis<AliasAnalysis>(), SI)) {
     Expander.clear();
     // If we generated new code for the base pointer, clean up.
-    deleteIfDeadInstruction(StoreBasePtr, *SE);
+    deleteIfDeadInstruction(StoreBasePtr, *SE, TLI);
     return false;
   }
 
@@ -594,8 +601,8 @@
                             StoreSize, getAnalysis<AliasAnalysis>(), SI)) {
     Expander.clear();
     // If we generated new code for the base pointer, clean up.
-    deleteIfDeadInstruction(LoadBasePtr, *SE);
-    deleteIfDeadInstruction(StoreBasePtr, *SE);
+    deleteIfDeadInstruction(LoadBasePtr, *SE, TLI);
+    deleteIfDeadInstruction(StoreBasePtr, *SE, TLI);
     return false;
   }
 
@@ -628,7 +635,7 @@
 
   // Okay, the memset has been formed.  Zap the original store and anything that
   // feeds into it.
-  deleteDeadInstruction(SI, *SE);
+  deleteDeadInstruction(SI, *SE, TLI);
   ++NumMemCpy;
   return true;
 }

Modified: llvm/branches/AMDILBackend/lib/Transforms/Scalar/LoopInstSimplify.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Scalar/LoopInstSimplify.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Scalar/LoopInstSimplify.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Scalar/LoopInstSimplify.cpp Tue Jan 15 11:16:16 2013
@@ -18,7 +18,7 @@
 #include "llvm/Analysis/LoopInfo.h"
 #include "llvm/Analysis/LoopPass.h"
 #include "llvm/Support/Debug.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Target/TargetLibraryInfo.h"
 #include "llvm/Transforms/Scalar.h"
 #include "llvm/Transforms/Utils/Local.h"
@@ -66,7 +66,7 @@
 bool LoopInstSimplify::runOnLoop(Loop *L, LPPassManager &LPM) {
   DominatorTree *DT = getAnalysisIfAvailable<DominatorTree>();
   LoopInfo *LI = &getAnalysis<LoopInfo>();
-  const TargetData *TD = getAnalysisIfAvailable<TargetData>();
+  const DataLayout *TD = getAnalysisIfAvailable<DataLayout>();
   const TargetLibraryInfo *TLI = &getAnalysis<TargetLibraryInfo>();
 
   SmallVector<BasicBlock*, 8> ExitBlocks;
@@ -120,7 +120,7 @@
             ++NumSimplified;
           }
         }
-        LocalChanged |= RecursivelyDeleteTriviallyDeadInstructions(I);
+        LocalChanged |= RecursivelyDeleteTriviallyDeadInstructions(I, TLI);
 
         if (IsSubloopHeader && !isa<PHINode>(I))
           break;

Modified: llvm/branches/AMDILBackend/lib/Transforms/Scalar/LoopRotation.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Scalar/LoopRotation.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Scalar/LoopRotation.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Scalar/LoopRotation.cpp Tue Jan 15 11:16:16 2013
@@ -24,6 +24,7 @@
 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
 #include "llvm/Transforms/Utils/SSAUpdater.h"
 #include "llvm/Transforms/Utils/ValueMapper.h"
+#include "llvm/Support/CFG.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/ADT/Statistic.h"
 using namespace llvm;
@@ -256,6 +257,7 @@
     return false;
 
   BasicBlock *OrigHeader = L->getHeader();
+  BasicBlock *OrigLatch = L->getLoopLatch();
 
   BranchInst *BI = dyn_cast<BranchInst>(OrigHeader->getTerminator());
   if (BI == 0 || BI->isUnconditional())
@@ -267,13 +269,9 @@
   if (!L->isLoopExiting(OrigHeader))
     return false;
 
-  // Updating PHInodes in loops with multiple exits adds complexity.
-  // Keep it simple, and restrict loop rotation to loops with one exit only.
-  // In future, lift this restriction and support for multiple exits if
-  // required.
-  SmallVector<BasicBlock*, 8> ExitBlocks;
-  L->getExitBlocks(ExitBlocks);
-  if (ExitBlocks.size() > 1)
+  // If the loop latch already contains a branch that leaves the loop then the
+  // loop is already rotated.
+  if (OrigLatch == 0 || L->isLoopExiting(OrigLatch))
     return false;
 
   // Check size of original header and reject loop if it is very big.
@@ -286,11 +284,10 @@
 
   // Now, this loop is suitable for rotation.
   BasicBlock *OrigPreheader = L->getLoopPreheader();
-  BasicBlock *OrigLatch = L->getLoopLatch();
 
   // If the loop could not be converted to canonical form, it must have an
   // indirectbr in it, just give up.
-  if (OrigPreheader == 0 || OrigLatch == 0)
+  if (OrigPreheader == 0)
     return false;
 
   // Anything ScalarEvolution may know about this loop or the PHI nodes
@@ -298,6 +295,8 @@
   if (ScalarEvolution *SE = getAnalysisIfAvailable<ScalarEvolution>())
     SE->forgetLoop(L);
 
+  DEBUG(dbgs() << "LoopRotation: rotating "; L->dump());
+
   // Find new Loop header. NewHeader is a Header's one and only successor
   // that is inside loop.  Header's other successor is outside the
   // loop.  Otherwise loop is not suitable for rotation.
@@ -408,10 +407,19 @@
     // Update DominatorTree to reflect the CFG change we just made.  Then split
     // edges as necessary to preserve LoopSimplify form.
     if (DominatorTree *DT = getAnalysisIfAvailable<DominatorTree>()) {
-      // Since OrigPreheader now has the conditional branch to Exit block, it is
-      // the dominator of Exit.
-      DT->changeImmediateDominator(Exit, OrigPreheader);
-      DT->changeImmediateDominator(NewHeader, OrigPreheader);
+      // Everything that was dominated by the old loop header is now dominated
+      // by the original loop preheader. Conceptually the header was merged
+      // into the preheader, even though we reuse the actual block as a new
+      // loop latch.
+      DomTreeNode *OrigHeaderNode = DT->getNode(OrigHeader);
+      SmallVector<DomTreeNode *, 8> HeaderChildren(OrigHeaderNode->begin(),
+                                                   OrigHeaderNode->end());
+      DomTreeNode *OrigPreheaderNode = DT->getNode(OrigPreheader);
+      for (unsigned I = 0, E = HeaderChildren.size(); I != E; ++I)
+        DT->changeImmediateDominator(HeaderChildren[I], OrigPreheaderNode);
+
+      assert(DT->getNode(Exit)->getIDom() == OrigPreheaderNode);
+      assert(DT->getNode(NewHeader)->getIDom() == OrigPreheaderNode);
 
       // Update OrigHeader to be dominated by the new header block.
       DT->changeImmediateDominator(OrigHeader, OrigLatch);
@@ -440,6 +448,35 @@
       // Update OrigHeader to be dominated by the new header block.
       DT->changeImmediateDominator(NewHeader, OrigPreheader);
       DT->changeImmediateDominator(OrigHeader, OrigLatch);
+
+      // Brute force incremental dominator tree update. Call
+      // findNearestCommonDominator on all CFG predecessors of each child of the
+      // original header.
+      DomTreeNode *OrigHeaderNode = DT->getNode(OrigHeader);
+      SmallVector<DomTreeNode *, 8> HeaderChildren(OrigHeaderNode->begin(),
+                                                   OrigHeaderNode->end());
+      bool Changed;
+      do {
+        Changed = false;
+        for (unsigned I = 0, E = HeaderChildren.size(); I != E; ++I) {
+          DomTreeNode *Node = HeaderChildren[I];
+          BasicBlock *BB = Node->getBlock();
+
+          pred_iterator PI = pred_begin(BB);
+          BasicBlock *NearestDom = *PI;
+          for (pred_iterator PE = pred_end(BB); PI != PE; ++PI)
+            NearestDom = DT->findNearestCommonDominator(NearestDom, *PI);
+
+          // Remember if this changes the DomTree.
+          if (Node->getIDom()->getBlock() != NearestDom) {
+            DT->changeImmediateDominator(BB, NearestDom);
+            Changed = true;
+          }
+        }
+
+      // If the dominator changed, this may have an effect on other
+      // predecessors, continue until we reach a fixpoint.
+      } while (Changed);
     }
   }
 
@@ -452,6 +489,8 @@
   // emitted code isn't too gross in this common case.
   MergeBlockIntoPredecessor(OrigHeader, this);
 
+  DEBUG(dbgs() << "LoopRotation: into "; L->dump());
+
   ++NumRotated;
   return true;
 }

Modified: llvm/branches/AMDILBackend/lib/Transforms/Scalar/LoopStrengthReduce.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Scalar/LoopStrengthReduce.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Scalar/LoopStrengthReduce.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Scalar/LoopStrengthReduce.cpp Tue Jan 15 11:16:16 2013
@@ -54,7 +54,7 @@
 //===----------------------------------------------------------------------===//
 
 #define DEBUG_TYPE "loop-reduce"
-#include "llvm/Transforms/Scalar.h"
+#include "llvm/AddressingMode.h"
 #include "llvm/Constants.h"
 #include "llvm/Instructions.h"
 #include "llvm/IntrinsicInst.h"
@@ -64,6 +64,7 @@
 #include "llvm/Analysis/LoopPass.h"
 #include "llvm/Analysis/ScalarEvolutionExpander.h"
 #include "llvm/Assembly/Writer.h"
+#include "llvm/Transforms/Scalar.h"
 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
 #include "llvm/Transforms/Utils/Local.h"
 #include "llvm/ADT/SmallBitVector.h"
@@ -121,9 +122,11 @@
   OS << "[NumUses=" << UsedByIndices.count() << ']';
 }
 
+#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
 void RegSortData::dump() const {
   print(errs()); errs() << '\n';
 }
+#endif
 
 namespace {
 
@@ -223,7 +226,7 @@
 struct Formula {
   /// AM - This is used to represent complex addressing, as well as other kinds
   /// of interesting uses.
-  TargetLowering::AddrMode AM;
+  AddrMode AM;
 
   /// BaseRegs - The list of "base" registers for this use. When this is
   /// non-empty, AM.HasBaseReg should be set to true.
@@ -414,9 +417,11 @@
   }
 }
 
+#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
 void Formula::dump() const {
   print(errs()); errs() << '\n';
 }
+#endif
 
 /// isAddRecSExtable - Return true if the given addrec can be sign-extended
 /// without changing its value.
@@ -738,7 +743,8 @@
   bool Changed = false;
 
   while (!DeadInsts.empty()) {
-    Instruction *I = dyn_cast_or_null<Instruction>(&*DeadInsts.pop_back_val());
+    Value *V = DeadInsts.pop_back_val();
+    Instruction *I = dyn_cast_or_null<Instruction>(V);
 
     if (I == 0 || !isInstructionTriviallyDead(I))
       continue;
@@ -973,9 +979,11 @@
     OS << ", plus " << SetupCost << " setup cost";
 }
 
+#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
 void Cost::dump() const {
   print(errs()); errs() << '\n';
 }
+#endif
 
 namespace {
 
@@ -1059,9 +1067,11 @@
     OS << ", Offset=" << Offset;
 }
 
+#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
 void LSRFixup::dump() const {
   print(errs()); errs() << '\n';
 }
+#endif
 
 namespace {
 
@@ -1251,14 +1261,16 @@
     OS << ", widest fixup type: " << *WidestFixupType;
 }
 
+#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
 void LSRUse::dump() const {
   print(errs()); errs() << '\n';
 }
+#endif
 
 /// isLegalUse - Test whether the use described by AM is "legal", meaning it can
 /// be completely folded into the user instruction at isel time. This includes
 /// address-mode folding and special icmp tricks.
-static bool isLegalUse(const TargetLowering::AddrMode &AM,
+static bool isLegalUse(const AddrMode &AM,
                        LSRUse::KindType Kind, Type *AccessTy,
                        const TargetLowering *TLI) {
   switch (Kind) {
@@ -1315,7 +1327,7 @@
   llvm_unreachable("Invalid LSRUse Kind!");
 }
 
-static bool isLegalUse(TargetLowering::AddrMode AM,
+static bool isLegalUse(AddrMode AM,
                        int64_t MinOffset, int64_t MaxOffset,
                        LSRUse::KindType Kind, Type *AccessTy,
                        const TargetLowering *TLI) {
@@ -1346,7 +1358,7 @@
 
   // Conservatively, create an address with an immediate and a
   // base and a scale.
-  TargetLowering::AddrMode AM;
+  AddrMode AM;
   AM.BaseOffs = BaseOffs;
   AM.BaseGV = BaseGV;
   AM.HasBaseReg = HasBaseReg;
@@ -1384,7 +1396,7 @@
 
   // Conservatively, create an address with an immediate and a
   // base and a scale.
-  TargetLowering::AddrMode AM;
+  AddrMode AM;
   AM.BaseOffs = BaseOffs;
   AM.BaseGV = BaseGV;
   AM.HasBaseReg = HasBaseReg;
@@ -2009,7 +2021,7 @@
               goto decline_post_inc;
             // Check for possible scaled-address reuse.
             Type *AccessTy = getAccessType(UI->getUser());
-            TargetLowering::AddrMode AM;
+            AddrMode AM;
             AM.Scale = C->getSExtValue();
             if (TLI->isLegalAddressingMode(AM, AccessTy))
               goto decline_post_inc;
@@ -3435,9 +3447,11 @@
      << " , add offset " << Imm;
 }
 
+#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
 void WorkItem::dump() const {
   print(errs()); errs() << '\n';
 }
+#endif
 
 /// GenerateCrossUseConstantOffsets - Look for registers which are a constant
 /// distance apart and try to form reuse opportunities between them.
@@ -4451,17 +4465,21 @@
             SplitLandingPadPredecessors(Parent, BB, "", "", P, NewBBs);
             NewBB = NewBBs[0];
           }
-
-          // If PN is outside of the loop and BB is in the loop, we want to
-          // move the block to be immediately before the PHI block, not
-          // immediately after BB.
-          if (L->contains(BB) && !L->contains(PN))
-            NewBB->moveBefore(PN->getParent());
-
-          // Splitting the edge can reduce the number of PHI entries we have.
-          e = PN->getNumIncomingValues();
-          BB = NewBB;
-          i = PN->getBasicBlockIndex(BB);
+          // If NewBB==NULL, then SplitCriticalEdge refused to split because all
+          // phi predecessors are identical. The simple thing to do is skip
+          // splitting in this case rather than complicate the API.
+          if (NewBB) {
+            // If PN is outside of the loop and BB is in the loop, we want to
+            // move the block to be immediately before the PHI block, not
+            // immediately after BB.
+            if (L->contains(BB) && !L->contains(PN))
+              NewBB->moveBefore(PN->getParent());
+
+            // Splitting the edge can reduce the number of PHI entries we have.
+            e = PN->getNumIncomingValues();
+            BB = NewBB;
+            i = PN->getBasicBlockIndex(BB);
+          }
         }
       }
 
@@ -4730,9 +4748,11 @@
   print_uses(OS);
 }
 
+#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
 void LSRInstance::dump() const {
   print(errs()); errs() << '\n';
 }
+#endif
 
 namespace {
 

Modified: llvm/branches/AMDILBackend/lib/Transforms/Scalar/LoopUnrollPass.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Scalar/LoopUnrollPass.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Scalar/LoopUnrollPass.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Scalar/LoopUnrollPass.cpp Tue Jan 15 11:16:16 2013
@@ -22,7 +22,7 @@
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/raw_ostream.h"
 #include "llvm/Transforms/Utils/UnrollLoop.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include <climits>
 
 using namespace llvm;
@@ -113,7 +113,7 @@
 
 /// ApproximateLoopSize - Approximate the size of the loop.
 static unsigned ApproximateLoopSize(const Loop *L, unsigned &NumCalls,
-                                    const TargetData *TD) {
+                                    const DataLayout *TD) {
   CodeMetrics Metrics;
   for (Loop::block_iterator I = L->block_begin(), E = L->block_end();
        I != E; ++I)
@@ -145,7 +145,8 @@
   // not user specified.
   unsigned Threshold = CurrentThreshold;
   if (!UserThreshold &&
-      Header->getParent()->hasFnAttr(Attribute::OptimizeForSize))
+      Header->getParent()->getFnAttributes().
+        hasAttribute(Attributes::OptimizeForSize))
     Threshold = OptSizeUnrollThreshold;
 
   // Find trip count and trip multiple if count is not available
@@ -178,7 +179,7 @@
 
   // Enforce the threshold.
   if (Threshold != NoThreshold) {
-    const TargetData *TD = getAnalysisIfAvailable<TargetData>();
+    const DataLayout *TD = getAnalysisIfAvailable<DataLayout>();
     unsigned NumInlineCandidates;
     unsigned LoopSize = ApproximateLoopSize(L, NumInlineCandidates, TD);
     DEBUG(dbgs() << "  Loop Size = " << LoopSize << "\n");

Modified: llvm/branches/AMDILBackend/lib/Transforms/Scalar/LoopUnswitch.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Scalar/LoopUnswitch.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Scalar/LoopUnswitch.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Scalar/LoopUnswitch.cpp Tue Jan 15 11:16:16 2013
@@ -638,7 +638,8 @@
   // Check to see if it would be profitable to unswitch current loop.
 
   // Do not do non-trivial unswitch while optimizing for size.
-  if (OptimizeForSize || F->hasFnAttr(Attribute::OptimizeForSize))
+  if (OptimizeForSize ||
+      F->getFnAttributes().hasAttribute(Attributes::OptimizeForSize))
     return false;
 
   UnswitchNontrivialCondition(LoopCond, Val, currentLoop);
@@ -906,13 +907,9 @@
 /// specified.
 static void RemoveFromWorklist(Instruction *I,
                                std::vector<Instruction*> &Worklist) {
-  std::vector<Instruction*>::iterator WI = std::find(Worklist.begin(),
-                                                     Worklist.end(), I);
-  while (WI != Worklist.end()) {
-    unsigned Offset = WI-Worklist.begin();
-    Worklist.erase(WI);
-    WI = std::find(Worklist.begin()+Offset, Worklist.end(), I);
-  }
+
+  Worklist.erase(std::remove(Worklist.begin(), Worklist.end(), I),
+                 Worklist.end());
 }
 
 /// ReplaceUsesOfWith - When we find that I really equals V, remove I from the

Modified: llvm/branches/AMDILBackend/lib/Transforms/Scalar/MemCpyOptimizer.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Scalar/MemCpyOptimizer.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Scalar/MemCpyOptimizer.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Scalar/MemCpyOptimizer.cpp Tue Jan 15 11:16:16 2013
@@ -27,7 +27,7 @@
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/GetElementPtrTypeIterator.h"
 #include "llvm/Support/raw_ostream.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Target/TargetLibraryInfo.h"
 #include "llvm/Transforms/Utils/Local.h"
 #include <list>
@@ -38,8 +38,8 @@
 STATISTIC(NumMoveToCpy,   "Number of memmoves converted to memcpy");
 STATISTIC(NumCpyToSet,    "Number of memcpys converted to memset");
 
-static int64_t GetOffsetFromIndex(const GetElementPtrInst *GEP, unsigned Idx,
-                                  bool &VariableIdxFound, const TargetData &TD){
+static int64_t GetOffsetFromIndex(const GEPOperator *GEP, unsigned Idx,
+                                  bool &VariableIdxFound, const DataLayout &TD){
   // Skip over the first indices.
   gep_type_iterator GTI = gep_type_begin(GEP);
   for (unsigned i = 1; i != Idx; ++i, ++GTI)
@@ -72,11 +72,11 @@
 /// constant offset, and return that constant offset.  For example, Ptr1 might
 /// be &A[42], and Ptr2 might be &A[40].  In this case offset would be -8.
 static bool IsPointerOffset(Value *Ptr1, Value *Ptr2, int64_t &Offset,
-                            const TargetData &TD) {
+                            const DataLayout &TD) {
   Ptr1 = Ptr1->stripPointerCasts();
   Ptr2 = Ptr2->stripPointerCasts();
-  GetElementPtrInst *GEP1 = dyn_cast<GetElementPtrInst>(Ptr1);
-  GetElementPtrInst *GEP2 = dyn_cast<GetElementPtrInst>(Ptr2);
+  GEPOperator *GEP1 = dyn_cast<GEPOperator>(Ptr1);
+  GEPOperator *GEP2 = dyn_cast<GEPOperator>(Ptr2);
 
   bool VariableIdxFound = false;
 
@@ -141,12 +141,12 @@
   /// TheStores - The actual stores that make up this range.
   SmallVector<Instruction*, 16> TheStores;
 
-  bool isProfitableToUseMemset(const TargetData &TD) const;
+  bool isProfitableToUseMemset(const DataLayout &TD) const;
 
 };
 } // end anon namespace
 
-bool MemsetRange::isProfitableToUseMemset(const TargetData &TD) const {
+bool MemsetRange::isProfitableToUseMemset(const DataLayout &TD) const {
   // If we found more than 4 stores to merge or 16 bytes, use memset.
   if (TheStores.size() >= 4 || End-Start >= 16) return true;
 
@@ -192,9 +192,9 @@
   /// because each element is relatively large and expensive to copy.
   std::list<MemsetRange> Ranges;
   typedef std::list<MemsetRange>::iterator range_iterator;
-  const TargetData &TD;
+  const DataLayout &TD;
 public:
-  MemsetRanges(const TargetData &td) : TD(td) {}
+  MemsetRanges(const DataLayout &td) : TD(td) {}
 
   typedef std::list<MemsetRange>::const_iterator const_iterator;
   const_iterator begin() const { return Ranges.begin(); }
@@ -302,7 +302,7 @@
   class MemCpyOpt : public FunctionPass {
     MemoryDependenceAnalysis *MD;
     TargetLibraryInfo *TLI;
-    const TargetData *TD;
+    const DataLayout *TD;
   public:
     static char ID; // Pass identification, replacement for typeid
     MemCpyOpt() : FunctionPass(ID) {
@@ -332,7 +332,7 @@
     bool processMemCpy(MemCpyInst *M);
     bool processMemMove(MemMoveInst *M);
     bool performCallSlotOptzn(Instruction *cpy, Value *cpyDst, Value *cpySrc,
-                              uint64_t cpyLen, CallInst *C);
+                              uint64_t cpyLen, unsigned cpyAlign, CallInst *C);
     bool processMemCpyMemCpyDependence(MemCpyInst *M, MemCpyInst *MDep,
                                        uint64_t MSize);
     bool processByValArgument(CallSite CS, unsigned ArgNo);
@@ -509,10 +509,18 @@
       }
 
       if (C) {
+        unsigned storeAlign = SI->getAlignment();
+        if (!storeAlign)
+          storeAlign = TD->getABITypeAlignment(SI->getOperand(0)->getType());
+        unsigned loadAlign = LI->getAlignment();
+        if (!loadAlign)
+          loadAlign = TD->getABITypeAlignment(LI->getType());
+
         bool changed = performCallSlotOptzn(LI,
                         SI->getPointerOperand()->stripPointerCasts(),
                         LI->getPointerOperand()->stripPointerCasts(),
-                        TD->getTypeStoreSize(SI->getOperand(0)->getType()), C);
+                        TD->getTypeStoreSize(SI->getOperand(0)->getType()),
+                        std::min(storeAlign, loadAlign), C);
         if (changed) {
           MD->removeInstruction(SI);
           SI->eraseFromParent();
@@ -559,7 +567,8 @@
 /// the call write its result directly into the destination of the memcpy.
 bool MemCpyOpt::performCallSlotOptzn(Instruction *cpy,
                                      Value *cpyDest, Value *cpySrc,
-                                     uint64_t cpyLen, CallInst *C) {
+                                     uint64_t cpyLen, unsigned cpyAlign,
+                                     CallInst *C) {
   // The general transformation to keep in mind is
   //
   //   call @func(..., src, ...)
@@ -625,6 +634,16 @@
     return false;
   }
 
+  // Check that dest points to memory that is at least as aligned as src.
+  unsigned srcAlign = srcAlloca->getAlignment();
+  if (!srcAlign)
+    srcAlign = TD->getABITypeAlignment(srcAlloca->getAllocatedType());
+  bool isDestSufficientlyAligned = srcAlign <= cpyAlign;
+  // If dest is not aligned enough and we can't increase its alignment then
+  // bail out.
+  if (!isDestSufficientlyAligned && !isa<AllocaInst>(cpyDest))
+    return false;
+
   // Check that src is not accessed except via the call and the memcpy.  This
   // guarantees that it holds only undefined values when passed in (so the final
   // memcpy can be dropped), that it is not read or written between the call and
@@ -673,20 +692,26 @@
   bool changedArgument = false;
   for (unsigned i = 0; i < CS.arg_size(); ++i)
     if (CS.getArgument(i)->stripPointerCasts() == cpySrc) {
-      if (cpySrc->getType() != cpyDest->getType())
-        cpyDest = CastInst::CreatePointerCast(cpyDest, cpySrc->getType(),
-                                              cpyDest->getName(), C);
+      Value *Dest = cpySrc->getType() == cpyDest->getType() ?  cpyDest
+        : CastInst::CreatePointerCast(cpyDest, cpySrc->getType(),
+                                      cpyDest->getName(), C);
       changedArgument = true;
-      if (CS.getArgument(i)->getType() == cpyDest->getType())
-        CS.setArgument(i, cpyDest);
+      if (CS.getArgument(i)->getType() == Dest->getType())
+        CS.setArgument(i, Dest);
       else
-        CS.setArgument(i, CastInst::CreatePointerCast(cpyDest,
-                          CS.getArgument(i)->getType(), cpyDest->getName(), C));
+        CS.setArgument(i, CastInst::CreatePointerCast(Dest,
+                          CS.getArgument(i)->getType(), Dest->getName(), C));
     }
 
   if (!changedArgument)
     return false;
 
+  // If the destination wasn't sufficiently aligned then increase its alignment.
+  if (!isDestSufficientlyAligned) {
+    assert(isa<AllocaInst>(cpyDest) && "Can only increase alloca alignment!");
+    cast<AllocaInst>(cpyDest)->setAlignment(srcAlign);
+  }
+
   // Drop any cached information about the call, because we may have changed
   // its dependence information by changing its parameter.
   MD->removeInstruction(C);
@@ -813,7 +838,8 @@
   if (DepInfo.isClobber()) {
     if (CallInst *C = dyn_cast<CallInst>(DepInfo.getInst())) {
       if (performCallSlotOptzn(M, M->getDest(), M->getSource(),
-                               CopySize->getZExtValue(), C)) {
+                               CopySize->getZExtValue(), M->getAlignment(),
+                               C)) {
         MD->removeInstruction(M);
         M->eraseFromParent();
         return true;
@@ -974,7 +1000,7 @@
 bool MemCpyOpt::runOnFunction(Function &F) {
   bool MadeChange = false;
   MD = &getAnalysis<MemoryDependenceAnalysis>();
-  TD = getAnalysisIfAvailable<TargetData>();
+  TD = getAnalysisIfAvailable<DataLayout>();
   TLI = &getAnalysis<TargetLibraryInfo>();
 
   // If we don't have at least memset and memcpy, there is little point of doing

Modified: llvm/branches/AMDILBackend/lib/Transforms/Scalar/ObjCARC.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Scalar/ObjCARC.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Scalar/ObjCARC.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Scalar/ObjCARC.cpp Tue Jan 15 11:16:16 2013
@@ -29,6 +29,7 @@
 //===----------------------------------------------------------------------===//
 
 #define DEBUG_TYPE "objc-arc"
+#include "llvm/Support/raw_ostream.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/ADT/DenseMap.h"
 using namespace llvm;
@@ -1120,9 +1121,8 @@
     bool relatedSelect(const SelectInst *A, const Value *B);
     bool relatedPHI(const PHINode *A, const Value *B);
 
-    // Do not implement.
-    void operator=(const ProvenanceAnalysis &);
-    ProvenanceAnalysis(const ProvenanceAnalysis &);
+    void operator=(const ProvenanceAnalysis &) LLVM_DELETED_FUNCTION;
+    ProvenanceAnalysis(const ProvenanceAnalysis &) LLVM_DELETED_FUNCTION;
 
   public:
     ProvenanceAnalysis() {}
@@ -1236,16 +1236,19 @@
 
   // An ObjC-Identified object can't alias a load if it is never locally stored.
   if (AIsIdentified) {
+    // Check for an obvious escape.
+    if (isa<LoadInst>(B))
+      return isStoredObjCPointer(A);
     if (BIsIdentified) {
-      // If both pointers have provenance, they can be directly compared.
-      if (A != B)
-        return false;
-    } else {
-      if (isa<LoadInst>(B))
-        return isStoredObjCPointer(A);
+      // Check for an obvious escape.
+      if (isa<LoadInst>(A))
+        return isStoredObjCPointer(B);
+      // Both pointers are identified and escapes aren't an evident problem.
+      return false;
     }
-  } else {
-    if (BIsIdentified && isa<LoadInst>(A))
+  } else if (BIsIdentified) {
+    // Check for an obvious escape.
+    if (isa<LoadInst>(A))
       return isStoredObjCPointer(B);
   }
 
@@ -1381,9 +1384,6 @@
   /// PtrState - This class summarizes several per-pointer runtime properties
   /// which are propogated through the flow graph.
   class PtrState {
-    /// NestCount - The known minimum level of retain+release nesting.
-    unsigned NestCount;
-
     /// KnownPositiveRefCount - True if the reference count is known to
     /// be incremented.
     bool KnownPositiveRefCount;
@@ -1401,7 +1401,7 @@
     /// TODO: Encapsulate this better.
     RRInfo RRI;
 
-    PtrState() : NestCount(0), KnownPositiveRefCount(false), Partial(false),
+    PtrState() : KnownPositiveRefCount(false), Partial(false),
                  Seq(S_None) {}
 
     void SetKnownPositiveRefCount() {
@@ -1416,18 +1416,6 @@
       return KnownPositiveRefCount;
     }
 
-    void IncrementNestCount() {
-      if (NestCount != UINT_MAX) ++NestCount;
-    }
-
-    void DecrementNestCount() {
-      if (NestCount != 0) --NestCount;
-    }
-
-    bool IsKnownNested() const {
-      return NestCount > 0;
-    }
-
     void SetSeq(Sequence NewSeq) {
       Seq = NewSeq;
     }
@@ -1454,7 +1442,6 @@
 PtrState::Merge(const PtrState &Other, bool TopDown) {
   Seq = MergeSeqs(Seq, Other.Seq, TopDown);
   KnownPositiveRefCount = KnownPositiveRefCount && Other.KnownPositiveRefCount;
-  NestCount = std::min(NestCount, Other.NestCount);
 
   // We can't merge a plain objc_retain with an objc_retainBlock.
   if (RRI.IsRetainBlock != Other.RRI.IsRetainBlock)
@@ -1610,6 +1597,12 @@
   // loop backedge. Loop backedges are special.
   TopDownPathCount += Other.TopDownPathCount;
 
+  // Check for overflow. If we have overflow, fall back to conservative behavior.
+  if (TopDownPathCount < Other.TopDownPathCount) {
+    clearTopDownPointers();
+    return;
+  }
+
   // For each entry in the other set, if our set has an entry with the same key,
   // merge the entries. Otherwise, copy the entry and merge it with an empty
   // entry.
@@ -1635,6 +1628,12 @@
   // loop backedge. Loop backedges are special.
   BottomUpPathCount += Other.BottomUpPathCount;
 
+  // Check for overflow. If we have overflow, fall back to conservative behavior.
+  if (BottomUpPathCount < Other.BottomUpPathCount) {
+    clearBottomUpPointers();
+    return;
+  }
+
   // For each entry in the other set, if our set has an entry with the
   // same key, merge the entries. Otherwise, copy the entry and merge
   // it with an empty entry.
@@ -1789,7 +1788,9 @@
     Type *I8X = PointerType::getUnqual(Type::getInt8Ty(C));
     Type *Params[] = { I8X };
     FunctionType *FTy = FunctionType::get(I8X, Params, /*isVarArg=*/false);
-    AttrListPtr Attributes = AttrListPtr().addAttr(~0u, Attribute::NoUnwind);
+    AttrListPtr Attributes =
+      AttrListPtr().addAttr(M->getContext(), AttrListPtr::FunctionIndex,
+                            Attributes::get(C, Attributes::NoUnwind));
     RetainRVCallee =
       M->getOrInsertFunction("objc_retainAutoreleasedReturnValue", FTy,
                              Attributes);
@@ -1803,7 +1804,9 @@
     Type *I8X = PointerType::getUnqual(Type::getInt8Ty(C));
     Type *Params[] = { I8X };
     FunctionType *FTy = FunctionType::get(I8X, Params, /*isVarArg=*/false);
-    AttrListPtr Attributes = AttrListPtr().addAttr(~0u, Attribute::NoUnwind);
+    AttrListPtr Attributes =
+      AttrListPtr().addAttr(M->getContext(), AttrListPtr::FunctionIndex,
+                            Attributes::get(C, Attributes::NoUnwind));
     AutoreleaseRVCallee =
       M->getOrInsertFunction("objc_autoreleaseReturnValue", FTy,
                              Attributes);
@@ -1815,7 +1818,9 @@
   if (!ReleaseCallee) {
     LLVMContext &C = M->getContext();
     Type *Params[] = { PointerType::getUnqual(Type::getInt8Ty(C)) };
-    AttrListPtr Attributes = AttrListPtr().addAttr(~0u, Attribute::NoUnwind);
+    AttrListPtr Attributes =
+      AttrListPtr().addAttr(M->getContext(), AttrListPtr::FunctionIndex,
+                            Attributes::get(C, Attributes::NoUnwind));
     ReleaseCallee =
       M->getOrInsertFunction(
         "objc_release",
@@ -1829,7 +1834,9 @@
   if (!RetainCallee) {
     LLVMContext &C = M->getContext();
     Type *Params[] = { PointerType::getUnqual(Type::getInt8Ty(C)) };
-    AttrListPtr Attributes = AttrListPtr().addAttr(~0u, Attribute::NoUnwind);
+    AttrListPtr Attributes =
+      AttrListPtr().addAttr(M->getContext(), AttrListPtr::FunctionIndex,
+                            Attributes::get(C, Attributes::NoUnwind));
     RetainCallee =
       M->getOrInsertFunction(
         "objc_retain",
@@ -1858,7 +1865,9 @@
   if (!AutoreleaseCallee) {
     LLVMContext &C = M->getContext();
     Type *Params[] = { PointerType::getUnqual(Type::getInt8Ty(C)) };
-    AttrListPtr Attributes = AttrListPtr().addAttr(~0u, Attribute::NoUnwind);
+    AttrListPtr Attributes =
+      AttrListPtr().addAttr(M->getContext(), AttrListPtr::FunctionIndex,
+                            Attributes::get(C, Attributes::NoUnwind));
     AutoreleaseCallee =
       M->getOrInsertFunction(
         "objc_autorelease",
@@ -1868,6 +1877,26 @@
   return AutoreleaseCallee;
 }
 
+/// IsPotentialUse - Test whether the given value is possible a
+/// reference-counted pointer, including tests which utilize AliasAnalysis.
+static bool IsPotentialUse(const Value *Op, AliasAnalysis &AA) {
+  // First make the rudimentary check.
+  if (!IsPotentialUse(Op))
+    return false;
+
+  // Objects in constant memory are not reference-counted.
+  if (AA.pointsToConstantMemory(Op))
+    return false;
+
+  // Pointers in constant memory are not pointing to reference-counted objects.
+  if (const LoadInst *LI = dyn_cast<LoadInst>(Op))
+    if (AA.pointsToConstantMemory(LI->getPointerOperand()))
+      return false;
+
+  // Otherwise assume the worst.
+  return true;
+}
+
 /// CanAlterRefCount - Test whether the given instruction can result in a
 /// reference count modification (positive or negative) for the pointer's
 /// object.
@@ -1894,7 +1923,7 @@
     for (ImmutableCallSite::arg_iterator I = CS.arg_begin(), E = CS.arg_end();
          I != E; ++I) {
       const Value *Op = *I;
-      if (IsPotentialUse(Op) && PA.related(Ptr, Op))
+      if (IsPotentialUse(Op, *PA.getAA()) && PA.related(Ptr, Op))
         return true;
     }
     return false;
@@ -1919,14 +1948,14 @@
     // Comparing a pointer with null, or any other constant, isn't really a use,
     // because we don't care what the pointer points to, or about the values
     // of any other dynamic reference-counted pointers.
-    if (!IsPotentialUse(ICI->getOperand(1)))
+    if (!IsPotentialUse(ICI->getOperand(1), *PA.getAA()))
       return false;
   } else if (ImmutableCallSite CS = static_cast<const Value *>(Inst)) {
     // For calls, just check the arguments (and not the callee operand).
     for (ImmutableCallSite::arg_iterator OI = CS.arg_begin(),
          OE = CS.arg_end(); OI != OE; ++OI) {
       const Value *Op = *OI;
-      if (IsPotentialUse(Op) && PA.related(Ptr, Op))
+      if (IsPotentialUse(Op, *PA.getAA()) && PA.related(Ptr, Op))
         return true;
     }
     return false;
@@ -1936,14 +1965,14 @@
     const Value *Op = GetUnderlyingObjCPtr(SI->getPointerOperand());
     // If we can't tell what the underlying object was, assume there is a
     // dependence.
-    return IsPotentialUse(Op) && PA.related(Op, Ptr);
+    return IsPotentialUse(Op, *PA.getAA()) && PA.related(Op, Ptr);
   }
 
   // Check each operand for a match.
   for (User::const_op_iterator OI = Inst->op_begin(), OE = Inst->op_end();
        OI != OE; ++OI) {
     const Value *Op = *OI;
-    if (IsPotentialUse(Op) && PA.related(Ptr, Op))
+    if (IsPotentialUse(Op, *PA.getAA()) && PA.related(Ptr, Op))
       return true;
   }
   return false;
@@ -2612,11 +2641,11 @@
     MDNode *ReleaseMetadata = Inst->getMetadata(ImpreciseReleaseMDKind);
     S.ResetSequenceProgress(ReleaseMetadata ? S_MovableRelease : S_Release);
     S.RRI.ReleaseMetadata = ReleaseMetadata;
-    S.RRI.KnownSafe = S.IsKnownNested() || S.IsKnownIncremented();
+    S.RRI.KnownSafe = S.IsKnownIncremented();
     S.RRI.IsTailCallRelease = cast<CallInst>(Inst)->isTailCall();
     S.RRI.Calls.insert(Inst);
 
-    S.IncrementNestCount();
+    S.SetKnownPositiveRefCount();
     break;
   }
   case IC_RetainBlock:
@@ -2631,7 +2660,6 @@
 
     PtrState &S = MyStates.getPtrBottomUpState(Arg);
     S.SetKnownPositiveRefCount();
-    S.DecrementNestCount();
 
     switch (S.GetSeq()) {
     case S_Stop:
@@ -2747,8 +2775,9 @@
 
   // Merge the states from each successor to compute the initial state
   // for the current block.
-  for (BBState::edge_iterator SI(MyStates.succ_begin()),
-       SE(MyStates.succ_end()); SI != SE; ++SI) {
+  BBState::edge_iterator SI(MyStates.succ_begin()),
+                         SE(MyStates.succ_end());
+  if (SI != SE) {
     const BasicBlock *Succ = *SI;
     DenseMap<const BasicBlock *, BBState>::iterator I = BBStates.find(Succ);
     assert(I != BBStates.end());
@@ -2760,7 +2789,6 @@
       assert(I != BBStates.end());
       MyStates.MergeSucc(I->second);
     }
-    break;
   }
 
   // Visit all the instructions, bottom-up.
@@ -2823,12 +2851,11 @@
 
       S.ResetSequenceProgress(S_Retain);
       S.RRI.IsRetainBlock = Class == IC_RetainBlock;
-      // Don't check S.IsKnownIncremented() here because it's not sufficient.
-      S.RRI.KnownSafe = S.IsKnownNested();
+      S.RRI.KnownSafe = S.IsKnownIncremented();
       S.RRI.Calls.insert(Inst);
     }
 
-    S.IncrementNestCount();
+    S.SetKnownPositiveRefCount();
 
     // A retain can be a potential use; procede to the generic checking
     // code below.
@@ -2838,7 +2865,7 @@
     Arg = GetObjCArg(Inst);
 
     PtrState &S = MyStates.getPtrTopDownState(Arg);
-    S.DecrementNestCount();
+    S.ClearRefCount();
 
     switch (S.GetSeq()) {
     case S_Retain:
@@ -2935,8 +2962,9 @@
 
   // Merge the states from each predecessor to compute the initial state
   // for the current block.
-  for (BBState::edge_iterator PI(MyStates.pred_begin()),
-       PE(MyStates.pred_end()); PI != PE; ++PI) {
+  BBState::edge_iterator PI(MyStates.pred_begin()),
+                         PE(MyStates.pred_end());
+  if (PI != PE) {
     const BasicBlock *Pred = *PI;
     DenseMap<const BasicBlock *, BBState>::iterator I = BBStates.find(Pred);
     assert(I != BBStates.end());
@@ -2948,7 +2976,6 @@
       assert(I != BBStates.end());
       MyStates.MergePred(I->second);
     }
-    break;
   }
 
   // Visit all the instructions, top-down.
@@ -3532,19 +3559,19 @@
 }
 
 /// OptimizeReturns - Look for this pattern:
-///
+/// \code
 ///    %call = call i8* @something(...)
 ///    %2 = call i8* @objc_retain(i8* %call)
 ///    %3 = call i8* @objc_autorelease(i8* %2)
 ///    ret i8* %3
-///
+/// \endcode
 /// And delete the retain and autorelease.
 ///
 /// Otherwise if it's just this:
-///
+/// \code
 ///    %3 = call i8* @objc_autorelease(i8* %2)
 ///    ret i8* %3
-///
+/// \endcode
 /// convert the autorelease to autoreleaseRV.
 void ObjCARCOpt::OptimizeReturns(Function &F) {
   if (!F.getReturnType()->isPointerTy())
@@ -3814,8 +3841,9 @@
     Type *Params[] = { I8XX, I8X };
 
     AttrListPtr Attributes = AttrListPtr()
-      .addAttr(~0u, Attribute::NoUnwind)
-      .addAttr(1, Attribute::NoCapture);
+      .addAttr(M->getContext(), AttrListPtr::FunctionIndex,
+               Attributes::get(C, Attributes::NoUnwind))
+      .addAttr(M->getContext(), 1, Attributes::get(C, Attributes::NoCapture));
 
     StoreStrongCallee =
       M->getOrInsertFunction(
@@ -3832,7 +3860,9 @@
     Type *I8X = PointerType::getUnqual(Type::getInt8Ty(C));
     Type *Params[] = { I8X };
     FunctionType *FTy = FunctionType::get(I8X, Params, /*isVarArg=*/false);
-    AttrListPtr Attributes = AttrListPtr().addAttr(~0u, Attribute::NoUnwind);
+    AttrListPtr Attributes =
+      AttrListPtr().addAttr(M->getContext(), AttrListPtr::FunctionIndex,
+                            Attributes::get(C, Attributes::NoUnwind));
     RetainAutoreleaseCallee =
       M->getOrInsertFunction("objc_retainAutorelease", FTy, Attributes);
   }
@@ -3845,7 +3875,9 @@
     Type *I8X = PointerType::getUnqual(Type::getInt8Ty(C));
     Type *Params[] = { I8X };
     FunctionType *FTy = FunctionType::get(I8X, Params, /*isVarArg=*/false);
-    AttrListPtr Attributes = AttrListPtr().addAttr(~0u, Attribute::NoUnwind);
+    AttrListPtr Attributes =
+      AttrListPtr().addAttr(M->getContext(), AttrListPtr::FunctionIndex,
+                            Attributes::get(C, Attributes::NoUnwind));
     RetainAutoreleaseRVCallee =
       M->getOrInsertFunction("objc_retainAutoreleaseReturnValue", FTy,
                              Attributes);

Modified: llvm/branches/AMDILBackend/lib/Transforms/Scalar/Reassociate.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Scalar/Reassociate.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Scalar/Reassociate.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Scalar/Reassociate.cpp Tue Jan 15 11:16:16 2013
@@ -339,36 +339,6 @@
   }
 }
 
-/// EvaluateRepeatedConstant - Compute C op C op ... op C where the constant C
-/// is repeated Weight times.
-static Constant *EvaluateRepeatedConstant(unsigned Opcode, Constant *C,
-                                          APInt Weight) {
-  // For addition the result can be efficiently computed as the product of the
-  // constant and the weight.
-  if (Opcode == Instruction::Add)
-    return ConstantExpr::getMul(C, ConstantInt::get(C->getContext(), Weight));
-
-  // The weight might be huge, so compute by repeated squaring to ensure that
-  // compile time is proportional to the logarithm of the weight.
-  Constant *Result = 0;
-  Constant *Power = C; // Successively C, C op C, (C op C) op (C op C) etc.
-  // Visit the bits in Weight.
-  while (Weight != 0) {
-    // If the current bit in Weight is non-zero do Result = Result op Power.
-    if (Weight[0])
-      Result = Result ? ConstantExpr::get(Opcode, Result, Power) : Power;
-    // Move on to the next bit if any more are non-zero.
-    Weight = Weight.lshr(1);
-    if (Weight.isMinValue())
-      break;
-    // Square the power.
-    Power = ConstantExpr::get(Opcode, Power, Power);
-  }
-
-  assert(Result && "Only positive weights supported!");
-  return Result;
-}
-
 typedef std::pair<Value*, APInt> RepeatedValue;
 
 /// LinearizeExprTree - Given an associative binary expression, return the leaf
@@ -382,9 +352,7 @@
 /// op
 ///   (Ops[N].first op Ops[N].first op ... Ops[N].first)  <- Ops[N].second times
 ///
-/// Note that the values Ops[0].first, ..., Ops[N].first are all distinct, and
-/// they are all non-constant except possibly for the last one, which if it is
-/// constant will have weight one (Ops[N].second === 1).
+/// Note that the values Ops[0].first, ..., Ops[N].first are all distinct.
 ///
 /// This routine may modify the function, in which case it returns 'true'.  The
 /// changes it makes may well be destructive, changing the value computed by 'I'
@@ -604,7 +572,6 @@
 
   // The leaves, repeated according to their weights, represent the linearized
   // form of the expression.
-  Constant *Cst = 0; // Accumulate constants here.
   for (unsigned i = 0, e = LeafOrder.size(); i != e; ++i) {
     Value *V = LeafOrder[i];
     LeafMap::iterator It = Leaves.find(V);
@@ -618,31 +585,14 @@
       continue;
     // Ensure the leaf is only output once.
     It->second = 0;
-    // Glob all constants together into Cst.
-    if (Constant *C = dyn_cast<Constant>(V)) {
-      C = EvaluateRepeatedConstant(Opcode, C, Weight);
-      Cst = Cst ? ConstantExpr::get(Opcode, Cst, C) : C;
-      continue;
-    }
-    // Add non-constant
     Ops.push_back(std::make_pair(V, Weight));
   }
 
-  // Add any constants back into Ops, all globbed together and reduced to having
-  // weight 1 for the convenience of users.
-  Constant *Identity = ConstantExpr::getBinOpIdentity(Opcode, I->getType());
-  if (Cst && Cst != Identity) {
-    // If combining multiple constants resulted in the absorber then the entire
-    // expression must evaluate to the absorber.
-    if (Cst == Absorber)
-      Ops.clear();
-    Ops.push_back(std::make_pair(Cst, APInt(Bitwidth, 1)));
-  }
-
   // For nilpotent operations or addition there may be no operands, for example
   // because the expression was "X xor X" or consisted of 2^Bitwidth additions:
   // in both cases the weight reduces to 0 causing the value to be skipped.
   if (Ops.empty()) {
+    Constant *Identity = ConstantExpr::getBinOpIdentity(Opcode, I->getType());
     assert(Identity && "Associative operation without identity!");
     Ops.push_back(std::make_pair(Identity, APInt(Bitwidth, 1)));
   }
@@ -656,8 +606,8 @@
                                   SmallVectorImpl<ValueEntry> &Ops) {
   assert(Ops.size() > 1 && "Single values should be used directly!");
 
-  // Since our optimizations never increase the number of operations, the new
-  // expression can always be written by reusing the existing binary operators
+  // Since our optimizations should never increase the number of operations, the
+  // new expression can usually be written reusing the existing binary operators
   // from the original expression tree, without creating any new instructions,
   // though the rewritten expression may have a completely different topology.
   // We take care to not change anything if the new expression will be the same
@@ -671,6 +621,20 @@
   unsigned Opcode = I->getOpcode();
   BinaryOperator *Op = I;
 
+  /// NotRewritable - The operands being written will be the leaves of the new
+  /// expression and must not be used as inner nodes (via NodesToRewrite) by
+  /// mistake.  Inner nodes are always reassociable, and usually leaves are not
+  /// (if they were they would have been incorporated into the expression and so
+  /// would not be leaves), so most of the time there is no danger of this.  But
+  /// in rare cases a leaf may become reassociable if an optimization kills uses
+  /// of it, or it may momentarily become reassociable during rewriting (below)
+  /// due it being removed as an operand of one of its uses.  Ensure that misuse
+  /// of leaf nodes as inner nodes cannot occur by remembering all of the future
+  /// leaves and refusing to reuse any of them as inner nodes.
+  SmallPtrSet<Value*, 8> NotRewritable;
+  for (unsigned i = 0, e = Ops.size(); i != e; ++i)
+    NotRewritable.insert(Ops[i].Op);
+
   // ExpressionChanged - Non-null if the rewritten expression differs from the
   // original in some non-trivial way, requiring the clearing of optional flags.
   // Flags are cleared from the operator in ExpressionChanged up to I inclusive.
@@ -703,12 +667,14 @@
       // the old operands with the new ones.
       DEBUG(dbgs() << "RA: " << *Op << '\n');
       if (NewLHS != OldLHS) {
-        if (BinaryOperator *BO = isReassociableOp(OldLHS, Opcode))
+        BinaryOperator *BO = isReassociableOp(OldLHS, Opcode);
+        if (BO && !NotRewritable.count(BO))
           NodesToRewrite.push_back(BO);
         Op->setOperand(0, NewLHS);
       }
       if (NewRHS != OldRHS) {
-        if (BinaryOperator *BO = isReassociableOp(OldRHS, Opcode))
+        BinaryOperator *BO = isReassociableOp(OldRHS, Opcode);
+        if (BO && !NotRewritable.count(BO))
           NodesToRewrite.push_back(BO);
         Op->setOperand(1, NewRHS);
       }
@@ -732,7 +698,8 @@
         Op->swapOperands();
       } else {
         // Overwrite with the new right-hand side.
-        if (BinaryOperator *BO = isReassociableOp(Op->getOperand(1), Opcode))
+        BinaryOperator *BO = isReassociableOp(Op->getOperand(1), Opcode);
+        if (BO && !NotRewritable.count(BO))
           NodesToRewrite.push_back(BO);
         Op->setOperand(1, NewRHS);
         ExpressionChanged = Op;
@@ -745,7 +712,8 @@
     // Now deal with the left-hand side.  If this is already an operation node
     // from the original expression then just rewrite the rest of the expression
     // into it.
-    if (BinaryOperator *BO = isReassociableOp(Op->getOperand(0), Opcode)) {
+    BinaryOperator *BO = isReassociableOp(Op->getOperand(0), Opcode);
+    if (BO && !NotRewritable.count(BO)) {
       Op = BO;
       continue;
     }
@@ -1446,9 +1414,26 @@
                                        SmallVectorImpl<ValueEntry> &Ops) {
   // Now that we have the linearized expression tree, try to optimize it.
   // Start by folding any constants that we found.
-  if (Ops.size() == 1) return Ops[0].Op;
-
+  Constant *Cst = 0;
   unsigned Opcode = I->getOpcode();
+  while (!Ops.empty() && isa<Constant>(Ops.back().Op)) {
+    Constant *C = cast<Constant>(Ops.pop_back_val().Op);
+    Cst = Cst ? ConstantExpr::get(Opcode, C, Cst) : C;
+  }
+  // If there was nothing but constants then we are done.
+  if (Ops.empty())
+    return Cst;
+
+  // Put the combined constant back at the end of the operand list, except if
+  // there is no point.  For example, an add of 0 gets dropped here, while a
+  // multiplication by zero turns the whole expression into zero.
+  if (Cst && Cst != ConstantExpr::getBinOpIdentity(Opcode, I->getType())) {
+    if (Cst == ConstantExpr::getBinOpAbsorber(Opcode, I->getType()))
+      return Cst;
+    Ops.push_back(ValueEntry(0, Cst));
+  }
+
+  if (Ops.size() == 1) return Ops[0].Op;
 
   // Handle destructive annihilation due to identities between elements in the
   // argument list here.

Modified: llvm/branches/AMDILBackend/lib/Transforms/Scalar/SCCP.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Scalar/SCCP.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Scalar/SCCP.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Scalar/SCCP.cpp Tue Jan 15 11:16:16 2013
@@ -26,7 +26,7 @@
 #include "llvm/Pass.h"
 #include "llvm/Analysis/ConstantFolding.h"
 #include "llvm/Transforms/Utils/Local.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Target/TargetLibraryInfo.h"
 #include "llvm/Support/CallSite.h"
 #include "llvm/Support/Debug.h"
@@ -153,7 +153,7 @@
 /// Constant Propagation.
 ///
 class SCCPSolver : public InstVisitor<SCCPSolver> {
-  const TargetData *TD;
+  const DataLayout *TD;
   const TargetLibraryInfo *TLI;
   SmallPtrSet<BasicBlock*, 8> BBExecutable; // The BBs that are executable.
   DenseMap<Value*, LatticeVal> ValueState;  // The state each value is in.
@@ -205,7 +205,7 @@
   typedef std::pair<BasicBlock*, BasicBlock*> Edge;
   DenseSet<Edge> KnownFeasibleEdges;
 public:
-  SCCPSolver(const TargetData *td, const TargetLibraryInfo *tli)
+  SCCPSolver(const DataLayout *td, const TargetLibraryInfo *tli)
     : TD(td), TLI(tli) {}
 
   /// MarkBlockExecutable - This method can be used by clients to mark all of
@@ -1564,7 +1564,7 @@
 //
 bool SCCP::runOnFunction(Function &F) {
   DEBUG(dbgs() << "SCCP on function '" << F.getName() << "'\n");
-  const TargetData *TD = getAnalysisIfAvailable<TargetData>();
+  const DataLayout *TD = getAnalysisIfAvailable<DataLayout>();
   const TargetLibraryInfo *TLI = &getAnalysis<TargetLibraryInfo>();
   SCCPSolver Solver(TD, TLI);
 
@@ -1693,7 +1693,7 @@
 }
 
 bool IPSCCP::runOnModule(Module &M) {
-  const TargetData *TD = getAnalysisIfAvailable<TargetData>();
+  const DataLayout *TD = getAnalysisIfAvailable<DataLayout>();
   const TargetLibraryInfo *TLI = &getAnalysis<TargetLibraryInfo>();
   SCCPSolver Solver(TD, TLI);
 

Copied: llvm/branches/AMDILBackend/lib/Transforms/Scalar/SROA.cpp (from r167701, llvm/trunk/lib/Transforms/Scalar/SROA.cpp)
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Scalar/SROA.cpp?p2=llvm/branches/AMDILBackend/lib/Transforms/Scalar/SROA.cpp&p1=llvm/trunk/lib/Transforms/Scalar/SROA.cpp&r1=167701&r2=172541&rev=172541&view=diff
==============================================================================
--- llvm/trunk/lib/Transforms/Scalar/SROA.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Scalar/SROA.cpp Tue Jan 15 11:16:16 2013
@@ -568,6 +568,10 @@
 
     // Clamp the end offset to the end of the allocation. Note that this is
     // formulated to handle even the case where "BeginOffset + Size" overflows.
+    // NOTE! This may appear superficially to be something we could ignore
+    // entirely, but that is not so! There may be PHI-node uses where some
+    // instructions are dead but not others. We can't completely ignore the
+    // PHI node, and so have to record at least the information here.
     assert(AllocSize >= BeginOffset); // Established above.
     if (Size > AllocSize - BeginOffset) {
       DEBUG(dbgs() << "WARNING: Clamping a " << Size << " byte use @" << Offset
@@ -1382,11 +1386,7 @@
   /// \brief A collection of instructions to delete.
   /// We try to batch deletions to simplify code and make things a bit more
   /// efficient.
-  SmallVector<Instruction *, 8> DeadInsts;
-
-  /// \brief A set to prevent repeatedly marking an instruction split into many
-  /// uses as dead. Only used to guard insertion into DeadInsts.
-  SmallPtrSet<Instruction *, 4> DeadSplitInsts;
+  SetVector<Instruction *, SmallVector<Instruction *, 8> > DeadInsts;
 
   /// \brief Post-promotion worklist.
   ///
@@ -1573,7 +1573,7 @@
     do {
       LoadInst *LI = Loads.pop_back_val();
       LI->replaceAllUsesWith(NewPN);
-      Pass.DeadInsts.push_back(LI);
+      Pass.DeadInsts.insert(LI);
     } while (!Loads.empty());
 
     // Inject loads into all of the pred blocks.
@@ -1717,7 +1717,7 @@
 
       DEBUG(dbgs() << "          speculated to: " << *V << "\n");
       LI->replaceAllUsesWith(V);
-      Pass.DeadInsts.push_back(LI);
+      Pass.DeadInsts.insert(LI);
     }
   }
 };
@@ -2134,8 +2134,13 @@
     } else if (I->U->get()->getType()->getPointerElementType()->isStructTy()) {
       // Disable vector promotion when there are loads or stores of an FCA.
       return false;
-    } else if (!isa<LoadInst>(I->U->getUser()) &&
-               !isa<StoreInst>(I->U->getUser())) {
+    } else if (LoadInst *LI = dyn_cast<LoadInst>(I->U->getUser())) {
+      if (LI->isVolatile())
+        return false;
+    } else if (StoreInst *SI = dyn_cast<StoreInst>(I->U->getUser())) {
+      if (SI->isVolatile())
+        return false;
+    } else {
       return false;
     }
   }
@@ -2155,6 +2160,9 @@
                                     AllocaPartitioning::const_use_iterator I,
                                     AllocaPartitioning::const_use_iterator E) {
   uint64_t SizeInBits = TD.getTypeSizeInBits(AllocaTy);
+  // Don't create integer types larger than the maximum bitwidth.
+  if (SizeInBits > IntegerType::MAX_INT_BITS)
+    return false;
 
   // Don't try to handle allocas with bit-padding.
   if (SizeInBits != TD.getTypeStoreSizeInBits(AllocaTy))
@@ -2193,7 +2201,7 @@
       if (RelBegin == 0 && RelEnd == Size)
         WholeAllocaOp = true;
       if (IntegerType *ITy = dyn_cast<IntegerType>(LI->getType())) {
-        if (ITy->getBitWidth() < TD.getTypeStoreSize(ITy))
+        if (ITy->getBitWidth() < TD.getTypeStoreSizeInBits(ITy))
           return false;
         continue;
       }
@@ -2209,7 +2217,7 @@
       if (RelBegin == 0 && RelEnd == Size)
         WholeAllocaOp = true;
       if (IntegerType *ITy = dyn_cast<IntegerType>(ValueTy)) {
-        if (ITy->getBitWidth() < TD.getTypeStoreSize(ITy))
+        if (ITy->getBitWidth() < TD.getTypeStoreSizeInBits(ITy))
           return false;
         continue;
       }
@@ -2241,18 +2249,23 @@
 static Value *extractInteger(const DataLayout &DL, IRBuilder<> &IRB, Value *V,
                              IntegerType *Ty, uint64_t Offset,
                              const Twine &Name) {
+  DEBUG(dbgs() << "       start: " << *V << "\n");
   IntegerType *IntTy = cast<IntegerType>(V->getType());
   assert(DL.getTypeStoreSize(Ty) + Offset <= DL.getTypeStoreSize(IntTy) &&
          "Element extends past full value");
   uint64_t ShAmt = 8*Offset;
   if (DL.isBigEndian())
     ShAmt = 8*(DL.getTypeStoreSize(IntTy) - DL.getTypeStoreSize(Ty) - Offset);
-  if (ShAmt)
+  if (ShAmt) {
     V = IRB.CreateLShr(V, ShAmt, Name + ".shift");
+    DEBUG(dbgs() << "     shifted: " << *V << "\n");
+  }
   assert(Ty->getBitWidth() <= IntTy->getBitWidth() &&
          "Cannot extract to a larger integer!");
-  if (Ty != IntTy)
+  if (Ty != IntTy) {
     V = IRB.CreateTrunc(V, Ty, Name + ".trunc");
+    DEBUG(dbgs() << "     trunced: " << *V << "\n");
+  }
   return V;
 }
 
@@ -2262,20 +2275,27 @@
   IntegerType *Ty = cast<IntegerType>(V->getType());
   assert(Ty->getBitWidth() <= IntTy->getBitWidth() &&
          "Cannot insert a larger integer!");
-  if (Ty != IntTy)
+  DEBUG(dbgs() << "       start: " << *V << "\n");
+  if (Ty != IntTy) {
     V = IRB.CreateZExt(V, IntTy, Name + ".ext");
+    DEBUG(dbgs() << "    extended: " << *V << "\n");
+  }
   assert(DL.getTypeStoreSize(Ty) + Offset <= DL.getTypeStoreSize(IntTy) &&
          "Element store outside of alloca store");
   uint64_t ShAmt = 8*Offset;
   if (DL.isBigEndian())
     ShAmt = 8*(DL.getTypeStoreSize(IntTy) - DL.getTypeStoreSize(Ty) - Offset);
-  if (ShAmt)
+  if (ShAmt) {
     V = IRB.CreateShl(V, ShAmt, Name + ".shift");
+    DEBUG(dbgs() << "     shifted: " << *V << "\n");
+  }
 
   if (ShAmt || Ty->getBitWidth() < IntTy->getBitWidth()) {
     APInt Mask = ~Ty->getMask().zext(IntTy->getBitWidth()).shl(ShAmt);
     Old = IRB.CreateAnd(Old, Mask, Name + ".mask");
+    DEBUG(dbgs() << "      masked: " << *Old << "\n");
     V = IRB.CreateOr(Old, V, Name + ".insert");
+    DEBUG(dbgs() << "    inserted: " << *V << "\n");
   }
   return V;
 }
@@ -2442,30 +2462,21 @@
   void deleteIfTriviallyDead(Value *V) {
     Instruction *I = cast<Instruction>(V);
     if (isInstructionTriviallyDead(I))
-      Pass.DeadInsts.push_back(I);
+      Pass.DeadInsts.insert(I);
   }
 
-  bool rewriteVectorizedLoadInst(IRBuilder<> &IRB, LoadInst &LI, Value *OldOp) {
-    Value *Result;
+  Value *rewriteVectorizedLoadInst(IRBuilder<> &IRB, LoadInst &LI, Value *OldOp) {
+    Value *V = IRB.CreateAlignedLoad(&NewAI, NewAI.getAlignment(),
+                                     getName(".load"));
     if (LI.getType() == VecTy->getElementType() ||
         BeginOffset > NewAllocaBeginOffset || EndOffset < NewAllocaEndOffset) {
-      Result = IRB.CreateExtractElement(
-        IRB.CreateAlignedLoad(&NewAI, NewAI.getAlignment(), getName(".load")),
-        getIndex(IRB, BeginOffset), getName(".extract"));
-    } else {
-      Result = IRB.CreateAlignedLoad(&NewAI, NewAI.getAlignment(),
-                                     getName(".load"));
+      V = IRB.CreateExtractElement(V, getIndex(IRB, BeginOffset),
+                                   getName(".extract"));
     }
-    if (Result->getType() != LI.getType())
-      Result = convertValue(TD, IRB, Result, LI.getType());
-    LI.replaceAllUsesWith(Result);
-    Pass.DeadInsts.push_back(&LI);
-
-    DEBUG(dbgs() << "          to: " << *Result << "\n");
-    return true;
+    return V;
   }
 
-  bool rewriteIntegerLoad(IRBuilder<> &IRB, LoadInst &LI) {
+  Value *rewriteIntegerLoad(IRBuilder<> &IRB, LoadInst &LI) {
     assert(IntTy && "We cannot insert an integer to the alloca");
     assert(!LI.isVolatile());
     Value *V = IRB.CreateAlignedLoad(&NewAI, NewAI.getAlignment(),
@@ -2473,12 +2484,10 @@
     V = convertValue(TD, IRB, V, IntTy);
     assert(BeginOffset >= NewAllocaBeginOffset && "Out of bounds offset");
     uint64_t Offset = BeginOffset - NewAllocaBeginOffset;
-    V = extractInteger(TD, IRB, V, cast<IntegerType>(LI.getType()), Offset,
-                       getName(".extract"));
-    LI.replaceAllUsesWith(V);
-    Pass.DeadInsts.push_back(&LI);
-    DEBUG(dbgs() << "          to: " << *V << "\n");
-    return true;
+    if (Offset > 0 || EndOffset < NewAllocaEndOffset)
+      V = extractInteger(TD, IRB, V, cast<IntegerType>(LI.getType()), Offset,
+                         getName(".extract"));
+    return V;
   }
 
   bool visitLoadInst(LoadInst &LI) {
@@ -2488,7 +2497,46 @@
     IRBuilder<> IRB(&LI);
 
     uint64_t Size = EndOffset - BeginOffset;
-    if (Size < TD.getTypeStoreSize(LI.getType())) {
+    bool IsSplitIntLoad = Size < TD.getTypeStoreSize(LI.getType());
+
+    // If this memory access can be shown to *statically* extend outside the
+    // bounds of the original allocation it's behavior is undefined. Rather
+    // than trying to transform it, just replace it with undef.
+    // FIXME: We should do something more clever for functions being
+    // instrumented by asan.
+    // FIXME: Eventually, once ASan and friends can flush out bugs here, this
+    // should be transformed to a load of null making it unreachable.
+    uint64_t OldAllocSize = TD.getTypeAllocSize(OldAI.getAllocatedType());
+    if (TD.getTypeStoreSize(LI.getType()) > OldAllocSize) {
+      LI.replaceAllUsesWith(UndefValue::get(LI.getType()));
+      Pass.DeadInsts.insert(&LI);
+      deleteIfTriviallyDead(OldOp);
+      DEBUG(dbgs() << "          to: undef!!\n");
+      return true;
+    }
+
+    Type *TargetTy = IsSplitIntLoad ? Type::getIntNTy(LI.getContext(), Size * 8)
+                                    : LI.getType();
+    bool IsPtrAdjusted = false;
+    Value *V;
+    if (VecTy) {
+      V = rewriteVectorizedLoadInst(IRB, LI, OldOp);
+    } else if (IntTy && LI.getType()->isIntegerTy()) {
+      V = rewriteIntegerLoad(IRB, LI);
+    } else if (BeginOffset == NewAllocaBeginOffset &&
+               canConvertValue(TD, NewAllocaTy, LI.getType())) {
+      V = IRB.CreateAlignedLoad(&NewAI, NewAI.getAlignment(),
+                                LI.isVolatile(), getName(".load"));
+    } else {
+      Type *LTy = TargetTy->getPointerTo();
+      V = IRB.CreateAlignedLoad(getAdjustedAllocaPtr(IRB, LTy),
+                                getPartitionTypeAlign(TargetTy),
+                                LI.isVolatile(), getName(".load"));
+      IsPtrAdjusted = true;
+    }
+    V = convertValue(TD, IRB, V, TargetTy);
+
+    if (IsSplitIntLoad) {
       assert(!LI.isVolatile());
       assert(LI.getType()->isIntegerTy() &&
              "Only integer type loads and stores are split");
@@ -2498,21 +2546,8 @@
       assert(LI.getType()->getIntegerBitWidth() ==
              TD.getTypeAllocSizeInBits(OldAI.getAllocatedType()) &&
              "Only alloca-wide loads can be split and recomposed");
-      IntegerType *NarrowTy = Type::getIntNTy(LI.getContext(), Size * 8);
-      bool IsConvertable = (BeginOffset - NewAllocaBeginOffset == 0) &&
-                           canConvertValue(TD, NewAllocaTy, NarrowTy);
-      Value *V;
       // Move the insertion point just past the load so that we can refer to it.
       IRB.SetInsertPoint(llvm::next(BasicBlock::iterator(&LI)));
-      if (IsConvertable)
-        V = convertValue(TD, IRB,
-                         IRB.CreateAlignedLoad(&NewAI, NewAI.getAlignment(),
-                                               getName(".load")),
-                         NarrowTy);
-      else
-        V = IRB.CreateAlignedLoad(
-          getAdjustedAllocaPtr(IRB, NarrowTy->getPointerTo()),
-          getPartitionTypeAlign(NarrowTy), getName(".load"));
       // Create a placeholder value with the same type as LI to use as the
       // basis for the new value. This allows us to replace the uses of LI with
       // the computed value, and then replace the placeholder with LI, leaving
@@ -2524,44 +2559,18 @@
       LI.replaceAllUsesWith(V);
       Placeholder->replaceAllUsesWith(&LI);
       delete Placeholder;
-      if (Pass.DeadSplitInsts.insert(&LI))
-        Pass.DeadInsts.push_back(&LI);
-      DEBUG(dbgs() << "          to: " << *V << "\n");
-      return IsConvertable;
+    } else {
+      LI.replaceAllUsesWith(V);
     }
 
-    if (VecTy)
-      return rewriteVectorizedLoadInst(IRB, LI, OldOp);
-    if (IntTy && LI.getType()->isIntegerTy())
-      return rewriteIntegerLoad(IRB, LI);
-
-    if (BeginOffset == NewAllocaBeginOffset &&
-        canConvertValue(TD, NewAllocaTy, LI.getType())) {
-      Value *NewLI = IRB.CreateAlignedLoad(&NewAI, NewAI.getAlignment(),
-                                           LI.isVolatile(), getName(".load"));
-      Value *NewV = convertValue(TD, IRB, NewLI, LI.getType());
-      LI.replaceAllUsesWith(NewV);
-      Pass.DeadInsts.push_back(&LI);
-
-      DEBUG(dbgs() << "          to: " << *NewLI << "\n");
-      return !LI.isVolatile();
-    }
-
-    assert(!IntTy && "Invalid load found with int-op widening enabled");
-
-    Value *NewPtr = getAdjustedAllocaPtr(IRB,
-                                         LI.getPointerOperand()->getType());
-    LI.setOperand(0, NewPtr);
-    LI.setAlignment(getPartitionTypeAlign(LI.getType()));
-    DEBUG(dbgs() << "          to: " << LI << "\n");
-
+    Pass.DeadInsts.insert(&LI);
     deleteIfTriviallyDead(OldOp);
-    return NewPtr == &NewAI && !LI.isVolatile();
+    DEBUG(dbgs() << "          to: " << *V << "\n");
+    return !LI.isVolatile() && !IsPtrAdjusted;
   }
 
-  bool rewriteVectorizedStoreInst(IRBuilder<> &IRB, StoreInst &SI,
-                                  Value *OldOp) {
-    Value *V = SI.getValueOperand();
+  bool rewriteVectorizedStoreInst(IRBuilder<> &IRB, Value *V,
+                                  StoreInst &SI, Value *OldOp) {
     if (V->getType() == ElementTy ||
         BeginOffset > NewAllocaBeginOffset || EndOffset < NewAllocaEndOffset) {
       if (V->getType() != ElementTy)
@@ -2574,17 +2583,16 @@
       V = convertValue(TD, IRB, V, VecTy);
     }
     StoreInst *Store = IRB.CreateAlignedStore(V, &NewAI, NewAI.getAlignment());
-    Pass.DeadInsts.push_back(&SI);
+    Pass.DeadInsts.insert(&SI);
 
     (void)Store;
     DEBUG(dbgs() << "          to: " << *Store << "\n");
     return true;
   }
 
-  bool rewriteIntegerStore(IRBuilder<> &IRB, StoreInst &SI) {
+  bool rewriteIntegerStore(IRBuilder<> &IRB, Value *V, StoreInst &SI) {
     assert(IntTy && "We cannot extract an integer from the alloca");
     assert(!SI.isVolatile());
-    Value *V = SI.getValueOperand();
     if (TD.getTypeSizeInBits(V->getType()) != IntTy->getBitWidth()) {
       Value *Old = IRB.CreateAlignedLoad(&NewAI, NewAI.getAlignment(),
                                          getName(".oldload"));
@@ -2596,7 +2604,7 @@
     }
     V = convertValue(TD, IRB, V, NewAllocaTy);
     StoreInst *Store = IRB.CreateAlignedStore(V, &NewAI, NewAI.getAlignment());
-    Pass.DeadInsts.push_back(&SI);
+    Pass.DeadInsts.insert(&SI);
     (void)Store;
     DEBUG(dbgs() << "          to: " << *Store << "\n");
     return true;
@@ -2608,74 +2616,53 @@
     assert(OldOp == OldPtr);
     IRBuilder<> IRB(&SI);
 
-    if (VecTy)
-      return rewriteVectorizedStoreInst(IRB, SI, OldOp);
-    Type *ValueTy = SI.getValueOperand()->getType();
+    Value *V = SI.getValueOperand();
+
+    // Strip all inbounds GEPs and pointer casts to try to dig out any root
+    // alloca that should be re-examined after promoting this alloca.
+    if (V->getType()->isPointerTy())
+      if (AllocaInst *AI = dyn_cast<AllocaInst>(V->stripInBoundsOffsets()))
+        Pass.PostPromotionWorklist.insert(AI);
 
     uint64_t Size = EndOffset - BeginOffset;
-    if (Size < TD.getTypeStoreSize(ValueTy)) {
+    if (Size < TD.getTypeStoreSize(V->getType())) {
       assert(!SI.isVolatile());
-      assert(ValueTy->isIntegerTy() &&
+      assert(V->getType()->isIntegerTy() &&
              "Only integer type loads and stores are split");
-      assert(ValueTy->getIntegerBitWidth() ==
-             TD.getTypeStoreSizeInBits(ValueTy) &&
+      assert(V->getType()->getIntegerBitWidth() ==
+             TD.getTypeStoreSizeInBits(V->getType()) &&
              "Non-byte-multiple bit width");
-      assert(ValueTy->getIntegerBitWidth() ==
+      assert(V->getType()->getIntegerBitWidth() ==
              TD.getTypeSizeInBits(OldAI.getAllocatedType()) &&
              "Only alloca-wide stores can be split and recomposed");
       IntegerType *NarrowTy = Type::getIntNTy(SI.getContext(), Size * 8);
-      Value *V = extractInteger(TD, IRB, SI.getValueOperand(), NarrowTy,
-                                BeginOffset, getName(".extract"));
-      StoreInst *NewSI;
-      bool IsConvertable = (BeginOffset - NewAllocaBeginOffset == 0) &&
-                           canConvertValue(TD, NarrowTy, NewAllocaTy);
-      if (IsConvertable)
-        NewSI = IRB.CreateAlignedStore(convertValue(TD, IRB, V, NewAllocaTy),
-                                       &NewAI, NewAI.getAlignment());
-      else
-        NewSI = IRB.CreateAlignedStore(
-          V, getAdjustedAllocaPtr(IRB, NarrowTy->getPointerTo()),
-          getPartitionTypeAlign(NarrowTy));
-      (void)NewSI;
-      if (Pass.DeadSplitInsts.insert(&SI))
-        Pass.DeadInsts.push_back(&SI);
-
-      DEBUG(dbgs() << "          to: " << *NewSI << "\n");
-      return IsConvertable;
+      V = extractInteger(TD, IRB, V, NarrowTy, BeginOffset,
+                         getName(".extract"));
     }
 
-    if (IntTy && ValueTy->isIntegerTy())
-      return rewriteIntegerStore(IRB, SI);
-
-    // Strip all inbounds GEPs and pointer casts to try to dig out any root
-    // alloca that should be re-examined after promoting this alloca.
-    if (ValueTy->isPointerTy())
-      if (AllocaInst *AI = dyn_cast<AllocaInst>(SI.getValueOperand()
-                                                  ->stripInBoundsOffsets()))
-        Pass.PostPromotionWorklist.insert(AI);
+    if (VecTy)
+      return rewriteVectorizedStoreInst(IRB, V, SI, OldOp);
+    if (IntTy && V->getType()->isIntegerTy())
+      return rewriteIntegerStore(IRB, V, SI);
 
+    StoreInst *NewSI;
     if (BeginOffset == NewAllocaBeginOffset &&
-        canConvertValue(TD, ValueTy, NewAllocaTy)) {
-      Value *NewV = convertValue(TD, IRB, SI.getValueOperand(), NewAllocaTy);
-      StoreInst *NewSI = IRB.CreateAlignedStore(NewV, &NewAI, NewAI.getAlignment(),
-                                                SI.isVolatile());
-      (void)NewSI;
-      Pass.DeadInsts.push_back(&SI);
-
-      DEBUG(dbgs() << "          to: " << *NewSI << "\n");
-      return !SI.isVolatile();
-    }
-
-    assert(!IntTy && "Invalid store found with int-op widening enabled");
-
-    Value *NewPtr = getAdjustedAllocaPtr(IRB,
-                                         SI.getPointerOperand()->getType());
-    SI.setOperand(1, NewPtr);
-    SI.setAlignment(getPartitionTypeAlign(SI.getValueOperand()->getType()));
-    DEBUG(dbgs() << "          to: " << SI << "\n");
-
+        canConvertValue(TD, V->getType(), NewAllocaTy)) {
+      V = convertValue(TD, IRB, V, NewAllocaTy);
+      NewSI = IRB.CreateAlignedStore(V, &NewAI, NewAI.getAlignment(),
+                                     SI.isVolatile());
+    } else {
+      Value *NewPtr = getAdjustedAllocaPtr(IRB, V->getType()->getPointerTo());
+      NewSI = IRB.CreateAlignedStore(V, NewPtr,
+                                     getPartitionTypeAlign(V->getType()),
+                                     SI.isVolatile());
+    }
+    (void)NewSI;
+    Pass.DeadInsts.insert(&SI);
     deleteIfTriviallyDead(OldOp);
-    return NewPtr == &NewAI && !SI.isVolatile();
+
+    DEBUG(dbgs() << "          to: " << *NewSI << "\n");
+    return NewSI->getPointerOperand() == &NewAI && !SI.isVolatile();
   }
 
   bool visitMemSetInst(MemSetInst &II) {
@@ -2695,8 +2682,7 @@
     }
 
     // Record this instruction for deletion.
-    if (Pass.DeadSplitInsts.insert(&II))
-      Pass.DeadInsts.push_back(&II);
+    Pass.DeadInsts.insert(&II);
 
     Type *AllocaTy = NewAI.getAllocatedType();
     Type *ScalarTy = AllocaTy->getScalarType();
@@ -2852,8 +2838,7 @@
       return false;
     }
     // Record this instruction for deletion.
-    if (Pass.DeadSplitInsts.insert(&II))
-      Pass.DeadInsts.push_back(&II);
+    Pass.DeadInsts.insert(&II);
 
     bool IsWholeAlloca = BeginOffset == NewAllocaBeginOffset &&
                          EndOffset == NewAllocaEndOffset;
@@ -2963,8 +2948,7 @@
     assert(II.getArgOperand(1) == OldPtr);
 
     // Record this instruction for deletion.
-    if (Pass.DeadSplitInsts.insert(&II))
-      Pass.DeadInsts.push_back(&II);
+    Pass.DeadInsts.insert(&II);
 
     ConstantInt *Size
       = ConstantInt::get(cast<IntegerType>(II.getArgOperand(0)->getType()),
@@ -3533,7 +3517,7 @@
        DI != DE; ++DI) {
     Changed = true;
     (*DI)->replaceAllUsesWith(UndefValue::get((*DI)->getType()));
-    DeadInsts.push_back(*DI);
+    DeadInsts.insert(*DI);
   }
   for (AllocaPartitioning::dead_op_iterator DO = P.dead_op_begin(),
                                             DE = P.dead_op_end();
@@ -3544,7 +3528,7 @@
     if (Instruction *OldI = dyn_cast<Instruction>(OldV))
       if (isInstructionTriviallyDead(OldI)) {
         Changed = true;
-        DeadInsts.push_back(OldI);
+        DeadInsts.insert(OldI);
       }
   }
 
@@ -3565,7 +3549,6 @@
 /// We also record the alloca instructions deleted here so that they aren't
 /// subsequently handed to mem2reg to promote.
 void SROA::deleteDeadInstructions(SmallPtrSet<AllocaInst*, 4> &DeletedAllocas) {
-  DeadSplitInsts.clear();
   while (!DeadInsts.empty()) {
     Instruction *I = DeadInsts.pop_back_val();
     DEBUG(dbgs() << "Deleting dead instruction: " << *I << "\n");
@@ -3577,7 +3560,7 @@
         // Zero out the operand and see if it becomes trivially dead.
         *OI = 0;
         if (isInstructionTriviallyDead(U))
-          DeadInsts.push_back(U);
+          DeadInsts.insert(U);
       }
 
     if (AllocaInst *AI = dyn_cast<AllocaInst>(I))

Modified: llvm/branches/AMDILBackend/lib/Transforms/Scalar/Scalar.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Scalar/Scalar.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Scalar/Scalar.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Scalar/Scalar.cpp Tue Jan 15 11:16:16 2013
@@ -19,7 +19,7 @@
 #include "llvm/PassManager.h"
 #include "llvm/Analysis/Passes.h"
 #include "llvm/Analysis/Verifier.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Transforms/Scalar.h"
 
 using namespace llvm;
@@ -59,6 +59,7 @@
   initializeRegToMemPass(Registry);
   initializeSCCPPass(Registry);
   initializeIPSCCPPass(Registry);
+  initializeSROAPass(Registry);
   initializeSROA_DTPass(Registry);
   initializeSROA_SSAUpPass(Registry);
   initializeCFGSimplifyPassPass(Registry);

Modified: llvm/branches/AMDILBackend/lib/Transforms/Scalar/ScalarReplAggregates.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Scalar/ScalarReplAggregates.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Scalar/ScalarReplAggregates.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Scalar/ScalarReplAggregates.cpp Tue Jan 15 11:16:16 2013
@@ -46,7 +46,7 @@
 #include "llvm/Support/GetElementPtrTypeIterator.h"
 #include "llvm/Support/MathExtras.h"
 #include "llvm/Support/raw_ostream.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Transforms/Utils/Local.h"
 #include "llvm/Transforms/Utils/PromoteMemToReg.h"
 #include "llvm/Transforms/Utils/SSAUpdater.h"
@@ -56,7 +56,6 @@
 STATISTIC(NumPromoted,  "Number of allocas promoted");
 STATISTIC(NumAdjusted,  "Number of scalar allocas adjusted to allow promotion");
 STATISTIC(NumConverted, "Number of aggregates converted to scalar");
-STATISTIC(NumGlobals,   "Number of allocas copied from constant global");
 
 namespace {
   struct SROA : public FunctionPass {
@@ -88,7 +87,7 @@
 
   private:
     bool HasDomTree;
-    TargetData *TD;
+    DataLayout *TD;
 
     /// DeadInsts - Keep track of instructions we have made dead, so that
     /// we can remove them after we are done working.
@@ -183,9 +182,6 @@
     void RewriteLoadUserOfWholeAlloca(LoadInst *LI, AllocaInst *AI,
                                       SmallVector<AllocaInst*, 32> &NewElts);
     bool ShouldAttemptScalarRepl(AllocaInst *AI);
-
-    static MemTransferInst *isOnlyCopiedFromConstantGlobal(
-        AllocaInst *AI, SmallVector<Instruction*, 4> &ToDelete);
   };
 
   // SROA_DT - SROA that uses DominatorTree.
@@ -262,7 +258,7 @@
 class ConvertToScalarInfo {
   /// AllocaSize - The size of the alloca being considered in bytes.
   unsigned AllocaSize;
-  const TargetData &TD;
+  const DataLayout &TD;
   unsigned ScalarLoadThreshold;
 
   /// IsNotTrivial - This is set to true if there is some access to the object
@@ -305,7 +301,7 @@
   bool HadDynamicAccess;
 
 public:
-  explicit ConvertToScalarInfo(unsigned Size, const TargetData &td,
+  explicit ConvertToScalarInfo(unsigned Size, const DataLayout &td,
                                unsigned SLT)
     : AllocaSize(Size), TD(td), ScalarLoadThreshold(SLT), IsNotTrivial(false),
     ScalarKind(Unknown), VectorTy(0), HadNonMemTransferAccess(false),
@@ -1024,11 +1020,11 @@
 
 
 bool SROA::runOnFunction(Function &F) {
-  TD = getAnalysisIfAvailable<TargetData>();
+  TD = getAnalysisIfAvailable<DataLayout>();
 
   bool Changed = performPromotion(F);
 
-  // FIXME: ScalarRepl currently depends on TargetData more than it
+  // FIXME: ScalarRepl currently depends on DataLayout more than it
   // theoretically needs to. It should be refactored in order to support
   // target-independent IR. Until this is done, just skip the actual
   // scalar-replacement portion of this pass.
@@ -1138,7 +1134,7 @@
 ///
 /// We can do this to a select if its only uses are loads and if the operand to
 /// the select can be loaded unconditionally.
-static bool isSafeSelectToSpeculate(SelectInst *SI, const TargetData *TD) {
+static bool isSafeSelectToSpeculate(SelectInst *SI, const DataLayout *TD) {
   bool TDerefable = SI->getTrueValue()->isDereferenceablePointer();
   bool FDerefable = SI->getFalseValue()->isDereferenceablePointer();
 
@@ -1176,7 +1172,7 @@
 ///
 /// We can do this to a select if its only uses are loads and if the operand to
 /// the select can be loaded unconditionally.
-static bool isSafePHIToSpeculate(PHINode *PN, const TargetData *TD) {
+static bool isSafePHIToSpeculate(PHINode *PN, const DataLayout *TD) {
   // For now, we can only do this promotion if the load is in the same block as
   // the PHI, and if there are no stores between the phi and load.
   // TODO: Allow recursive phi users.
@@ -1240,7 +1236,7 @@
 /// direct (non-volatile) loads and stores to it.  If the alloca is close but
 /// not quite there, this will transform the code to allow promotion.  As such,
 /// it is a non-pure predicate.
-static bool tryToMakeAllocaBePromotable(AllocaInst *AI, const TargetData *TD) {
+static bool tryToMakeAllocaBePromotable(AllocaInst *AI, const DataLayout *TD) {
   SetVector<Instruction*, SmallVector<Instruction*, 4>,
             SmallPtrSet<Instruction*, 4> > InstsToRewrite;
 
@@ -1465,26 +1461,6 @@
   return false;
 }
 
-/// getPointeeAlignment - Compute the minimum alignment of the value pointed
-/// to by the given pointer.
-static unsigned getPointeeAlignment(Value *V, const TargetData &TD) {
-  if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V))
-    if (CE->getOpcode() == Instruction::BitCast ||
-        (CE->getOpcode() == Instruction::GetElementPtr &&
-         cast<GEPOperator>(CE)->hasAllZeroIndices()))
-      return getPointeeAlignment(CE->getOperand(0), TD);
-
-  if (GlobalVariable *GV = dyn_cast<GlobalVariable>(V))
-    if (!GV->isDeclaration())
-      return TD.getPreferredAlignment(GV);
-
-  if (PointerType *PT = dyn_cast<PointerType>(V->getType()))
-    return TD.getABITypeAlignment(PT->getElementType());
-
-  return 0;
-}
-
-
 // performScalarRepl - This algorithm is a simple worklist driven algorithm,
 // which runs on all of the alloca instructions in the function, removing them
 // if they are only used by getelementptr instructions.
@@ -1516,29 +1492,6 @@
     if (AI->isArrayAllocation() || !AI->getAllocatedType()->isSized())
       continue;
 
-    // Check to see if this allocation is only modified by a memcpy/memmove from
-    // a constant global whose alignment is equal to or exceeds that of the
-    // allocation.  If this is the case, we can change all users to use
-    // the constant global instead.  This is commonly produced by the CFE by
-    // constructs like "void foo() { int A[] = {1,2,3,4,5,6,7,8,9...}; }" if 'A'
-    // is only subsequently read.
-    SmallVector<Instruction *, 4> ToDelete;
-    if (MemTransferInst *Copy = isOnlyCopiedFromConstantGlobal(AI, ToDelete)) {
-      if (AI->getAlignment() <= getPointeeAlignment(Copy->getSource(), *TD)) {
-        DEBUG(dbgs() << "Found alloca equal to global: " << *AI << '\n');
-        DEBUG(dbgs() << "  memcpy = " << *Copy << '\n');
-        for (unsigned i = 0, e = ToDelete.size(); i != e; ++i)
-          ToDelete[i]->eraseFromParent();
-        Constant *TheSrc = cast<Constant>(Copy->getSource());
-        AI->replaceAllUsesWith(ConstantExpr::getBitCast(TheSrc, AI->getType()));
-        Copy->eraseFromParent();  // Don't mutate the global.
-        AI->eraseFromParent();
-        ++NumGlobals;
-        Changed = true;
-        continue;
-      }
-    }
-
     // Check to see if we can perform the core SROA transformation.  We cannot
     // transform the allocation instruction if it is an array allocation
     // (allocations OF arrays are ok though), and an allocation of a scalar
@@ -2584,7 +2537,7 @@
 /// HasPadding - Return true if the specified type has any structure or
 /// alignment padding in between the elements that would be split apart
 /// by SROA; return false otherwise.
-static bool HasPadding(Type *Ty, const TargetData &TD) {
+static bool HasPadding(Type *Ty, const DataLayout &TD) {
   if (ArrayType *ATy = dyn_cast<ArrayType>(Ty)) {
     Ty = ATy->getElementType();
     return TD.getTypeSizeInBits(Ty) != TD.getTypeAllocSizeInBits(Ty);
@@ -2656,134 +2609,3 @@
 
   return true;
 }
-
-
-
-/// PointsToConstantGlobal - Return true if V (possibly indirectly) points to
-/// some part of a constant global variable.  This intentionally only accepts
-/// constant expressions because we don't can't rewrite arbitrary instructions.
-static bool PointsToConstantGlobal(Value *V) {
-  if (GlobalVariable *GV = dyn_cast<GlobalVariable>(V))
-    return GV->isConstant();
-  if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V))
-    if (CE->getOpcode() == Instruction::BitCast ||
-        CE->getOpcode() == Instruction::GetElementPtr)
-      return PointsToConstantGlobal(CE->getOperand(0));
-  return false;
-}
-
-/// isOnlyCopiedFromConstantGlobal - Recursively walk the uses of a (derived)
-/// pointer to an alloca.  Ignore any reads of the pointer, return false if we
-/// see any stores or other unknown uses.  If we see pointer arithmetic, keep
-/// track of whether it moves the pointer (with isOffset) but otherwise traverse
-/// the uses.  If we see a memcpy/memmove that targets an unoffseted pointer to
-/// the alloca, and if the source pointer is a pointer to a constant global, we
-/// can optimize this.
-static bool
-isOnlyCopiedFromConstantGlobal(Value *V, MemTransferInst *&TheCopy,
-                               bool isOffset,
-                               SmallVector<Instruction *, 4> &LifetimeMarkers) {
-  // We track lifetime intrinsics as we encounter them.  If we decide to go
-  // ahead and replace the value with the global, this lets the caller quickly
-  // eliminate the markers.
-
-  for (Value::use_iterator UI = V->use_begin(), E = V->use_end(); UI!=E; ++UI) {
-    User *U = cast<Instruction>(*UI);
-
-    if (LoadInst *LI = dyn_cast<LoadInst>(U)) {
-      // Ignore non-volatile loads, they are always ok.
-      if (!LI->isSimple()) return false;
-      continue;
-    }
-
-    if (BitCastInst *BCI = dyn_cast<BitCastInst>(U)) {
-      // If uses of the bitcast are ok, we are ok.
-      if (!isOnlyCopiedFromConstantGlobal(BCI, TheCopy, isOffset,
-                                          LifetimeMarkers))
-        return false;
-      continue;
-    }
-    if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(U)) {
-      // If the GEP has all zero indices, it doesn't offset the pointer.  If it
-      // doesn't, it does.
-      if (!isOnlyCopiedFromConstantGlobal(GEP, TheCopy,
-                                          isOffset || !GEP->hasAllZeroIndices(),
-                                          LifetimeMarkers))
-        return false;
-      continue;
-    }
-
-    if (CallSite CS = U) {
-      // If this is the function being called then we treat it like a load and
-      // ignore it.
-      if (CS.isCallee(UI))
-        continue;
-
-      // If this is a readonly/readnone call site, then we know it is just a
-      // load (but one that potentially returns the value itself), so we can
-      // ignore it if we know that the value isn't captured.
-      unsigned ArgNo = CS.getArgumentNo(UI);
-      if (CS.onlyReadsMemory() &&
-          (CS.getInstruction()->use_empty() || CS.doesNotCapture(ArgNo)))
-        continue;
-
-      // If this is being passed as a byval argument, the caller is making a
-      // copy, so it is only a read of the alloca.
-      if (CS.isByValArgument(ArgNo))
-        continue;
-    }
-
-    // Lifetime intrinsics can be handled by the caller.
-    if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(U)) {
-      if (II->getIntrinsicID() == Intrinsic::lifetime_start ||
-          II->getIntrinsicID() == Intrinsic::lifetime_end) {
-        assert(II->use_empty() && "Lifetime markers have no result to use!");
-        LifetimeMarkers.push_back(II);
-        continue;
-      }
-    }
-
-    // If this is isn't our memcpy/memmove, reject it as something we can't
-    // handle.
-    MemTransferInst *MI = dyn_cast<MemTransferInst>(U);
-    if (MI == 0)
-      return false;
-
-    // If the transfer is using the alloca as a source of the transfer, then
-    // ignore it since it is a load (unless the transfer is volatile).
-    if (UI.getOperandNo() == 1) {
-      if (MI->isVolatile()) return false;
-      continue;
-    }
-
-    // If we already have seen a copy, reject the second one.
-    if (TheCopy) return false;
-
-    // If the pointer has been offset from the start of the alloca, we can't
-    // safely handle this.
-    if (isOffset) return false;
-
-    // If the memintrinsic isn't using the alloca as the dest, reject it.
-    if (UI.getOperandNo() != 0) return false;
-
-    // If the source of the memcpy/move is not a constant global, reject it.
-    if (!PointsToConstantGlobal(MI->getSource()))
-      return false;
-
-    // Otherwise, the transform is safe.  Remember the copy instruction.
-    TheCopy = MI;
-  }
-  return true;
-}
-
-/// isOnlyCopiedFromConstantGlobal - Return true if the specified alloca is only
-/// modified by a copy from a constant global.  If we can prove this, we can
-/// replace any uses of the alloca with uses of the global directly.
-MemTransferInst *
-SROA::isOnlyCopiedFromConstantGlobal(AllocaInst *AI,
-                                     SmallVector<Instruction*, 4> &ToDelete) {
-  MemTransferInst *TheCopy = 0;
-  if (::isOnlyCopiedFromConstantGlobal(AI, TheCopy, false, ToDelete))
-    return TheCopy;
-  return 0;
-}

Modified: llvm/branches/AMDILBackend/lib/Transforms/Scalar/SimplifyCFGPass.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Scalar/SimplifyCFGPass.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Scalar/SimplifyCFGPass.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Scalar/SimplifyCFGPass.cpp Tue Jan 15 11:16:16 2013
@@ -31,10 +31,11 @@
 #include "llvm/Attributes.h"
 #include "llvm/Support/CFG.h"
 #include "llvm/Pass.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/SmallPtrSet.h"
 #include "llvm/ADT/Statistic.h"
+#include "llvm/TargetTransformInfo.h"
 using namespace llvm;
 
 STATISTIC(NumSimpl, "Number of blocks simplified");
@@ -59,9 +60,9 @@
   return new CFGSimplifyPass();
 }
 
-/// ChangeToUnreachable - Insert an unreachable instruction before the specified
+/// changeToUnreachable - Insert an unreachable instruction before the specified
 /// instruction, making it and the rest of the code in the block dead.
-static void ChangeToUnreachable(Instruction *I, bool UseLLVMTrap) {
+static void changeToUnreachable(Instruction *I, bool UseLLVMTrap) {
   BasicBlock *BB = I->getParent();
   // Loop over all of the successors, removing BB's entry from any PHI
   // nodes.
@@ -87,8 +88,8 @@
   }
 }
 
-/// ChangeToCall - Convert the specified invoke into a normal call.
-static void ChangeToCall(InvokeInst *II) {
+/// changeToCall - Convert the specified invoke into a normal call.
+static void changeToCall(InvokeInst *II) {
   SmallVector<Value*, 8> Args(II->op_begin(), II->op_end() - 3);
   CallInst *NewCall = CallInst::Create(II->getCalledValue(), Args, "", II);
   NewCall->takeName(II);
@@ -105,7 +106,7 @@
   II->eraseFromParent();
 }
 
-static bool MarkAliveBlocks(BasicBlock *BB,
+static bool markAliveBlocks(BasicBlock *BB,
                             SmallPtrSet<BasicBlock*, 128> &Reachable) {
 
   SmallVector<BasicBlock*, 128> Worklist;
@@ -129,7 +130,7 @@
           ++BBI;
           if (!isa<UnreachableInst>(BBI)) {
             // Don't insert a call to llvm.trap right before the unreachable.
-            ChangeToUnreachable(BBI, false);
+            changeToUnreachable(BBI, false);
             Changed = true;
           }
           break;
@@ -148,7 +149,7 @@
         if (isa<UndefValue>(Ptr) ||
             (isa<ConstantPointerNull>(Ptr) &&
              SI->getPointerAddressSpace() == 0)) {
-          ChangeToUnreachable(SI, true);
+          changeToUnreachable(SI, true);
           Changed = true;
           break;
         }
@@ -159,7 +160,7 @@
     if (InvokeInst *II = dyn_cast<InvokeInst>(BB->getTerminator())) {
       Value *Callee = II->getCalledValue();
       if (isa<ConstantPointerNull>(Callee) || isa<UndefValue>(Callee)) {
-        ChangeToUnreachable(II, true);
+        changeToUnreachable(II, true);
         Changed = true;
       } else if (II->doesNotThrow()) {
         if (II->use_empty() && II->onlyReadsMemory()) {
@@ -168,7 +169,7 @@
           II->getUnwindDest()->removePredecessor(II->getParent());
           II->eraseFromParent();
         } else
-          ChangeToCall(II);
+          changeToCall(II);
         Changed = true;
       }
     }
@@ -180,12 +181,12 @@
   return Changed;
 }
 
-/// RemoveUnreachableBlocksFromFn - Remove blocks that are not reachable, even
+/// removeUnreachableBlocksFromFn - Remove blocks that are not reachable, even
 /// if they are in a dead cycle.  Return true if a change was made, false
 /// otherwise.
-static bool RemoveUnreachableBlocksFromFn(Function &F) {
+static bool removeUnreachableBlocksFromFn(Function &F) {
   SmallPtrSet<BasicBlock*, 128> Reachable;
-  bool Changed = MarkAliveBlocks(F.begin(), Reachable);
+  bool Changed = markAliveBlocks(F.begin(), Reachable);
 
   // If there are unreachable blocks in the CFG...
   if (Reachable.size() == F.size())
@@ -215,9 +216,9 @@
   return true;
 }
 
-/// MergeEmptyReturnBlocks - If we have more than one empty (other than phi
+/// mergeEmptyReturnBlocks - If we have more than one empty (other than phi
 /// node) return blocks, merge them together to promote recursive block merging.
-static bool MergeEmptyReturnBlocks(Function &F) {
+static bool mergeEmptyReturnBlocks(Function &F) {
   bool Changed = false;
 
   BasicBlock *RetBlock = 0;
@@ -291,9 +292,10 @@
   return Changed;
 }
 
-/// IterativeSimplifyCFG - Call SimplifyCFG on all the blocks in the function,
+/// iterativelySimplifyCFG - Call SimplifyCFG on all the blocks in the function,
 /// iterating until no more changes are made.
-static bool IterativeSimplifyCFG(Function &F, const TargetData *TD) {
+static bool iterativelySimplifyCFG(Function &F, const DataLayout *TD,
+                                   const TargetTransformInfo *TTI) {
   bool Changed = false;
   bool LocalChange = true;
   while (LocalChange) {
@@ -302,7 +304,7 @@
     // Loop over all of the basic blocks and remove them if they are unneeded...
     //
     for (Function::iterator BBIt = F.begin(); BBIt != F.end(); ) {
-      if (SimplifyCFG(BBIt++, TD)) {
+      if (SimplifyCFG(BBIt++, TD, TTI)) {
         LocalChange = true;
         ++NumSimpl;
       }
@@ -316,25 +318,27 @@
 // simplify the CFG.
 //
 bool CFGSimplifyPass::runOnFunction(Function &F) {
-  const TargetData *TD = getAnalysisIfAvailable<TargetData>();
-  bool EverChanged = RemoveUnreachableBlocksFromFn(F);
-  EverChanged |= MergeEmptyReturnBlocks(F);
-  EverChanged |= IterativeSimplifyCFG(F, TD);
+  const DataLayout *TD = getAnalysisIfAvailable<DataLayout>();
+  const TargetTransformInfo *TTI =
+      getAnalysisIfAvailable<TargetTransformInfo>();
+  bool EverChanged = removeUnreachableBlocksFromFn(F);
+  EverChanged |= mergeEmptyReturnBlocks(F);
+  EverChanged |= iterativelySimplifyCFG(F, TD, TTI);
 
   // If neither pass changed anything, we're done.
   if (!EverChanged) return false;
 
-  // IterativeSimplifyCFG can (rarely) make some loops dead.  If this happens,
-  // RemoveUnreachableBlocksFromFn is needed to nuke them, which means we should
+  // iterativelySimplifyCFG can (rarely) make some loops dead.  If this happens,
+  // removeUnreachableBlocksFromFn is needed to nuke them, which means we should
   // iterate between the two optimizations.  We structure the code like this to
-  // avoid reruning IterativeSimplifyCFG if the second pass of
-  // RemoveUnreachableBlocksFromFn doesn't do anything.
-  if (!RemoveUnreachableBlocksFromFn(F))
+  // avoid reruning iterativelySimplifyCFG if the second pass of
+  // removeUnreachableBlocksFromFn doesn't do anything.
+  if (!removeUnreachableBlocksFromFn(F))
     return true;
 
   do {
-    EverChanged = IterativeSimplifyCFG(F, TD);
-    EverChanged |= RemoveUnreachableBlocksFromFn(F);
+    EverChanged = iterativelySimplifyCFG(F, TD, TTI);
+    EverChanged |= removeUnreachableBlocksFromFn(F);
   } while (EverChanged);
 
   return true;

Modified: llvm/branches/AMDILBackend/lib/Transforms/Scalar/SimplifyLibCalls.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Scalar/SimplifyLibCalls.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Scalar/SimplifyLibCalls.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Scalar/SimplifyLibCalls.cpp Tue Jan 15 11:16:16 2013
@@ -28,9 +28,10 @@
 #include "llvm/ADT/Statistic.h"
 #include "llvm/ADT/StringMap.h"
 #include "llvm/Analysis/ValueTracking.h"
+#include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/raw_ostream.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Target/TargetLibraryInfo.h"
 #include "llvm/Config/config.h"            // FIXME: Shouldn't depend on host!
 using namespace llvm;
@@ -38,6 +39,10 @@
 STATISTIC(NumSimplified, "Number of library calls simplified");
 STATISTIC(NumAnnotated, "Number of attributes added to library functions");
 
+static cl::opt<bool> UnsafeFPShrink("enable-double-float-shrink", cl::Hidden,
+                                   cl::init(false),
+                                   cl::desc("Enable unsafe double to float "
+                                            "shrinking for math lib calls"));
 //===----------------------------------------------------------------------===//
 // Optimizer Base Class
 //===----------------------------------------------------------------------===//
@@ -48,7 +53,7 @@
 class LibCallOptimization {
 protected:
   Function *Caller;
-  const TargetData *TD;
+  const DataLayout *TD;
   const TargetLibraryInfo *TLI;
   LLVMContext* Context;
 public:
@@ -63,7 +68,7 @@
   virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B)
     =0;
 
-  Value *OptimizeCall(CallInst *CI, const TargetData *TD,
+  Value *OptimizeCall(CallInst *CI, const DataLayout *TD,
                       const TargetLibraryInfo *TLI, IRBuilder<> &B) {
     Caller = CI->getParent()->getParent();
     this->TD = TD;
@@ -85,22 +90,6 @@
 // Helper Functions
 //===----------------------------------------------------------------------===//
 
-/// 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();
-       UI != E; ++UI) {
-    if (ICmpInst *IC = dyn_cast<ICmpInst>(*UI))
-      if (IC->isEquality())
-        if (Constant *C = dyn_cast<Constant>(IC->getOperand(1)))
-          if (C->isNullValue())
-            continue;
-    // Unknown instruction.
-    return false;
-  }
-  return true;
-}
-
 static bool CallHasFloatingPointArgument(const CallInst *CI) {
   for (CallInst::const_op_iterator it = CI->op_begin(), e = CI->op_end();
        it != e; ++it) {
@@ -110,799 +99,62 @@
   return false;
 }
 
-/// IsOnlyUsedInEqualityComparison - Return true if it is only used in equality
-/// comparisons with With.
-static bool IsOnlyUsedInEqualityComparison(Value *V, Value *With) {
-  for (Value::use_iterator UI = V->use_begin(), E = V->use_end();
-       UI != E; ++UI) {
-    if (ICmpInst *IC = dyn_cast<ICmpInst>(*UI))
-      if (IC->isEquality() && IC->getOperand(1) == With)
-        continue;
-    // Unknown instruction.
-    return false;
-  }
-  return true;
-}
-
+namespace {
 //===----------------------------------------------------------------------===//
-// String and Memory LibCall Optimizations
+// Math Library Optimizations
 //===----------------------------------------------------------------------===//
 
 //===---------------------------------------===//
-// 'strcat' Optimizations
-namespace {
-struct StrCatOpt : public LibCallOptimization {
-  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
-    // Verify the "strcat" function prototype.
-    FunctionType *FT = Callee->getFunctionType();
-    if (FT->getNumParams() != 2 ||
-        FT->getReturnType() != B.getInt8PtrTy() ||
-        FT->getParamType(0) != FT->getReturnType() ||
-        FT->getParamType(1) != FT->getReturnType())
-      return 0;
-
-    // Extract some information from the instruction
-    Value *Dst = CI->getArgOperand(0);
-    Value *Src = CI->getArgOperand(1);
-
-    // See if we can get the length of the input string.
-    uint64_t Len = GetStringLength(Src);
-    if (Len == 0) return 0;
-    --Len;  // Unbias length.
-
-    // Handle the simple, do-nothing case: strcat(x, "") -> x
-    if (Len == 0)
-      return Dst;
-
-    // These optimizations require TargetData.
-    if (!TD) return 0;
-
-    return EmitStrLenMemCpy(Src, Dst, Len, B);
-  }
-
-  Value *EmitStrLenMemCpy(Value *Src, Value *Dst, uint64_t Len, IRBuilder<> &B) {
-    // We need to find the end of the destination string.  That's where the
-    // memory is to be moved to. We just generate a call to strlen.
-    Value *DstLen = EmitStrLen(Dst, B, TD, TLI);
-    if (!DstLen)
-      return 0;
-
-    // Now that we have the destination's length, we must index into the
-    // destination's pointer to get the actual memcpy destination (end of
-    // the string .. we're concatenating).
-    Value *CpyDst = B.CreateGEP(Dst, DstLen, "endptr");
-
-    // We have enough information to now generate the memcpy call to do the
-    // concatenation for us.  Make a memcpy to copy the nul byte with align = 1.
-    B.CreateMemCpy(CpyDst, Src,
-                   ConstantInt::get(TD->getIntPtrType(*Context), Len + 1), 1);
-    return Dst;
-  }
-};
-
-//===---------------------------------------===//
-// 'strncat' Optimizations
-
-struct StrNCatOpt : public StrCatOpt {
-  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
-    // Verify the "strncat" function prototype.
-    FunctionType *FT = Callee->getFunctionType();
-    if (FT->getNumParams() != 3 ||
-        FT->getReturnType() != B.getInt8PtrTy() ||
-        FT->getParamType(0) != FT->getReturnType() ||
-        FT->getParamType(1) != FT->getReturnType() ||
-        !FT->getParamType(2)->isIntegerTy())
-      return 0;
-
-    // Extract some information from the instruction
-    Value *Dst = CI->getArgOperand(0);
-    Value *Src = CI->getArgOperand(1);
-    uint64_t Len;
-
-    // We don't do anything if length is not constant
-    if (ConstantInt *LengthArg = dyn_cast<ConstantInt>(CI->getArgOperand(2)))
-      Len = LengthArg->getZExtValue();
-    else
-      return 0;
-
-    // See if we can get the length of the input string.
-    uint64_t SrcLen = GetStringLength(Src);
-    if (SrcLen == 0) return 0;
-    --SrcLen;  // Unbias length.
-
-    // Handle the simple, do-nothing cases:
-    // strncat(x, "", c) -> x
-    // strncat(x,  c, 0) -> x
-    if (SrcLen == 0 || Len == 0) return Dst;
-
-    // These optimizations require TargetData.
-    if (!TD) return 0;
-
-    // We don't optimize this case
-    if (Len < SrcLen) return 0;
-
-    // strncat(x, s, c) -> strcat(x, s)
-    // s is constant so the strcat can be optimized further
-    return EmitStrLenMemCpy(Src, Dst, SrcLen, B);
-  }
-};
-
-//===---------------------------------------===//
-// 'strchr' Optimizations
-
-struct StrChrOpt : public LibCallOptimization {
-  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
-    // Verify the "strchr" function prototype.
-    FunctionType *FT = Callee->getFunctionType();
-    if (FT->getNumParams() != 2 ||
-        FT->getReturnType() != B.getInt8PtrTy() ||
-        FT->getParamType(0) != FT->getReturnType() ||
-        !FT->getParamType(1)->isIntegerTy(32))
-      return 0;
-
-    Value *SrcStr = CI->getArgOperand(0);
-
-    // If the second operand is non-constant, see if we can compute the length
-    // of the input string and turn this into memchr.
-    ConstantInt *CharC = dyn_cast<ConstantInt>(CI->getArgOperand(1));
-    if (CharC == 0) {
-      // These optimizations require TargetData.
-      if (!TD) return 0;
-
-      uint64_t Len = GetStringLength(SrcStr);
-      if (Len == 0 || !FT->getParamType(1)->isIntegerTy(32))// memchr needs i32.
-        return 0;
-
-      return EmitMemChr(SrcStr, CI->getArgOperand(1), // include nul.
-                        ConstantInt::get(TD->getIntPtrType(*Context), Len),
-                        B, TD, TLI);
-    }
-
-    // Otherwise, the character is a constant, see if the first argument is
-    // a string literal.  If so, we can constant fold.
-    StringRef Str;
-    if (!getConstantStringInfo(SrcStr, Str))
-      return 0;
-
-    // Compute the offset, make sure to handle the case when we're searching for
-    // zero (a weird way to spell strlen).
-    size_t I = CharC->getSExtValue() == 0 ?
-        Str.size() : Str.find(CharC->getSExtValue());
-    if (I == StringRef::npos) // Didn't find the char.  strchr returns null.
-      return Constant::getNullValue(CI->getType());
-
-    // strchr(s+n,c)  -> gep(s+n+i,c)
-    return B.CreateGEP(SrcStr, B.getInt64(I), "strchr");
-  }
-};
-
-//===---------------------------------------===//
-// 'strrchr' Optimizations
-
-struct StrRChrOpt : public LibCallOptimization {
-  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
-    // Verify the "strrchr" function prototype.
-    FunctionType *FT = Callee->getFunctionType();
-    if (FT->getNumParams() != 2 ||
-        FT->getReturnType() != B.getInt8PtrTy() ||
-        FT->getParamType(0) != FT->getReturnType() ||
-        !FT->getParamType(1)->isIntegerTy(32))
-      return 0;
-
-    Value *SrcStr = CI->getArgOperand(0);
-    ConstantInt *CharC = dyn_cast<ConstantInt>(CI->getArgOperand(1));
-
-    // Cannot fold anything if we're not looking for a constant.
-    if (!CharC)
-      return 0;
-
-    StringRef Str;
-    if (!getConstantStringInfo(SrcStr, Str)) {
-      // strrchr(s, 0) -> strchr(s, 0)
-      if (TD && CharC->isZero())
-        return EmitStrChr(SrcStr, '\0', B, TD, TLI);
-      return 0;
-    }
-
-    // Compute the offset.
-    size_t I = CharC->getSExtValue() == 0 ?
-        Str.size() : Str.rfind(CharC->getSExtValue());
-    if (I == StringRef::npos) // Didn't find the char. Return null.
-      return Constant::getNullValue(CI->getType());
-
-    // strrchr(s+n,c) -> gep(s+n+i,c)
-    return B.CreateGEP(SrcStr, B.getInt64(I), "strrchr");
-  }
-};
-
-//===---------------------------------------===//
-// 'strcmp' Optimizations
-
-struct StrCmpOpt : public LibCallOptimization {
-  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
-    // Verify the "strcmp" function prototype.
-    FunctionType *FT = Callee->getFunctionType();
-    if (FT->getNumParams() != 2 ||
-        !FT->getReturnType()->isIntegerTy(32) ||
-        FT->getParamType(0) != FT->getParamType(1) ||
-        FT->getParamType(0) != B.getInt8PtrTy())
-      return 0;
-
-    Value *Str1P = CI->getArgOperand(0), *Str2P = CI->getArgOperand(1);
-    if (Str1P == Str2P)      // strcmp(x,x)  -> 0
-      return ConstantInt::get(CI->getType(), 0);
-
-    StringRef Str1, Str2;
-    bool HasStr1 = getConstantStringInfo(Str1P, Str1);
-    bool HasStr2 = getConstantStringInfo(Str2P, Str2);
-
-    // strcmp(x, y)  -> cnst  (if both x and y are constant strings)
-    if (HasStr1 && HasStr2)
-      return ConstantInt::get(CI->getType(), Str1.compare(Str2));
-
-    if (HasStr1 && Str1.empty()) // strcmp("", x) -> -*x
-      return B.CreateNeg(B.CreateZExt(B.CreateLoad(Str2P, "strcmpload"),
-                                      CI->getType()));
-
-    if (HasStr2 && Str2.empty()) // strcmp(x,"") -> *x
-      return B.CreateZExt(B.CreateLoad(Str1P, "strcmpload"), CI->getType());
-
-    // strcmp(P, "x") -> memcmp(P, "x", 2)
-    uint64_t Len1 = GetStringLength(Str1P);
-    uint64_t Len2 = GetStringLength(Str2P);
-    if (Len1 && Len2) {
-      // These optimizations require TargetData.
-      if (!TD) return 0;
-
-      return EmitMemCmp(Str1P, Str2P,
-                        ConstantInt::get(TD->getIntPtrType(*Context),
-                        std::min(Len1, Len2)), B, TD, TLI);
-    }
-
-    return 0;
-  }
-};
-
-//===---------------------------------------===//
-// 'strncmp' Optimizations
-
-struct StrNCmpOpt : public LibCallOptimization {
-  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
-    // Verify the "strncmp" function prototype.
-    FunctionType *FT = Callee->getFunctionType();
-    if (FT->getNumParams() != 3 ||
-        !FT->getReturnType()->isIntegerTy(32) ||
-        FT->getParamType(0) != FT->getParamType(1) ||
-        FT->getParamType(0) != B.getInt8PtrTy() ||
-        !FT->getParamType(2)->isIntegerTy())
-      return 0;
-
-    Value *Str1P = CI->getArgOperand(0), *Str2P = CI->getArgOperand(1);
-    if (Str1P == Str2P)      // strncmp(x,x,n)  -> 0
-      return ConstantInt::get(CI->getType(), 0);
-
-    // Get the length argument if it is constant.
-    uint64_t Length;
-    if (ConstantInt *LengthArg = dyn_cast<ConstantInt>(CI->getArgOperand(2)))
-      Length = LengthArg->getZExtValue();
-    else
-      return 0;
-
-    if (Length == 0) // strncmp(x,y,0)   -> 0
-      return ConstantInt::get(CI->getType(), 0);
-
-    if (TD && Length == 1) // strncmp(x,y,1) -> memcmp(x,y,1)
-      return EmitMemCmp(Str1P, Str2P, CI->getArgOperand(2), B, TD, TLI);
-
-    StringRef Str1, Str2;
-    bool HasStr1 = getConstantStringInfo(Str1P, Str1);
-    bool HasStr2 = getConstantStringInfo(Str2P, Str2);
-
-    // strncmp(x, y)  -> cnst  (if both x and y are constant strings)
-    if (HasStr1 && HasStr2) {
-      StringRef SubStr1 = Str1.substr(0, Length);
-      StringRef SubStr2 = Str2.substr(0, Length);
-      return ConstantInt::get(CI->getType(), SubStr1.compare(SubStr2));
-    }
-
-    if (HasStr1 && Str1.empty())  // strncmp("", x, n) -> -*x
-      return B.CreateNeg(B.CreateZExt(B.CreateLoad(Str2P, "strcmpload"),
-                                      CI->getType()));
-
-    if (HasStr2 && Str2.empty())  // strncmp(x, "", n) -> *x
-      return B.CreateZExt(B.CreateLoad(Str1P, "strcmpload"), CI->getType());
-
-    return 0;
-  }
-};
-
-
-//===---------------------------------------===//
-// 'strcpy' Optimizations
-
-struct StrCpyOpt : public LibCallOptimization {
-  bool OptChkCall;  // True if it's optimizing a __strcpy_chk libcall.
-
-  StrCpyOpt(bool c) : OptChkCall(c) {}
-
-  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
-    // Verify the "strcpy" function prototype.
-    unsigned NumParams = OptChkCall ? 3 : 2;
-    FunctionType *FT = Callee->getFunctionType();
-    if (FT->getNumParams() != NumParams ||
-        FT->getReturnType() != FT->getParamType(0) ||
-        FT->getParamType(0) != FT->getParamType(1) ||
-        FT->getParamType(0) != B.getInt8PtrTy())
-      return 0;
-
-    Value *Dst = CI->getArgOperand(0), *Src = CI->getArgOperand(1);
-    if (Dst == Src)      // strcpy(x,x)  -> x
-      return Src;
-
-    // These optimizations require TargetData.
-    if (!TD) return 0;
-
-    // See if we can get the length of the input string.
-    uint64_t Len = GetStringLength(Src);
-    if (Len == 0) return 0;
-
-    // We have enough information to now generate the memcpy call to do the
-    // concatenation for us.  Make a memcpy to copy the nul byte with align = 1.
-    if (!OptChkCall ||
-        !EmitMemCpyChk(Dst, Src,
-                       ConstantInt::get(TD->getIntPtrType(*Context), Len),
-                       CI->getArgOperand(2), B, TD, TLI))
-      B.CreateMemCpy(Dst, Src,
-                     ConstantInt::get(TD->getIntPtrType(*Context), Len), 1);
-    return Dst;
-  }
-};
-
-//===---------------------------------------===//
-// 'stpcpy' Optimizations
-
-struct StpCpyOpt: public LibCallOptimization {
-  bool OptChkCall;  // True if it's optimizing a __stpcpy_chk libcall.
-
-  StpCpyOpt(bool c) : OptChkCall(c) {}
-
-  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
-    // Verify the "stpcpy" function prototype.
-    unsigned NumParams = OptChkCall ? 3 : 2;
-    FunctionType *FT = Callee->getFunctionType();
-    if (FT->getNumParams() != NumParams ||
-        FT->getReturnType() != FT->getParamType(0) ||
-        FT->getParamType(0) != FT->getParamType(1) ||
-        FT->getParamType(0) != B.getInt8PtrTy())
-      return 0;
-
-    // These optimizations require TargetData.
-    if (!TD) return 0;
-
-    Value *Dst = CI->getArgOperand(0), *Src = CI->getArgOperand(1);
-    if (Dst == Src) {  // stpcpy(x,x)  -> x+strlen(x)
-      Value *StrLen = EmitStrLen(Src, B, TD, TLI);
-      return StrLen ? B.CreateInBoundsGEP(Dst, StrLen) : 0;
-    }
-
-    // See if we can get the length of the input string.
-    uint64_t Len = GetStringLength(Src);
-    if (Len == 0) return 0;
-
-    Value *LenV = ConstantInt::get(TD->getIntPtrType(*Context), Len);
-    Value *DstEnd = B.CreateGEP(Dst,
-                                ConstantInt::get(TD->getIntPtrType(*Context),
-                                                 Len - 1));
-
-    // We have enough information to now generate the memcpy call to do the
-    // copy for us.  Make a memcpy to copy the nul byte with align = 1.
-    if (!OptChkCall || !EmitMemCpyChk(Dst, Src, LenV, CI->getArgOperand(2), B,
-                                      TD, TLI))
-      B.CreateMemCpy(Dst, Src, LenV, 1);
-    return DstEnd;
-  }
-};
-
-//===---------------------------------------===//
-// 'strncpy' Optimizations
-
-struct StrNCpyOpt : public LibCallOptimization {
-  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
-    FunctionType *FT = Callee->getFunctionType();
-    if (FT->getNumParams() != 3 || FT->getReturnType() != FT->getParamType(0) ||
-        FT->getParamType(0) != FT->getParamType(1) ||
-        FT->getParamType(0) != B.getInt8PtrTy() ||
-        !FT->getParamType(2)->isIntegerTy())
-      return 0;
-
-    Value *Dst = CI->getArgOperand(0);
-    Value *Src = CI->getArgOperand(1);
-    Value *LenOp = CI->getArgOperand(2);
-
-    // See if we can get the length of the input string.
-    uint64_t SrcLen = GetStringLength(Src);
-    if (SrcLen == 0) return 0;
-    --SrcLen;
-
-    if (SrcLen == 0) {
-      // strncpy(x, "", y) -> memset(x, '\0', y, 1)
-      B.CreateMemSet(Dst, B.getInt8('\0'), LenOp, 1);
-      return Dst;
-    }
-
-    uint64_t Len;
-    if (ConstantInt *LengthArg = dyn_cast<ConstantInt>(LenOp))
-      Len = LengthArg->getZExtValue();
-    else
-      return 0;
-
-    if (Len == 0) return Dst; // strncpy(x, y, 0) -> x
-
-    // These optimizations require TargetData.
-    if (!TD) return 0;
-
-    // Let strncpy handle the zero padding
-    if (Len > SrcLen+1) return 0;
-
-    // strncpy(x, s, c) -> memcpy(x, s, c, 1) [s and c are constant]
-    B.CreateMemCpy(Dst, Src,
-                   ConstantInt::get(TD->getIntPtrType(*Context), Len), 1);
-
-    return Dst;
-  }
-};
-
-//===---------------------------------------===//
-// 'strlen' Optimizations
-
-struct StrLenOpt : public LibCallOptimization {
-  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
-    FunctionType *FT = Callee->getFunctionType();
-    if (FT->getNumParams() != 1 ||
-        FT->getParamType(0) != B.getInt8PtrTy() ||
-        !FT->getReturnType()->isIntegerTy())
-      return 0;
-
-    Value *Src = CI->getArgOperand(0);
-
-    // Constant folding: strlen("xyz") -> 3
-    if (uint64_t Len = GetStringLength(Src))
-      return ConstantInt::get(CI->getType(), Len-1);
-
-    // strlen(x) != 0 --> *x != 0
-    // strlen(x) == 0 --> *x == 0
-    if (IsOnlyUsedInZeroEqualityComparison(CI))
-      return B.CreateZExt(B.CreateLoad(Src, "strlenfirst"), CI->getType());
-    return 0;
-  }
-};
-
-
-//===---------------------------------------===//
-// 'strpbrk' Optimizations
-
-struct StrPBrkOpt : public LibCallOptimization {
-  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
-    FunctionType *FT = Callee->getFunctionType();
-    if (FT->getNumParams() != 2 ||
-        FT->getParamType(0) != B.getInt8PtrTy() ||
-        FT->getParamType(1) != FT->getParamType(0) ||
-        FT->getReturnType() != FT->getParamType(0))
-      return 0;
-
-    StringRef S1, S2;
-    bool HasS1 = getConstantStringInfo(CI->getArgOperand(0), S1);
-    bool HasS2 = getConstantStringInfo(CI->getArgOperand(1), S2);
-
-    // strpbrk(s, "") -> NULL
-    // strpbrk("", s) -> NULL
-    if ((HasS1 && S1.empty()) || (HasS2 && S2.empty()))
-      return Constant::getNullValue(CI->getType());
-
-    // Constant folding.
-    if (HasS1 && HasS2) {
-      size_t I = S1.find_first_of(S2);
-      if (I == std::string::npos) // No match.
-        return Constant::getNullValue(CI->getType());
-
-      return B.CreateGEP(CI->getArgOperand(0), B.getInt64(I), "strpbrk");
-    }
-
-    // strpbrk(s, "a") -> strchr(s, 'a')
-    if (TD && HasS2 && S2.size() == 1)
-      return EmitStrChr(CI->getArgOperand(0), S2[0], B, TD, TLI);
-
-    return 0;
-  }
-};
-
-//===---------------------------------------===//
-// 'strto*' Optimizations.  This handles strtol, strtod, strtof, strtoul, etc.
-
-struct StrToOpt : public LibCallOptimization {
-  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
-    FunctionType *FT = Callee->getFunctionType();
-    if ((FT->getNumParams() != 2 && FT->getNumParams() != 3) ||
-        !FT->getParamType(0)->isPointerTy() ||
-        !FT->getParamType(1)->isPointerTy())
-      return 0;
-
-    Value *EndPtr = CI->getArgOperand(1);
-    if (isa<ConstantPointerNull>(EndPtr)) {
-      // With a null EndPtr, this function won't capture the main argument.
-      // It would be readonly too, except that it still may write to errno.
-      CI->addAttribute(1, Attribute::NoCapture);
-    }
-
-    return 0;
-  }
-};
-
-//===---------------------------------------===//
-// 'strspn' Optimizations
-
-struct StrSpnOpt : public LibCallOptimization {
-  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
-    FunctionType *FT = Callee->getFunctionType();
-    if (FT->getNumParams() != 2 ||
-        FT->getParamType(0) != B.getInt8PtrTy() ||
-        FT->getParamType(1) != FT->getParamType(0) ||
-        !FT->getReturnType()->isIntegerTy())
-      return 0;
-
-    StringRef S1, S2;
-    bool HasS1 = getConstantStringInfo(CI->getArgOperand(0), S1);
-    bool HasS2 = getConstantStringInfo(CI->getArgOperand(1), S2);
-
-    // strspn(s, "") -> 0
-    // strspn("", s) -> 0
-    if ((HasS1 && S1.empty()) || (HasS2 && S2.empty()))
-      return Constant::getNullValue(CI->getType());
-
-    // Constant folding.
-    if (HasS1 && HasS2) {
-      size_t Pos = S1.find_first_not_of(S2);
-      if (Pos == StringRef::npos) Pos = S1.size();
-      return ConstantInt::get(CI->getType(), Pos);
-    }
-
-    return 0;
-  }
-};
-
-//===---------------------------------------===//
-// 'strcspn' Optimizations
-
-struct StrCSpnOpt : public LibCallOptimization {
-  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
-    FunctionType *FT = Callee->getFunctionType();
-    if (FT->getNumParams() != 2 ||
-        FT->getParamType(0) != B.getInt8PtrTy() ||
-        FT->getParamType(1) != FT->getParamType(0) ||
-        !FT->getReturnType()->isIntegerTy())
-      return 0;
-
-    StringRef S1, S2;
-    bool HasS1 = getConstantStringInfo(CI->getArgOperand(0), S1);
-    bool HasS2 = getConstantStringInfo(CI->getArgOperand(1), S2);
-
-    // strcspn("", s) -> 0
-    if (HasS1 && S1.empty())
-      return Constant::getNullValue(CI->getType());
-
-    // Constant folding.
-    if (HasS1 && HasS2) {
-      size_t Pos = S1.find_first_of(S2);
-      if (Pos == StringRef::npos) Pos = S1.size();
-      return ConstantInt::get(CI->getType(), Pos);
-    }
-
-    // strcspn(s, "") -> strlen(s)
-    if (TD && HasS2 && S2.empty())
-      return EmitStrLen(CI->getArgOperand(0), B, TD, TLI);
-
-    return 0;
-  }
-};
-
-//===---------------------------------------===//
-// 'strstr' Optimizations
+// Double -> Float Shrinking Optimizations for Unary Functions like 'floor'
 
-struct StrStrOpt : public LibCallOptimization {
+struct UnaryDoubleFPOpt : public LibCallOptimization {
+  bool CheckRetType;
+  UnaryDoubleFPOpt(bool CheckReturnType): CheckRetType(CheckReturnType) {}
   virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
     FunctionType *FT = Callee->getFunctionType();
-    if (FT->getNumParams() != 2 ||
-        !FT->getParamType(0)->isPointerTy() ||
-        !FT->getParamType(1)->isPointerTy() ||
-        !FT->getReturnType()->isPointerTy())
+    if (FT->getNumParams() != 1 || !FT->getReturnType()->isDoubleTy() ||
+        !FT->getParamType(0)->isDoubleTy())
       return 0;
 
-    // fold strstr(x, x) -> x.
-    if (CI->getArgOperand(0) == CI->getArgOperand(1))
-      return B.CreateBitCast(CI->getArgOperand(0), CI->getType());
-
-    // fold strstr(a, b) == a -> strncmp(a, b, strlen(b)) == 0
-    if (TD && IsOnlyUsedInEqualityComparison(CI, CI->getArgOperand(0))) {
-      Value *StrLen = EmitStrLen(CI->getArgOperand(1), B, TD, TLI);
-      if (!StrLen)
-        return 0;
-      Value *StrNCmp = EmitStrNCmp(CI->getArgOperand(0), CI->getArgOperand(1),
-                                   StrLen, B, TD, TLI);
-      if (!StrNCmp)
-        return 0;
-      for (Value::use_iterator UI = CI->use_begin(), UE = CI->use_end();
-           UI != UE; ) {
-        ICmpInst *Old = cast<ICmpInst>(*UI++);
-        Value *Cmp = B.CreateICmp(Old->getPredicate(), StrNCmp,
-                                  ConstantInt::getNullValue(StrNCmp->getType()),
-                                  "cmp");
-        Old->replaceAllUsesWith(Cmp);
-        Old->eraseFromParent();
+    if (CheckRetType) {
+      // Check if all the uses for function like 'sin' are converted to float.
+      for (Value::use_iterator UseI = CI->use_begin(); UseI != CI->use_end();
+          ++UseI) {
+        FPTruncInst *Cast = dyn_cast<FPTruncInst>(*UseI);
+        if (Cast == 0 || !Cast->getType()->isFloatTy())
+          return 0;
       }
-      return CI;
-    }
-
-    // See if either input string is a constant string.
-    StringRef SearchStr, ToFindStr;
-    bool HasStr1 = getConstantStringInfo(CI->getArgOperand(0), SearchStr);
-    bool HasStr2 = getConstantStringInfo(CI->getArgOperand(1), ToFindStr);
-
-    // fold strstr(x, "") -> x.
-    if (HasStr2 && ToFindStr.empty())
-      return B.CreateBitCast(CI->getArgOperand(0), CI->getType());
-
-    // If both strings are known, constant fold it.
-    if (HasStr1 && HasStr2) {
-      std::string::size_type Offset = SearchStr.find(ToFindStr);
-
-      if (Offset == StringRef::npos) // strstr("foo", "bar") -> null
-        return Constant::getNullValue(CI->getType());
-
-      // strstr("abcd", "bc") -> gep((char*)"abcd", 1)
-      Value *Result = CastToCStr(CI->getArgOperand(0), B);
-      Result = B.CreateConstInBoundsGEP1_64(Result, Offset, "strstr");
-      return B.CreateBitCast(Result, CI->getType());
-    }
-
-    // fold strstr(x, "y") -> strchr(x, 'y').
-    if (HasStr2 && ToFindStr.size() == 1) {
-      Value *StrChr= EmitStrChr(CI->getArgOperand(0), ToFindStr[0], B, TD, TLI);
-      return StrChr ? B.CreateBitCast(StrChr, CI->getType()) : 0;
-    }
-    return 0;
-  }
-};
-
-
-//===---------------------------------------===//
-// 'memcmp' Optimizations
-
-struct MemCmpOpt : public LibCallOptimization {
-  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
-    FunctionType *FT = Callee->getFunctionType();
-    if (FT->getNumParams() != 3 || !FT->getParamType(0)->isPointerTy() ||
-        !FT->getParamType(1)->isPointerTy() ||
-        !FT->getReturnType()->isIntegerTy(32))
-      return 0;
-
-    Value *LHS = CI->getArgOperand(0), *RHS = CI->getArgOperand(1);
-
-    if (LHS == RHS)  // memcmp(s,s,x) -> 0
-      return Constant::getNullValue(CI->getType());
-
-    // Make sure we have a constant length.
-    ConstantInt *LenC = dyn_cast<ConstantInt>(CI->getArgOperand(2));
-    if (!LenC) return 0;
-    uint64_t Len = LenC->getZExtValue();
-
-    if (Len == 0) // memcmp(s1,s2,0) -> 0
-      return Constant::getNullValue(CI->getType());
-
-    // memcmp(S1,S2,1) -> *(unsigned char*)LHS - *(unsigned char*)RHS
-    if (Len == 1) {
-      Value *LHSV = B.CreateZExt(B.CreateLoad(CastToCStr(LHS, B), "lhsc"),
-                                 CI->getType(), "lhsv");
-      Value *RHSV = B.CreateZExt(B.CreateLoad(CastToCStr(RHS, B), "rhsc"),
-                                 CI->getType(), "rhsv");
-      return B.CreateSub(LHSV, RHSV, "chardiff");
-    }
-
-    // Constant folding: memcmp(x, y, l) -> cnst (all arguments are constant)
-    StringRef LHSStr, RHSStr;
-    if (getConstantStringInfo(LHS, LHSStr) &&
-        getConstantStringInfo(RHS, RHSStr)) {
-      // Make sure we're not reading out-of-bounds memory.
-      if (Len > LHSStr.size() || Len > RHSStr.size())
-        return 0;
-      uint64_t Ret = memcmp(LHSStr.data(), RHSStr.data(), Len);
-      return ConstantInt::get(CI->getType(), Ret);
     }
 
-    return 0;
-  }
-};
-
-//===---------------------------------------===//
-// 'memcpy' Optimizations
-
-struct MemCpyOpt : public LibCallOptimization {
-  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
-    // These optimizations require TargetData.
-    if (!TD) return 0;
-
-    FunctionType *FT = Callee->getFunctionType();
-    if (FT->getNumParams() != 3 || FT->getReturnType() != FT->getParamType(0) ||
-        !FT->getParamType(0)->isPointerTy() ||
-        !FT->getParamType(1)->isPointerTy() ||
-        FT->getParamType(2) != TD->getIntPtrType(*Context))
-      return 0;
-
-    // memcpy(x, y, n) -> llvm.memcpy(x, y, n, 1)
-    B.CreateMemCpy(CI->getArgOperand(0), CI->getArgOperand(1),
-                   CI->getArgOperand(2), 1);
-    return CI->getArgOperand(0);
-  }
-};
-
-//===---------------------------------------===//
-// 'memmove' Optimizations
-
-struct MemMoveOpt : public LibCallOptimization {
-  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
-    // These optimizations require TargetData.
-    if (!TD) return 0;
-
-    FunctionType *FT = Callee->getFunctionType();
-    if (FT->getNumParams() != 3 || FT->getReturnType() != FT->getParamType(0) ||
-        !FT->getParamType(0)->isPointerTy() ||
-        !FT->getParamType(1)->isPointerTy() ||
-        FT->getParamType(2) != TD->getIntPtrType(*Context))
+    // If this is something like 'floor((double)floatval)', convert to floorf.
+    FPExtInst *Cast = dyn_cast<FPExtInst>(CI->getArgOperand(0));
+    if (Cast == 0 || !Cast->getOperand(0)->getType()->isFloatTy())
       return 0;
 
-    // memmove(x, y, n) -> llvm.memmove(x, y, n, 1)
-    B.CreateMemMove(CI->getArgOperand(0), CI->getArgOperand(1),
-                    CI->getArgOperand(2), 1);
-    return CI->getArgOperand(0);
+    // floor((double)floatval) -> (double)floorf(floatval)
+    Value *V = Cast->getOperand(0);
+    V = EmitUnaryFloatFnCall(V, Callee->getName(), B, Callee->getAttributes());
+    return B.CreateFPExt(V, B.getDoubleTy());
   }
 };
 
 //===---------------------------------------===//
-// 'memset' Optimizations
-
-struct MemSetOpt : public LibCallOptimization {
-  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
-    // These optimizations require TargetData.
-    if (!TD) return 0;
-
-    FunctionType *FT = Callee->getFunctionType();
-    if (FT->getNumParams() != 3 || FT->getReturnType() != FT->getParamType(0) ||
-        !FT->getParamType(0)->isPointerTy() ||
-        !FT->getParamType(1)->isIntegerTy() ||
-        FT->getParamType(2) != TD->getIntPtrType(*Context))
-      return 0;
-
-    // memset(p, v, n) -> llvm.memset(p, v, n, 1)
-    Value *Val = B.CreateIntCast(CI->getArgOperand(1), B.getInt8Ty(), false);
-    B.CreateMemSet(CI->getArgOperand(0), Val, CI->getArgOperand(2), 1);
-    return CI->getArgOperand(0);
-  }
-};
-
-//===----------------------------------------------------------------------===//
-// Math Library Optimizations
-//===----------------------------------------------------------------------===//
-
-//===---------------------------------------===//
 // 'cos*' Optimizations
-
 struct CosOpt : public LibCallOptimization {
   virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
+    Value *Ret = NULL;
+    if (UnsafeFPShrink && Callee->getName() == "cos" &&
+        TLI->has(LibFunc::cosf)) {
+      UnaryDoubleFPOpt UnsafeUnaryDoubleFP(true);
+      Ret = UnsafeUnaryDoubleFP.CallOptimizer(Callee, CI, B);
+    }
+
     FunctionType *FT = Callee->getFunctionType();
     // Just make sure this has 1 argument of FP type, which matches the
     // result type.
     if (FT->getNumParams() != 1 || FT->getReturnType() != FT->getParamType(0) ||
         !FT->getParamType(0)->isFloatingPointTy())
-      return 0;
+      return Ret;
 
     // cos(-x) -> cos(x)
     Value *Op1 = CI->getArgOperand(0);
@@ -910,7 +162,7 @@
       BinaryOperator *BinExpr = cast<BinaryOperator>(Op1);
       return B.CreateCall(Callee, BinExpr->getOperand(1), "cos");
     }
-    return 0;
+    return Ret;
   }
 };
 
@@ -919,13 +171,20 @@
 
 struct PowOpt : public LibCallOptimization {
   virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
+    Value *Ret = NULL;
+    if (UnsafeFPShrink && Callee->getName() == "pow" &&
+        TLI->has(LibFunc::powf)) {
+      UnaryDoubleFPOpt UnsafeUnaryDoubleFP(true);
+      Ret = UnsafeUnaryDoubleFP.CallOptimizer(Callee, CI, B);
+    }
+
     FunctionType *FT = Callee->getFunctionType();
     // Just make sure this has 2 arguments of the same FP type, which match the
     // result type.
     if (FT->getNumParams() != 2 || FT->getReturnType() != FT->getParamType(0) ||
         FT->getParamType(0) != FT->getParamType(1) ||
         !FT->getParamType(0)->isFloatingPointTy())
-      return 0;
+      return Ret;
 
     Value *Op1 = CI->getArgOperand(0), *Op2 = CI->getArgOperand(1);
     if (ConstantFP *Op1C = dyn_cast<ConstantFP>(Op1)) {
@@ -936,7 +195,7 @@
     }
 
     ConstantFP *Op2C = dyn_cast<ConstantFP>(Op2);
-    if (Op2C == 0) return 0;
+    if (Op2C == 0) return Ret;
 
     if (Op2C->getValueAPF().isZero())  // pow(x, 0.0) -> 1.0
       return ConstantFP::get(CI->getType(), 1.0);
@@ -974,12 +233,19 @@
 
 struct Exp2Opt : public LibCallOptimization {
   virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
+    Value *Ret = NULL;
+    if (UnsafeFPShrink && Callee->getName() == "exp2" &&
+        TLI->has(LibFunc::exp2)) {
+      UnaryDoubleFPOpt UnsafeUnaryDoubleFP(true);
+      Ret = UnsafeUnaryDoubleFP.CallOptimizer(Callee, CI, B);
+    }
+
     FunctionType *FT = Callee->getFunctionType();
     // Just make sure this has 1 argument of FP type, which matches the
     // result type.
     if (FT->getNumParams() != 1 || FT->getReturnType() != FT->getParamType(0) ||
         !FT->getParamType(0)->isFloatingPointTy())
-      return 0;
+      return Ret;
 
     Value *Op = CI->getArgOperand(0);
     // Turn exp2(sitofp(x)) -> ldexp(1.0, sext(x))  if sizeof(x) <= 32
@@ -1016,29 +282,7 @@
 
       return CI;
     }
-    return 0;
-  }
-};
-
-//===---------------------------------------===//
-// Double -> Float Shrinking Optimizations for Unary Functions like 'floor'
-
-struct UnaryDoubleFPOpt : public LibCallOptimization {
-  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
-    FunctionType *FT = Callee->getFunctionType();
-    if (FT->getNumParams() != 1 || !FT->getReturnType()->isDoubleTy() ||
-        !FT->getParamType(0)->isDoubleTy())
-      return 0;
-
-    // If this is something like 'floor((double)floatval)', convert to floorf.
-    FPExtInst *Cast = dyn_cast<FPExtInst>(CI->getArgOperand(0));
-    if (Cast == 0 || !Cast->getOperand(0)->getType()->isFloatTy())
-      return 0;
-
-    // floor((double)floatval) -> (double)floorf(floatval)
-    Value *V = Cast->getOperand(0);
-    V = EmitUnaryFloatFnCall(V, Callee->getName(), B, Callee->getAttributes());
-    return B.CreateFPExt(V, B.getDoubleTy());
+    return Ret;
   }
 };
 
@@ -1063,8 +307,8 @@
 
     // Constant fold.
     if (ConstantInt *CI = dyn_cast<ConstantInt>(Op)) {
-      if (CI->getValue() == 0)  // ffs(0) -> 0.
-        return Constant::getNullValue(CI->getType());
+      if (CI->isZero()) // ffs(0) -> 0.
+        return B.getInt32(0);
       // ffs(c) -> cttz(c)+1
       return B.getInt32(CI->getValue().countTrailingZeros() + 1);
     }
@@ -1267,7 +511,7 @@
         if (FormatStr[i] == '%')
           return 0; // we found a format specifier, bail out.
 
-      // These optimizations require TargetData.
+      // These optimizations require DataLayout.
       if (!TD) return 0;
 
       // sprintf(str, fmt) -> llvm.memcpy(str, fmt, strlen(fmt)+1, 1)
@@ -1297,7 +541,7 @@
     }
 
     if (FormatStr[1] == 's') {
-      // These optimizations require TargetData.
+      // These optimizations require DataLayout.
       if (!TD) return 0;
 
       // sprintf(dest, "%s", str) -> llvm.memcpy(dest, str, strlen(str)+1, 1)
@@ -1385,7 +629,7 @@
 
 struct FPutsOpt : public LibCallOptimization {
   virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
-    // These optimizations require TargetData.
+    // These optimizations require DataLayout.
     if (!TD) return 0;
 
     // Require two pointers.  Also, we can't optimize if return value is used.
@@ -1422,7 +666,7 @@
         if (FormatStr[i] == '%')  // Could handle %% -> % if we cared.
           return 0; // We found a format specifier.
 
-      // These optimizations require TargetData.
+      // These optimizations require DataLayout.
       if (!TD) return 0;
 
       Value *NewCI = EmitFWrite(CI->getArgOperand(1),
@@ -1524,17 +768,9 @@
     TargetLibraryInfo *TLI;
 
     StringMap<LibCallOptimization*> Optimizations;
-    // String and Memory LibCall Optimizations
-    StrCatOpt StrCat; StrNCatOpt StrNCat; StrChrOpt StrChr; StrRChrOpt StrRChr;
-    StrCmpOpt StrCmp; StrNCmpOpt StrNCmp;
-    StrCpyOpt StrCpy; StrCpyOpt StrCpyChk;
-    StpCpyOpt StpCpy; StpCpyOpt StpCpyChk;
-    StrNCpyOpt StrNCpy;
-    StrLenOpt StrLen; StrPBrkOpt StrPBrk;
-    StrToOpt StrTo; StrSpnOpt StrSpn; StrCSpnOpt StrCSpn; StrStrOpt StrStr;
-    MemCmpOpt MemCmp; MemCpyOpt MemCpy; MemMoveOpt MemMove; MemSetOpt MemSet;
     // Math Library Optimizations
-    CosOpt Cos; PowOpt Pow; Exp2Opt Exp2; UnaryDoubleFPOpt UnaryDoubleFP;
+    CosOpt Cos; PowOpt Pow; Exp2Opt Exp2;
+    UnaryDoubleFPOpt UnaryDoubleFP, UnsafeUnaryDoubleFP;
     // Integer Optimizations
     FFSOpt FFS; AbsOpt Abs; IsDigitOpt IsDigit; IsAsciiOpt IsAscii;
     ToAsciiOpt ToAscii;
@@ -1546,11 +782,13 @@
     bool Modified;  // This is only used by doInitialization.
   public:
     static char ID; // Pass identification
-    SimplifyLibCalls() : FunctionPass(ID), StrCpy(false), StrCpyChk(true),
-                         StpCpy(false), StpCpyChk(true) {
+    SimplifyLibCalls() : FunctionPass(ID), UnaryDoubleFP(false),
+                         UnsafeUnaryDoubleFP(true) {
       initializeSimplifyLibCallsPass(*PassRegistry::getPassRegistry());
     }
     void AddOpt(LibFunc::Func F, LibCallOptimization* Opt);
+    void AddOpt(LibFunc::Func F1, LibFunc::Func F2, LibCallOptimization* Opt);
+
     void InitOptimizations();
     bool runOnFunction(Function &F);
 
@@ -1586,40 +824,15 @@
     Optimizations[TLI->getName(F)] = Opt;
 }
 
+void SimplifyLibCalls::AddOpt(LibFunc::Func F1, LibFunc::Func F2,
+                              LibCallOptimization* Opt) {
+  if (TLI->has(F1) && TLI->has(F2))
+    Optimizations[TLI->getName(F1)] = Opt;
+}
+
 /// Optimizations - Populate the Optimizations map with all the optimizations
 /// we know.
 void SimplifyLibCalls::InitOptimizations() {
-  // String and Memory LibCall Optimizations
-  Optimizations["strcat"] = &StrCat;
-  Optimizations["strncat"] = &StrNCat;
-  Optimizations["strchr"] = &StrChr;
-  Optimizations["strrchr"] = &StrRChr;
-  Optimizations["strcmp"] = &StrCmp;
-  Optimizations["strncmp"] = &StrNCmp;
-  Optimizations["strcpy"] = &StrCpy;
-  Optimizations["strncpy"] = &StrNCpy;
-  Optimizations["stpcpy"] = &StpCpy;
-  Optimizations["strlen"] = &StrLen;
-  Optimizations["strpbrk"] = &StrPBrk;
-  Optimizations["strtol"] = &StrTo;
-  Optimizations["strtod"] = &StrTo;
-  Optimizations["strtof"] = &StrTo;
-  Optimizations["strtoul"] = &StrTo;
-  Optimizations["strtoll"] = &StrTo;
-  Optimizations["strtold"] = &StrTo;
-  Optimizations["strtoull"] = &StrTo;
-  Optimizations["strspn"] = &StrSpn;
-  Optimizations["strcspn"] = &StrCSpn;
-  Optimizations["strstr"] = &StrStr;
-  Optimizations["memcmp"] = &MemCmp;
-  AddOpt(LibFunc::memcpy, &MemCpy);
-  Optimizations["memmove"] = &MemMove;
-  AddOpt(LibFunc::memset, &MemSet);
-
-  // _chk variants of String and Memory LibCall Optimizations.
-  Optimizations["__strcpy_chk"] = &StrCpyChk;
-  Optimizations["__stpcpy_chk"] = &StpCpyChk;
-
   // Math Library Optimizations
   Optimizations["cosf"] = &Cos;
   Optimizations["cos"] = &Cos;
@@ -1641,16 +854,37 @@
   Optimizations["llvm.exp2.f64"] = &Exp2;
   Optimizations["llvm.exp2.f32"] = &Exp2;
 
-  if (TLI->has(LibFunc::floor) && TLI->has(LibFunc::floorf))
-    Optimizations["floor"] = &UnaryDoubleFP;
-  if (TLI->has(LibFunc::ceil) && TLI->has(LibFunc::ceilf))
-    Optimizations["ceil"] = &UnaryDoubleFP;
-  if (TLI->has(LibFunc::round) && TLI->has(LibFunc::roundf))
-    Optimizations["round"] = &UnaryDoubleFP;
-  if (TLI->has(LibFunc::rint) && TLI->has(LibFunc::rintf))
-    Optimizations["rint"] = &UnaryDoubleFP;
-  if (TLI->has(LibFunc::nearbyint) && TLI->has(LibFunc::nearbyintf))
-    Optimizations["nearbyint"] = &UnaryDoubleFP;
+  AddOpt(LibFunc::ceil, LibFunc::ceilf, &UnaryDoubleFP);
+  AddOpt(LibFunc::fabs, LibFunc::fabsf, &UnaryDoubleFP);
+  AddOpt(LibFunc::floor, LibFunc::floorf, &UnaryDoubleFP);
+  AddOpt(LibFunc::rint, LibFunc::rintf, &UnaryDoubleFP);
+  AddOpt(LibFunc::round, LibFunc::roundf, &UnaryDoubleFP);
+  AddOpt(LibFunc::nearbyint, LibFunc::nearbyintf, &UnaryDoubleFP);
+  AddOpt(LibFunc::trunc, LibFunc::truncf, &UnaryDoubleFP);
+
+  if(UnsafeFPShrink) {
+    AddOpt(LibFunc::acos, LibFunc::acosf, &UnsafeUnaryDoubleFP);
+    AddOpt(LibFunc::acosh, LibFunc::acoshf, &UnsafeUnaryDoubleFP);
+    AddOpt(LibFunc::asin, LibFunc::asinf, &UnsafeUnaryDoubleFP);
+    AddOpt(LibFunc::asinh, LibFunc::asinhf, &UnsafeUnaryDoubleFP);
+    AddOpt(LibFunc::atan, LibFunc::atanf, &UnsafeUnaryDoubleFP);
+    AddOpt(LibFunc::atanh, LibFunc::atanhf, &UnsafeUnaryDoubleFP);
+    AddOpt(LibFunc::cbrt, LibFunc::cbrtf, &UnsafeUnaryDoubleFP);
+    AddOpt(LibFunc::cosh, LibFunc::coshf, &UnsafeUnaryDoubleFP);
+    AddOpt(LibFunc::exp, LibFunc::expf, &UnsafeUnaryDoubleFP);
+    AddOpt(LibFunc::exp10, LibFunc::exp10f, &UnsafeUnaryDoubleFP);
+    AddOpt(LibFunc::expm1, LibFunc::expm1f, &UnsafeUnaryDoubleFP);
+    AddOpt(LibFunc::log, LibFunc::logf, &UnsafeUnaryDoubleFP);
+    AddOpt(LibFunc::log10, LibFunc::log10f, &UnsafeUnaryDoubleFP);
+    AddOpt(LibFunc::log1p, LibFunc::log1pf, &UnsafeUnaryDoubleFP);
+    AddOpt(LibFunc::log2, LibFunc::log2f, &UnsafeUnaryDoubleFP);
+    AddOpt(LibFunc::logb, LibFunc::logbf, &UnsafeUnaryDoubleFP);
+    AddOpt(LibFunc::sin, LibFunc::sinf, &UnsafeUnaryDoubleFP);
+    AddOpt(LibFunc::sinh, LibFunc::sinhf, &UnsafeUnaryDoubleFP);
+    AddOpt(LibFunc::sqrt, LibFunc::sqrtf, &UnsafeUnaryDoubleFP);
+    AddOpt(LibFunc::tan, LibFunc::tanf, &UnsafeUnaryDoubleFP);
+    AddOpt(LibFunc::tanh, LibFunc::tanhf, &UnsafeUnaryDoubleFP);
+  }
 
   // Integer Optimizations
   Optimizations["ffs"] = &FFS;
@@ -1681,7 +915,7 @@
   if (Optimizations.empty())
     InitOptimizations();
 
-  const TargetData *TD = getAnalysisIfAvailable<TargetData>();
+  const DataLayout *TD = getAnalysisIfAvailable<DataLayout>();
 
   IRBuilder<> Builder(F.getContext());
 

Modified: llvm/branches/AMDILBackend/lib/Transforms/Utils/AddrModeMatcher.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Utils/AddrModeMatcher.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Utils/AddrModeMatcher.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Utils/AddrModeMatcher.cpp Tue Jan 15 11:16:16 2013
@@ -16,7 +16,7 @@
 #include "llvm/GlobalValue.h"
 #include "llvm/Instruction.h"
 #include "llvm/Assembly/Writer.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/GetElementPtrTypeIterator.h"
 #include "llvm/Support/PatternMatch.h"
@@ -55,10 +55,12 @@
   OS << ']';
 }
 
+#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
 void ExtAddrMode::dump() const {
   print(dbgs());
   dbgs() << '\n';
 }
+#endif
 
 
 /// MatchScaledValue - Try adding ScaleReg*Scale to the current addressing mode.
@@ -219,7 +221,7 @@
     unsigned VariableScale = 0;
     
     int64_t ConstantOffset = 0;
-    const TargetData *TD = TLI.getTargetData();
+    const DataLayout *TD = TLI.getDataLayout();
     gep_type_iterator GTI = gep_type_begin(AddrInst);
     for (unsigned i = 1, e = AddrInst->getNumOperands(); i != e; ++i, ++GTI) {
       if (StructType *STy = dyn_cast<StructType>(*GTI)) {

Modified: llvm/branches/AMDILBackend/lib/Transforms/Utils/BasicBlockUtils.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Utils/BasicBlockUtils.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Utils/BasicBlockUtils.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Utils/BasicBlockUtils.cpp Tue Jan 15 11:16:16 2013
@@ -22,7 +22,7 @@
 #include "llvm/Analysis/Dominators.h"
 #include "llvm/Analysis/LoopInfo.h"
 #include "llvm/Analysis/MemoryDependenceAnalysis.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Transforms/Utils/Local.h"
 #include "llvm/Transforms/Scalar.h"
 #include "llvm/Support/ErrorHandling.h"
@@ -94,7 +94,7 @@
 /// is dead. Also recursively delete any operands that become dead as
 /// a result. This includes tracing the def-use list from the PHI to see if
 /// it is ultimately unused or if it reaches an unused cycle.
-bool llvm::DeleteDeadPHIs(BasicBlock *BB) {
+bool llvm::DeleteDeadPHIs(BasicBlock *BB, const TargetLibraryInfo *TLI) {
   // Recursively deleting a PHI may cause multiple PHIs to be deleted
   // or RAUW'd undef, so use an array of WeakVH for the PHIs to delete.
   SmallVector<WeakVH, 8> PHIs;
@@ -105,7 +105,7 @@
   bool Changed = false;
   for (unsigned i = 0, e = PHIs.size(); i != e; ++i)
     if (PHINode *PN = dyn_cast_or_null<PHINode>(PHIs[i].operator Value*()))
-      Changed |= RecursivelyDeleteDeadPHINode(PN);
+      Changed |= RecursivelyDeleteDeadPHINode(PN, TLI);
 
   return Changed;
 }
@@ -687,3 +687,42 @@
   return cast<ReturnInst>(NewRet);
 }
 
+/// SplitBlockAndInsertIfThen - Split the containing block at the
+/// specified instruction - everything before and including Cmp stays
+/// in the old basic block, and everything after Cmp is moved to a
+/// new block. The two blocks are connected by a conditional branch
+/// (with value of Cmp being the condition).
+/// Before:
+///   Head
+///   Cmp
+///   Tail
+/// After:
+///   Head
+///   Cmp
+///   if (Cmp)
+///     ThenBlock
+///   Tail
+///
+/// If Unreachable is true, then ThenBlock ends with
+/// UnreachableInst, otherwise it branches to Tail.
+/// Returns the NewBasicBlock's terminator.
+
+TerminatorInst *llvm::SplitBlockAndInsertIfThen(Instruction *Cmp,
+    bool Unreachable, MDNode *BranchWeights) {
+  Instruction *SplitBefore = Cmp->getNextNode();
+  BasicBlock *Head = SplitBefore->getParent();
+  BasicBlock *Tail = Head->splitBasicBlock(SplitBefore);
+  TerminatorInst *HeadOldTerm = Head->getTerminator();
+  LLVMContext &C = Head->getContext();
+  BasicBlock *ThenBlock = BasicBlock::Create(C, "", Head->getParent(), Tail);
+  TerminatorInst *CheckTerm;
+  if (Unreachable)
+    CheckTerm = new UnreachableInst(C, ThenBlock);
+  else
+    CheckTerm = BranchInst::Create(Tail, ThenBlock);
+  BranchInst *HeadNewTerm =
+    BranchInst::Create(/*ifTrue*/ThenBlock, /*ifFalse*/Tail, Cmp);
+  HeadNewTerm->setMetadata(LLVMContext::MD_prof, BranchWeights);
+  ReplaceInstWithInst(HeadOldTerm, HeadNewTerm);
+  return CheckTerm;
+}

Modified: llvm/branches/AMDILBackend/lib/Transforms/Utils/BuildLibCalls.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Utils/BuildLibCalls.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Utils/BuildLibCalls.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Utils/BuildLibCalls.cpp Tue Jan 15 11:16:16 2013
@@ -22,7 +22,7 @@
 #include "llvm/Module.h"
 #include "llvm/Type.h"
 #include "llvm/ADT/SmallString.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Target/TargetLibraryInfo.h"
 
 using namespace llvm;
@@ -34,19 +34,22 @@
 
 /// EmitStrLen - Emit a call to the strlen function to the builder, for the
 /// specified pointer.  This always returns an integer value of size intptr_t.
-Value *llvm::EmitStrLen(Value *Ptr, IRBuilder<> &B, const TargetData *TD,
+Value *llvm::EmitStrLen(Value *Ptr, IRBuilder<> &B, const DataLayout *TD,
                         const TargetLibraryInfo *TLI) {
   if (!TLI->has(LibFunc::strlen))
     return 0;
 
   Module *M = B.GetInsertBlock()->getParent()->getParent();
   AttributeWithIndex AWI[2];
-  AWI[0] = AttributeWithIndex::get(1, Attribute::NoCapture);
-  AWI[1] = AttributeWithIndex::get(~0u, Attribute::ReadOnly |
-                                   Attribute::NoUnwind);
+  AWI[0] = AttributeWithIndex::get(M->getContext(), 1, Attributes::NoCapture);
+  Attributes::AttrVal AVs[2] = { Attributes::ReadOnly, Attributes::NoUnwind };
+  AWI[1] = AttributeWithIndex::get(M->getContext(), AttrListPtr::FunctionIndex,
+                                   ArrayRef<Attributes::AttrVal>(AVs, 2));
 
   LLVMContext &Context = B.GetInsertBlock()->getContext();
-  Constant *StrLen = M->getOrInsertFunction("strlen", AttrListPtr::get(AWI),
+  Constant *StrLen = M->getOrInsertFunction("strlen",
+                                            AttrListPtr::get(M->getContext(),
+                                                             AWI),
                                             TD->getIntPtrType(Context),
                                             B.getInt8PtrTy(),
                                             NULL);
@@ -61,18 +64,21 @@
 /// specified pointer.  Ptr is required to be some pointer type, MaxLen must
 /// be of size_t type, and the return value has 'intptr_t' type.
 Value *llvm::EmitStrNLen(Value *Ptr, Value *MaxLen, IRBuilder<> &B,
-                         const TargetData *TD, const TargetLibraryInfo *TLI) {
+                         const DataLayout *TD, const TargetLibraryInfo *TLI) {
   if (!TLI->has(LibFunc::strnlen))
     return 0;
 
   Module *M = B.GetInsertBlock()->getParent()->getParent();
   AttributeWithIndex AWI[2];
-  AWI[0] = AttributeWithIndex::get(1, Attribute::NoCapture);
-  AWI[1] = AttributeWithIndex::get(~0u, Attribute::ReadOnly |
-                                   Attribute::NoUnwind);
+  AWI[0] = AttributeWithIndex::get(M->getContext(), 1, Attributes::NoCapture);
+  Attributes::AttrVal AVs[2] = { Attributes::ReadOnly, Attributes::NoUnwind };
+  AWI[1] = AttributeWithIndex::get(M->getContext(), AttrListPtr::FunctionIndex,
+                                   ArrayRef<Attributes::AttrVal>(AVs, 2));
 
   LLVMContext &Context = B.GetInsertBlock()->getContext();
-  Constant *StrNLen = M->getOrInsertFunction("strnlen", AttrListPtr::get(AWI),
+  Constant *StrNLen = M->getOrInsertFunction("strnlen",
+                                             AttrListPtr::get(M->getContext(),
+                                                              AWI),
                                              TD->getIntPtrType(Context),
                                              B.getInt8PtrTy(),
                                              TD->getIntPtrType(Context),
@@ -88,17 +94,21 @@
 /// specified pointer and character.  Ptr is required to be some pointer type,
 /// and the return value has 'i8*' type.
 Value *llvm::EmitStrChr(Value *Ptr, char C, IRBuilder<> &B,
-                        const TargetData *TD, const TargetLibraryInfo *TLI) {
+                        const DataLayout *TD, const TargetLibraryInfo *TLI) {
   if (!TLI->has(LibFunc::strchr))
     return 0;
 
   Module *M = B.GetInsertBlock()->getParent()->getParent();
+  Attributes::AttrVal AVs[2] = { Attributes::ReadOnly, Attributes::NoUnwind };
   AttributeWithIndex AWI =
-    AttributeWithIndex::get(~0u, Attribute::ReadOnly | Attribute::NoUnwind);
+    AttributeWithIndex::get(M->getContext(), AttrListPtr::FunctionIndex,
+                            ArrayRef<Attributes::AttrVal>(AVs, 2));
 
   Type *I8Ptr = B.getInt8PtrTy();
   Type *I32Ty = B.getInt32Ty();
-  Constant *StrChr = M->getOrInsertFunction("strchr", AttrListPtr::get(AWI),
+  Constant *StrChr = M->getOrInsertFunction("strchr",
+                                            AttrListPtr::get(M->getContext(),
+                                                             AWI),
                                             I8Ptr, I8Ptr, I32Ty, NULL);
   CallInst *CI = B.CreateCall2(StrChr, CastToCStr(Ptr, B),
                                ConstantInt::get(I32Ty, C), "strchr");
@@ -109,20 +119,23 @@
 
 /// EmitStrNCmp - Emit a call to the strncmp function to the builder.
 Value *llvm::EmitStrNCmp(Value *Ptr1, Value *Ptr2, Value *Len,
-                         IRBuilder<> &B, const TargetData *TD,
+                         IRBuilder<> &B, const DataLayout *TD,
                          const TargetLibraryInfo *TLI) {
   if (!TLI->has(LibFunc::strncmp))
     return 0;
 
   Module *M = B.GetInsertBlock()->getParent()->getParent();
   AttributeWithIndex AWI[3];
-  AWI[0] = AttributeWithIndex::get(1, Attribute::NoCapture);
-  AWI[1] = AttributeWithIndex::get(2, Attribute::NoCapture);
-  AWI[2] = AttributeWithIndex::get(~0u, Attribute::ReadOnly |
-                                   Attribute::NoUnwind);
+  AWI[0] = AttributeWithIndex::get(M->getContext(), 1, Attributes::NoCapture);
+  AWI[1] = AttributeWithIndex::get(M->getContext(), 2, Attributes::NoCapture);
+  Attributes::AttrVal AVs[2] = { Attributes::ReadOnly, Attributes::NoUnwind };
+  AWI[2] = AttributeWithIndex::get(M->getContext(), AttrListPtr::FunctionIndex,
+                                   ArrayRef<Attributes::AttrVal>(AVs, 2));
 
   LLVMContext &Context = B.GetInsertBlock()->getContext();
-  Value *StrNCmp = M->getOrInsertFunction("strncmp", AttrListPtr::get(AWI),
+  Value *StrNCmp = M->getOrInsertFunction("strncmp",
+                                          AttrListPtr::get(M->getContext(),
+                                                           AWI),
                                           B.getInt32Ty(),
                                           B.getInt8PtrTy(),
                                           B.getInt8PtrTy(),
@@ -139,17 +152,19 @@
 /// EmitStrCpy - Emit a call to the strcpy function to the builder, for the
 /// specified pointer arguments.
 Value *llvm::EmitStrCpy(Value *Dst, Value *Src, IRBuilder<> &B,
-                        const TargetData *TD, const TargetLibraryInfo *TLI,
+                        const DataLayout *TD, const TargetLibraryInfo *TLI,
                         StringRef Name) {
   if (!TLI->has(LibFunc::strcpy))
     return 0;
 
   Module *M = B.GetInsertBlock()->getParent()->getParent();
   AttributeWithIndex AWI[2];
-  AWI[0] = AttributeWithIndex::get(2, Attribute::NoCapture);
-  AWI[1] = AttributeWithIndex::get(~0u, Attribute::NoUnwind);
+  AWI[0] = AttributeWithIndex::get(M->getContext(), 2, Attributes::NoCapture);
+  AWI[1] = AttributeWithIndex::get(M->getContext(), AttrListPtr::FunctionIndex,
+                                   Attributes::NoUnwind);
   Type *I8Ptr = B.getInt8PtrTy();
-  Value *StrCpy = M->getOrInsertFunction(Name, AttrListPtr::get(AWI),
+  Value *StrCpy = M->getOrInsertFunction(Name,
+                                         AttrListPtr::get(M->getContext(), AWI),
                                          I8Ptr, I8Ptr, I8Ptr, NULL);
   CallInst *CI = B.CreateCall2(StrCpy, CastToCStr(Dst, B), CastToCStr(Src, B),
                                Name);
@@ -161,17 +176,20 @@
 /// EmitStrNCpy - Emit a call to the strncpy function to the builder, for the
 /// specified pointer arguments.
 Value *llvm::EmitStrNCpy(Value *Dst, Value *Src, Value *Len,
-                         IRBuilder<> &B, const TargetData *TD,
+                         IRBuilder<> &B, const DataLayout *TD,
                          const TargetLibraryInfo *TLI, StringRef Name) {
   if (!TLI->has(LibFunc::strncpy))
     return 0;
 
   Module *M = B.GetInsertBlock()->getParent()->getParent();
   AttributeWithIndex AWI[2];
-  AWI[0] = AttributeWithIndex::get(2, Attribute::NoCapture);
-  AWI[1] = AttributeWithIndex::get(~0u, Attribute::NoUnwind);
+  AWI[0] = AttributeWithIndex::get(M->getContext(), 2, Attributes::NoCapture);
+  AWI[1] = AttributeWithIndex::get(M->getContext(), AttrListPtr::FunctionIndex,
+                                   Attributes::NoUnwind);
   Type *I8Ptr = B.getInt8PtrTy();
-  Value *StrNCpy = M->getOrInsertFunction(Name, AttrListPtr::get(AWI),
+  Value *StrNCpy = M->getOrInsertFunction(Name,
+                                          AttrListPtr::get(M->getContext(),
+                                                           AWI),
                                           I8Ptr, I8Ptr, I8Ptr,
                                           Len->getType(), NULL);
   CallInst *CI = B.CreateCall3(StrNCpy, CastToCStr(Dst, B), CastToCStr(Src, B),
@@ -185,17 +203,18 @@
 /// This expects that the Len and ObjSize have type 'intptr_t' and Dst/Src
 /// are pointers.
 Value *llvm::EmitMemCpyChk(Value *Dst, Value *Src, Value *Len, Value *ObjSize,
-                           IRBuilder<> &B, const TargetData *TD,
+                           IRBuilder<> &B, const DataLayout *TD,
                            const TargetLibraryInfo *TLI) {
   if (!TLI->has(LibFunc::memcpy_chk))
     return 0;
 
   Module *M = B.GetInsertBlock()->getParent()->getParent();
   AttributeWithIndex AWI;
-  AWI = AttributeWithIndex::get(~0u, Attribute::NoUnwind);
+  AWI = AttributeWithIndex::get(M->getContext(), AttrListPtr::FunctionIndex,
+                                Attributes::NoUnwind);
   LLVMContext &Context = B.GetInsertBlock()->getContext();
   Value *MemCpy = M->getOrInsertFunction("__memcpy_chk",
-                                         AttrListPtr::get(AWI),
+                                         AttrListPtr::get(M->getContext(), AWI),
                                          B.getInt8PtrTy(),
                                          B.getInt8PtrTy(),
                                          B.getInt8PtrTy(),
@@ -212,16 +231,19 @@
 /// EmitMemChr - Emit a call to the memchr function.  This assumes that Ptr is
 /// a pointer, Val is an i32 value, and Len is an 'intptr_t' value.
 Value *llvm::EmitMemChr(Value *Ptr, Value *Val,
-                        Value *Len, IRBuilder<> &B, const TargetData *TD,
+                        Value *Len, IRBuilder<> &B, const DataLayout *TD,
                         const TargetLibraryInfo *TLI) {
   if (!TLI->has(LibFunc::memchr))
     return 0;
 
   Module *M = B.GetInsertBlock()->getParent()->getParent();
   AttributeWithIndex AWI;
-  AWI = AttributeWithIndex::get(~0u, Attribute::ReadOnly | Attribute::NoUnwind);
+  Attributes::AttrVal AVs[2] = { Attributes::ReadOnly, Attributes::NoUnwind };
+  AWI = AttributeWithIndex::get(M->getContext(), AttrListPtr::FunctionIndex,
+                                ArrayRef<Attributes::AttrVal>(AVs, 2));
   LLVMContext &Context = B.GetInsertBlock()->getContext();
-  Value *MemChr = M->getOrInsertFunction("memchr", AttrListPtr::get(AWI),
+  Value *MemChr = M->getOrInsertFunction("memchr",
+                                         AttrListPtr::get(M->getContext(), AWI),
                                          B.getInt8PtrTy(),
                                          B.getInt8PtrTy(),
                                          B.getInt32Ty(),
@@ -237,20 +259,22 @@
 
 /// EmitMemCmp - Emit a call to the memcmp function.
 Value *llvm::EmitMemCmp(Value *Ptr1, Value *Ptr2,
-                        Value *Len, IRBuilder<> &B, const TargetData *TD,
+                        Value *Len, IRBuilder<> &B, const DataLayout *TD,
                         const TargetLibraryInfo *TLI) {
   if (!TLI->has(LibFunc::memcmp))
     return 0;
 
   Module *M = B.GetInsertBlock()->getParent()->getParent();
   AttributeWithIndex AWI[3];
-  AWI[0] = AttributeWithIndex::get(1, Attribute::NoCapture);
-  AWI[1] = AttributeWithIndex::get(2, Attribute::NoCapture);
-  AWI[2] = AttributeWithIndex::get(~0u, Attribute::ReadOnly |
-                                   Attribute::NoUnwind);
+  AWI[0] = AttributeWithIndex::get(M->getContext(), 1, Attributes::NoCapture);
+  AWI[1] = AttributeWithIndex::get(M->getContext(), 2, Attributes::NoCapture);
+  Attributes::AttrVal AVs[2] = { Attributes::ReadOnly, Attributes::NoUnwind };
+  AWI[2] = AttributeWithIndex::get(M->getContext(), AttrListPtr::FunctionIndex,
+                                   ArrayRef<Attributes::AttrVal>(AVs, 2));
 
   LLVMContext &Context = B.GetInsertBlock()->getContext();
-  Value *MemCmp = M->getOrInsertFunction("memcmp", AttrListPtr::get(AWI),
+  Value *MemCmp = M->getOrInsertFunction("memcmp",
+                                         AttrListPtr::get(M->getContext(), AWI),
                                          B.getInt32Ty(),
                                          B.getInt8PtrTy(),
                                          B.getInt8PtrTy(),
@@ -294,7 +318,7 @@
 
 /// EmitPutChar - Emit a call to the putchar function.  This assumes that Char
 /// is an integer.
-Value *llvm::EmitPutChar(Value *Char, IRBuilder<> &B, const TargetData *TD,
+Value *llvm::EmitPutChar(Value *Char, IRBuilder<> &B, const DataLayout *TD,
                          const TargetLibraryInfo *TLI) {
   if (!TLI->has(LibFunc::putchar))
     return 0;
@@ -316,17 +340,19 @@
 
 /// EmitPutS - Emit a call to the puts function.  This assumes that Str is
 /// some pointer.
-Value *llvm::EmitPutS(Value *Str, IRBuilder<> &B, const TargetData *TD,
+Value *llvm::EmitPutS(Value *Str, IRBuilder<> &B, const DataLayout *TD,
                       const TargetLibraryInfo *TLI) {
   if (!TLI->has(LibFunc::puts))
     return 0;
 
   Module *M = B.GetInsertBlock()->getParent()->getParent();
   AttributeWithIndex AWI[2];
-  AWI[0] = AttributeWithIndex::get(1, Attribute::NoCapture);
-  AWI[1] = AttributeWithIndex::get(~0u, Attribute::NoUnwind);
+  AWI[0] = AttributeWithIndex::get(M->getContext(), 1, Attributes::NoCapture);
+  AWI[1] = AttributeWithIndex::get(M->getContext(), AttrListPtr::FunctionIndex,
+                                   Attributes::NoUnwind);
 
-  Value *PutS = M->getOrInsertFunction("puts", AttrListPtr::get(AWI),
+  Value *PutS = M->getOrInsertFunction("puts",
+                                       AttrListPtr::get(M->getContext(), AWI),
                                        B.getInt32Ty(),
                                        B.getInt8PtrTy(),
                                        NULL);
@@ -339,17 +365,19 @@
 /// EmitFPutC - Emit a call to the fputc function.  This assumes that Char is
 /// an integer and File is a pointer to FILE.
 Value *llvm::EmitFPutC(Value *Char, Value *File, IRBuilder<> &B,
-                       const TargetData *TD, const TargetLibraryInfo *TLI) {
+                       const DataLayout *TD, const TargetLibraryInfo *TLI) {
   if (!TLI->has(LibFunc::fputc))
     return 0;
 
   Module *M = B.GetInsertBlock()->getParent()->getParent();
   AttributeWithIndex AWI[2];
-  AWI[0] = AttributeWithIndex::get(2, Attribute::NoCapture);
-  AWI[1] = AttributeWithIndex::get(~0u, Attribute::NoUnwind);
+  AWI[0] = AttributeWithIndex::get(M->getContext(), 2, Attributes::NoCapture);
+  AWI[1] = AttributeWithIndex::get(M->getContext(), AttrListPtr::FunctionIndex,
+                                   Attributes::NoUnwind);
   Constant *F;
   if (File->getType()->isPointerTy())
-    F = M->getOrInsertFunction("fputc", AttrListPtr::get(AWI),
+    F = M->getOrInsertFunction("fputc",
+                               AttrListPtr::get(M->getContext(), AWI),
                                B.getInt32Ty(),
                                B.getInt32Ty(), File->getType(),
                                NULL);
@@ -370,19 +398,21 @@
 /// EmitFPutS - Emit a call to the puts function.  Str is required to be a
 /// pointer and File is a pointer to FILE.
 Value *llvm::EmitFPutS(Value *Str, Value *File, IRBuilder<> &B,
-                       const TargetData *TD, const TargetLibraryInfo *TLI) {
+                       const DataLayout *TD, const TargetLibraryInfo *TLI) {
   if (!TLI->has(LibFunc::fputs))
     return 0;
 
   Module *M = B.GetInsertBlock()->getParent()->getParent();
   AttributeWithIndex AWI[3];
-  AWI[0] = AttributeWithIndex::get(1, Attribute::NoCapture);
-  AWI[1] = AttributeWithIndex::get(2, Attribute::NoCapture);
-  AWI[2] = AttributeWithIndex::get(~0u, Attribute::NoUnwind);
+  AWI[0] = AttributeWithIndex::get(M->getContext(), 1, Attributes::NoCapture);
+  AWI[1] = AttributeWithIndex::get(M->getContext(), 2, Attributes::NoCapture);
+  AWI[2] = AttributeWithIndex::get(M->getContext(), AttrListPtr::FunctionIndex,
+                                   Attributes::NoUnwind);
   StringRef FPutsName = TLI->getName(LibFunc::fputs);
   Constant *F;
   if (File->getType()->isPointerTy())
-    F = M->getOrInsertFunction(FPutsName, AttrListPtr::get(AWI),
+    F = M->getOrInsertFunction(FPutsName,
+                               AttrListPtr::get(M->getContext(), AWI),
                                B.getInt32Ty(),
                                B.getInt8PtrTy(),
                                File->getType(), NULL);
@@ -400,21 +430,23 @@
 /// EmitFWrite - Emit a call to the fwrite function.  This assumes that Ptr is
 /// a pointer, Size is an 'intptr_t', and File is a pointer to FILE.
 Value *llvm::EmitFWrite(Value *Ptr, Value *Size, Value *File,
-                        IRBuilder<> &B, const TargetData *TD,
+                        IRBuilder<> &B, const DataLayout *TD,
                         const TargetLibraryInfo *TLI) {
   if (!TLI->has(LibFunc::fwrite))
     return 0;
 
   Module *M = B.GetInsertBlock()->getParent()->getParent();
   AttributeWithIndex AWI[3];
-  AWI[0] = AttributeWithIndex::get(1, Attribute::NoCapture);
-  AWI[1] = AttributeWithIndex::get(4, Attribute::NoCapture);
-  AWI[2] = AttributeWithIndex::get(~0u, Attribute::NoUnwind);
+  AWI[0] = AttributeWithIndex::get(M->getContext(), 1, Attributes::NoCapture);
+  AWI[1] = AttributeWithIndex::get(M->getContext(), 4, Attributes::NoCapture);
+  AWI[2] = AttributeWithIndex::get(M->getContext(), AttrListPtr::FunctionIndex,
+                                   Attributes::NoUnwind);
   LLVMContext &Context = B.GetInsertBlock()->getContext();
   StringRef FWriteName = TLI->getName(LibFunc::fwrite);
   Constant *F;
   if (File->getType()->isPointerTy())
-    F = M->getOrInsertFunction(FWriteName, AttrListPtr::get(AWI),
+    F = M->getOrInsertFunction(FWriteName,
+                               AttrListPtr::get(M->getContext(), AWI),
                                TD->getIntPtrType(Context),
                                B.getInt8PtrTy(),
                                TD->getIntPtrType(Context),
@@ -436,9 +468,9 @@
 
 SimplifyFortifiedLibCalls::~SimplifyFortifiedLibCalls() { }
 
-bool SimplifyFortifiedLibCalls::fold(CallInst *CI, const TargetData *TD,
+bool SimplifyFortifiedLibCalls::fold(CallInst *CI, const DataLayout *TD,
                                      const TargetLibraryInfo *TLI) {
-  // We really need TargetData for later.
+  // We really need DataLayout for later.
   if (!TD) return false;
   
   this->CI = CI;

Modified: llvm/branches/AMDILBackend/lib/Transforms/Utils/CMakeLists.txt
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Utils/CMakeLists.txt?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Utils/CMakeLists.txt (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Utils/CMakeLists.txt Tue Jan 15 11:16:16 2013
@@ -3,6 +3,7 @@
   BasicBlockUtils.cpp
   BreakCriticalEdges.cpp
   BuildLibCalls.cpp
+  BypassSlowDivision.cpp
   CloneFunction.cpp
   CloneModule.cpp
   CmpInstAnalysis.cpp
@@ -10,6 +11,7 @@
   DemoteRegToStack.cpp
   InlineFunction.cpp
   InstructionNamer.cpp
+  IntegerDivision.cpp
   LCSSA.cpp
   Local.cpp
   LoopSimplify.cpp
@@ -19,12 +21,14 @@
   LowerInvoke.cpp
   LowerSwitch.cpp
   Mem2Reg.cpp
+  MetaRenamer.cpp
   ModuleUtils.cpp
   PromoteMemoryToRegister.cpp
   SSAUpdater.cpp
   SimplifyCFG.cpp
   SimplifyIndVar.cpp
   SimplifyInstructions.cpp
+  SimplifyLibCalls.cpp
   UnifyFunctionExitNodes.cpp
   Utils.cpp
   ValueMapper.cpp

Modified: llvm/branches/AMDILBackend/lib/Transforms/Utils/CloneFunction.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Utils/CloneFunction.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Utils/CloneFunction.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Utils/CloneFunction.cpp Tue Jan 15 11:16:16 2013
@@ -98,10 +98,14 @@
         Anew->addAttr( OldFunc->getAttributes()
                        .getParamAttributes(I->getArgNo() + 1));
     NewFunc->setAttributes(NewFunc->getAttributes()
-                           .addAttr(0, OldFunc->getAttributes()
+                           .addAttr(NewFunc->getContext(),
+                                    AttrListPtr::ReturnIndex,
+                                    OldFunc->getAttributes()
                                      .getRetAttributes()));
     NewFunc->setAttributes(NewFunc->getAttributes()
-                           .addAttr(~0, OldFunc->getAttributes()
+                           .addAttr(NewFunc->getContext(),
+                                    AttrListPtr::FunctionIndex,
+                                    OldFunc->getAttributes()
                                      .getFnAttributes()));
 
   }
@@ -202,14 +206,14 @@
     bool ModuleLevelChanges;
     const char *NameSuffix;
     ClonedCodeInfo *CodeInfo;
-    const TargetData *TD;
+    const DataLayout *TD;
   public:
     PruningFunctionCloner(Function *newFunc, const Function *oldFunc,
                           ValueToValueMapTy &valueMap,
                           bool moduleLevelChanges,
                           const char *nameSuffix, 
                           ClonedCodeInfo *codeInfo,
-                          const TargetData *td)
+                          const DataLayout *td)
     : NewFunc(newFunc), OldFunc(oldFunc),
       VMap(valueMap), ModuleLevelChanges(moduleLevelChanges),
       NameSuffix(nameSuffix), CodeInfo(codeInfo), TD(td) {
@@ -365,7 +369,7 @@
                                      SmallVectorImpl<ReturnInst*> &Returns,
                                      const char *NameSuffix, 
                                      ClonedCodeInfo *CodeInfo,
-                                     const TargetData *TD,
+                                     const DataLayout *TD,
                                      Instruction *TheCall) {
   assert(NameSuffix && "NameSuffix cannot be null!");
   

Modified: llvm/branches/AMDILBackend/lib/Transforms/Utils/CodeExtractor.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Utils/CodeExtractor.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Utils/CodeExtractor.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Utils/CodeExtractor.cpp Tue Jan 15 11:16:16 2013
@@ -346,7 +346,7 @@
                                            header->getName(), M);
   // If the old function is no-throw, so is the new one.
   if (oldFunction->doesNotThrow())
-    newFunction->setDoesNotThrow(true);
+    newFunction->setDoesNotThrow();
   
   newFunction->getBasicBlockList().push_back(newRootNode);
 

Modified: llvm/branches/AMDILBackend/lib/Transforms/Utils/InlineFunction.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Utils/InlineFunction.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Utils/InlineFunction.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Utils/InlineFunction.cpp Tue Jan 15 11:16:16 2013
@@ -27,7 +27,7 @@
 #include "llvm/Analysis/CallGraph.h"
 #include "llvm/Analysis/InstructionSimplify.h"
 #include "llvm/Support/CallSite.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Transforms/Utils/Local.h"
 using namespace llvm;
 
@@ -357,7 +357,7 @@
 
   Type *VoidPtrTy = Type::getInt8PtrTy(Context);
   
-  // Create the alloca.  If we have TargetData, use nice alignment.
+  // Create the alloca.  If we have DataLayout, use nice alignment.
   unsigned Align = 1;
   if (IFI.TD)
     Align = IFI.TD->getPrefTypeAlignment(AggTy);

Modified: llvm/branches/AMDILBackend/lib/Transforms/Utils/LCSSA.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Utils/LCSSA.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Utils/LCSSA.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Utils/LCSSA.cpp Tue Jan 15 11:16:16 2013
@@ -53,6 +53,8 @@
 
     // Cached analysis information for the current function.
     DominatorTree *DT;
+    LoopInfo *LI;
+    ScalarEvolution *SE;
     std::vector<BasicBlock*> LoopBlocks;
     PredIteratorCache PredCache;
     Loop *L;
@@ -117,6 +119,8 @@
   L = TheLoop;
   
   DT = &getAnalysis<DominatorTree>();
+  LI = &getAnalysis<LoopInfo>();
+  SE = getAnalysisIfAvailable<ScalarEvolution>();
 
   // Get the set of exiting blocks.
   SmallVector<BasicBlock*, 8> ExitBlocks;
@@ -156,6 +160,12 @@
       MadeChange |= ProcessInstruction(I, ExitBlocks);
     }
   }
+
+  // If we modified the code, remove any caches about the loop from SCEV to
+  // avoid dangling entries.
+  // FIXME: This is a big hammer, can we clear the cache more selectively?
+  if (SE && MadeChange)
+    SE->forgetLoop(L);
   
   assert(L->isLCSSAForm(*DT));
   PredCache.clear();
@@ -245,7 +255,7 @@
     // Remember that this phi makes the value alive in this block.
     SSAUpdate.AddAvailableValue(ExitBB, PN);
   }
-  
+
   // Rewrite all uses outside the loop in terms of the new PHIs we just
   // inserted.
   for (unsigned i = 0, e = UsesToRewrite.size(); i != e; ++i) {
@@ -260,6 +270,9 @@
 
     if (isa<PHINode>(UserBB->begin()) &&
         isExitBlock(UserBB, ExitBlocks)) {
+      // Tell the VHs that the uses changed. This updates SCEV's caches.
+      if (UsesToRewrite[i]->get()->hasValueHandle())
+        ValueHandleBase::ValueIsRAUWd(*UsesToRewrite[i], UserBB->begin());
       UsesToRewrite[i]->set(UserBB->begin());
       continue;
     }

Modified: llvm/branches/AMDILBackend/lib/Transforms/Utils/Local.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Utils/Local.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Utils/Local.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Utils/Local.cpp Tue Jan 15 11:16:16 2013
@@ -23,6 +23,7 @@
 #include "llvm/Instructions.h"
 #include "llvm/IntrinsicInst.h"
 #include "llvm/Intrinsics.h"
+#include "llvm/MDBuilder.h"
 #include "llvm/Metadata.h"
 #include "llvm/Operator.h"
 #include "llvm/ADT/DenseMap.h"
@@ -38,7 +39,7 @@
 #include "llvm/Support/MathExtras.h"
 #include "llvm/Support/ValueHandle.h"
 #include "llvm/Support/raw_ostream.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 using namespace llvm;
 
 //===----------------------------------------------------------------------===//
@@ -52,7 +53,8 @@
 /// Also calls RecursivelyDeleteTriviallyDeadInstructions() on any branch/switch
 /// conditions and indirectbr addresses this might make dead if
 /// DeleteDeadConditions is true.
-bool llvm::ConstantFoldTerminator(BasicBlock *BB, bool DeleteDeadConditions) {
+bool llvm::ConstantFoldTerminator(BasicBlock *BB, bool DeleteDeadConditions,
+                                  const TargetLibraryInfo *TLI) {
   TerminatorInst *T = BB->getTerminator();
   IRBuilder<> Builder(T);
 
@@ -96,7 +98,7 @@
       Value *Cond = BI->getCondition();
       BI->eraseFromParent();
       if (DeleteDeadConditions)
-        RecursivelyDeleteTriviallyDeadInstructions(Cond);
+        RecursivelyDeleteTriviallyDeadInstructions(Cond, TLI);
       return true;
     }
     return false;
@@ -121,6 +123,27 @@
       // Check to see if this branch is going to the same place as the default
       // dest.  If so, eliminate it as an explicit compare.
       if (i.getCaseSuccessor() == DefaultDest) {
+        MDNode* MD = SI->getMetadata(LLVMContext::MD_prof);
+        // MD should have 2 + NumCases operands.
+        if (MD && MD->getNumOperands() == 2 + SI->getNumCases()) {
+          // Collect branch weights into a vector.
+          SmallVector<uint32_t, 8> Weights;
+          for (unsigned MD_i = 1, MD_e = MD->getNumOperands(); MD_i < MD_e;
+               ++MD_i) {
+            ConstantInt* CI = dyn_cast<ConstantInt>(MD->getOperand(MD_i));
+            assert(CI);
+            Weights.push_back(CI->getValue().getZExtValue());
+          }
+          // Merge weight of this case to the default weight.
+          unsigned idx = i.getCaseIndex();
+          Weights[0] += Weights[idx+1];
+          // Remove weight for this case.
+          std::swap(Weights[idx+1], Weights.back());
+          Weights.pop_back();
+          SI->setMetadata(LLVMContext::MD_prof,
+                          MDBuilder(BB->getContext()).
+                          createBranchWeights(Weights));
+        }
         // Remove this entry.
         DefaultDest->removePredecessor(SI->getParent());
         SI->removeCase(i);
@@ -161,7 +184,7 @@
       Value *Cond = SI->getCondition();
       SI->eraseFromParent();
       if (DeleteDeadConditions)
-        RecursivelyDeleteTriviallyDeadInstructions(Cond);
+        RecursivelyDeleteTriviallyDeadInstructions(Cond, TLI);
       return true;
     }
     
@@ -177,8 +200,20 @@
             "cond");
 
         // Insert the new branch.
-        Builder.CreateCondBr(Cond, FirstCase.getCaseSuccessor(),
-                             SI->getDefaultDest());
+        BranchInst *NewBr = Builder.CreateCondBr(Cond,
+                                FirstCase.getCaseSuccessor(),
+                                SI->getDefaultDest());
+        MDNode* MD = SI->getMetadata(LLVMContext::MD_prof);
+        if (MD && MD->getNumOperands() == 3) {
+          ConstantInt *SICase = dyn_cast<ConstantInt>(MD->getOperand(2));
+          ConstantInt *SIDef = dyn_cast<ConstantInt>(MD->getOperand(1));
+          assert(SICase && SIDef);
+          // The TrueWeight should be the weight for the single case of SI.
+          NewBr->setMetadata(LLVMContext::MD_prof,
+                 MDBuilder(BB->getContext()).
+                 createBranchWeights(SICase->getValue().getZExtValue(),
+                                     SIDef->getValue().getZExtValue()));
+        }
 
         // Delete the old switch.
         SI->eraseFromParent();
@@ -205,7 +240,7 @@
       Value *Address = IBI->getAddress();
       IBI->eraseFromParent();
       if (DeleteDeadConditions)
-        RecursivelyDeleteTriviallyDeadInstructions(Address);
+        RecursivelyDeleteTriviallyDeadInstructions(Address, TLI);
       
       // If we didn't find our destination in the IBI successor list, then we
       // have undefined behavior.  Replace the unconditional branch with an
@@ -230,7 +265,8 @@
 /// isInstructionTriviallyDead - Return true if the result produced by the
 /// instruction is not used, and the instruction has no side effects.
 ///
-bool llvm::isInstructionTriviallyDead(Instruction *I) {
+bool llvm::isInstructionTriviallyDead(Instruction *I,
+                                      const TargetLibraryInfo *TLI) {
   if (!I->use_empty() || isa<TerminatorInst>(I)) return false;
 
   // We don't want the landingpad instruction removed by anything this general.
@@ -265,9 +301,9 @@
       return isa<UndefValue>(II->getArgOperand(1));
   }
 
-  if (isAllocLikeFn(I)) return true;
+  if (isAllocLikeFn(I, TLI)) return true;
 
-  if (CallInst *CI = isFreeCall(I))
+  if (CallInst *CI = isFreeCall(I, TLI))
     if (Constant *C = dyn_cast<Constant>(CI->getArgOperand(0)))
       return C->isNullValue() || isa<UndefValue>(C);
 
@@ -278,9 +314,11 @@
 /// trivially dead instruction, delete it.  If that makes any of its operands
 /// trivially dead, delete them too, recursively.  Return true if any
 /// instructions were deleted.
-bool llvm::RecursivelyDeleteTriviallyDeadInstructions(Value *V) {
+bool
+llvm::RecursivelyDeleteTriviallyDeadInstructions(Value *V,
+                                                 const TargetLibraryInfo *TLI) {
   Instruction *I = dyn_cast<Instruction>(V);
-  if (!I || !I->use_empty() || !isInstructionTriviallyDead(I))
+  if (!I || !I->use_empty() || !isInstructionTriviallyDead(I, TLI))
     return false;
   
   SmallVector<Instruction*, 16> DeadInsts;
@@ -301,7 +339,7 @@
       // operand, and if it is 'trivially' dead, delete it in a future loop
       // iteration.
       if (Instruction *OpI = dyn_cast<Instruction>(OpV))
-        if (isInstructionTriviallyDead(OpI))
+        if (isInstructionTriviallyDead(OpI, TLI))
           DeadInsts.push_back(OpI);
     }
     
@@ -334,19 +372,20 @@
 /// either forms a cycle or is terminated by a trivially dead instruction,
 /// delete it.  If that makes any of its operands trivially dead, delete them
 /// too, recursively.  Return true if a change was made.
-bool llvm::RecursivelyDeleteDeadPHINode(PHINode *PN) {
+bool llvm::RecursivelyDeleteDeadPHINode(PHINode *PN,
+                                        const TargetLibraryInfo *TLI) {
   SmallPtrSet<Instruction*, 4> Visited;
   for (Instruction *I = PN; areAllUsesEqual(I) && !I->mayHaveSideEffects();
        I = cast<Instruction>(*I->use_begin())) {
     if (I->use_empty())
-      return RecursivelyDeleteTriviallyDeadInstructions(I);
+      return RecursivelyDeleteTriviallyDeadInstructions(I, TLI);
 
     // If we find an instruction more than once, we're on a cycle that
     // won't prove fruitful.
     if (!Visited.insert(I)) {
       // Break the cycle and delete the instruction and its operands.
       I->replaceAllUsesWith(UndefValue::get(I->getType()));
-      (void)RecursivelyDeleteTriviallyDeadInstructions(I);
+      (void)RecursivelyDeleteTriviallyDeadInstructions(I, TLI);
       return true;
     }
   }
@@ -358,7 +397,8 @@
 ///
 /// This returns true if it changed the code, note that it can delete
 /// instructions in other blocks as well in this block.
-bool llvm::SimplifyInstructionsInBlock(BasicBlock *BB, const TargetData *TD) {
+bool llvm::SimplifyInstructionsInBlock(BasicBlock *BB, const DataLayout *TD,
+                                       const TargetLibraryInfo *TLI) {
   bool MadeChange = false;
 
 #ifndef NDEBUG
@@ -381,7 +421,7 @@
       continue;
     }
 
-    MadeChange |= RecursivelyDeleteTriviallyDeadInstructions(Inst);
+    MadeChange |= RecursivelyDeleteTriviallyDeadInstructions(Inst, TLI);
     if (BIHandle != BI)
       BI = BB->begin();
   }
@@ -405,7 +445,7 @@
 /// .. and delete the predecessor corresponding to the '1', this will attempt to
 /// recursively fold the and to 0.
 void llvm::RemovePredecessorAndSimplify(BasicBlock *BB, BasicBlock *Pred,
-                                        TargetData *TD) {
+                                        DataLayout *TD) {
   // This only adjusts blocks with PHI nodes.
   if (!isa<PHINode>(BB->begin()))
     return;
@@ -720,7 +760,7 @@
 /// their preferred alignment from the beginning.
 ///
 static unsigned enforceKnownAlignment(Value *V, unsigned Align,
-                                      unsigned PrefAlign, const TargetData *TD) {
+                                      unsigned PrefAlign, const DataLayout *TD) {
   V = V->stripPointerCasts();
 
   if (AllocaInst *AI = dyn_cast<AllocaInst>(V)) {
@@ -763,7 +803,7 @@
 /// and it is more than the alignment of the ultimate object, see if we can
 /// increase the alignment of the ultimate object, making this check succeed.
 unsigned llvm::getOrEnforceKnownAlignment(Value *V, unsigned PrefAlign,
-                                          const TargetData *TD) {
+                                          const DataLayout *TD) {
   assert(V->getType()->isPointerTy() &&
          "getOrEnforceKnownAlignment expects a pointer!");
   unsigned BitWidth = TD ? TD->getPointerSizeInBits() : 64;

Modified: llvm/branches/AMDILBackend/lib/Transforms/Utils/LoopSimplify.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Utils/LoopSimplify.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Utils/LoopSimplify.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Utils/LoopSimplify.cpp Tue Jan 15 11:16:16 2013
@@ -46,6 +46,7 @@
 #include "llvm/LLVMContext.h"
 #include "llvm/Type.h"
 #include "llvm/Analysis/AliasAnalysis.h"
+#include "llvm/Analysis/DependenceAnalysis.h"
 #include "llvm/Analysis/Dominators.h"
 #include "llvm/Analysis/InstructionSimplify.h"
 #include "llvm/Analysis/LoopPass.h"
@@ -89,6 +90,7 @@
 
       AU.addPreserved<AliasAnalysis>();
       AU.addPreserved<ScalarEvolution>();
+      AU.addPreserved<DependenceAnalysis>();
       AU.addPreservedID(BreakCriticalEdgesID);  // No critical edges added.
     }
 
@@ -194,6 +196,11 @@
 
           BI->setCondition(ConstantInt::get(Cond->getType(),
                                             !L->contains(BI->getSuccessor(0))));
+
+          // This may make the loop analyzable, force SCEV recomputation.
+          if (SE)
+            SE->forgetLoop(L);
+
           Changed = true;
         }
       }

Modified: llvm/branches/AMDILBackend/lib/Transforms/Utils/PromoteMemoryToRegister.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Utils/PromoteMemoryToRegister.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Utils/PromoteMemoryToRegister.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Utils/PromoteMemoryToRegister.cpp Tue Jan 15 11:16:16 2013
@@ -212,9 +212,13 @@
     ///
     DenseMap<AllocaInst*, unsigned>  AllocaLookup;
 
-    /// NewPhiNodes - The PhiNodes we're adding.
+    /// NewPhiNodes - The PhiNodes we're adding.  That map is used to simplify
+    /// some Phi nodes as we iterate over it, so it should have deterministic
+    /// iterators.  We could use a MapVector, but since we already maintain a
+    /// map from BasicBlock* to a stable numbering (BBNumbers), the DenseMap is
+    /// more efficient (also supports removal).
     ///
-    DenseMap<std::pair<BasicBlock*, unsigned>, PHINode*> NewPhiNodes;
+    DenseMap<std::pair<unsigned, unsigned>, PHINode*> NewPhiNodes;
     
     /// PhiToAllocaMap - For each PHI node, keep track of which entry in Allocas
     /// it corresponds to.
@@ -588,7 +592,11 @@
   while (EliminatedAPHI) {
     EliminatedAPHI = false;
     
-    for (DenseMap<std::pair<BasicBlock*, unsigned>, PHINode*>::iterator I =
+    // Iterating over NewPhiNodes is deterministic, so it is safe to try to
+    // simplify and RAUW them as we go.  If it was not, we could add uses to
+    // the values we replace with in a non deterministic order, thus creating
+    // non deterministic def->use chains.
+    for (DenseMap<std::pair<unsigned, unsigned>, PHINode*>::iterator I =
            NewPhiNodes.begin(), E = NewPhiNodes.end(); I != E;) {
       PHINode *PN = I->second;
 
@@ -612,7 +620,7 @@
   // have incoming values for all predecessors.  Loop over all PHI nodes we have
   // created, inserting undef values if they are missing any incoming values.
   //
-  for (DenseMap<std::pair<BasicBlock*, unsigned>, PHINode*>::iterator I =
+  for (DenseMap<std::pair<unsigned, unsigned>, PHINode*>::iterator I =
          NewPhiNodes.begin(), E = NewPhiNodes.end(); I != E; ++I) {
     // We want to do this once per basic block.  As such, only process a block
     // when we find the PHI that is the first entry in the block.
@@ -992,7 +1000,7 @@
 bool PromoteMem2Reg::QueuePhiNode(BasicBlock *BB, unsigned AllocaNo,
                                   unsigned &Version) {
   // Look up the basic-block in question.
-  PHINode *&PN = NewPhiNodes[std::make_pair(BB, AllocaNo)];
+  PHINode *&PN = NewPhiNodes[std::make_pair(BBNumbers[BB], AllocaNo)];
 
   // If the BB already has a phi node added for the i'th alloca then we're done!
   if (PN) return false;

Modified: llvm/branches/AMDILBackend/lib/Transforms/Utils/SSAUpdater.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Utils/SSAUpdater.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Utils/SSAUpdater.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Utils/SSAUpdater.cpp Tue Jan 15 11:16:16 2013
@@ -39,7 +39,7 @@
   : AV(0), ProtoType(0), ProtoName(), InsertedPHIs(NewPHI) {}
 
 SSAUpdater::~SSAUpdater() {
-  delete &getAvailableVals(AV);
+  delete static_cast<AvailableValsTy*>(AV);
 }
 
 /// Initialize - Reset this object to get ready for a new set of SSA
@@ -214,6 +214,11 @@
   else
     V = GetValueInMiddleOfBlock(User->getParent());
 
+  // Notify that users of the existing value that it is being replaced.
+  Value *OldVal = U.get();
+  if (OldVal != V && OldVal->hasValueHandle())
+    ValueHandleBase::ValueIsRAUWd(OldVal, V);
+
   U.set(V);
 }
 

Modified: llvm/branches/AMDILBackend/lib/Transforms/Utils/SimplifyCFG.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Utils/SimplifyCFG.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Utils/SimplifyCFG.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Utils/SimplifyCFG.cpp Tue Jan 15 11:16:16 2013
@@ -14,6 +14,7 @@
 #define DEBUG_TYPE "simplifycfg"
 #include "llvm/Transforms/Utils/Local.h"
 #include "llvm/Constants.h"
+#include "llvm/DataLayout.h"
 #include "llvm/DerivedTypes.h"
 #include "llvm/GlobalVariable.h"
 #include "llvm/IRBuilder.h"
@@ -22,6 +23,7 @@
 #include "llvm/LLVMContext.h"
 #include "llvm/MDBuilder.h"
 #include "llvm/Metadata.h"
+#include "llvm/Module.h"
 #include "llvm/Operator.h"
 #include "llvm/Type.h"
 #include "llvm/ADT/DenseMap.h"
@@ -38,7 +40,7 @@
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/NoFolder.h"
 #include "llvm/Support/raw_ostream.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/TargetTransformInfo.h"
 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
 #include <algorithm>
 #include <set>
@@ -53,6 +55,13 @@
 DupRet("simplifycfg-dup-ret", cl::Hidden, cl::init(false),
        cl::desc("Duplicate return instructions into unconditional branches"));
 
+static cl::opt<bool>
+SinkCommon("simplifycfg-sink-common", cl::Hidden, cl::init(true),
+       cl::desc("Sink common instructions down to the end block"));
+
+STATISTIC(NumBitMaps, "Number of switch instructions turned into bitmaps");
+STATISTIC(NumLookupTables, "Number of switch instructions turned into lookup tables");
+STATISTIC(NumSinkCommons, "Number of common instructions sunk down to the end block");
 STATISTIC(NumSpeculations, "Number of speculative executed instructions");
 
 namespace {
@@ -68,10 +77,13 @@
       // Comparing pointers is ok as we only rely on the order for uniquing.
       return Value < RHS.Value;
     }
+
+    bool operator==(BasicBlock *RHSDest) const { return Dest == RHSDest; }
   };
 
 class SimplifyCFGOpt {
-  const TargetData *const TD;
+  const DataLayout *const TD;
+  const TargetTransformInfo *const TTI;
 
   Value *isValueEqualityComparison(TerminatorInst *TI);
   BasicBlock *GetValueEqualityComparisonCases(TerminatorInst *TI,
@@ -91,7 +103,8 @@
   bool SimplifyCondBranch(BranchInst *BI, IRBuilder <>&Builder);
 
 public:
-  explicit SimplifyCFGOpt(const TargetData *td) : TD(td) {}
+  SimplifyCFGOpt(const DataLayout *td, const TargetTransformInfo *tti)
+      : TD(td), TTI(tti) {}
   bool run(BasicBlock *BB);
 };
 }
@@ -101,14 +114,14 @@
 ///
 static bool SafeToMergeTerminators(TerminatorInst *SI1, TerminatorInst *SI2) {
   if (SI1 == SI2) return false;  // Can't merge with self!
-  
+
   // It is not safe to merge these two switch instructions if they have a common
   // successor, and if that successor has a PHI node, and if *that* PHI node has
   // conflicting incoming values from the two switch blocks.
   BasicBlock *SI1BB = SI1->getParent();
   BasicBlock *SI2BB = SI2->getParent();
   SmallPtrSet<BasicBlock*, 16> SI1Succs(succ_begin(SI1BB), succ_end(SI1BB));
-  
+
   for (succ_iterator I = succ_begin(SI2BB), E = succ_end(SI2BB); I != E; ++I)
     if (SI1Succs.count(*I))
       for (BasicBlock::iterator BBI = (*I)->begin();
@@ -118,7 +131,7 @@
             PN->getIncomingValueForBlock(SI2BB))
           return false;
       }
-        
+
   return true;
 }
 
@@ -135,7 +148,7 @@
   assert(SI1->isUnconditional() && SI2->isConditional());
 
   // We fold the unconditional branch if we can easily update all PHI nodes in
-  // common successors: 
+  // common successors:
   // 1> We have a constant incoming value for the conditional branch;
   // 2> We have "Cond" as the incoming value for the unconditional branch;
   // 3> SI2->getCondition() and Cond have same operands.
@@ -170,7 +183,7 @@
 static void AddPredecessorToBlock(BasicBlock *Succ, BasicBlock *NewPred,
                                   BasicBlock *ExistPred) {
   if (!isa<PHINode>(Succ->begin())) return; // Quick exit if nothing to do
-  
+
   PHINode *PN;
   for (BasicBlock::iterator I = Succ->begin();
        (PN = dyn_cast<PHINode>(I)); ++I)
@@ -222,7 +235,7 @@
     // doesn't dominate BB.
     if (Pred2->getSinglePredecessor() == 0)
       return 0;
-    
+
     // If we found a conditional branch predecessor, make sure that it branches
     // to BB and Pred2Br.  If it doesn't, this isn't an "if statement".
     if (Pred1Br->getSuccessor(0) == BB &&
@@ -252,7 +265,7 @@
   // Otherwise, if this is a conditional branch, then we can use it!
   BranchInst *BI = dyn_cast<BranchInst>(CommonPred->getTerminator());
   if (BI == 0) return 0;
-  
+
   assert(BI->isConditional() && "Two successors but not conditional?");
   if (BI->getSuccessor(0) == Pred1) {
     IfTrue = Pred1;
@@ -345,7 +358,7 @@
   // If we aren't allowing aggressive promotion anymore, then don't consider
   // instructions in the 'if region'.
   if (AggressiveInsts == 0) return false;
-  
+
   // If we have seen this instruction before, don't count it again.
   if (AggressiveInsts->count(I)) return true;
 
@@ -374,7 +387,7 @@
 
 /// GetConstantInt - Extract ConstantInt from value, looking through IntToPtr
 /// and PointerNullValue. Return NULL if value is not a constant int.
-static ConstantInt *GetConstantInt(Value *V, const TargetData *TD) {
+static ConstantInt *GetConstantInt(Value *V, const DataLayout *TD) {
   // Normal constant int.
   ConstantInt *CI = dyn_cast<ConstantInt>(V);
   if (CI || !TD || !isa<Constant>(V) || !V->getType()->isPointerTy())
@@ -382,7 +395,7 @@
 
   // This is some kind of pointer constant. Turn it into a pointer-sized
   // ConstantInt if possible.
-  IntegerType *PtrTy = TD->getIntPtrType(V->getContext());
+  IntegerType *PtrTy = cast<IntegerType>(TD->getIntPtrType(V->getType()));
 
   // Null pointer means 0, see SelectionDAGBuilder::getValue(const Value*).
   if (isa<ConstantPointerNull>(V))
@@ -408,10 +421,10 @@
 /// Values vector.
 static Value *
 GatherConstantCompares(Value *V, std::vector<ConstantInt*> &Vals, Value *&Extra,
-                       const TargetData *TD, bool isEQ, unsigned &UsedICmps) {
+                       const DataLayout *TD, bool isEQ, unsigned &UsedICmps) {
   Instruction *I = dyn_cast<Instruction>(V);
   if (I == 0) return 0;
-  
+
   // If this is an icmp against a constant, handle this as one of the cases.
   if (ICmpInst *ICI = dyn_cast<ICmpInst>(I)) {
     if (ConstantInt *C = GetConstantInt(I->getOperand(1), TD)) {
@@ -420,21 +433,21 @@
         Vals.push_back(C);
         return I->getOperand(0);
       }
-      
+
       // If we have "x ult 3" comparison, for example, then we can add 0,1,2 to
       // the set.
       ConstantRange Span =
         ConstantRange::makeICmpRegion(ICI->getPredicate(), C->getValue());
-      
+
       // If this is an and/!= check then we want to optimize "x ugt 2" into
       // x != 0 && x != 1.
       if (!isEQ)
         Span = Span.inverse();
-      
+
       // If there are a ton of values, we don't want to make a ginormous switch.
       if (Span.getSetSize().ugt(8) || Span.isEmptySet())
         return 0;
-      
+
       for (APInt Tmp = Span.getLower(); Tmp != Span.getUpper(); ++Tmp)
         Vals.push_back(ConstantInt::get(V->getContext(), Tmp));
       UsedICmps++;
@@ -442,11 +455,11 @@
     }
     return 0;
   }
-  
+
   // Otherwise, we can only handle an | or &, depending on isEQ.
   if (I->getOpcode() != (isEQ ? Instruction::Or : Instruction::And))
     return 0;
-  
+
   unsigned NumValsBeforeLHS = Vals.size();
   unsigned UsedICmpsBeforeLHS = UsedICmps;
   if (Value *LHS = GatherConstantCompares(I->getOperand(0), Vals, Extra, TD,
@@ -467,12 +480,12 @@
       Extra = I->getOperand(1);
       return LHS;
     }
-    
+
     Vals.resize(NumValsBeforeLHS);
     UsedICmps = UsedICmpsBeforeLHS;
     return 0;
   }
-  
+
   // If the LHS can't be folded in, but Extra is available and RHS can, try to
   // use LHS as Extra.
   if (Extra == 0 || Extra == I->getOperand(0)) {
@@ -484,7 +497,7 @@
     assert(Vals.size() == NumValsBeforeLHS);
     Extra = OldExtra;
   }
-  
+
   return 0;
 }
 
@@ -556,11 +569,7 @@
 /// in the list that match the specified block.
 static void EliminateBlockCases(BasicBlock *BB,
                               std::vector<ValueEqualityComparisonCase> &Cases) {
-  for (unsigned i = 0, e = Cases.size(); i != e; ++i)
-    if (Cases[i].Dest == BB) {
-      Cases.erase(Cases.begin()+i);
-      --i; --e;
-    }
+  Cases.erase(std::remove(Cases.begin(), Cases.end(), BB), Cases.end());
 }
 
 /// ValuesOverlap - Return true if there are any keys in C1 that exist in C2 as
@@ -615,6 +624,9 @@
   assert(ThisVal && "This isn't a value comparison!!");
   if (ThisVal != PredVal) return false;  // Different predicates.
 
+  // TODO: Preserve branch weight metadata, similarly to how
+  // FoldValueComparisonIntoPredecessors preserves it.
+
   // Find out information about when control will move from Pred to TI's block.
   std::vector<ValueEqualityComparisonCase> PredCases;
   BasicBlock *PredDef = GetValueEqualityComparisonCases(Pred->getTerminator(),
@@ -634,7 +646,7 @@
     // can simplify TI.
     if (!ValuesOverlap(PredCases, ThisCases))
       return false;
-    
+
     if (isa<BranchInst>(TI)) {
       // Okay, one of the successors of this condbr is dead.  Convert it to a
       // uncond br.
@@ -652,7 +664,7 @@
       EraseTerminatorInstAndDCECond(TI);
       return true;
     }
-      
+
     SwitchInst *SI = cast<SwitchInst>(TI);
     // Okay, TI has cases that are statically dead, prune them away.
     SmallPtrSet<Constant*, 16> DeadCases;
@@ -662,18 +674,37 @@
     DEBUG(dbgs() << "Threading pred instr: " << *Pred->getTerminator()
                  << "Through successor TI: " << *TI);
 
+    // Collect branch weights into a vector.
+    SmallVector<uint32_t, 8> Weights;
+    MDNode* MD = SI->getMetadata(LLVMContext::MD_prof);
+    bool HasWeight = MD && (MD->getNumOperands() == 2 + SI->getNumCases());
+    if (HasWeight)
+      for (unsigned MD_i = 1, MD_e = MD->getNumOperands(); MD_i < MD_e;
+           ++MD_i) {
+        ConstantInt* CI = dyn_cast<ConstantInt>(MD->getOperand(MD_i));
+        assert(CI);
+        Weights.push_back(CI->getValue().getZExtValue());
+      }
     for (SwitchInst::CaseIt i = SI->case_end(), e = SI->case_begin(); i != e;) {
       --i;
       if (DeadCases.count(i.getCaseValue())) {
+        if (HasWeight) {
+          std::swap(Weights[i.getCaseIndex()+1], Weights.back());
+          Weights.pop_back();
+        }
         i.getCaseSuccessor()->removePredecessor(TI->getParent());
         SI->removeCase(i);
       }
     }
+    if (HasWeight && Weights.size() >= 2)
+      SI->setMetadata(LLVMContext::MD_prof,
+                      MDBuilder(SI->getParent()->getContext()).
+                      createBranchWeights(Weights));
 
     DEBUG(dbgs() << "Leaving: " << *TI << "\n");
     return true;
   }
-  
+
   // Otherwise, TI's block must correspond to some matched value.  Find out
   // which value (or set of values) this is.
   ConstantInt *TIV = 0;
@@ -729,8 +760,8 @@
 }
 
 static int ConstantIntSortPredicate(const void *P1, const void *P2) {
-  const ConstantInt *LHS = *(const ConstantInt**)P1;
-  const ConstantInt *RHS = *(const ConstantInt**)P2;
+  const ConstantInt *LHS = *(const ConstantInt*const*)P1;
+  const ConstantInt *RHS = *(const ConstantInt*const*)P2;
   if (LHS->getValue().ult(RHS->getValue()))
     return 1;
   if (LHS->getValue() == RHS->getValue())
@@ -738,6 +769,56 @@
   return -1;
 }
 
+static inline bool HasBranchWeights(const Instruction* I) {
+  MDNode* ProfMD = I->getMetadata(LLVMContext::MD_prof);
+  if (ProfMD && ProfMD->getOperand(0))
+    if (MDString* MDS = dyn_cast<MDString>(ProfMD->getOperand(0)))
+      return MDS->getString().equals("branch_weights");
+
+  return false;
+}
+
+/// Get Weights of a given TerminatorInst, the default weight is at the front
+/// of the vector. If TI is a conditional eq, we need to swap the branch-weight
+/// metadata.
+static void GetBranchWeights(TerminatorInst *TI,
+                             SmallVectorImpl<uint64_t> &Weights) {
+  MDNode* MD = TI->getMetadata(LLVMContext::MD_prof);
+  assert(MD);
+  for (unsigned i = 1, e = MD->getNumOperands(); i < e; ++i) {
+    ConstantInt* CI = dyn_cast<ConstantInt>(MD->getOperand(i));
+    assert(CI);
+    Weights.push_back(CI->getValue().getZExtValue());
+  }
+
+  // If TI is a conditional eq, the default case is the false case,
+  // and the corresponding branch-weight data is at index 2. We swap the
+  // default weight to be the first entry.
+  if (BranchInst* BI = dyn_cast<BranchInst>(TI)) {
+    assert(Weights.size() == 2);
+    ICmpInst *ICI = cast<ICmpInst>(BI->getCondition());
+    if (ICI->getPredicate() == ICmpInst::ICMP_EQ)
+      std::swap(Weights.front(), Weights.back());
+  }
+}
+
+/// Sees if any of the weights are too big for a uint32_t, and halves all the
+/// weights if any are.
+static void FitWeights(MutableArrayRef<uint64_t> Weights) {
+  bool Halve = false;
+  for (unsigned i = 0; i < Weights.size(); ++i)
+    if (Weights[i] > UINT_MAX) {
+      Halve = true;
+      break;
+    }
+
+  if (! Halve)
+    return;
+
+  for (unsigned i = 0; i < Weights.size(); ++i)
+    Weights[i] /= 2;
+}
+
 /// FoldValueComparisonIntoPredecessors - The specified terminator is a value
 /// equality comparison instruction (either a switch or a branch on "X == c").
 /// See if any of the predecessors of the terminator block are value comparisons
@@ -770,6 +851,31 @@
       // build.
       SmallVector<BasicBlock*, 8> NewSuccessors;
 
+      // Update the branch weight metadata along the way
+      SmallVector<uint64_t, 8> Weights;
+      bool PredHasWeights = HasBranchWeights(PTI);
+      bool SuccHasWeights = HasBranchWeights(TI);
+
+      if (PredHasWeights) {
+        GetBranchWeights(PTI, Weights);
+        // branch-weight metadata is inconsistant here.
+        if (Weights.size() != 1 + PredCases.size())
+          PredHasWeights = SuccHasWeights = false;
+      } else if (SuccHasWeights)
+        // If there are no predecessor weights but there are successor weights,
+        // populate Weights with 1, which will later be scaled to the sum of
+        // successor's weights
+        Weights.assign(1 + PredCases.size(), 1);
+
+      SmallVector<uint64_t, 8> SuccWeights;
+      if (SuccHasWeights) {
+        GetBranchWeights(TI, SuccWeights);
+        // branch-weight metadata is inconsistant here.
+        if (SuccWeights.size() != 1 + BBCases.size())
+          PredHasWeights = SuccHasWeights = false;
+      } else if (PredHasWeights)
+        SuccWeights.assign(1 + BBCases.size(), 1);
+
       if (PredDefault == BB) {
         // If this is the default destination from PTI, only the edges in TI
         // that don't occur in PTI, or that branch to BB will be activated.
@@ -780,6 +886,14 @@
           else {
             // The default destination is BB, we don't need explicit targets.
             std::swap(PredCases[i], PredCases.back());
+
+            if (PredHasWeights || SuccHasWeights) {
+              // Increase weight for the default case.
+              Weights[0] += Weights[i+1];
+              std::swap(Weights[i+1], Weights.back());
+              Weights.pop_back();
+            }
+
             PredCases.pop_back();
             --i; --e;
           }
@@ -790,21 +904,47 @@
           PredDefault = BBDefault;
           NewSuccessors.push_back(BBDefault);
         }
+
+        unsigned CasesFromPred = Weights.size();
+        uint64_t ValidTotalSuccWeight = 0;
         for (unsigned i = 0, e = BBCases.size(); i != e; ++i)
           if (!PTIHandled.count(BBCases[i].Value) &&
               BBCases[i].Dest != BBDefault) {
             PredCases.push_back(BBCases[i]);
             NewSuccessors.push_back(BBCases[i].Dest);
+            if (SuccHasWeights || PredHasWeights) {
+              // The default weight is at index 0, so weight for the ith case
+              // should be at index i+1. Scale the cases from successor by
+              // PredDefaultWeight (Weights[0]).
+              Weights.push_back(Weights[0] * SuccWeights[i+1]);
+              ValidTotalSuccWeight += SuccWeights[i+1];
+            }
           }
 
+        if (SuccHasWeights || PredHasWeights) {
+          ValidTotalSuccWeight += SuccWeights[0];
+          // Scale the cases from predecessor by ValidTotalSuccWeight.
+          for (unsigned i = 1; i < CasesFromPred; ++i)
+            Weights[i] *= ValidTotalSuccWeight;
+          // Scale the default weight by SuccDefaultWeight (SuccWeights[0]).
+          Weights[0] *= SuccWeights[0];
+        }
       } else {
         // If this is not the default destination from PSI, only the edges
         // in SI that occur in PSI with a destination of BB will be
         // activated.
         std::set<ConstantInt*, ConstantIntOrdering> PTIHandled;
+        std::map<ConstantInt*, uint64_t> WeightsForHandled;
         for (unsigned i = 0, e = PredCases.size(); i != e; ++i)
           if (PredCases[i].Dest == BB) {
             PTIHandled.insert(PredCases[i].Value);
+
+            if (PredHasWeights || SuccHasWeights) {
+              WeightsForHandled[PredCases[i].Value] = Weights[i+1];
+              std::swap(Weights[i+1], Weights.back());
+              Weights.pop_back();
+            }
+
             std::swap(PredCases[i], PredCases.back());
             PredCases.pop_back();
             --i; --e;
@@ -815,6 +955,8 @@
         for (unsigned i = 0, e = BBCases.size(); i != e; ++i)
           if (PTIHandled.count(BBCases[i].Value)) {
             // If this is one we are capable of getting...
+            if (PredHasWeights || SuccHasWeights)
+              Weights.push_back(WeightsForHandled[BBCases[i].Value]);
             PredCases.push_back(BBCases[i]);
             NewSuccessors.push_back(BBCases[i].Dest);
             PTIHandled.erase(BBCases[i].Value);// This constant is taken care of
@@ -822,9 +964,11 @@
 
         // If there are any constants vectored to BB that TI doesn't handle,
         // they must go to the default destination of TI.
-        for (std::set<ConstantInt*, ConstantIntOrdering>::iterator I = 
+        for (std::set<ConstantInt*, ConstantIntOrdering>::iterator I =
                                     PTIHandled.begin(),
                E = PTIHandled.end(); I != E; ++I) {
+          if (PredHasWeights || SuccHasWeights) 
+            Weights.push_back(WeightsForHandled[*I]); 
           PredCases.push_back(ValueEqualityComparisonCase(*I, BBDefault));
           NewSuccessors.push_back(BBDefault);
         }
@@ -839,7 +983,7 @@
       Builder.SetInsertPoint(PTI);
       // Convert pointer to int before we switch.
       if (CV->getType()->isPointerTy()) {
-        assert(TD && "Cannot switch on pointer without TargetData");
+        assert(TD && "Cannot switch on pointer without DataLayout");
         CV = Builder.CreatePtrToInt(CV, TD->getIntPtrType(CV->getContext()),
                                     "magicptr");
       }
@@ -851,6 +995,17 @@
       for (unsigned i = 0, e = PredCases.size(); i != e; ++i)
         NewSI->addCase(PredCases[i].Value, PredCases[i].Dest);
 
+      if (PredHasWeights || SuccHasWeights) {
+        // Halve the weights if any of them cannot fit in an uint32_t
+        FitWeights(Weights);
+
+        SmallVector<uint32_t, 8> MDWeights(Weights.begin(), Weights.end());
+
+        NewSI->setMetadata(LLVMContext::MD_prof,
+                           MDBuilder(BB->getContext()).
+                           createBranchWeights(MDWeights));
+      }
+
       EraseTerminatorInstAndDCECond(PTI);
 
       // Okay, last check.  If BB is still a successor of PSI, then we must
@@ -984,11 +1139,11 @@
       Value *BB1V = PN->getIncomingValueForBlock(BB1);
       Value *BB2V = PN->getIncomingValueForBlock(BB2);
       if (BB1V == BB2V) continue;
-      
+
       // These values do not agree.  Insert a select instruction before NT
       // that determines the right value.
       SelectInst *&SI = InsertedSelects[std::make_pair(BB1V, BB2V)];
-      if (SI == 0) 
+      if (SI == 0)
         SI = cast<SelectInst>
           (Builder.CreateSelect(BI->getCondition(), BB1V, BB2V,
                                 BB1V->getName()+"."+BB2V->getName()));
@@ -1008,6 +1163,175 @@
   return true;
 }
 
+/// SinkThenElseCodeToEnd - Given an unconditional branch that goes to BBEnd,
+/// check whether BBEnd has only two predecessors and the other predecessor
+/// ends with an unconditional branch. If it is true, sink any common code
+/// in the two predecessors to BBEnd.
+static bool SinkThenElseCodeToEnd(BranchInst *BI1) {
+  assert(BI1->isUnconditional());
+  BasicBlock *BB1 = BI1->getParent();
+  BasicBlock *BBEnd = BI1->getSuccessor(0);
+
+  // Check that BBEnd has two predecessors and the other predecessor ends with
+  // an unconditional branch.
+  pred_iterator PI = pred_begin(BBEnd), PE = pred_end(BBEnd);
+  BasicBlock *Pred0 = *PI++;
+  if (PI == PE) // Only one predecessor.
+    return false;
+  BasicBlock *Pred1 = *PI++;
+  if (PI != PE) // More than two predecessors.
+    return false;
+  BasicBlock *BB2 = (Pred0 == BB1) ? Pred1 : Pred0;
+  BranchInst *BI2 = dyn_cast<BranchInst>(BB2->getTerminator());
+  if (!BI2 || !BI2->isUnconditional())
+    return false;
+
+  // Gather the PHI nodes in BBEnd.
+  std::map<Value*, std::pair<Value*, PHINode*> > MapValueFromBB1ToBB2;
+  Instruction *FirstNonPhiInBBEnd = 0;
+  for (BasicBlock::iterator I = BBEnd->begin(), E = BBEnd->end();
+       I != E; ++I) {
+    if (PHINode *PN = dyn_cast<PHINode>(I)) {
+      Value *BB1V = PN->getIncomingValueForBlock(BB1);
+      Value *BB2V = PN->getIncomingValueForBlock(BB2); 
+      MapValueFromBB1ToBB2[BB1V] = std::make_pair(BB2V, PN);
+    } else {
+      FirstNonPhiInBBEnd = &*I;
+      break;
+    }
+  }
+  if (!FirstNonPhiInBBEnd)
+    return false;
+  
+
+  // This does very trivial matching, with limited scanning, to find identical
+  // instructions in the two blocks.  We scan backward for obviously identical
+  // instructions in an identical order.
+  BasicBlock::InstListType::reverse_iterator RI1 = BB1->getInstList().rbegin(),
+      RE1 = BB1->getInstList().rend(), RI2 = BB2->getInstList().rbegin(),
+      RE2 = BB2->getInstList().rend();
+  // Skip debug info.
+  while (RI1 != RE1 && isa<DbgInfoIntrinsic>(&*RI1)) ++RI1;
+  if (RI1 == RE1)
+    return false;
+  while (RI2 != RE2 && isa<DbgInfoIntrinsic>(&*RI2)) ++RI2;
+  if (RI2 == RE2)
+    return false;
+  // Skip the unconditional branches.
+  ++RI1;
+  ++RI2;
+
+  bool Changed = false;
+  while (RI1 != RE1 && RI2 != RE2) {
+    // Skip debug info.
+    while (RI1 != RE1 && isa<DbgInfoIntrinsic>(&*RI1)) ++RI1;
+    if (RI1 == RE1)
+      return Changed;
+    while (RI2 != RE2 && isa<DbgInfoIntrinsic>(&*RI2)) ++RI2;
+    if (RI2 == RE2)
+      return Changed;
+
+    Instruction *I1 = &*RI1, *I2 = &*RI2;
+    // I1 and I2 should have a single use in the same PHI node, and they
+    // perform the same operation.
+    // Cannot move control-flow-involving, volatile loads, vaarg, etc.
+    if (isa<PHINode>(I1) || isa<PHINode>(I2) ||
+        isa<TerminatorInst>(I1) || isa<TerminatorInst>(I2) ||
+        isa<LandingPadInst>(I1) || isa<LandingPadInst>(I2) ||
+        isa<AllocaInst>(I1) || isa<AllocaInst>(I2) ||
+        I1->mayHaveSideEffects() || I2->mayHaveSideEffects() ||
+        I1->mayReadOrWriteMemory() || I2->mayReadOrWriteMemory() ||
+        !I1->hasOneUse() || !I2->hasOneUse() ||
+        MapValueFromBB1ToBB2.find(I1) == MapValueFromBB1ToBB2.end() ||
+        MapValueFromBB1ToBB2[I1].first != I2)
+      return Changed;
+
+    // Check whether we should swap the operands of ICmpInst.
+    ICmpInst *ICmp1 = dyn_cast<ICmpInst>(I1), *ICmp2 = dyn_cast<ICmpInst>(I2);
+    bool SwapOpnds = false;
+    if (ICmp1 && ICmp2 &&
+        ICmp1->getOperand(0) != ICmp2->getOperand(0) &&
+        ICmp1->getOperand(1) != ICmp2->getOperand(1) &&
+        (ICmp1->getOperand(0) == ICmp2->getOperand(1) ||
+         ICmp1->getOperand(1) == ICmp2->getOperand(0))) {
+      ICmp2->swapOperands();
+      SwapOpnds = true;
+    }
+    if (!I1->isSameOperationAs(I2)) {
+      if (SwapOpnds)
+        ICmp2->swapOperands();
+      return Changed;
+    }
+
+    // The operands should be either the same or they need to be generated
+    // with a PHI node after sinking. We only handle the case where there is
+    // a single pair of different operands.
+    Value *DifferentOp1 = 0, *DifferentOp2 = 0;
+    unsigned Op1Idx = 0;
+    for (unsigned I = 0, E = I1->getNumOperands(); I != E; ++I) {
+      if (I1->getOperand(I) == I2->getOperand(I))
+        continue;
+      // Early exit if we have more-than one pair of different operands or
+      // the different operand is already in MapValueFromBB1ToBB2.
+      // Early exit if we need a PHI node to replace a constant.
+      if (DifferentOp1 ||
+          MapValueFromBB1ToBB2.find(I1->getOperand(I)) !=
+          MapValueFromBB1ToBB2.end() ||
+          isa<Constant>(I1->getOperand(I)) ||
+          isa<Constant>(I2->getOperand(I))) {
+        // If we can't sink the instructions, undo the swapping.
+        if (SwapOpnds)
+          ICmp2->swapOperands();
+        return Changed;
+      }
+      DifferentOp1 = I1->getOperand(I);
+      Op1Idx = I;
+      DifferentOp2 = I2->getOperand(I);
+    }
+
+    // We insert the pair of different operands to MapValueFromBB1ToBB2 and
+    // remove (I1, I2) from MapValueFromBB1ToBB2.
+    if (DifferentOp1) {
+      PHINode *NewPN = PHINode::Create(DifferentOp1->getType(), 2,
+                                       DifferentOp1->getName() + ".sink",
+                                       BBEnd->begin());
+      MapValueFromBB1ToBB2[DifferentOp1] = std::make_pair(DifferentOp2, NewPN);
+      // I1 should use NewPN instead of DifferentOp1.
+      I1->setOperand(Op1Idx, NewPN);
+      NewPN->addIncoming(DifferentOp1, BB1);
+      NewPN->addIncoming(DifferentOp2, BB2);
+      DEBUG(dbgs() << "Create PHI node " << *NewPN << "\n";);
+    }
+    PHINode *OldPN = MapValueFromBB1ToBB2[I1].second;
+    MapValueFromBB1ToBB2.erase(I1);
+
+    DEBUG(dbgs() << "SINK common instructions " << *I1 << "\n";);
+    DEBUG(dbgs() << "                         " << *I2 << "\n";);
+    // We need to update RE1 and RE2 if we are going to sink the first
+    // instruction in the basic block down.
+    bool UpdateRE1 = (I1 == BB1->begin()), UpdateRE2 = (I2 == BB2->begin());
+    // Sink the instruction.
+    BBEnd->getInstList().splice(FirstNonPhiInBBEnd, BB1->getInstList(), I1);
+    if (!OldPN->use_empty())
+      OldPN->replaceAllUsesWith(I1);
+    OldPN->eraseFromParent();
+
+    if (!I2->use_empty())
+      I2->replaceAllUsesWith(I1);
+    I1->intersectOptionalDataWith(I2);
+    I2->eraseFromParent();
+
+    if (UpdateRE1)
+      RE1 = BB1->getInstList().rend();
+    if (UpdateRE2)
+      RE2 = BB2->getInstList().rend();
+    FirstNonPhiInBBEnd = I1;
+    NumSinkCommons++;
+    Changed = true;
+  }
+  return Changed;
+}
+
 /// SpeculativelyExecuteBB - Given a conditional branch that goes to BB1
 /// and an BB2 and the only successor of BB1 is BB2, hoist simple code
 /// (for now, restricted to a single instruction that's side effect free) from
@@ -1056,7 +1380,7 @@
     // Do not hoist the instruction if any of its operands are defined but not
     // used in this BB. The transformation will prevent the operand from
     // being sunk into the use block.
-    for (User::op_iterator i = HInst->op_begin(), e = HInst->op_end(); 
+    for (User::op_iterator i = HInst->op_begin(), e = HInst->op_end();
          i != e; ++i) {
       Instruction *OpI = dyn_cast<Instruction>(*i);
       if (OpI && OpI->getParent() == BIParent &&
@@ -1112,7 +1436,7 @@
   // as well.
   if (PHIs.empty())
     return false;
-  
+
   // If we get here, we can hoist the instruction and if-convert.
   DEBUG(dbgs() << "SPECULATIVELY EXECUTING BB" << *BB1 << "\n";);
 
@@ -1162,13 +1486,13 @@
 static bool BlockIsSimpleEnoughToThreadThrough(BasicBlock *BB) {
   BranchInst *BI = cast<BranchInst>(BB->getTerminator());
   unsigned Size = 0;
-  
+
   for (BasicBlock::iterator BBI = BB->begin(); &*BBI != BI; ++BBI) {
     if (isa<DbgInfoIntrinsic>(BBI))
       continue;
     if (Size > 10) return false;  // Don't clone large BB's.
     ++Size;
-    
+
     // We can only support instructions that do not define values that are
     // live outside of the current basic block.
     for (Value::use_iterator UI = BBI->use_begin(), E = BBI->use_end();
@@ -1176,7 +1500,7 @@
       Instruction *U = cast<Instruction>(*UI);
       if (U->getParent() != BB || isa<PHINode>(U)) return false;
     }
-    
+
     // Looks ok, continue checking.
   }
 
@@ -1187,38 +1511,38 @@
 /// that is defined in the same block as the branch and if any PHI entries are
 /// constants, thread edges corresponding to that entry to be branches to their
 /// ultimate destination.
-static bool FoldCondBranchOnPHI(BranchInst *BI, const TargetData *TD) {
+static bool FoldCondBranchOnPHI(BranchInst *BI, const DataLayout *TD) {
   BasicBlock *BB = BI->getParent();
   PHINode *PN = dyn_cast<PHINode>(BI->getCondition());
   // NOTE: we currently cannot transform this case if the PHI node is used
   // outside of the block.
   if (!PN || PN->getParent() != BB || !PN->hasOneUse())
     return false;
-  
+
   // Degenerate case of a single entry PHI.
   if (PN->getNumIncomingValues() == 1) {
     FoldSingleEntryPHINodes(PN->getParent());
-    return true;    
+    return true;
   }
 
   // Now we know that this block has multiple preds and two succs.
   if (!BlockIsSimpleEnoughToThreadThrough(BB)) return false;
-  
+
   // Okay, this is a simple enough basic block.  See if any phi values are
   // constants.
   for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) {
     ConstantInt *CB = dyn_cast<ConstantInt>(PN->getIncomingValue(i));
     if (CB == 0 || !CB->getType()->isIntegerTy(1)) continue;
-    
+
     // Okay, we now know that all edges from PredBB should be revectored to
     // branch to RealDest.
     BasicBlock *PredBB = PN->getIncomingBlock(i);
     BasicBlock *RealDest = BI->getSuccessor(!CB->getZExtValue());
-    
+
     if (RealDest == BB) continue;  // Skip self loops.
     // Skip if the predecessor's terminator is an indirect branch.
     if (isa<IndirectBrInst>(PredBB->getTerminator())) continue;
-    
+
     // The dest block might have PHI nodes, other predecessors and other
     // difficult cases.  Instead of being smart about this, just insert a new
     // block that jumps to the destination block, effectively splitting
@@ -1227,7 +1551,7 @@
                                             RealDest->getName()+".critedge",
                                             RealDest->getParent(), RealDest);
     BranchInst::Create(RealDest, EdgeBB);
-    
+
     // Update PHI nodes.
     AddPredecessorToBlock(RealDest, EdgeBB, BB);
 
@@ -1244,7 +1568,7 @@
       // Clone the instruction.
       Instruction *N = BBI->clone();
       if (BBI->hasName()) N->setName(BBI->getName()+".c");
-      
+
       // Update operands due to translation.
       for (User::op_iterator i = N->op_begin(), e = N->op_end();
            i != e; ++i) {
@@ -1252,7 +1576,7 @@
         if (PI != TranslateMap.end())
           *i = PI->second;
       }
-      
+
       // Check for trivial simplification.
       if (Value *V = SimplifyInstruction(N, TD)) {
         TranslateMap[BBI] = V;
@@ -1283,7 +1607,7 @@
 
 /// FoldTwoEntryPHINode - Given a BB that starts with the specified two-entry
 /// PHI node, see if we can eliminate it.
-static bool FoldTwoEntryPHINode(PHINode *PN, const TargetData *TD) {
+static bool FoldTwoEntryPHINode(PHINode *PN, const DataLayout *TD) {
   // Ok, this is a two entry PHI node.  Check to see if this is a simple "if
   // statement", which has a very simple dominance structure.  Basically, we
   // are trying to find the condition that is being branched on, which
@@ -1297,7 +1621,7 @@
       // Don't bother if the branch will be constant folded trivially.
       isa<ConstantInt>(IfCond))
     return false;
-  
+
   // Okay, we found that we can merge this two-entry phi node into a select.
   // Doing so would require us to fold *all* two entry phi nodes in this block.
   // At some point this becomes non-profitable (particularly if the target
@@ -1307,14 +1631,14 @@
   for (BasicBlock::iterator I = BB->begin(); isa<PHINode>(I); ++NumPhis, ++I)
     if (NumPhis > 2)
       return false;
-  
+
   // Loop over the PHI's seeing if we can promote them all to select
   // instructions.  While we are at it, keep track of the instructions
   // that need to be moved to the dominating block.
   SmallPtrSet<Instruction*, 4> AggressiveInsts;
   unsigned MaxCostVal0 = PHINodeFoldingThreshold,
            MaxCostVal1 = PHINodeFoldingThreshold;
-  
+
   for (BasicBlock::iterator II = BB->begin(); isa<PHINode>(II);) {
     PHINode *PN = cast<PHINode>(II++);
     if (Value *V = SimplifyInstruction(PN, TD)) {
@@ -1322,19 +1646,19 @@
       PN->eraseFromParent();
       continue;
     }
-    
+
     if (!DominatesMergePoint(PN->getIncomingValue(0), BB, &AggressiveInsts,
                              MaxCostVal0) ||
         !DominatesMergePoint(PN->getIncomingValue(1), BB, &AggressiveInsts,
                              MaxCostVal1))
       return false;
   }
-  
+
   // If we folded the first phi, PN dangles at this point.  Refresh it.  If
   // we ran out of PHIs then we simplified them all.
   PN = dyn_cast<PHINode>(BB->begin());
   if (PN == 0) return true;
-  
+
   // Don't fold i1 branches on PHIs which contain binary operators.  These can
   // often be turned into switches and other things.
   if (PN->getType()->isIntegerTy(1) &&
@@ -1342,7 +1666,7 @@
        isa<BinaryOperator>(PN->getIncomingValue(1)) ||
        isa<BinaryOperator>(IfCond)))
     return false;
-  
+
   // If we all PHI nodes are promotable, check to make sure that all
   // instructions in the predecessor blocks can be promoted as well.  If
   // not, we won't be able to get rid of the control flow, so it's not
@@ -1362,7 +1686,7 @@
         return false;
       }
   }
-    
+
   if (cast<BranchInst>(IfBlock2->getTerminator())->isConditional()) {
     IfBlock2 = 0;
   } else {
@@ -1375,15 +1699,15 @@
         return false;
       }
   }
-  
+
   DEBUG(dbgs() << "FOUND IF CONDITION!  " << *IfCond << "  T: "
                << IfTrue->getName() << "  F: " << IfFalse->getName() << "\n");
-      
+
   // If we can still promote the PHI nodes after this gauntlet of tests,
   // do all of the PHI's now.
   Instruction *InsertPt = DomBlock->getTerminator();
   IRBuilder<true, NoFolder> Builder(InsertPt);
-  
+
   // Move all 'aggressive' instructions, which are defined in the
   // conditional parts of the if's up to the dominating block.
   if (IfBlock1)
@@ -1394,19 +1718,19 @@
     DomBlock->getInstList().splice(InsertPt,
                                    IfBlock2->getInstList(), IfBlock2->begin(),
                                    IfBlock2->getTerminator());
-  
+
   while (PHINode *PN = dyn_cast<PHINode>(BB->begin())) {
     // Change the PHI node into a select instruction.
     Value *TrueVal  = PN->getIncomingValue(PN->getIncomingBlock(0) == IfFalse);
     Value *FalseVal = PN->getIncomingValue(PN->getIncomingBlock(0) == IfTrue);
-    
-    SelectInst *NV = 
+
+    SelectInst *NV =
       cast<SelectInst>(Builder.CreateSelect(IfCond, TrueVal, FalseVal, ""));
     PN->replaceAllUsesWith(NV);
     NV->takeName(PN);
     PN->eraseFromParent();
   }
-  
+
   // At this point, IfBlock1 and IfBlock2 are both empty, so our if statement
   // has been flattened.  Change DomBlock to jump directly to our new block to
   // avoid other simplifycfg's kicking in on the diamond.
@@ -1420,14 +1744,14 @@
 /// SimplifyCondBranchToTwoReturns - If we found a conditional branch that goes
 /// to two returning blocks, try to merge them together into one return,
 /// introducing a select if the return values disagree.
-static bool SimplifyCondBranchToTwoReturns(BranchInst *BI, 
+static bool SimplifyCondBranchToTwoReturns(BranchInst *BI,
                                            IRBuilder<> &Builder) {
   assert(BI->isConditional() && "Must be a conditional branch");
   BasicBlock *TrueSucc = BI->getSuccessor(0);
   BasicBlock *FalseSucc = BI->getSuccessor(1);
   ReturnInst *TrueRet = cast<ReturnInst>(TrueSucc->getTerminator());
   ReturnInst *FalseRet = cast<ReturnInst>(FalseSucc->getTerminator());
-  
+
   // Check to ensure both blocks are empty (just a return) or optionally empty
   // with PHI nodes.  If there are other instructions, merging would cause extra
   // computation on one path or the other.
@@ -1447,12 +1771,12 @@
     EraseTerminatorInstAndDCECond(BI);
     return true;
   }
-    
+
   // Otherwise, figure out what the true and false return values are
   // so we can insert a new select instruction.
   Value *TrueValue = TrueRet->getReturnValue();
   Value *FalseValue = FalseRet->getReturnValue();
-  
+
   // Unwrap any PHI nodes in the return blocks.
   if (PHINode *TVPN = dyn_cast_or_null<PHINode>(TrueValue))
     if (TVPN->getParent() == TrueSucc)
@@ -1460,7 +1784,7 @@
   if (PHINode *FVPN = dyn_cast_or_null<PHINode>(FalseValue))
     if (FVPN->getParent() == FalseSucc)
       FalseValue = FVPN->getIncomingValueForBlock(BI->getParent());
-  
+
   // In order for this transformation to be safe, we must be able to
   // unconditionally execute both operands to the return.  This is
   // normally the case, but we could have a potentially-trapping
@@ -1472,12 +1796,12 @@
   if (ConstantExpr *FCV = dyn_cast_or_null<ConstantExpr>(FalseValue))
     if (FCV->canTrap())
       return false;
-  
+
   // Okay, we collected all the mapped values and checked them for sanity, and
   // defined to really do this transformation.  First, update the CFG.
   TrueSucc->removePredecessor(BI->getParent());
   FalseSucc->removePredecessor(BI->getParent());
-  
+
   // Insert select instructions where needed.
   Value *BrCond = BI->getCondition();
   if (TrueValue) {
@@ -1491,15 +1815,15 @@
     }
   }
 
-  Value *RI = !TrueValue ? 
+  Value *RI = !TrueValue ?
     Builder.CreateRetVoid() : Builder.CreateRet(TrueValue);
 
   (void) RI;
-      
+
   DEBUG(dbgs() << "\nCHANGING BRANCH TO TWO RETURNS INTO SELECT:"
                << "\n  " << *BI << "NewRet = " << *RI
                << "TRUEBLOCK: " << *TrueSucc << "FALSEBLOCK: "<< *FalseSucc);
-      
+
   EraseTerminatorInstAndDCECond(BI);
 
   return true;
@@ -1510,7 +1834,7 @@
 /// parameters and return true, or returns false if no or invalid metadata was
 /// found.
 static bool ExtractBranchMetadata(BranchInst *BI,
-                                  APInt &ProbTrue, APInt &ProbFalse) {
+                                  uint64_t &ProbTrue, uint64_t &ProbFalse) {
   assert(BI->isConditional() &&
          "Looking for probabilities on unconditional branch?");
   MDNode *ProfileData = BI->getMetadata(LLVMContext::MD_prof);
@@ -1518,35 +1842,11 @@
   ConstantInt *CITrue = dyn_cast<ConstantInt>(ProfileData->getOperand(1));
   ConstantInt *CIFalse = dyn_cast<ConstantInt>(ProfileData->getOperand(2));
   if (!CITrue || !CIFalse) return false;
-  ProbTrue = CITrue->getValue();
-  ProbFalse = CIFalse->getValue();
-  assert(ProbTrue.getBitWidth() == 32 && ProbFalse.getBitWidth() == 32 &&
-         "Branch probability metadata must be 32-bit integers");
+  ProbTrue = CITrue->getValue().getZExtValue();
+  ProbFalse = CIFalse->getValue().getZExtValue();
   return true;
 }
 
-/// MultiplyAndLosePrecision - Multiplies A and B, then returns the result. In
-/// the event of overflow, logically-shifts all four inputs right until the
-/// multiply fits.
-static APInt MultiplyAndLosePrecision(APInt &A, APInt &B, APInt &C, APInt &D,
-                                      unsigned &BitsLost) {
-  BitsLost = 0;
-  bool Overflow = false;
-  APInt Result = A.umul_ov(B, Overflow);
-  if (Overflow) {
-    APInt MaxB = APInt::getMaxValue(A.getBitWidth()).udiv(A);
-    do {
-      B = B.lshr(1);
-      ++BitsLost;
-    } while (B.ugt(MaxB));
-    A = A.lshr(BitsLost);
-    C = C.lshr(BitsLost);
-    D = D.lshr(BitsLost);
-    Result = A * B;
-  }
-  return Result;
-}
-
 /// checkCSEInPredecessor - Return true if the given instruction is available
 /// in its predecessor block. If yes, the instruction will be removed.
 ///
@@ -1600,7 +1900,7 @@
     if (Cond == 0)
       return false;
   }
-     
+
   if (Cond == 0 || (!isa<CmpInst>(Cond) && !isa<BinaryOperator>(Cond)) ||
     Cond->getParent() != BB || !Cond->hasOneUse())
   return false;
@@ -1623,7 +1923,7 @@
       isSafeToSpeculativelyExecute(FrontIt)) {
     BonusInst = &*FrontIt;
     ++FrontIt;
-    
+
     // Ignore dbg intrinsics.
     while (isa<DbgInfoIntrinsic>(FrontIt)) ++FrontIt;
   }
@@ -1631,13 +1931,13 @@
   // Only a single bonus inst is allowed.
   if (&*FrontIt != Cond)
     return false;
-  
+
   // Make sure the instruction after the condition is the cond branch.
   BasicBlock::iterator CondIt = Cond; ++CondIt;
 
   // Ingore dbg intrinsics.
   while (isa<DbgInfoIntrinsic>(CondIt)) ++CondIt;
-  
+
   if (&*CondIt != BI)
     return false;
 
@@ -1649,7 +1949,7 @@
   if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Cond->getOperand(1)))
     if (CE->canTrap())
       return false;
-  
+
   // Finally, don't infinitely unroll conditional loops.
   BasicBlock *TrueDest  = BI->getSuccessor(0);
   BasicBlock *FalseDest = (BI->isConditional()) ? BI->getSuccessor(1) : 0;
@@ -1659,22 +1959,22 @@
   for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
     BasicBlock *PredBlock = *PI;
     BranchInst *PBI = dyn_cast<BranchInst>(PredBlock->getTerminator());
-    
+
     // Check that we have two conditional branches.  If there is a PHI node in
     // the common successor, verify that the same value flows in from both
     // blocks.
     SmallVector<PHINode*, 4> PHIs;
     if (PBI == 0 || PBI->isUnconditional() ||
-        (BI->isConditional() && 
+        (BI->isConditional() &&
          !SafeToMergeTerminators(BI, PBI)) ||
         (!BI->isConditional() &&
          !isProfitableToFoldUnconditional(BI, PBI, Cond, PHIs)))
       continue;
-    
+
     // Determine if the two branches share a common destination.
-    Instruction::BinaryOps Opc;
+    Instruction::BinaryOps Opc = Instruction::BinaryOpsEnd;
     bool InvertPredCond = false;
-    
+
     if (BI->isConditional()) {
       if (PBI->getSuccessor(0) == TrueDest)
         Opc = Instruction::Or;
@@ -1693,7 +1993,7 @@
 
     // Ensure that any values used in the bonus instruction are also used
     // by the terminator of the predecessor.  This means that those values
-    // must already have been resolved, so we won't be inhibiting the 
+    // must already have been resolved, so we won't be inhibiting the
     // out-of-order core by speculating them earlier.
     if (BonusInst) {
       // Collect the values used by the bonus inst
@@ -1707,47 +2007,47 @@
 
       SmallVector<std::pair<Value*, unsigned>, 4> Worklist;
       Worklist.push_back(std::make_pair(PBI->getOperand(0), 0));
-      
+
       // Walk up to four levels back up the use-def chain of the predecessor's
       // terminator to see if all those values were used.  The choice of four
       // levels is arbitrary, to provide a compile-time-cost bound.
       while (!Worklist.empty()) {
         std::pair<Value*, unsigned> Pair = Worklist.back();
         Worklist.pop_back();
-        
+
         if (Pair.second >= 4) continue;
         UsedValues.erase(Pair.first);
         if (UsedValues.empty()) break;
-        
+
         if (Instruction *I = dyn_cast<Instruction>(Pair.first)) {
           for (Instruction::op_iterator OI = I->op_begin(), OE = I->op_end();
                OI != OE; ++OI)
             Worklist.push_back(std::make_pair(OI->get(), Pair.second+1));
-        }       
+        }
       }
-      
+
       if (!UsedValues.empty()) return false;
     }
 
     DEBUG(dbgs() << "FOLDING BRANCH TO COMMON DEST:\n" << *PBI << *BB);
-    IRBuilder<> Builder(PBI);    
+    IRBuilder<> Builder(PBI);
 
     // If we need to invert the condition in the pred block to match, do so now.
     if (InvertPredCond) {
       Value *NewCond = PBI->getCondition();
-      
+
       if (NewCond->hasOneUse() && isa<CmpInst>(NewCond)) {
         CmpInst *CI = cast<CmpInst>(NewCond);
         CI->setPredicate(CI->getInversePredicate());
       } else {
-        NewCond = Builder.CreateNot(NewCond, 
+        NewCond = Builder.CreateNot(NewCond,
                                     PBI->getCondition()->getName()+".not");
       }
-      
+
       PBI->setCondition(NewCond);
       PBI->swapSuccessors();
     }
-    
+
     // If we have a bonus inst, clone it into the predecessor block.
     Instruction *NewBonus = 0;
     if (BonusInst) {
@@ -1756,7 +2056,7 @@
       NewBonus->takeName(BonusInst);
       BonusInst->setName(BonusInst->getName()+".old");
     }
-    
+
     // Clone Cond into the predecessor basic block, and or/and the
     // two conditions together.
     Instruction *New = Cond->clone();
@@ -1764,21 +2064,60 @@
     PredBlock->getInstList().insert(PBI, New);
     New->takeName(Cond);
     Cond->setName(New->getName()+".old");
-    
+
     if (BI->isConditional()) {
-      Instruction *NewCond = 
+      Instruction *NewCond =
         cast<Instruction>(Builder.CreateBinOp(Opc, PBI->getCondition(),
                                             New, "or.cond"));
       PBI->setCondition(NewCond);
 
+      uint64_t PredTrueWeight, PredFalseWeight, SuccTrueWeight, SuccFalseWeight;
+      bool PredHasWeights = ExtractBranchMetadata(PBI, PredTrueWeight,
+                                                  PredFalseWeight);
+      bool SuccHasWeights = ExtractBranchMetadata(BI, SuccTrueWeight,
+                                                  SuccFalseWeight);
+      SmallVector<uint64_t, 8> NewWeights;
+
       if (PBI->getSuccessor(0) == BB) {
+        if (PredHasWeights && SuccHasWeights) {
+          // PBI: br i1 %x, BB, FalseDest
+          // BI:  br i1 %y, TrueDest, FalseDest
+          //TrueWeight is TrueWeight for PBI * TrueWeight for BI.
+          NewWeights.push_back(PredTrueWeight * SuccTrueWeight);
+          //FalseWeight is FalseWeight for PBI * TotalWeight for BI +
+          //               TrueWeight for PBI * FalseWeight for BI.
+          // We assume that total weights of a BranchInst can fit into 32 bits.
+          // Therefore, we will not have overflow using 64-bit arithmetic.
+          NewWeights.push_back(PredFalseWeight * (SuccFalseWeight +
+               SuccTrueWeight) + PredTrueWeight * SuccFalseWeight);
+        }
         AddPredecessorToBlock(TrueDest, PredBlock, BB);
         PBI->setSuccessor(0, TrueDest);
       }
       if (PBI->getSuccessor(1) == BB) {
+        if (PredHasWeights && SuccHasWeights) {
+          // PBI: br i1 %x, TrueDest, BB
+          // BI:  br i1 %y, TrueDest, FalseDest
+          //TrueWeight is TrueWeight for PBI * TotalWeight for BI +
+          //              FalseWeight for PBI * TrueWeight for BI.
+          NewWeights.push_back(PredTrueWeight * (SuccFalseWeight +
+              SuccTrueWeight) + PredFalseWeight * SuccTrueWeight);
+          //FalseWeight is FalseWeight for PBI * FalseWeight for BI.
+          NewWeights.push_back(PredFalseWeight * SuccFalseWeight);
+        }
         AddPredecessorToBlock(FalseDest, PredBlock, BB);
         PBI->setSuccessor(1, FalseDest);
       }
+      if (NewWeights.size() == 2) {
+        // Halve the weights if any of them cannot fit in an uint32_t
+        FitWeights(NewWeights);
+
+        SmallVector<uint32_t, 8> MDWeights(NewWeights.begin(),NewWeights.end());
+        PBI->setMetadata(LLVMContext::MD_prof,
+                         MDBuilder(BI->getContext()).
+                         createBranchWeights(MDWeights));
+      } else
+        PBI->setMetadata(LLVMContext::MD_prof, NULL);
     } else {
       // Update PHI nodes in the common successors.
       for (unsigned i = 0, e = PHIs.size(); i != e; ++i) {
@@ -1806,7 +2145,7 @@
           // Create (PBI_Cond and BI_Value) or (!PBI_Cond and PBI_C)
           // PBI_C is true: (PBI_Cond and BI_Value) or (!PBI_Cond)
           //       is false: PBI_Cond and BI_Value
-          MergedCond = 
+          MergedCond =
             cast<Instruction>(Builder.CreateBinOp(Instruction::And,
                                 PBI->getCondition(), New,
                                 "and.cond"));
@@ -1814,7 +2153,7 @@
             Instruction *NotCond =
               cast<Instruction>(Builder.CreateNot(PBI->getCondition(),
                                   "not.cond"));
-            MergedCond = 
+            MergedCond =
               cast<Instruction>(Builder.CreateBinOp(Instruction::Or,
                                   NotCond, MergedCond,
                                   "or.cond"));
@@ -1833,95 +2172,11 @@
     // TODO: If BB is reachable from all paths through PredBlock, then we
     // could replace PBI's branch probabilities with BI's.
 
-    // Merge probability data into PredBlock's branch.
-    APInt A, B, C, D;
-    if (PBI->isConditional() && BI->isConditional() &&
-        ExtractBranchMetadata(PBI, C, D) && ExtractBranchMetadata(BI, A, B)) {
-      // Given IR which does:
-      //   bbA:
-      //     br i1 %x, label %bbB, label %bbC
-      //   bbB:
-      //     br i1 %y, label %bbD, label %bbC
-      // Let's call the probability that we take the edge from %bbA to %bbB
-      // 'a', from %bbA to %bbC, 'b', from %bbB to %bbD 'c' and from %bbB to
-      // %bbC probability 'd'.
-      //
-      // We transform the IR into:
-      //   bbA:
-      //     br i1 %z, label %bbD, label %bbC
-      // where the probability of going to %bbD is (a*c) and going to bbC is
-      // (b+a*d).
-      //
-      // Probabilities aren't stored as ratios directly. Using branch weights,
-      // we get:
-      // (a*c)% = A*C, (b+(a*d))% = A*D+B*C+B*D.
-
-      // In the event of overflow, we want to drop the LSB of the input
-      // probabilities.
-      unsigned BitsLost;
-
-      // Ignore overflow result on ProbTrue.
-      APInt ProbTrue = MultiplyAndLosePrecision(A, C, B, D, BitsLost);
-
-      APInt Tmp1 = MultiplyAndLosePrecision(B, D, A, C, BitsLost);
-      if (BitsLost) {
-        ProbTrue = ProbTrue.lshr(BitsLost*2);
-      }
-
-      APInt Tmp2 = MultiplyAndLosePrecision(A, D, C, B, BitsLost);
-      if (BitsLost) {
-        ProbTrue = ProbTrue.lshr(BitsLost*2);
-        Tmp1 = Tmp1.lshr(BitsLost*2);
-      }
-
-      APInt Tmp3 = MultiplyAndLosePrecision(B, C, A, D, BitsLost);
-      if (BitsLost) {
-        ProbTrue = ProbTrue.lshr(BitsLost*2);
-        Tmp1 = Tmp1.lshr(BitsLost*2);
-        Tmp2 = Tmp2.lshr(BitsLost*2);
-      }
-
-      bool Overflow1 = false, Overflow2 = false;
-      APInt Tmp4 = Tmp2.uadd_ov(Tmp3, Overflow1);
-      APInt ProbFalse = Tmp4.uadd_ov(Tmp1, Overflow2);
-
-      if (Overflow1 || Overflow2) {
-        ProbTrue = ProbTrue.lshr(1);
-        Tmp1 = Tmp1.lshr(1);
-        Tmp2 = Tmp2.lshr(1);
-        Tmp3 = Tmp3.lshr(1);
-        Tmp4 = Tmp2 + Tmp3;
-        ProbFalse = Tmp4 + Tmp1;
-      }
-
-      // The sum of branch weights must fit in 32-bits.
-      if (ProbTrue.isNegative() && ProbFalse.isNegative()) {
-        ProbTrue = ProbTrue.lshr(1);
-        ProbFalse = ProbFalse.lshr(1);
-      }
-
-      if (ProbTrue != ProbFalse) {
-        // Normalize the result.
-        APInt GCD = APIntOps::GreatestCommonDivisor(ProbTrue, ProbFalse);
-        ProbTrue = ProbTrue.udiv(GCD);
-        ProbFalse = ProbFalse.udiv(GCD);
-
-        MDBuilder MDB(BI->getContext());
-        MDNode *N = MDB.createBranchWeights(ProbTrue.getZExtValue(),
-                                            ProbFalse.getZExtValue());
-        PBI->setMetadata(LLVMContext::MD_prof, N);
-      } else {
-        PBI->setMetadata(LLVMContext::MD_prof, NULL);
-      }
-    } else {
-      PBI->setMetadata(LLVMContext::MD_prof, NULL);
-    }
-
     // Copy any debug value intrinsics into the end of PredBlock.
     for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
       if (isa<DbgInfoIntrinsic>(*I))
         I->clone()->insertBefore(PBI);
-      
+
     return true;
   }
   return false;
@@ -1936,7 +2191,7 @@
   BasicBlock *BB = BI->getParent();
 
   // If this block ends with a branch instruction, and if there is a
-  // predecessor that ends on a branch of the same condition, make 
+  // predecessor that ends on a branch of the same condition, make
   // this conditional branch redundant.
   if (PBI->getCondition() == BI->getCondition() &&
       PBI->getSuccessor(0) != PBI->getSuccessor(1)) {
@@ -1945,11 +2200,11 @@
     if (BB->getSinglePredecessor()) {
       // Turn this into a branch on constant.
       bool CondIsTrue = PBI->getSuccessor(0) == BB;
-      BI->setCondition(ConstantInt::get(Type::getInt1Ty(BB->getContext()), 
+      BI->setCondition(ConstantInt::get(Type::getInt1Ty(BB->getContext()),
                                         CondIsTrue));
       return true;  // Nuke the branch on constant.
     }
-    
+
     // Otherwise, if there are multiple predecessors, insert a PHI that merges
     // in the constant and simplify the block result.  Subsequent passes of
     // simplifycfg will thread the block.
@@ -1969,18 +2224,18 @@
             PBI->getCondition() == BI->getCondition() &&
             PBI->getSuccessor(0) != PBI->getSuccessor(1)) {
           bool CondIsTrue = PBI->getSuccessor(0) == BB;
-          NewPN->addIncoming(ConstantInt::get(Type::getInt1Ty(BB->getContext()), 
+          NewPN->addIncoming(ConstantInt::get(Type::getInt1Ty(BB->getContext()),
                                               CondIsTrue), P);
         } else {
           NewPN->addIncoming(BI->getCondition(), P);
         }
       }
-      
+
       BI->setCondition(NewPN);
       return true;
     }
   }
-  
+
   // If this is a conditional branch in an empty block, and if any
   // predecessors is a conditional branch to one of our destinations,
   // fold the conditions into logical ops and one cond br.
@@ -1991,11 +2246,11 @@
   if (&*BBI != BI)
     return false;
 
-  
+
   if (ConstantExpr *CE = dyn_cast<ConstantExpr>(BI->getCondition()))
     if (CE->canTrap())
       return false;
-  
+
   int PBIOp, BIOp;
   if (PBI->getSuccessor(0) == BI->getSuccessor(0))
     PBIOp = BIOp = 0;
@@ -2007,31 +2262,31 @@
     PBIOp = BIOp = 1;
   else
     return false;
-    
+
   // Check to make sure that the other destination of this branch
   // isn't BB itself.  If so, this is an infinite loop that will
   // keep getting unwound.
   if (PBI->getSuccessor(PBIOp) == BB)
     return false;
-    
-  // Do not perform this transformation if it would require 
+
+  // Do not perform this transformation if it would require
   // insertion of a large number of select instructions. For targets
   // without predication/cmovs, this is a big pessimization.
   BasicBlock *CommonDest = PBI->getSuccessor(PBIOp);
-      
+
   unsigned NumPhis = 0;
   for (BasicBlock::iterator II = CommonDest->begin();
        isa<PHINode>(II); ++II, ++NumPhis)
     if (NumPhis > 2) // Disable this xform.
       return false;
-    
+
   // Finally, if everything is ok, fold the branches to logical ops.
   BasicBlock *OtherDest  = BI->getSuccessor(BIOp ^ 1);
-  
+
   DEBUG(dbgs() << "FOLDING BRs:" << *PBI->getParent()
                << "AND: " << *BI->getParent());
-  
-  
+
+
   // If OtherDest *is* BB, then BB is a basic block with a single conditional
   // branch in it, where one edge (OtherDest) goes back to itself but the other
   // exits.  We don't *know* that the program avoids the infinite loop
@@ -2046,13 +2301,13 @@
                                                   "infloop", BB->getParent());
     BranchInst::Create(InfLoopBlock, InfLoopBlock);
     OtherDest = InfLoopBlock;
-  }  
-  
+  }
+
   DEBUG(dbgs() << *PBI->getParent()->getParent());
 
   // BI may have other predecessors.  Because of this, we leave
   // it alone, but modify PBI.
-  
+
   // Make sure we get to CommonDest on True&True directions.
   Value *PBICond = PBI->getCondition();
   IRBuilder<true, NoFolder> Builder(PBI);
@@ -2065,16 +2320,43 @@
 
   // Merge the conditions.
   Value *Cond = Builder.CreateOr(PBICond, BICond, "brmerge");
-  
+
   // Modify PBI to branch on the new condition to the new dests.
   PBI->setCondition(Cond);
   PBI->setSuccessor(0, CommonDest);
   PBI->setSuccessor(1, OtherDest);
-  
+
+  // Update branch weight for PBI.
+  uint64_t PredTrueWeight, PredFalseWeight, SuccTrueWeight, SuccFalseWeight;
+  bool PredHasWeights = ExtractBranchMetadata(PBI, PredTrueWeight,
+                                              PredFalseWeight);
+  bool SuccHasWeights = ExtractBranchMetadata(BI, SuccTrueWeight,
+                                              SuccFalseWeight);
+  if (PredHasWeights && SuccHasWeights) {
+    uint64_t PredCommon = PBIOp ? PredFalseWeight : PredTrueWeight;
+    uint64_t PredOther = PBIOp ?PredTrueWeight : PredFalseWeight;
+    uint64_t SuccCommon = BIOp ? SuccFalseWeight : SuccTrueWeight;
+    uint64_t SuccOther = BIOp ? SuccTrueWeight : SuccFalseWeight;
+    // The weight to CommonDest should be PredCommon * SuccTotal +
+    //                                    PredOther * SuccCommon.
+    // The weight to OtherDest should be PredOther * SuccOther.
+    SmallVector<uint64_t, 2> NewWeights;
+    NewWeights.push_back(PredCommon * (SuccCommon + SuccOther) +
+                         PredOther * SuccCommon);
+    NewWeights.push_back(PredOther * SuccOther);
+    // Halve the weights if any of them cannot fit in an uint32_t
+    FitWeights(NewWeights);
+
+    SmallVector<uint32_t, 2> MDWeights(NewWeights.begin(),NewWeights.end());
+    PBI->setMetadata(LLVMContext::MD_prof,
+                     MDBuilder(BI->getContext()).
+                     createBranchWeights(MDWeights));
+  }
+
   // OtherDest may have phi nodes.  If so, add an entry from PBI's
   // block that are identical to the entries for BI's block.
   AddPredecessorToBlock(OtherDest, PBI->getParent(), BB);
-  
+
   // We know that the CommonDest already had an edge from PBI to
   // it.  If it has PHIs though, the PHIs may have different
   // entries for BB and PBI's BB.  If so, insert a select to make
@@ -2092,10 +2374,10 @@
       PN->setIncomingValue(PBBIdx, NV);
     }
   }
-  
+
   DEBUG(dbgs() << "INTO: " << *PBI->getParent());
   DEBUG(dbgs() << *PBI->getParent()->getParent());
-  
+
   // This basic block is probably dead.  We know it has at least
   // one fewer predecessor.
   return true;
@@ -2107,7 +2389,9 @@
 // Also makes sure not to introduce new successors by assuming that edges to
 // non-successor TrueBBs and FalseBBs aren't reachable.
 static bool SimplifyTerminatorOnSelect(TerminatorInst *OldTerm, Value *Cond,
-                                       BasicBlock *TrueBB, BasicBlock *FalseBB){
+                                       BasicBlock *TrueBB, BasicBlock *FalseBB,
+                                       uint32_t TrueWeight,
+                                       uint32_t FalseWeight){
   // Remove any superfluous successor edges from the CFG.
   // First, figure out which successors to preserve.
   // If TrueBB and FalseBB are equal, only try to preserve one copy of that
@@ -2136,10 +2420,15 @@
       // We were only looking for one successor, and it was present.
       // Create an unconditional branch to it.
       Builder.CreateBr(TrueBB);
-    else
+    else {
       // We found both of the successors we were looking for.
       // Create a conditional branch sharing the condition of the select.
-      Builder.CreateCondBr(Cond, TrueBB, FalseBB);
+      BranchInst *NewBI = Builder.CreateCondBr(Cond, TrueBB, FalseBB);
+      if (TrueWeight != FalseWeight)
+        NewBI->setMetadata(LLVMContext::MD_prof,
+                           MDBuilder(OldTerm->getContext()).
+                           createBranchWeights(TrueWeight, FalseWeight));
+    }
   } else if (KeepEdge1 && (KeepEdge2 || TrueBB == FalseBB)) {
     // Neither of the selected blocks were successors, so this
     // terminator must be unreachable.
@@ -2176,8 +2465,23 @@
   BasicBlock *TrueBB = SI->findCaseValue(TrueVal).getCaseSuccessor();
   BasicBlock *FalseBB = SI->findCaseValue(FalseVal).getCaseSuccessor();
 
+  // Get weight for TrueBB and FalseBB.
+  uint32_t TrueWeight = 0, FalseWeight = 0;
+  SmallVector<uint64_t, 8> Weights;
+  bool HasWeights = HasBranchWeights(SI);
+  if (HasWeights) {
+    GetBranchWeights(SI, Weights);
+    if (Weights.size() == 1 + SI->getNumCases()) {
+      TrueWeight = (uint32_t)Weights[SI->findCaseValue(TrueVal).
+                                     getSuccessorIndex()];
+      FalseWeight = (uint32_t)Weights[SI->findCaseValue(FalseVal).
+                                      getSuccessorIndex()];
+    }
+  }
+
   // Perform the actual simplification.
-  return SimplifyTerminatorOnSelect(SI, Condition, TrueBB, FalseBB);
+  return SimplifyTerminatorOnSelect(SI, Condition, TrueBB, FalseBB,
+                                    TrueWeight, FalseWeight);
 }
 
 // SimplifyIndirectBrOnSelect - Replaces
@@ -2197,7 +2501,8 @@
   BasicBlock *FalseBB = FBA->getBasicBlock();
 
   // Perform the actual simplification.
-  return SimplifyTerminatorOnSelect(IBI, SI->getCondition(), TrueBB, FalseBB);
+  return SimplifyTerminatorOnSelect(IBI, SI->getCondition(), TrueBB, FalseBB,
+                                    0, 0);
 }
 
 /// TryToSimplifyUncondBranchWithICmpInIt - This is called when we find an icmp
@@ -2214,11 +2519,11 @@
 ///   br label %end
 /// end:
 ///   ... = phi i1 [ true, %entry ], [ %tmp, %DEFAULT ], [ true, %entry ]
-/// 
+///
 /// We prefer to split the edge to 'end' so that there is a true/false entry to
 /// the PHI, merging the third icmp into the switch.
 static bool TryToSimplifyUncondBranchWithICmpInIt(ICmpInst *ICI,
-                                                  const TargetData *TD,
+                                                  const DataLayout *TD,
                                                   IRBuilder<> &Builder) {
   BasicBlock *BB = ICI->getParent();
 
@@ -2228,17 +2533,17 @@
 
   Value *V = ICI->getOperand(0);
   ConstantInt *Cst = cast<ConstantInt>(ICI->getOperand(1));
-  
+
   // The pattern we're looking for is where our only predecessor is a switch on
   // 'V' and this block is the default case for the switch.  In this case we can
   // fold the compared value into the switch to simplify things.
   BasicBlock *Pred = BB->getSinglePredecessor();
   if (Pred == 0 || !isa<SwitchInst>(Pred->getTerminator())) return false;
-  
+
   SwitchInst *SI = cast<SwitchInst>(Pred->getTerminator());
   if (SI->getCondition() != V)
     return false;
-  
+
   // If BB is reachable on a non-default case, then we simply know the value of
   // V in this block.  Substitute it and constant fold the icmp instruction
   // away.
@@ -2246,7 +2551,7 @@
     ConstantInt *VVal = SI->findCaseDest(BB);
     assert(VVal && "Should have a unique destination value");
     ICI->setOperand(0, VVal);
-    
+
     if (Value *V = SimplifyInstruction(ICI, TD)) {
       ICI->replaceAllUsesWith(V);
       ICI->eraseFromParent();
@@ -2254,7 +2559,7 @@
     // BB is now empty, so it is likely to simplify away.
     return SimplifyCFG(BB) | true;
   }
-  
+
   // Ok, the block is reachable from the default dest.  If the constant we're
   // comparing exists in one of the other edges, then we can constant fold ICI
   // and zap it.
@@ -2264,13 +2569,13 @@
       V = ConstantInt::getFalse(BB->getContext());
     else
       V = ConstantInt::getTrue(BB->getContext());
-    
+
     ICI->replaceAllUsesWith(V);
     ICI->eraseFromParent();
     // BB is now empty, so it is likely to simplify away.
     return SimplifyCFG(BB) | true;
   }
-  
+
   // The use of the icmp has to be in the 'end' block, by the only PHI node in
   // the block.
   BasicBlock *SuccBlock = BB->getTerminator()->getSuccessor(0);
@@ -2296,8 +2601,23 @@
   // the switch to the merge point on the compared value.
   BasicBlock *NewBB = BasicBlock::Create(BB->getContext(), "switch.edge",
                                          BB->getParent(), BB);
+  SmallVector<uint64_t, 8> Weights;
+  bool HasWeights = HasBranchWeights(SI);
+  if (HasWeights) {
+    GetBranchWeights(SI, Weights);
+    if (Weights.size() == 1 + SI->getNumCases()) {
+      // Split weight for default case to case for "Cst".
+      Weights[0] = (Weights[0]+1) >> 1;
+      Weights.push_back(Weights[0]);
+
+      SmallVector<uint32_t, 8> MDWeights(Weights.begin(), Weights.end());
+      SI->setMetadata(LLVMContext::MD_prof,
+                      MDBuilder(SI->getContext()).
+                      createBranchWeights(MDWeights));
+    }
+  }
   SI->addCase(Cst, NewBB);
-  
+
   // NewBB branches to the phi block, add the uncond branch and the phi entry.
   Builder.SetInsertPoint(NewBB);
   Builder.SetCurrentDebugLocation(SI->getDebugLoc());
@@ -2309,12 +2629,12 @@
 /// SimplifyBranchOnICmpChain - The specified branch is a conditional branch.
 /// Check to see if it is branching on an or/and chain of icmp instructions, and
 /// fold it into a switch instruction if so.
-static bool SimplifyBranchOnICmpChain(BranchInst *BI, const TargetData *TD,
+static bool SimplifyBranchOnICmpChain(BranchInst *BI, const DataLayout *TD,
                                       IRBuilder<> &Builder) {
   Instruction *Cond = dyn_cast<Instruction>(BI->getCondition());
   if (Cond == 0) return false;
-  
-  
+
+
   // Change br (X == 0 | X == 1), T, F into a switch instruction.
   // If this is a bunch of seteq's or'd together, or if it's a bunch of
   // 'setne's and'ed together, collect them.
@@ -2323,7 +2643,7 @@
   bool TrueWhenEqual = true;
   Value *ExtraCase = 0;
   unsigned UsedICmps = 0;
-  
+
   if (Cond->getOpcode() == Instruction::Or) {
     CompVal = GatherConstantCompares(Cond, Values, ExtraCase, TD, true,
                                      UsedICmps);
@@ -2332,7 +2652,7 @@
                                      UsedICmps);
     TrueWhenEqual = false;
   }
-  
+
   // If we didn't have a multiply compared value, fail.
   if (CompVal == 0) return false;
 
@@ -2344,21 +2664,24 @@
   // instruction can't handle, remove them now.
   array_pod_sort(Values.begin(), Values.end(), ConstantIntSortPredicate);
   Values.erase(std::unique(Values.begin(), Values.end()), Values.end());
-  
+
   // If Extra was used, we require at least two switch values to do the
   // transformation.  A switch with one value is just an cond branch.
   if (ExtraCase && Values.size() < 2) return false;
-  
+
+  // TODO: Preserve branch weight metadata, similarly to how
+  // FoldValueComparisonIntoPredecessors preserves it.
+
   // Figure out which block is which destination.
   BasicBlock *DefaultBB = BI->getSuccessor(1);
   BasicBlock *EdgeBB    = BI->getSuccessor(0);
   if (!TrueWhenEqual) std::swap(DefaultBB, EdgeBB);
-  
+
   BasicBlock *BB = BI->getParent();
-  
+
   DEBUG(dbgs() << "Converting 'icmp' chain with " << Values.size()
                << " cases into SWITCH.  BB is:\n" << *BB);
-  
+
   // If there are any extra values that couldn't be folded into the switch
   // then we evaluate them with an explicit branch first.  Split the block
   // right before the condbr to handle it.
@@ -2372,13 +2695,13 @@
       Builder.CreateCondBr(ExtraCase, EdgeBB, NewBB);
     else
       Builder.CreateCondBr(ExtraCase, NewBB, EdgeBB);
-      
+
     OldTI->eraseFromParent();
-    
+
     // If there are PHI nodes in EdgeBB, then we need to add a new entry to them
     // for the edge we just added.
     AddPredecessorToBlock(EdgeBB, BB, NewBB);
-    
+
     DEBUG(dbgs() << "  ** 'icmp' chain unhandled condition: " << *ExtraCase
           << "\nEXTRABB = " << *BB);
     BB = NewBB;
@@ -2387,19 +2710,19 @@
   Builder.SetInsertPoint(BI);
   // Convert pointer to int before we switch.
   if (CompVal->getType()->isPointerTy()) {
-    assert(TD && "Cannot switch on pointer without TargetData");
+    assert(TD && "Cannot switch on pointer without DataLayout");
     CompVal = Builder.CreatePtrToInt(CompVal,
                                      TD->getIntPtrType(CompVal->getContext()),
                                      "magicptr");
   }
-  
+
   // Create the new switch instruction now.
   SwitchInst *New = Builder.CreateSwitch(CompVal, DefaultBB, Values.size());
 
   // Add all of the 'cases' to the switch instruction.
   for (unsigned i = 0, e = Values.size(); i != e; ++i)
     New->addCase(Values[i], EdgeBB);
-  
+
   // We added edges from PI to the EdgeBB.  As such, if there were any
   // PHI nodes in EdgeBB, they need entries to be added corresponding to
   // the number of edges added.
@@ -2410,10 +2733,10 @@
     for (unsigned i = 0, e = Values.size()-1; i != e; ++i)
       PN->addIncoming(InVal, BB);
   }
-  
+
   // Erase the old branch instruction.
   EraseTerminatorInstAndDCECond(BI);
-  
+
   DEBUG(dbgs() << "  ** 'icmp' chain result is:\n" << *BB << '\n');
   return true;
 }
@@ -2467,7 +2790,7 @@
 bool SimplifyCFGOpt::SimplifyReturn(ReturnInst *RI, IRBuilder<> &Builder) {
   BasicBlock *BB = RI->getParent();
   if (!BB->getFirstNonPHIOrDbg()->isTerminator()) return false;
-  
+
   // Find predecessors that end with branches.
   SmallVector<BasicBlock*, 8> UncondBranchPreds;
   SmallVector<BranchInst*, 8> CondBranchPreds;
@@ -2481,7 +2804,7 @@
         CondBranchPreds.push_back(BI);
     }
   }
-  
+
   // If we found some, do the transformation!
   if (!UncondBranchPreds.empty() && DupRet) {
     while (!UncondBranchPreds.empty()) {
@@ -2490,21 +2813,21 @@
             << "INTO UNCOND BRANCH PRED: " << *Pred);
       (void)FoldReturnIntoUncondBranch(RI, BB, Pred);
     }
-    
+
     // If we eliminated all predecessors of the block, delete the block now.
     if (pred_begin(BB) == pred_end(BB))
       // We know there are no successors, so just nuke the block.
       BB->eraseFromParent();
-    
+
     return true;
   }
-  
+
   // Check out all of the conditional branches going to this return
   // instruction.  If any of them just select between returns, change the
   // branch itself into a select/return pair.
   while (!CondBranchPreds.empty()) {
     BranchInst *BI = CondBranchPreds.pop_back_val();
-    
+
     // Check to see if the non-BB successor is also a return block.
     if (isa<ReturnInst>(BI->getSuccessor(0)->getTerminator()) &&
         isa<ReturnInst>(BI->getSuccessor(1)->getTerminator()) &&
@@ -2516,9 +2839,9 @@
 
 bool SimplifyCFGOpt::SimplifyUnreachable(UnreachableInst *UI) {
   BasicBlock *BB = UI->getParent();
-  
+
   bool Changed = false;
-  
+
   // If there are any instructions immediately before the unreachable that can
   // be removed, do so.
   while (UI != BB->begin()) {
@@ -2558,11 +2881,11 @@
     BBI->eraseFromParent();
     Changed = true;
   }
-  
+
   // If the unreachable instruction is the first in the block, take a gander
   // at all of the predecessors of this instruction, and simplify them.
   if (&BB->front() != UI) return Changed;
-  
+
   SmallVector<BasicBlock*, 8> Preds(pred_begin(BB), pred_end(BB));
   for (unsigned i = 0, e = Preds.size(); i != e; ++i) {
     TerminatorInst *TI = Preds[i]->getTerminator();
@@ -2615,7 +2938,7 @@
         BasicBlock *MaxBlock = 0;
         for (std::map<BasicBlock*, std::pair<unsigned, unsigned> >::iterator
              I = Popularity.begin(), E = Popularity.end(); I != E; ++I) {
-          if (I->second.first > MaxPop || 
+          if (I->second.first > MaxPop ||
               (I->second.first == MaxPop && MaxIndex > I->second.second)) {
             MaxPop = I->second.first;
             MaxIndex = I->second.second;
@@ -2627,13 +2950,13 @@
           // edges to it.
           SI->setDefaultDest(MaxBlock);
           Changed = true;
-          
+
           // If MaxBlock has phinodes in it, remove MaxPop-1 entries from
           // it.
           if (isa<PHINode>(MaxBlock->begin()))
             for (unsigned i = 0; i != MaxPop-1; ++i)
               MaxBlock->removePredecessor(SI->getParent());
-          
+
           for (SwitchInst::CaseIt i = SI->case_begin(), e = SI->case_end();
                i != e; ++i)
             if (i.getCaseSuccessor() == MaxBlock) {
@@ -2648,7 +2971,7 @@
         // place to note that the call does not throw though.
         BranchInst *BI = Builder.CreateBr(II->getNormalDest());
         II->removeFromParent();   // Take out of symbol table
-        
+
         // Insert the call now...
         SmallVector<Value*, 8> Args(II->op_begin(), II->op_end()-3);
         Builder.SetInsertPoint(BI);
@@ -2663,7 +2986,7 @@
       }
     }
   }
-  
+
   // If this block is now dead, remove it.
   if (pred_begin(BB) == pred_end(BB) &&
       BB != &BB->getParent()->getEntryBlock()) {
@@ -2706,9 +3029,28 @@
   if (!Offset->isNullValue())
     Sub = Builder.CreateAdd(Sub, Offset, Sub->getName()+".off");
   Value *Cmp = Builder.CreateICmpULT(Sub, NumCases, "switch");
-  Builder.CreateCondBr(
+  BranchInst *NewBI = Builder.CreateCondBr(
       Cmp, SI->case_begin().getCaseSuccessor(), SI->getDefaultDest());
 
+  // Update weight for the newly-created conditional branch.
+  SmallVector<uint64_t, 8> Weights;
+  bool HasWeights = HasBranchWeights(SI);
+  if (HasWeights) {
+    GetBranchWeights(SI, Weights);
+    if (Weights.size() == 1 + SI->getNumCases()) {
+      // Combine all weights for the cases to be the true weight of NewBI.
+      // We assume that the sum of all weights for a Terminator can fit into 32
+      // bits.
+      uint32_t NewTrueWeight = 0;
+      for (unsigned I = 1, E = Weights.size(); I != E; ++I)
+        NewTrueWeight += (uint32_t)Weights[I];
+      NewBI->setMetadata(LLVMContext::MD_prof,
+                         MDBuilder(SI->getContext()).
+                         createBranchWeights(NewTrueWeight,
+                                             (uint32_t)Weights[0]));
+    }
+  }
+
   // Prune obsolete incoming values off the successor's PHI nodes.
   for (BasicBlock::iterator BBI = SI->case_begin().getCaseSuccessor()->begin();
        isa<PHINode>(BBI); ++BBI) {
@@ -2739,15 +3081,33 @@
     }
   }
 
+  SmallVector<uint64_t, 8> Weights;
+  bool HasWeight = HasBranchWeights(SI);
+  if (HasWeight) {
+    GetBranchWeights(SI, Weights);
+    HasWeight = (Weights.size() == 1 + SI->getNumCases());
+  }
+
   // Remove dead cases from the switch.
   for (unsigned I = 0, E = DeadCases.size(); I != E; ++I) {
     SwitchInst::CaseIt Case = SI->findCaseValue(DeadCases[I]);
     assert(Case != SI->case_default() &&
            "Case was not found. Probably mistake in DeadCases forming.");
+    if (HasWeight) {
+      std::swap(Weights[Case.getCaseIndex()+1], Weights.back());
+      Weights.pop_back();
+    }
+
     // Prune unused values from PHI nodes.
     Case.getCaseSuccessor()->removePredecessor(SI->getParent());
     SI->removeCase(Case);
   }
+  if (HasWeight) {
+    SmallVector<uint32_t, 8> MDWeights(Weights.begin(), Weights.end());
+    SI->setMetadata(LLVMContext::MD_prof,
+                    MDBuilder(SI->getParent()->getContext()).
+                    createBranchWeights(MDWeights));
+  }
 
   return !DeadCases.empty();
 }
@@ -2823,33 +3183,512 @@
   return Changed;
 }
 
-bool SimplifyCFGOpt::SimplifySwitch(SwitchInst *SI, IRBuilder<> &Builder) {
-  // If this switch is too complex to want to look at, ignore it.
-  if (!isValueEqualityComparison(SI))
+/// ValidLookupTableConstant - Return true if the backend will be able to handle
+/// initializing an array of constants like C.
+static bool ValidLookupTableConstant(Constant *C) {
+  if (ConstantExpr *CE = dyn_cast<ConstantExpr>(C))
+    return CE->isGEPWithNoNotionalOverIndexing();
+
+  return isa<ConstantFP>(C) ||
+      isa<ConstantInt>(C) ||
+      isa<ConstantPointerNull>(C) ||
+      isa<GlobalValue>(C) ||
+      isa<UndefValue>(C);
+}
+
+/// LookupConstant - If V is a Constant, return it. Otherwise, try to look up
+/// its constant value in ConstantPool, returning 0 if it's not there.
+static Constant *LookupConstant(Value *V,
+                         const SmallDenseMap<Value*, Constant*>& ConstantPool) {
+  if (Constant *C = dyn_cast<Constant>(V))
+    return C;
+  return ConstantPool.lookup(V);
+}
+
+/// ConstantFold - Try to fold instruction I into a constant. This works for
+/// simple instructions such as binary operations where both operands are
+/// constant or can be replaced by constants from the ConstantPool. Returns the
+/// resulting constant on success, 0 otherwise.
+static Constant *ConstantFold(Instruction *I,
+                         const SmallDenseMap<Value*, Constant*>& ConstantPool) {
+  if (BinaryOperator *BO = dyn_cast<BinaryOperator>(I)) {
+    Constant *A = LookupConstant(BO->getOperand(0), ConstantPool);
+    if (!A)
+      return 0;
+    Constant *B = LookupConstant(BO->getOperand(1), ConstantPool);
+    if (!B)
+      return 0;
+    return ConstantExpr::get(BO->getOpcode(), A, B);
+  }
+
+  if (CmpInst *Cmp = dyn_cast<CmpInst>(I)) {
+    Constant *A = LookupConstant(I->getOperand(0), ConstantPool);
+    if (!A)
+      return 0;
+    Constant *B = LookupConstant(I->getOperand(1), ConstantPool);
+    if (!B)
+      return 0;
+    return ConstantExpr::getCompare(Cmp->getPredicate(), A, B);
+  }
+
+  if (SelectInst *Select = dyn_cast<SelectInst>(I)) {
+    Constant *A = LookupConstant(Select->getCondition(), ConstantPool);
+    if (!A)
+      return 0;
+    if (A->isAllOnesValue())
+      return LookupConstant(Select->getTrueValue(), ConstantPool);
+    if (A->isNullValue())
+      return LookupConstant(Select->getFalseValue(), ConstantPool);
+    return 0;
+  }
+
+  if (CastInst *Cast = dyn_cast<CastInst>(I)) {
+    Constant *A = LookupConstant(I->getOperand(0), ConstantPool);
+    if (!A)
+      return 0;
+    return ConstantExpr::getCast(Cast->getOpcode(), A, Cast->getDestTy());
+  }
+
+  return 0;
+}
+
+/// GetCaseResults - Try to determine the resulting constant values in phi nodes
+/// at the common destination basic block, *CommonDest, for one of the case
+/// destionations CaseDest corresponding to value CaseVal (0 for the default
+/// case), of a switch instruction SI.
+static bool GetCaseResults(SwitchInst *SI,
+                           ConstantInt *CaseVal,
+                           BasicBlock *CaseDest,
+                           BasicBlock **CommonDest,
+                           SmallVector<std::pair<PHINode*,Constant*>, 4> &Res) {
+  // The block from which we enter the common destination.
+  BasicBlock *Pred = SI->getParent();
+
+  // If CaseDest is empty except for some side-effect free instructions through
+  // which we can constant-propagate the CaseVal, continue to its successor.
+  SmallDenseMap<Value*, Constant*> ConstantPool;
+  ConstantPool.insert(std::make_pair(SI->getCondition(), CaseVal));
+  for (BasicBlock::iterator I = CaseDest->begin(), E = CaseDest->end(); I != E;
+       ++I) {
+    if (TerminatorInst *T = dyn_cast<TerminatorInst>(I)) {
+      // If the terminator is a simple branch, continue to the next block.
+      if (T->getNumSuccessors() != 1)
+        return false;
+      Pred = CaseDest;
+      CaseDest = T->getSuccessor(0);
+    } else if (isa<DbgInfoIntrinsic>(I)) {
+      // Skip debug intrinsic.
+      continue;
+    } else if (Constant *C = ConstantFold(I, ConstantPool)) {
+      // Instruction is side-effect free and constant.
+      ConstantPool.insert(std::make_pair(I, C));
+    } else {
+      break;
+    }
+  }
+
+  // If we did not have a CommonDest before, use the current one.
+  if (!*CommonDest)
+    *CommonDest = CaseDest;
+  // If the destination isn't the common one, abort.
+  if (CaseDest != *CommonDest)
     return false;
 
+  // Get the values for this case from phi nodes in the destination block.
+  BasicBlock::iterator I = (*CommonDest)->begin();
+  while (PHINode *PHI = dyn_cast<PHINode>(I++)) {
+    int Idx = PHI->getBasicBlockIndex(Pred);
+    if (Idx == -1)
+      continue;
+
+    Constant *ConstVal = LookupConstant(PHI->getIncomingValue(Idx),
+                                        ConstantPool);
+    if (!ConstVal)
+      return false;
+
+    // Note: If the constant comes from constant-propagating the case value
+    // through the CaseDest basic block, it will be safe to remove the
+    // instructions in that block. They cannot be used (except in the phi nodes
+    // we visit) outside CaseDest, because that block does not dominate its
+    // successor. If it did, we would not be in this phi node.
+
+    // Be conservative about which kinds of constants we support.
+    if (!ValidLookupTableConstant(ConstVal))
+      return false;
+
+    Res.push_back(std::make_pair(PHI, ConstVal));
+  }
+
+  return true;
+}
+
+namespace {
+  /// SwitchLookupTable - This class represents a lookup table that can be used
+  /// to replace a switch.
+  class SwitchLookupTable {
+  public:
+    /// SwitchLookupTable - Create a lookup table to use as a switch replacement
+    /// with the contents of Values, using DefaultValue to fill any holes in the
+    /// table.
+    SwitchLookupTable(Module &M,
+                      uint64_t TableSize,
+                      ConstantInt *Offset,
+               const SmallVector<std::pair<ConstantInt*, Constant*>, 4>& Values,
+                      Constant *DefaultValue,
+                      const DataLayout *TD);
+
+    /// BuildLookup - Build instructions with Builder to retrieve the value at
+    /// the position given by Index in the lookup table.
+    Value *BuildLookup(Value *Index, IRBuilder<> &Builder);
+
+    /// WouldFitInRegister - Return true if a table with TableSize elements of
+    /// type ElementType would fit in a target-legal register.
+    static bool WouldFitInRegister(const DataLayout *TD,
+                                   uint64_t TableSize,
+                                   const Type *ElementType);
+
+  private:
+    // Depending on the contents of the table, it can be represented in
+    // different ways.
+    enum {
+      // For tables where each element contains the same value, we just have to
+      // store that single value and return it for each lookup.
+      SingleValueKind,
+
+      // For small tables with integer elements, we can pack them into a bitmap
+      // that fits into a target-legal register. Values are retrieved by
+      // shift and mask operations.
+      BitMapKind,
+
+      // The table is stored as an array of values. Values are retrieved by load
+      // instructions from the table.
+      ArrayKind
+    } Kind;
+
+    // For SingleValueKind, this is the single value.
+    Constant *SingleValue;
+
+    // For BitMapKind, this is the bitmap.
+    ConstantInt *BitMap;
+    IntegerType *BitMapElementTy;
+
+    // For ArrayKind, this is the array.
+    GlobalVariable *Array;
+  };
+}
+
+SwitchLookupTable::SwitchLookupTable(Module &M,
+                                     uint64_t TableSize,
+                                     ConstantInt *Offset,
+               const SmallVector<std::pair<ConstantInt*, Constant*>, 4>& Values,
+                                     Constant *DefaultValue,
+                                     const DataLayout *TD) {
+  assert(Values.size() && "Can't build lookup table without values!");
+  assert(TableSize >= Values.size() && "Can't fit values in table!");
+
+  // If all values in the table are equal, this is that value.
+  SingleValue = Values.begin()->second;
+
+  // Build up the table contents.
+  SmallVector<Constant*, 64> TableContents(TableSize);
+  for (size_t I = 0, E = Values.size(); I != E; ++I) {
+    ConstantInt *CaseVal = Values[I].first;
+    Constant *CaseRes = Values[I].second;
+    assert(CaseRes->getType() == DefaultValue->getType());
+
+    uint64_t Idx = (CaseVal->getValue() - Offset->getValue())
+                   .getLimitedValue();
+    TableContents[Idx] = CaseRes;
+
+    if (CaseRes != SingleValue)
+      SingleValue = 0;
+  }
+
+  // Fill in any holes in the table with the default result.
+  if (Values.size() < TableSize) {
+    for (uint64_t I = 0; I < TableSize; ++I) {
+      if (!TableContents[I])
+        TableContents[I] = DefaultValue;
+    }
+
+    if (DefaultValue != SingleValue)
+      SingleValue = 0;
+  }
+
+  // If each element in the table contains the same value, we only need to store
+  // that single value.
+  if (SingleValue) {
+    Kind = SingleValueKind;
+    return;
+  }
+
+  // If the type is integer and the table fits in a register, build a bitmap.
+  if (WouldFitInRegister(TD, TableSize, DefaultValue->getType())) {
+    IntegerType *IT = cast<IntegerType>(DefaultValue->getType());
+    APInt TableInt(TableSize * IT->getBitWidth(), 0);
+    for (uint64_t I = TableSize; I > 0; --I) {
+      TableInt <<= IT->getBitWidth();
+      // Insert values into the bitmap. Undef values are set to zero.
+      if (!isa<UndefValue>(TableContents[I - 1])) {
+        ConstantInt *Val = cast<ConstantInt>(TableContents[I - 1]);
+        TableInt |= Val->getValue().zext(TableInt.getBitWidth());
+      }
+    }
+    BitMap = ConstantInt::get(M.getContext(), TableInt);
+    BitMapElementTy = IT;
+    Kind = BitMapKind;
+    ++NumBitMaps;
+    return;
+  }
+
+  // Store the table in an array.
+  ArrayType *ArrayTy = ArrayType::get(DefaultValue->getType(), TableSize);
+  Constant *Initializer = ConstantArray::get(ArrayTy, TableContents);
+
+  Array = new GlobalVariable(M, ArrayTy, /*constant=*/ true,
+                             GlobalVariable::PrivateLinkage,
+                             Initializer,
+                             "switch.table");
+  Array->setUnnamedAddr(true);
+  Kind = ArrayKind;
+}
+
+Value *SwitchLookupTable::BuildLookup(Value *Index, IRBuilder<> &Builder) {
+  switch (Kind) {
+    case SingleValueKind:
+      return SingleValue;
+    case BitMapKind: {
+      // Type of the bitmap (e.g. i59).
+      IntegerType *MapTy = BitMap->getType();
+
+      // Cast Index to the same type as the bitmap.
+      // Note: The Index is <= the number of elements in the table, so
+      // truncating it to the width of the bitmask is safe.
+      Value *ShiftAmt = Builder.CreateZExtOrTrunc(Index, MapTy, "switch.cast");
+
+      // Multiply the shift amount by the element width.
+      ShiftAmt = Builder.CreateMul(ShiftAmt,
+                      ConstantInt::get(MapTy, BitMapElementTy->getBitWidth()),
+                                   "switch.shiftamt");
+
+      // Shift down.
+      Value *DownShifted = Builder.CreateLShr(BitMap, ShiftAmt,
+                                              "switch.downshift");
+      // Mask off.
+      return Builder.CreateTrunc(DownShifted, BitMapElementTy,
+                                 "switch.masked");
+    }
+    case ArrayKind: {
+      Value *GEPIndices[] = { Builder.getInt32(0), Index };
+      Value *GEP = Builder.CreateInBoundsGEP(Array, GEPIndices,
+                                             "switch.gep");
+      return Builder.CreateLoad(GEP, "switch.load");
+    }
+  }
+  llvm_unreachable("Unknown lookup table kind!");
+}
+
+bool SwitchLookupTable::WouldFitInRegister(const DataLayout *TD,
+                                           uint64_t TableSize,
+                                           const Type *ElementType) {
+  if (!TD)
+    return false;
+  const IntegerType *IT = dyn_cast<IntegerType>(ElementType);
+  if (!IT)
+    return false;
+  // FIXME: If the type is wider than it needs to be, e.g. i8 but all values
+  // are <= 15, we could try to narrow the type.
+
+  // Avoid overflow, fitsInLegalInteger uses unsigned int for the width.
+  if (TableSize >= UINT_MAX/IT->getBitWidth())
+    return false;
+  return TD->fitsInLegalInteger(TableSize * IT->getBitWidth());
+}
+
+/// ShouldBuildLookupTable - Determine whether a lookup table should be built
+/// for this switch, based on the number of caes, size of the table and the
+/// types of the results.
+static bool ShouldBuildLookupTable(SwitchInst *SI,
+                                   uint64_t TableSize,
+                                   const DataLayout *TD,
+                            const SmallDenseMap<PHINode*, Type*>& ResultTypes) {
+  // The table density should be at least 40%. This is the same criterion as for
+  // jump tables, see SelectionDAGBuilder::handleJTSwitchCase.
+  // FIXME: Find the best cut-off.
+  if (SI->getNumCases() > TableSize || TableSize >= UINT64_MAX / 10)
+    return false; // TableSize overflowed, or mul below might overflow.
+  if (SI->getNumCases() * 10 >= TableSize * 4)
+    return true;
+
+  // If each table would fit in a register, we should build it anyway.
+  for (SmallDenseMap<PHINode*, Type*>::const_iterator I = ResultTypes.begin(),
+       E = ResultTypes.end(); I != E; ++I) {
+    if (!SwitchLookupTable::WouldFitInRegister(TD, TableSize, I->second))
+      return false;
+  }
+  return true;
+}
+
+/// SwitchToLookupTable - If the switch is only used to initialize one or more
+/// phi nodes in a common successor block with different constant values,
+/// replace the switch with lookup tables.
+static bool SwitchToLookupTable(SwitchInst *SI,
+                                IRBuilder<> &Builder,
+                                const DataLayout* TD,
+                                const TargetTransformInfo *TTI) {
+  assert(SI->getNumCases() > 1 && "Degenerate switch?");
+
+  // Only build lookup table when we have a target that supports it.
+  if (!TTI || !TTI->getScalarTargetTransformInfo() ||
+      !TTI->getScalarTargetTransformInfo()->shouldBuildLookupTables())
+    return false;
+
+  // FIXME: If the switch is too sparse for a lookup table, perhaps we could
+  // split off a dense part and build a lookup table for that.
+
+  // FIXME: This creates arrays of GEPs to constant strings, which means each
+  // GEP needs a runtime relocation in PIC code. We should just build one big
+  // string and lookup indices into that.
+
+  // Ignore the switch if the number of cases is too small.
+  // This is similar to the check when building jump tables in
+  // SelectionDAGBuilder::handleJTSwitchCase.
+  // FIXME: Determine the best cut-off.
+  if (SI->getNumCases() < 4)
+    return false;
+
+  // Figure out the corresponding result for each case value and phi node in the
+  // common destination, as well as the the min and max case values.
+  assert(SI->case_begin() != SI->case_end());
+  SwitchInst::CaseIt CI = SI->case_begin();
+  ConstantInt *MinCaseVal = CI.getCaseValue();
+  ConstantInt *MaxCaseVal = CI.getCaseValue();
+
+  BasicBlock *CommonDest = 0;
+  typedef SmallVector<std::pair<ConstantInt*, Constant*>, 4> ResultListTy;
+  SmallDenseMap<PHINode*, ResultListTy> ResultLists;
+  SmallDenseMap<PHINode*, Constant*> DefaultResults;
+  SmallDenseMap<PHINode*, Type*> ResultTypes;
+  SmallVector<PHINode*, 4> PHIs;
+
+  for (SwitchInst::CaseIt E = SI->case_end(); CI != E; ++CI) {
+    ConstantInt *CaseVal = CI.getCaseValue();
+    if (CaseVal->getValue().slt(MinCaseVal->getValue()))
+      MinCaseVal = CaseVal;
+    if (CaseVal->getValue().sgt(MaxCaseVal->getValue()))
+      MaxCaseVal = CaseVal;
+
+    // Resulting value at phi nodes for this case value.
+    typedef SmallVector<std::pair<PHINode*, Constant*>, 4> ResultsTy;
+    ResultsTy Results;
+    if (!GetCaseResults(SI, CaseVal, CI.getCaseSuccessor(), &CommonDest,
+                        Results))
+      return false;
+
+    // Append the result from this case to the list for each phi.
+    for (ResultsTy::iterator I = Results.begin(), E = Results.end(); I!=E; ++I) {
+      if (!ResultLists.count(I->first))
+        PHIs.push_back(I->first);
+      ResultLists[I->first].push_back(std::make_pair(CaseVal, I->second));
+    }
+  }
+
+  // Get the resulting values for the default case.
+  SmallVector<std::pair<PHINode*, Constant*>, 4> DefaultResultsList;
+  if (!GetCaseResults(SI, 0, SI->getDefaultDest(), &CommonDest,
+                      DefaultResultsList))
+    return false;
+  for (size_t I = 0, E = DefaultResultsList.size(); I != E; ++I) {
+    PHINode *PHI = DefaultResultsList[I].first;
+    Constant *Result = DefaultResultsList[I].second;
+    DefaultResults[PHI] = Result;
+    ResultTypes[PHI] = Result->getType();
+  }
+
+  APInt RangeSpread = MaxCaseVal->getValue() - MinCaseVal->getValue();
+  uint64_t TableSize = RangeSpread.getLimitedValue() + 1;
+  if (!ShouldBuildLookupTable(SI, TableSize, TD, ResultTypes))
+    return false;
+
+  // Create the BB that does the lookups.
+  Module &Mod = *CommonDest->getParent()->getParent();
+  BasicBlock *LookupBB = BasicBlock::Create(Mod.getContext(),
+                                            "switch.lookup",
+                                            CommonDest->getParent(),
+                                            CommonDest);
+
+  // Check whether the condition value is within the case range, and branch to
+  // the new BB.
+  Builder.SetInsertPoint(SI);
+  Value *TableIndex = Builder.CreateSub(SI->getCondition(), MinCaseVal,
+                                        "switch.tableidx");
+  Value *Cmp = Builder.CreateICmpULT(TableIndex, ConstantInt::get(
+      MinCaseVal->getType(), TableSize));
+  Builder.CreateCondBr(Cmp, LookupBB, SI->getDefaultDest());
+
+  // Populate the BB that does the lookups.
+  Builder.SetInsertPoint(LookupBB);
+  bool ReturnedEarly = false;
+  for (size_t I = 0, E = PHIs.size(); I != E; ++I) {
+    PHINode *PHI = PHIs[I];
+
+    SwitchLookupTable Table(Mod, TableSize, MinCaseVal, ResultLists[PHI],
+                            DefaultResults[PHI], TD);
+
+    Value *Result = Table.BuildLookup(TableIndex, Builder);
+
+    // If the result is used to return immediately from the function, we want to
+    // do that right here.
+    if (PHI->hasOneUse() && isa<ReturnInst>(*PHI->use_begin()) &&
+        *PHI->use_begin() == CommonDest->getFirstNonPHIOrDbg()) {
+      Builder.CreateRet(Result);
+      ReturnedEarly = true;
+      break;
+    }
+
+    PHI->addIncoming(Result, LookupBB);
+  }
+
+  if (!ReturnedEarly)
+    Builder.CreateBr(CommonDest);
+
+  // Remove the switch.
+  for (unsigned i = 0; i < SI->getNumSuccessors(); ++i) {
+    BasicBlock *Succ = SI->getSuccessor(i);
+    if (Succ == SI->getDefaultDest()) continue;
+    Succ->removePredecessor(SI->getParent());
+  }
+  SI->eraseFromParent();
+
+  ++NumLookupTables;
+  return true;
+}
+
+bool SimplifyCFGOpt::SimplifySwitch(SwitchInst *SI, IRBuilder<> &Builder) {
   BasicBlock *BB = SI->getParent();
 
-  // If we only have one predecessor, and if it is a branch on this value,
-  // see if that predecessor totally determines the outcome of this switch.
-  if (BasicBlock *OnlyPred = BB->getSinglePredecessor())
-    if (SimplifyEqualityComparisonWithOnlyPredecessor(SI, OnlyPred, Builder))
-      return SimplifyCFG(BB) | true;
+  if (isValueEqualityComparison(SI)) {
+    // If we only have one predecessor, and if it is a branch on this value,
+    // see if that predecessor totally determines the outcome of this switch.
+    if (BasicBlock *OnlyPred = BB->getSinglePredecessor())
+      if (SimplifyEqualityComparisonWithOnlyPredecessor(SI, OnlyPred, Builder))
+        return SimplifyCFG(BB) | true;
 
-  Value *Cond = SI->getCondition();
-  if (SelectInst *Select = dyn_cast<SelectInst>(Cond))
-    if (SimplifySwitchOnSelect(SI, Select))
-      return SimplifyCFG(BB) | true;
+    Value *Cond = SI->getCondition();
+    if (SelectInst *Select = dyn_cast<SelectInst>(Cond))
+      if (SimplifySwitchOnSelect(SI, Select))
+        return SimplifyCFG(BB) | true;
 
-  // If the block only contains the switch, see if we can fold the block
-  // away into any preds.
-  BasicBlock::iterator BBI = BB->begin();
-  // Ignore dbg intrinsics.
-  while (isa<DbgInfoIntrinsic>(BBI))
-    ++BBI;
-  if (SI == &*BBI)
-    if (FoldValueComparisonIntoPredecessors(SI, Builder))
-      return SimplifyCFG(BB) | true;
+    // If the block only contains the switch, see if we can fold the block
+    // away into any preds.
+    BasicBlock::iterator BBI = BB->begin();
+    // Ignore dbg intrinsics.
+    while (isa<DbgInfoIntrinsic>(BBI))
+      ++BBI;
+    if (SI == &*BBI)
+      if (FoldValueComparisonIntoPredecessors(SI, Builder))
+        return SimplifyCFG(BB) | true;
+  }
 
   // Try to transform the switch into an icmp and a branch.
   if (TurnSwitchRangeIntoICmp(SI, Builder))
@@ -2862,13 +3701,16 @@
   if (ForwardSwitchConditionToPHI(SI))
     return SimplifyCFG(BB) | true;
 
+  if (SwitchToLookupTable(SI, Builder, TD, TTI))
+    return SimplifyCFG(BB) | true;
+
   return false;
 }
 
 bool SimplifyCFGOpt::SimplifyIndirectBr(IndirectBrInst *IBI) {
   BasicBlock *BB = IBI->getParent();
   bool Changed = false;
-  
+
   // Eliminate redundant destinations.
   SmallPtrSet<Value *, 8> Succs;
   for (unsigned i = 0, e = IBI->getNumDestinations(); i != e; ++i) {
@@ -2879,7 +3721,7 @@
       --i; --e;
       Changed = true;
     }
-  } 
+  }
 
   if (IBI->getNumDestinations() == 0) {
     // If the indirectbr has no successors, change it to unreachable.
@@ -2887,14 +3729,14 @@
     EraseTerminatorInstAndDCECond(IBI);
     return true;
   }
-  
+
   if (IBI->getNumDestinations() == 1) {
     // If the indirectbr has one successor, change it to a direct branch.
     BranchInst::Create(IBI->getDestination(0), IBI);
     EraseTerminatorInstAndDCECond(IBI);
     return true;
   }
-  
+
   if (SelectInst *SI = dyn_cast<SelectInst>(IBI->getAddress())) {
     if (SimplifyIndirectBrOnSelect(IBI, SI))
       return SimplifyCFG(BB) | true;
@@ -2904,13 +3746,16 @@
 
 bool SimplifyCFGOpt::SimplifyUncondBranch(BranchInst *BI, IRBuilder<> &Builder){
   BasicBlock *BB = BI->getParent();
-  
+
+  if (SinkCommon && SinkThenElseCodeToEnd(BI))
+    return true;
+
   // If the Terminator is the only non-phi instruction, simplify the block.
   BasicBlock::iterator I = BB->getFirstNonPHIOrDbgOrLifetime();
   if (I->isTerminator() && BB != &BB->getParent()->getEntryBlock() &&
       TryToSimplifyUncondBranchFromEmptyBlock(BB))
     return true;
-  
+
   // If the only instruction in the block is a seteq/setne comparison
   // against a constant, try to simplify the block.
   if (ICmpInst *ICI = dyn_cast<ICmpInst>(I))
@@ -2921,7 +3766,7 @@
           TryToSimplifyUncondBranchWithICmpInIt(ICI, TD, Builder))
         return true;
     }
-  
+
   // If this basic block is ONLY a compare and a branch, and if a predecessor
   // branches to us and our successor, fold the comparison into the
   // predecessor and use logical operations to update the incoming value
@@ -2934,7 +3779,7 @@
 
 bool SimplifyCFGOpt::SimplifyCondBranch(BranchInst *BI, IRBuilder<> &Builder) {
   BasicBlock *BB = BI->getParent();
-  
+
   // Conditional branch
   if (isValueEqualityComparison(BI)) {
     // If we only have one predecessor, and if it is a branch on this value,
@@ -2943,7 +3788,7 @@
     if (BasicBlock *OnlyPred = BB->getSinglePredecessor())
       if (SimplifyEqualityComparisonWithOnlyPredecessor(BI, OnlyPred, Builder))
         return SimplifyCFG(BB) | true;
-    
+
     // This block must be empty, except for the setcond inst, if it exists.
     // Ignore dbg intrinsics.
     BasicBlock::iterator I = BB->begin();
@@ -2962,17 +3807,17 @@
         return SimplifyCFG(BB) | true;
     }
   }
-  
+
   // Try to turn "br (X == 0 | X == 1), T, F" into a switch instruction.
   if (SimplifyBranchOnICmpChain(BI, TD, Builder))
     return true;
-  
+
   // If this basic block is ONLY a compare and a branch, and if a predecessor
   // branches to us and one of our successors, fold the comparison into the
   // predecessor and use logical operations to pick the right destination.
   if (FoldBranchToCommonDest(BI))
     return SimplifyCFG(BB) | true;
-  
+
   // We have a conditional branch to two blocks that are only reachable
   // from BI.  We know that the condbr dominates the two blocks, so see if
   // there is any identical code in the "then" and "else" blocks.  If so, we
@@ -2999,14 +3844,14 @@
       if (SpeculativelyExecuteBB(BI, BI->getSuccessor(1)))
         return SimplifyCFG(BB) | true;
   }
-  
+
   // If this is a branch on a phi node in the current block, thread control
   // through this block if any PHI node entries are constants.
   if (PHINode *PN = dyn_cast<PHINode>(BI->getCondition()))
     if (PN->getParent() == BI->getParent())
       if (FoldCondBranchOnPHI(BI, TD))
         return SimplifyCFG(BB) | true;
-  
+
   // Scan predecessor blocks for conditional branches.
   for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI)
     if (BranchInst *PBI = dyn_cast<BranchInst>((*PI)->getTerminator()))
@@ -3023,11 +3868,12 @@
   if (!C)
     return false;
 
-  if (!I->hasOneUse()) // Only look at single-use instructions, for compile time
+  if (I->use_empty())
     return false;
 
   if (C->isNullValue()) {
-    Instruction *Use = I->use_back();
+    // Only look at the first use, avoid hurting compile time with long uselists
+    User *Use = *I->use_begin();
 
     // Now make sure that there are no instructions in between that can alter
     // control flow (eg. calls)
@@ -3114,7 +3960,7 @@
   //
   if (MergeBlockIntoPredecessor(BB))
     return true;
-  
+
   IRBuilder<> Builder(BB);
 
   // If there is a trivial two-entry PHI node in this basic block, and we can
@@ -3152,6 +3998,7 @@
 /// eliminates unreachable basic blocks, and does other "peephole" optimization
 /// of the CFG.  It returns true if a modification was made.
 ///
-bool llvm::SimplifyCFG(BasicBlock *BB, const TargetData *TD) {
-  return SimplifyCFGOpt(TD).run(BB);
+bool llvm::SimplifyCFG(BasicBlock *BB, const DataLayout *TD,
+                       const TargetTransformInfo *TTI) {
+  return SimplifyCFGOpt(TD, TTI).run(BB);
 }

Modified: llvm/branches/AMDILBackend/lib/Transforms/Utils/SimplifyIndVar.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Utils/SimplifyIndVar.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Utils/SimplifyIndVar.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Utils/SimplifyIndVar.cpp Tue Jan 15 11:16:16 2013
@@ -24,7 +24,7 @@
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/raw_ostream.h"
 #include "llvm/Transforms/Utils/SimplifyIndVar.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/Statistic.h"
 
@@ -44,7 +44,7 @@
     Loop             *L;
     LoopInfo         *LI;
     ScalarEvolution  *SE;
-    const TargetData *TD; // May be NULL
+    const DataLayout *TD; // May be NULL
 
     SmallVectorImpl<WeakVH> &DeadInsts;
 
@@ -56,7 +56,7 @@
       L(Loop),
       LI(LPM->getAnalysisIfAvailable<LoopInfo>()),
       SE(SE),
-      TD(LPM->getAnalysisIfAvailable<TargetData>()),
+      TD(LPM->getAnalysisIfAvailable<DataLayout>()),
       DeadInsts(Dead),
       Changed(false) {
       assert(LI && "IV simplification requires LoopInfo");

Modified: llvm/branches/AMDILBackend/lib/Transforms/Utils/SimplifyInstructions.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Utils/SimplifyInstructions.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Utils/SimplifyInstructions.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Utils/SimplifyInstructions.cpp Tue Jan 15 11:16:16 2013
@@ -23,7 +23,7 @@
 #include "llvm/ADT/Statistic.h"
 #include "llvm/Analysis/Dominators.h"
 #include "llvm/Analysis/InstructionSimplify.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Target/TargetLibraryInfo.h"
 #include "llvm/Transforms/Scalar.h"
 #include "llvm/Transforms/Utils/Local.h"
@@ -46,7 +46,7 @@
     /// runOnFunction - Remove instructions that simplify.
     bool runOnFunction(Function &F) {
       const DominatorTree *DT = getAnalysisIfAvailable<DominatorTree>();
-      const TargetData *TD = getAnalysisIfAvailable<TargetData>();
+      const DataLayout *TD = getAnalysisIfAvailable<DataLayout>();
       const TargetLibraryInfo *TLI = &getAnalysis<TargetLibraryInfo>();
       SmallPtrSet<const Instruction*, 8> S1, S2, *ToSimplify = &S1, *Next = &S2;
       bool Changed = false;
@@ -72,7 +72,7 @@
                 ++NumSimplified;
                 Changed = true;
               }
-            Changed |= RecursivelyDeleteTriviallyDeadInstructions(I);
+            Changed |= RecursivelyDeleteTriviallyDeadInstructions(I, TLI);
           }
 
         // Place the list of instructions to simplify on the next loop iteration

Modified: llvm/branches/AMDILBackend/lib/Transforms/Utils/Utils.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Utils/Utils.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Utils/Utils.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Utils/Utils.cpp Tue Jan 15 11:16:16 2013
@@ -29,6 +29,7 @@
   initializePromotePassPass(Registry);
   initializeUnifyFunctionExitNodesPass(Registry);
   initializeInstSimplifierPass(Registry);
+  initializeMetaRenamerPass(Registry);
 }
 
 /// LLVMInitializeTransformUtils - C binding for initializeTransformUtilsPasses.

Modified: llvm/branches/AMDILBackend/lib/Transforms/Utils/ValueMapper.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Utils/ValueMapper.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Utils/ValueMapper.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Utils/ValueMapper.cpp Tue Jan 15 11:16:16 2013
@@ -21,7 +21,7 @@
 using namespace llvm;
 
 // Out of line method to get vtable etc for class.
-void ValueMapTypeRemapper::Anchor() {}
+void ValueMapTypeRemapper::anchor() {}
 
 Value *llvm::MapValue(const Value *V, ValueToValueMapTy &VM, RemapFlags Flags,
                       ValueMapTypeRemapper *TypeMapper) {

Modified: llvm/branches/AMDILBackend/lib/Transforms/Vectorize/BBVectorize.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Vectorize/BBVectorize.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Vectorize/BBVectorize.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Vectorize/BBVectorize.cpp Tue Jan 15 11:16:16 2013
@@ -28,12 +28,14 @@
 #include "llvm/Type.h"
 #include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/DenseSet.h"
+#include "llvm/ADT/SmallSet.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/Statistic.h"
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/StringExtras.h"
 #include "llvm/Analysis/AliasAnalysis.h"
 #include "llvm/Analysis/AliasSetTracker.h"
+#include "llvm/Analysis/Dominators.h"
 #include "llvm/Analysis/ScalarEvolution.h"
 #include "llvm/Analysis/ScalarEvolutionExpressions.h"
 #include "llvm/Analysis/ValueTracking.h"
@@ -41,17 +43,27 @@
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/raw_ostream.h"
 #include "llvm/Support/ValueHandle.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
+#include "llvm/TargetTransformInfo.h"
 #include "llvm/Transforms/Utils/Local.h"
 #include "llvm/Transforms/Vectorize.h"
 #include <algorithm>
 #include <map>
 using namespace llvm;
 
+static cl::opt<bool>
+IgnoreTargetInfo("bb-vectorize-ignore-target-info",  cl::init(false),
+  cl::Hidden, cl::desc("Ignore target information"));
+
 static cl::opt<unsigned>
 ReqChainDepth("bb-vectorize-req-chain-depth", cl::init(6), cl::Hidden,
   cl::desc("The required chain depth for vectorization"));
 
+static cl::opt<bool>
+UseChainDepthWithTI("bb-vectorize-use-chain-depth",  cl::init(false),
+  cl::Hidden, cl::desc("Use the chain depth requirement with"
+                       " target information"));
+
 static cl::opt<unsigned>
 SearchLimit("bb-vectorize-search-limit", cl::init(400), cl::Hidden,
   cl::desc("The maximum search distance for instruction pairs"));
@@ -93,8 +105,9 @@
 NoFloats("bb-vectorize-no-floats", cl::init(false), cl::Hidden,
   cl::desc("Don't try to vectorize floating-point values"));
 
+// FIXME: This should default to false once pointer vector support works.
 static cl::opt<bool>
-NoPointers("bb-vectorize-no-pointers", cl::init(false), cl::Hidden,
+NoPointers("bb-vectorize-no-pointers", cl::init(/*false*/ true), cl::Hidden,
   cl::desc("Don't try to vectorize pointer values"));
 
 static cl::opt<bool>
@@ -159,6 +172,12 @@
   cl::init(false), cl::Hidden,
   cl::desc("When debugging is enabled, output information on the"
            " cycle-checking process"));
+
+static cl::opt<bool>
+PrintAfterEveryPair("bb-vectorize-debug-print-after-every-pair",
+  cl::init(false), cl::Hidden,
+  cl::desc("When debugging is enabled, dump the basic block after"
+           " every pair is fused"));
 #endif
 
 STATISTIC(NumFusedOps, "Number of operations fused by bb-vectorize");
@@ -177,13 +196,19 @@
     BBVectorize(Pass *P, const VectorizeConfig &C)
       : BasicBlockPass(ID), Config(C) {
       AA = &P->getAnalysis<AliasAnalysis>();
+      DT = &P->getAnalysis<DominatorTree>();
       SE = &P->getAnalysis<ScalarEvolution>();
-      TD = P->getAnalysisIfAvailable<TargetData>();
+      TD = P->getAnalysisIfAvailable<DataLayout>();
+      TTI = IgnoreTargetInfo ? 0 :
+        P->getAnalysisIfAvailable<TargetTransformInfo>();
+      VTTI = TTI ? TTI->getVectorTargetTransformInfo() : 0;
     }
 
     typedef std::pair<Value *, Value *> ValuePair;
+    typedef std::pair<ValuePair, int> ValuePairWithCost;
     typedef std::pair<ValuePair, size_t> ValuePairWithDepth;
     typedef std::pair<ValuePair, ValuePair> VPPair; // A ValuePair pair
+    typedef std::pair<VPPair, unsigned> VPPairWithType;
     typedef std::pair<std::multimap<Value *, Value *>::iterator,
               std::multimap<Value *, Value *>::iterator> VPIteratorPair;
     typedef std::pair<std::multimap<ValuePair, ValuePair>::iterator,
@@ -191,8 +216,11 @@
                 VPPIteratorPair;
 
     AliasAnalysis *AA;
+    DominatorTree *DT;
     ScalarEvolution *SE;
-    TargetData *TD;
+    DataLayout *TD;
+    TargetTransformInfo *TTI;
+    const VectorTargetTransformInfo *VTTI;
 
     // FIXME: const correct?
 
@@ -201,11 +229,23 @@
     bool getCandidatePairs(BasicBlock &BB,
                        BasicBlock::iterator &Start,
                        std::multimap<Value *, Value *> &CandidatePairs,
+                       DenseSet<ValuePair> &FixedOrderPairs,
+                       DenseMap<ValuePair, int> &CandidatePairCostSavings,
                        std::vector<Value *> &PairableInsts, bool NonPow2Len);
 
+    // FIXME: The current implementation does not account for pairs that
+    // are connected in multiple ways. For example:
+    //   C1 = A1 / A2; C2 = A2 / A1 (which may be both direct and a swap)
+    enum PairConnectionType {
+      PairConnectionDirect,
+      PairConnectionSwap,
+      PairConnectionSplat
+    };
+
     void computeConnectedPairs(std::multimap<Value *, Value *> &CandidatePairs,
                        std::vector<Value *> &PairableInsts,
-                       std::multimap<ValuePair, ValuePair> &ConnectedPairs);
+                       std::multimap<ValuePair, ValuePair> &ConnectedPairs,
+                       DenseMap<VPPair, unsigned> &PairConnectionTypes);
 
     void buildDepMap(BasicBlock &BB,
                        std::multimap<Value *, Value *> &CandidatePairs,
@@ -213,19 +253,29 @@
                        DenseSet<ValuePair> &PairableInstUsers);
 
     void choosePairs(std::multimap<Value *, Value *> &CandidatePairs,
+                        DenseMap<ValuePair, int> &CandidatePairCostSavings,
                         std::vector<Value *> &PairableInsts,
+                        DenseSet<ValuePair> &FixedOrderPairs,
+                        DenseMap<VPPair, unsigned> &PairConnectionTypes,
                         std::multimap<ValuePair, ValuePair> &ConnectedPairs,
+                        std::multimap<ValuePair, ValuePair> &ConnectedPairDeps,
                         DenseSet<ValuePair> &PairableInstUsers,
                         DenseMap<Value *, Value *>& ChosenPairs);
 
     void fuseChosenPairs(BasicBlock &BB,
                      std::vector<Value *> &PairableInsts,
-                     DenseMap<Value *, Value *>& ChosenPairs);
+                     DenseMap<Value *, Value *>& ChosenPairs,
+                     DenseSet<ValuePair> &FixedOrderPairs,
+                     DenseMap<VPPair, unsigned> &PairConnectionTypes,
+                     std::multimap<ValuePair, ValuePair> &ConnectedPairs,
+                     std::multimap<ValuePair, ValuePair> &ConnectedPairDeps);
+
 
     bool isInstVectorizable(Instruction *I, bool &IsSimpleLoadStore);
 
     bool areInstsCompatible(Instruction *I, Instruction *J,
-                       bool IsSimpleLoadStore, bool NonPow2Len);
+                       bool IsSimpleLoadStore, bool NonPow2Len,
+                       int &CostSavings, int &FixedOrder);
 
     bool trackUsesOfI(DenseSet<Value *> &Users,
                       AliasSetTracker &WriteSet, Instruction *I,
@@ -236,6 +286,7 @@
                       std::multimap<Value *, Value *> &CandidatePairs,
                       std::vector<Value *> &PairableInsts,
                       std::multimap<ValuePair, ValuePair> &ConnectedPairs,
+                      DenseMap<VPPair, unsigned> &PairConnectionTypes,
                       ValuePair P);
 
     bool pairsConflict(ValuePair P, ValuePair Q,
@@ -267,17 +318,21 @@
 
     void findBestTreeFor(
                       std::multimap<Value *, Value *> &CandidatePairs,
+                      DenseMap<ValuePair, int> &CandidatePairCostSavings,
                       std::vector<Value *> &PairableInsts,
+                      DenseSet<ValuePair> &FixedOrderPairs,
+                      DenseMap<VPPair, unsigned> &PairConnectionTypes,
                       std::multimap<ValuePair, ValuePair> &ConnectedPairs,
+                      std::multimap<ValuePair, ValuePair> &ConnectedPairDeps,
                       DenseSet<ValuePair> &PairableInstUsers,
                       std::multimap<ValuePair, ValuePair> &PairableInstUserMap,
                       DenseMap<Value *, Value *> &ChosenPairs,
                       DenseSet<ValuePair> &BestTree, size_t &BestMaxDepth,
-                      size_t &BestEffSize, VPIteratorPair ChoiceRange,
+                      int &BestEffSize, VPIteratorPair ChoiceRange,
                       bool UseCycleCheck);
 
     Value *getReplacementPointerInput(LLVMContext& Context, Instruction *I,
-                     Instruction *J, unsigned o, bool FlipMemInputs);
+                     Instruction *J, unsigned o);
 
     void fillNewShuffleMask(LLVMContext& Context, Instruction *J,
                      unsigned MaskOffset, unsigned NumInElem,
@@ -289,20 +344,20 @@
 
     bool expandIEChain(LLVMContext& Context, Instruction *I, Instruction *J,
                        unsigned o, Value *&LOp, unsigned numElemL,
-                       Type *ArgTypeL, Type *ArgTypeR,
+                       Type *ArgTypeL, Type *ArgTypeR, bool IBeforeJ,
                        unsigned IdxOff = 0);
 
     Value *getReplacementInput(LLVMContext& Context, Instruction *I,
-                     Instruction *J, unsigned o, bool FlipMemInputs);
+                     Instruction *J, unsigned o, bool IBeforeJ);
 
     void getReplacementInputsForPair(LLVMContext& Context, Instruction *I,
                      Instruction *J, SmallVector<Value *, 3> &ReplacedOperands,
-                     bool FlipMemInputs);
+                     bool IBeforeJ);
 
     void replaceOutputsOfPair(LLVMContext& Context, Instruction *I,
                      Instruction *J, Instruction *K,
                      Instruction *&InsertionPt, Instruction *&K1,
-                     Instruction *&K2, bool FlipMemInputs);
+                     Instruction *&K2);
 
     void collectPairLoadMoveSet(BasicBlock &BB,
                      DenseMap<Value *, Value *> &ChosenPairs,
@@ -314,10 +369,6 @@
                      DenseMap<Value *, Value *> &ChosenPairs,
                      std::multimap<Value *, Value *> &LoadMoveSet);
 
-    void collectPtrInfo(std::vector<Value *> &PairableInsts,
-                        DenseMap<Value *, Value *> &ChosenPairs,
-                        DenseSet<Value *> &LowPtrInsts);
-
     bool canMoveUsesOfIAfterJ(BasicBlock &BB,
                      std::multimap<Value *, Value *> &LoadMoveSet,
                      Instruction *I, Instruction *J);
@@ -330,13 +381,22 @@
     void combineMetadata(Instruction *K, const Instruction *J);
 
     bool vectorizeBB(BasicBlock &BB) {
+      if (!DT->isReachableFromEntry(&BB)) {
+        DEBUG(dbgs() << "BBV: skipping unreachable " << BB.getName() <<
+              " in " << BB.getParent()->getName() << "\n");
+        return false;
+      }
+
+      DEBUG(if (VTTI) dbgs() << "BBV: using target information\n");
+
       bool changed = false;
       // Iterate a sufficient number of times to merge types of size 1 bit,
       // then 2 bits, then 4, etc. up to half of the target vector width of the
       // target vector register.
       unsigned n = 1;
       for (unsigned v = 2;
-           v <= Config.VectorBits && (!Config.MaxIter || n <= Config.MaxIter);
+           (VTTI || v <= Config.VectorBits) &&
+           (!Config.MaxIter || n <= Config.MaxIter);
            v *= 2, ++n) {
         DEBUG(dbgs() << "BBV: fusing loop #" << n <<
               " for " << BB.getName() << " in " <<
@@ -363,8 +423,12 @@
 
     virtual bool runOnBasicBlock(BasicBlock &BB) {
       AA = &getAnalysis<AliasAnalysis>();
+      DT = &getAnalysis<DominatorTree>();
       SE = &getAnalysis<ScalarEvolution>();
-      TD = getAnalysisIfAvailable<TargetData>();
+      TD = getAnalysisIfAvailable<DataLayout>();
+      TTI = IgnoreTargetInfo ? 0 :
+        getAnalysisIfAvailable<TargetTransformInfo>();
+      VTTI = TTI ? TTI->getVectorTargetTransformInfo() : 0;
 
       return vectorizeBB(BB);
     }
@@ -372,8 +436,10 @@
     virtual void getAnalysisUsage(AnalysisUsage &AU) const {
       BasicBlockPass::getAnalysisUsage(AU);
       AU.addRequired<AliasAnalysis>();
+      AU.addRequired<DominatorTree>();
       AU.addRequired<ScalarEvolution>();
       AU.addPreserved<AliasAnalysis>();
+      AU.addPreserved<DominatorTree>();
       AU.addPreserved<ScalarEvolution>();
       AU.setPreservesCFG();
     }
@@ -415,6 +481,14 @@
         T2 = cast<CastInst>(I)->getSrcTy();
       else
         T2 = T1;
+
+      if (SelectInst *SI = dyn_cast<SelectInst>(I)) {
+        T2 = SI->getCondition()->getType();
+      } else if (ShuffleVectorInst *SI = dyn_cast<ShuffleVectorInst>(I)) {
+        T2 = SI->getOperand(0)->getType();
+      } else if (CmpInst *CI = dyn_cast<CmpInst>(I)) {
+        T2 = CI->getOperand(0)->getType();
+      }
     }
 
     // Returns the weight associated with the provided value. A chain of
@@ -446,6 +520,62 @@
       return 1;
     }
 
+    // Returns the cost of the provided instruction using VTTI.
+    // This does not handle loads and stores.
+    unsigned getInstrCost(unsigned Opcode, Type *T1, Type *T2) {
+      switch (Opcode) {
+      default: break;
+      case Instruction::GetElementPtr:
+        // We mark this instruction as zero-cost because scalar GEPs are usually
+        // lowered to the intruction addressing mode. At the moment we don't
+        // generate vector GEPs.
+        return 0;
+      case Instruction::Br:
+        return VTTI->getCFInstrCost(Opcode);
+      case Instruction::PHI:
+        return 0;
+      case Instruction::Add:
+      case Instruction::FAdd:
+      case Instruction::Sub:
+      case Instruction::FSub:
+      case Instruction::Mul:
+      case Instruction::FMul:
+      case Instruction::UDiv:
+      case Instruction::SDiv:
+      case Instruction::FDiv:
+      case Instruction::URem:
+      case Instruction::SRem:
+      case Instruction::FRem:
+      case Instruction::Shl:
+      case Instruction::LShr:
+      case Instruction::AShr:
+      case Instruction::And:
+      case Instruction::Or:
+      case Instruction::Xor:
+        return VTTI->getArithmeticInstrCost(Opcode, T1);
+      case Instruction::Select:
+      case Instruction::ICmp:
+      case Instruction::FCmp:
+        return VTTI->getCmpSelInstrCost(Opcode, T1, T2);
+      case Instruction::ZExt:
+      case Instruction::SExt:
+      case Instruction::FPToUI:
+      case Instruction::FPToSI:
+      case Instruction::FPExt:
+      case Instruction::PtrToInt:
+      case Instruction::IntToPtr:
+      case Instruction::SIToFP:
+      case Instruction::UIToFP:
+      case Instruction::Trunc:
+      case Instruction::FPTrunc:
+      case Instruction::BitCast:
+      case Instruction::ShuffleVector:
+        return VTTI->getCastInstrCost(Opcode, T1, T2);
+      }
+
+      return 1;
+    }
+
     // This determines the relative offset of two loads or stores, returning
     // true if the offset could be determined to be some constant value.
     // For example, if OffsetInElmts == 1, then J accesses the memory directly
@@ -453,20 +583,30 @@
     // directly after J.
     bool getPairPtrInfo(Instruction *I, Instruction *J,
         Value *&IPtr, Value *&JPtr, unsigned &IAlignment, unsigned &JAlignment,
-        int64_t &OffsetInElmts) {
+        unsigned &IAddressSpace, unsigned &JAddressSpace,
+        int64_t &OffsetInElmts, bool ComputeOffset = true) {
       OffsetInElmts = 0;
-      if (isa<LoadInst>(I)) {
-        IPtr = cast<LoadInst>(I)->getPointerOperand();
-        JPtr = cast<LoadInst>(J)->getPointerOperand();
-        IAlignment = cast<LoadInst>(I)->getAlignment();
-        JAlignment = cast<LoadInst>(J)->getAlignment();
+      if (LoadInst *LI = dyn_cast<LoadInst>(I)) {
+        LoadInst *LJ = cast<LoadInst>(J);
+        IPtr = LI->getPointerOperand();
+        JPtr = LJ->getPointerOperand();
+        IAlignment = LI->getAlignment();
+        JAlignment = LJ->getAlignment();
+        IAddressSpace = LI->getPointerAddressSpace();
+        JAddressSpace = LJ->getPointerAddressSpace();
       } else {
-        IPtr = cast<StoreInst>(I)->getPointerOperand();
-        JPtr = cast<StoreInst>(J)->getPointerOperand();
-        IAlignment = cast<StoreInst>(I)->getAlignment();
-        JAlignment = cast<StoreInst>(J)->getAlignment();
+        StoreInst *SI = cast<StoreInst>(I), *SJ = cast<StoreInst>(J);
+        IPtr = SI->getPointerOperand();
+        JPtr = SJ->getPointerOperand();
+        IAlignment = SI->getAlignment();
+        JAlignment = SJ->getAlignment();
+        IAddressSpace = SI->getPointerAddressSpace();
+        JAddressSpace = SJ->getPointerAddressSpace();
       }
 
+      if (!ComputeOffset)
+        return true;
+
       const SCEV *IPtrSCEV = SE->getSCEV(IPtr);
       const SCEV *JPtrSCEV = SE->getSCEV(JPtr);
 
@@ -536,6 +676,19 @@
 
       return false;
     }
+
+    bool isPureIEChain(InsertElementInst *IE) {
+      InsertElementInst *IENext = IE;
+      do {
+        if (!isa<UndefValue>(IENext->getOperand(0)) &&
+            !isa<InsertElementInst>(IENext->getOperand(0))) {
+          return false;
+        }
+      } while ((IENext =
+                 dyn_cast<InsertElementInst>(IENext->getOperand(0))));
+
+      return true;
+    }
   };
 
   // This function implements one vectorization iteration on the provided
@@ -546,11 +699,18 @@
 
     std::vector<Value *> AllPairableInsts;
     DenseMap<Value *, Value *> AllChosenPairs;
+    DenseSet<ValuePair> AllFixedOrderPairs;
+    DenseMap<VPPair, unsigned> AllPairConnectionTypes;
+    std::multimap<ValuePair, ValuePair> AllConnectedPairs, AllConnectedPairDeps;
 
     do {
       std::vector<Value *> PairableInsts;
       std::multimap<Value *, Value *> CandidatePairs;
+      DenseSet<ValuePair> FixedOrderPairs;
+      DenseMap<ValuePair, int> CandidatePairCostSavings;
       ShouldContinue = getCandidatePairs(BB, Start, CandidatePairs,
+                                         FixedOrderPairs,
+                                         CandidatePairCostSavings,
                                          PairableInsts, NonPow2Len);
       if (PairableInsts.empty()) continue;
 
@@ -563,10 +723,18 @@
       // Note that it only matters that both members of the second pair use some
       // element of the first pair (to allow for splatting).
 
-      std::multimap<ValuePair, ValuePair> ConnectedPairs;
-      computeConnectedPairs(CandidatePairs, PairableInsts, ConnectedPairs);
+      std::multimap<ValuePair, ValuePair> ConnectedPairs, ConnectedPairDeps;
+      DenseMap<VPPair, unsigned> PairConnectionTypes;
+      computeConnectedPairs(CandidatePairs, PairableInsts, ConnectedPairs,
+                            PairConnectionTypes);
       if (ConnectedPairs.empty()) continue;
 
+      for (std::multimap<ValuePair, ValuePair>::iterator
+           I = ConnectedPairs.begin(), IE = ConnectedPairs.end();
+           I != IE; ++I) {
+        ConnectedPairDeps.insert(VPPair(I->second, I->first));
+      }
+
       // Build the pairable-instruction dependency map
       DenseSet<ValuePair> PairableInstUsers;
       buildDepMap(BB, CandidatePairs, PairableInsts, PairableInstUsers);
@@ -578,13 +746,48 @@
       // variables.
 
       DenseMap<Value *, Value *> ChosenPairs;
-      choosePairs(CandidatePairs, PairableInsts, ConnectedPairs,
+      choosePairs(CandidatePairs, CandidatePairCostSavings,
+        PairableInsts, FixedOrderPairs, PairConnectionTypes,
+        ConnectedPairs, ConnectedPairDeps,
         PairableInstUsers, ChosenPairs);
 
       if (ChosenPairs.empty()) continue;
       AllPairableInsts.insert(AllPairableInsts.end(), PairableInsts.begin(),
                               PairableInsts.end());
       AllChosenPairs.insert(ChosenPairs.begin(), ChosenPairs.end());
+
+      // Only for the chosen pairs, propagate information on fixed-order pairs,
+      // pair connections, and their types to the data structures used by the
+      // pair fusion procedures.
+      for (DenseMap<Value *, Value *>::iterator I = ChosenPairs.begin(),
+           IE = ChosenPairs.end(); I != IE; ++I) {
+        if (FixedOrderPairs.count(*I))
+          AllFixedOrderPairs.insert(*I);
+        else if (FixedOrderPairs.count(ValuePair(I->second, I->first)))
+          AllFixedOrderPairs.insert(ValuePair(I->second, I->first));
+
+        for (DenseMap<Value *, Value *>::iterator J = ChosenPairs.begin();
+             J != IE; ++J) {
+          DenseMap<VPPair, unsigned>::iterator K =
+            PairConnectionTypes.find(VPPair(*I, *J));
+          if (K != PairConnectionTypes.end()) {
+            AllPairConnectionTypes.insert(*K);
+          } else {
+            K = PairConnectionTypes.find(VPPair(*J, *I));
+            if (K != PairConnectionTypes.end())
+              AllPairConnectionTypes.insert(*K);
+          }
+        }
+      }
+
+      for (std::multimap<ValuePair, ValuePair>::iterator
+           I = ConnectedPairs.begin(), IE = ConnectedPairs.end();
+           I != IE; ++I) {
+        if (AllPairConnectionTypes.count(*I)) {
+          AllConnectedPairs.insert(*I);
+          AllConnectedPairDeps.insert(VPPair(I->second, I->first));
+        }
+      }
     } while (ShouldContinue);
 
     if (AllChosenPairs.empty()) return false;
@@ -597,11 +800,13 @@
     // replaced with a vector_extract on the result.  Subsequent optimization
     // passes should coalesce the build/extract combinations.
 
-    fuseChosenPairs(BB, AllPairableInsts, AllChosenPairs);
+    fuseChosenPairs(BB, AllPairableInsts, AllChosenPairs, AllFixedOrderPairs,
+                    AllPairConnectionTypes,
+                    AllConnectedPairs, AllConnectedPairDeps);
 
     // It is important to cleanup here so that future iterations of this
     // function have less work to do.
-    (void) SimplifyInstructionsInBlock(&BB, TD);
+    (void) SimplifyInstructionsInBlock(&BB, TD, AA->getTargetLibraryInfo());
     return true;
   }
 
@@ -667,15 +872,22 @@
         !(VectorType::isValidElementType(T2) || T2->isVectorTy()))
       return false;
 
-    if (T1->getScalarSizeInBits() == 1 && T2->getScalarSizeInBits() == 1) {
+    if (T1->getScalarSizeInBits() == 1) {
       if (!Config.VectorizeBools)
         return false;
     } else {
-      if (!Config.VectorizeInts
-          && (T1->isIntOrIntVectorTy() || T2->isIntOrIntVectorTy()))
+      if (!Config.VectorizeInts && T1->isIntOrIntVectorTy())
         return false;
     }
-  
+
+    if (T2->getScalarSizeInBits() == 1) {
+      if (!Config.VectorizeBools)
+        return false;
+    } else {
+      if (!Config.VectorizeInts && T2->isIntOrIntVectorTy())
+        return false;
+    }
+
     if (!Config.VectorizeFloats
         && (T1->isFPOrFPVectorTy() || T2->isFPOrFPVectorTy()))
       return false;
@@ -691,8 +903,8 @@
          T2->getScalarType()->isPointerTy()))
       return false;
 
-    if (T1->getPrimitiveSizeInBits() >= Config.VectorBits ||
-        T2->getPrimitiveSizeInBits() >= Config.VectorBits)
+    if (!VTTI && (T1->getPrimitiveSizeInBits() >= Config.VectorBits ||
+                  T2->getPrimitiveSizeInBits() >= Config.VectorBits))
       return false;
 
     return true;
@@ -703,10 +915,14 @@
   // that I has already been determined to be vectorizable and that J is not
   // in the use tree of I.
   bool BBVectorize::areInstsCompatible(Instruction *I, Instruction *J,
-                       bool IsSimpleLoadStore, bool NonPow2Len) {
+                       bool IsSimpleLoadStore, bool NonPow2Len,
+                       int &CostSavings, int &FixedOrder) {
     DEBUG(if (DebugInstructionExamination) dbgs() << "BBV: looking at " << *I <<
                      " <-> " << *J << "\n");
 
+    CostSavings = 0;
+    FixedOrder = 0;
+
     // Loads and stores can be merged if they have different alignments,
     // but are otherwise the same.
     if (!J->isSameOperationAs(I, Instruction::CompareIgnoringAlignment |
@@ -719,38 +935,84 @@
     unsigned MaxTypeBits = std::max(
       IT1->getPrimitiveSizeInBits() + JT1->getPrimitiveSizeInBits(),
       IT2->getPrimitiveSizeInBits() + JT2->getPrimitiveSizeInBits());
-    if (MaxTypeBits > Config.VectorBits)
+    if (!VTTI && MaxTypeBits > Config.VectorBits)
       return false;
 
     // FIXME: handle addsub-type operations!
 
     if (IsSimpleLoadStore) {
       Value *IPtr, *JPtr;
-      unsigned IAlignment, JAlignment;
+      unsigned IAlignment, JAlignment, IAddressSpace, JAddressSpace;
       int64_t OffsetInElmts = 0;
       if (getPairPtrInfo(I, J, IPtr, JPtr, IAlignment, JAlignment,
+            IAddressSpace, JAddressSpace,
             OffsetInElmts) && abs64(OffsetInElmts) == 1) {
-        if (Config.AlignedOnly) {
-          Type *aTypeI = isa<StoreInst>(I) ?
-            cast<StoreInst>(I)->getValueOperand()->getType() : I->getType();
-          Type *aTypeJ = isa<StoreInst>(J) ?
-            cast<StoreInst>(J)->getValueOperand()->getType() : J->getType();
+        FixedOrder = (int) OffsetInElmts;
+        unsigned BottomAlignment = IAlignment;
+        if (OffsetInElmts < 0) BottomAlignment = JAlignment;
+
+        Type *aTypeI = isa<StoreInst>(I) ?
+          cast<StoreInst>(I)->getValueOperand()->getType() : I->getType();
+        Type *aTypeJ = isa<StoreInst>(J) ?
+          cast<StoreInst>(J)->getValueOperand()->getType() : J->getType();
+        Type *VType = getVecTypeForPair(aTypeI, aTypeJ);
 
+        if (Config.AlignedOnly) {
           // An aligned load or store is possible only if the instruction
           // with the lower offset has an alignment suitable for the
           // vector type.
 
-          unsigned BottomAlignment = IAlignment;
-          if (OffsetInElmts < 0) BottomAlignment = JAlignment;
-
-          Type *VType = getVecTypeForPair(aTypeI, aTypeJ);
           unsigned VecAlignment = TD->getPrefTypeAlignment(VType);
           if (BottomAlignment < VecAlignment)
             return false;
         }
+
+        if (VTTI) {
+          unsigned ICost = VTTI->getMemoryOpCost(I->getOpcode(), I->getType(),
+                                                 IAlignment, IAddressSpace);
+          unsigned JCost = VTTI->getMemoryOpCost(J->getOpcode(), J->getType(),
+                                                 JAlignment, JAddressSpace);
+          unsigned VCost = VTTI->getMemoryOpCost(I->getOpcode(), VType,
+                                                 BottomAlignment,
+                                                 IAddressSpace);
+          if (VCost > ICost + JCost)
+            return false;
+
+          // We don't want to fuse to a type that will be split, even
+          // if the two input types will also be split and there is no other
+          // associated cost.
+          unsigned VParts = VTTI->getNumberOfParts(VType);
+          if (VParts > 1)
+            return false;
+          else if (!VParts && VCost == ICost + JCost)
+            return false;
+
+          CostSavings = ICost + JCost - VCost;
+        }
       } else {
         return false;
       }
+    } else if (VTTI) {
+      unsigned ICost = getInstrCost(I->getOpcode(), IT1, IT2);
+      unsigned JCost = getInstrCost(J->getOpcode(), JT1, JT2);
+      Type *VT1 = getVecTypeForPair(IT1, JT1),
+           *VT2 = getVecTypeForPair(IT2, JT2);
+      unsigned VCost = getInstrCost(I->getOpcode(), VT1, VT2);
+
+      if (VCost > ICost + JCost)
+        return false;
+
+      // We don't want to fuse to a type that will be split, even
+      // if the two input types will also be split and there is no other
+      // associated cost.
+      unsigned VParts1 = VTTI->getNumberOfParts(VT1),
+               VParts2 = VTTI->getNumberOfParts(VT2);
+      if (VParts1 > 1 || VParts2 > 1)
+        return false;
+      else if ((!VParts1 || !VParts2) && VCost == ICost + JCost)
+        return false;
+
+      CostSavings = ICost + JCost - VCost;
     }
 
     // The powi intrinsic is special because only the first argument is
@@ -833,6 +1095,8 @@
   bool BBVectorize::getCandidatePairs(BasicBlock &BB,
                        BasicBlock::iterator &Start,
                        std::multimap<Value *, Value *> &CandidatePairs,
+                       DenseSet<ValuePair> &FixedOrderPairs,
+                       DenseMap<ValuePair, int> &CandidatePairCostSavings,
                        std::vector<Value *> &PairableInsts, bool NonPow2Len) {
     BasicBlock::iterator E = BB.end();
     if (Start == E) return false;
@@ -869,7 +1133,9 @@
 
         // J does not use I, and comes before the first use of I, so it can be
         // merged with I if the instructions are compatible.
-        if (!areInstsCompatible(I, J, IsSimpleLoadStore, NonPow2Len)) continue;
+        int CostSavings, FixedOrder;
+        if (!areInstsCompatible(I, J, IsSimpleLoadStore, NonPow2Len,
+            CostSavings, FixedOrder)) continue;
 
         // J is a candidate for merging with I.
         if (!PairableInsts.size() ||
@@ -878,6 +1144,14 @@
         }
 
         CandidatePairs.insert(ValuePair(I, J));
+        if (VTTI)
+          CandidatePairCostSavings.insert(ValuePairWithCost(ValuePair(I, J),
+                                                            CostSavings));
+
+        if (FixedOrder == 1)
+          FixedOrderPairs.insert(ValuePair(I, J));
+        else if (FixedOrder == -1)
+          FixedOrderPairs.insert(ValuePair(J, I));
 
         // The next call to this function must start after the last instruction
         // selected during this invocation.
@@ -887,7 +1161,8 @@
         }
 
         DEBUG(if (DebugCandidateSelection) dbgs() << "BBV: candidate pair "
-                     << *I << " <-> " << *J << "\n");
+                     << *I << " <-> " << *J << " (cost savings: " <<
+                     CostSavings << ")\n");
 
         // If we have already found too many pairs, break here and this function
         // will be called again starting after the last instruction selected
@@ -915,6 +1190,7 @@
                       std::multimap<Value *, Value *> &CandidatePairs,
                       std::vector<Value *> &PairableInsts,
                       std::multimap<ValuePair, ValuePair> &ConnectedPairs,
+                      DenseMap<VPPair, unsigned> &PairConnectionTypes,
                       ValuePair P) {
     StoreInst *SI, *SJ;
 
@@ -946,12 +1222,18 @@
         VPIteratorPair JPairRange = CandidatePairs.equal_range(*J);
 
         // Look for <I, J>:
-        if (isSecondInIteratorPair<Value*>(*J, IPairRange))
-          ConnectedPairs.insert(VPPair(P, ValuePair(*I, *J)));
+        if (isSecondInIteratorPair<Value*>(*J, IPairRange)) {
+          VPPair VP(P, ValuePair(*I, *J));
+          ConnectedPairs.insert(VP);
+          PairConnectionTypes.insert(VPPairWithType(VP, PairConnectionDirect));
+        }
 
         // Look for <J, I>:
-        if (isSecondInIteratorPair<Value*>(*I, JPairRange))
-          ConnectedPairs.insert(VPPair(P, ValuePair(*J, *I)));
+        if (isSecondInIteratorPair<Value*>(*I, JPairRange)) {
+          VPPair VP(P, ValuePair(*J, *I));
+          ConnectedPairs.insert(VP);
+          PairConnectionTypes.insert(VPPairWithType(VP, PairConnectionSwap));
+        }
       }
 
       if (Config.SplatBreaksChain) continue;
@@ -962,8 +1244,11 @@
             P.first == SJ->getPointerOperand())
           continue;
 
-        if (isSecondInIteratorPair<Value*>(*J, IPairRange))
-          ConnectedPairs.insert(VPPair(P, ValuePair(*I, *J)));
+        if (isSecondInIteratorPair<Value*>(*J, IPairRange)) {
+          VPPair VP(P, ValuePair(*I, *J));
+          ConnectedPairs.insert(VP);
+          PairConnectionTypes.insert(VPPairWithType(VP, PairConnectionSplat));
+        }
       }
     }
 
@@ -985,8 +1270,11 @@
             P.second == SJ->getPointerOperand())
           continue;
 
-        if (isSecondInIteratorPair<Value*>(*J, IPairRange))
-          ConnectedPairs.insert(VPPair(P, ValuePair(*I, *J)));
+        if (isSecondInIteratorPair<Value*>(*J, IPairRange)) {
+          VPPair VP(P, ValuePair(*I, *J));
+          ConnectedPairs.insert(VP);
+          PairConnectionTypes.insert(VPPairWithType(VP, PairConnectionSplat));
+        }
       }
     }
   }
@@ -997,7 +1285,8 @@
   void BBVectorize::computeConnectedPairs(
                       std::multimap<Value *, Value *> &CandidatePairs,
                       std::vector<Value *> &PairableInsts,
-                      std::multimap<ValuePair, ValuePair> &ConnectedPairs) {
+                      std::multimap<ValuePair, ValuePair> &ConnectedPairs,
+                      DenseMap<VPPair, unsigned> &PairConnectionTypes) {
 
     for (std::vector<Value *>::iterator PI = PairableInsts.begin(),
          PE = PairableInsts.end(); PI != PE; ++PI) {
@@ -1006,7 +1295,7 @@
       for (std::multimap<Value *, Value *>::iterator P = choiceRange.first;
            P != choiceRange.second; ++P)
         computePairsConnectedTo(CandidatePairs, PairableInsts,
-                                ConnectedPairs, *P);
+                                ConnectedPairs, PairConnectionTypes, *P);
     }
 
     DEBUG(dbgs() << "BBV: found " << ConnectedPairs.size()
@@ -1196,7 +1485,7 @@
       PrunedTree.insert(QTop.first);
 
       // Visit each child, pruning as necessary...
-      DenseMap<ValuePair, size_t> BestChildren;
+      SmallVector<ValuePairWithDepth, 8> BestChildren;
       VPPIteratorPair QTopRange = ConnectedPairs.equal_range(QTop.first);
       for (std::multimap<ValuePair, ValuePair>::iterator K = QTopRange.first;
            K != QTopRange.second; ++K) {
@@ -1228,7 +1517,7 @@
         DenseSet<ValuePair> CurrentPairs;
 
         bool CanAdd = true;
-        for (DenseMap<ValuePair, size_t>::iterator C2
+        for (SmallVector<ValuePairWithDepth, 8>::iterator C2
               = BestChildren.begin(), E2 = BestChildren.end();
              C2 != E2; ++C2) {
           if (C2->first.first == C->first.first ||
@@ -1313,22 +1602,22 @@
         // to an already-selected child. Check for this here, and if a
         // conflict is found, then remove the previously-selected child
         // before adding this one in its place.
-        for (DenseMap<ValuePair, size_t>::iterator C2
+        for (SmallVector<ValuePairWithDepth, 8>::iterator C2
               = BestChildren.begin(); C2 != BestChildren.end();) {
           if (C2->first.first == C->first.first ||
               C2->first.first == C->first.second ||
               C2->first.second == C->first.first ||
               C2->first.second == C->first.second ||
               pairsConflict(C2->first, C->first, PairableInstUsers))
-            BestChildren.erase(C2++);
+            C2 = BestChildren.erase(C2);
           else
             ++C2;
         }
 
-        BestChildren.insert(ValuePairWithDepth(C->first, C->second));
+        BestChildren.push_back(ValuePairWithDepth(C->first, C->second));
       }
 
-      for (DenseMap<ValuePair, size_t>::iterator C
+      for (SmallVector<ValuePairWithDepth, 8>::iterator C
             = BestChildren.begin(), E2 = BestChildren.end();
            C != E2; ++C) {
         size_t DepthF = getDepthFactor(C->first.first);
@@ -1341,13 +1630,17 @@
   // pairs, given the choice of root pairs as an iterator range.
   void BBVectorize::findBestTreeFor(
                       std::multimap<Value *, Value *> &CandidatePairs,
+                      DenseMap<ValuePair, int> &CandidatePairCostSavings,
                       std::vector<Value *> &PairableInsts,
+                      DenseSet<ValuePair> &FixedOrderPairs,
+                      DenseMap<VPPair, unsigned> &PairConnectionTypes,
                       std::multimap<ValuePair, ValuePair> &ConnectedPairs,
+                      std::multimap<ValuePair, ValuePair> &ConnectedPairDeps,
                       DenseSet<ValuePair> &PairableInstUsers,
                       std::multimap<ValuePair, ValuePair> &PairableInstUserMap,
                       DenseMap<Value *, Value *> &ChosenPairs,
                       DenseSet<ValuePair> &BestTree, size_t &BestMaxDepth,
-                      size_t &BestEffSize, VPIteratorPair ChoiceRange,
+                      int &BestEffSize, VPIteratorPair ChoiceRange,
                       bool UseCycleCheck) {
     for (std::multimap<Value *, Value *>::iterator J = ChoiceRange.first;
          J != ChoiceRange.second; ++J) {
@@ -1397,17 +1690,289 @@
                    PairableInstUsers, PairableInstUserMap, ChosenPairs, Tree,
                    PrunedTree, *J, UseCycleCheck);
 
-      size_t EffSize = 0;
-      for (DenseSet<ValuePair>::iterator S = PrunedTree.begin(),
-           E = PrunedTree.end(); S != E; ++S)
-        EffSize += getDepthFactor(S->first);
+      int EffSize = 0;
+      if (VTTI) {
+        DenseSet<Value *> PrunedTreeInstrs;
+        for (DenseSet<ValuePair>::iterator S = PrunedTree.begin(),
+             E = PrunedTree.end(); S != E; ++S) {
+          PrunedTreeInstrs.insert(S->first);
+          PrunedTreeInstrs.insert(S->second);
+        }
+
+        // The set of pairs that have already contributed to the total cost.
+        DenseSet<ValuePair> IncomingPairs;
+
+        // If the cost model were perfect, this might not be necessary; but we
+        // need to make sure that we don't get stuck vectorizing our own
+        // shuffle chains.
+        bool HasNontrivialInsts = false;
+
+        // The node weights represent the cost savings associated with
+        // fusing the pair of instructions.
+        for (DenseSet<ValuePair>::iterator S = PrunedTree.begin(),
+             E = PrunedTree.end(); S != E; ++S) {
+          if (!isa<ShuffleVectorInst>(S->first) &&
+              !isa<InsertElementInst>(S->first) &&
+              !isa<ExtractElementInst>(S->first))
+            HasNontrivialInsts = true;
+
+          bool FlipOrder = false;
+
+          if (getDepthFactor(S->first)) {
+            int ESContrib = CandidatePairCostSavings.find(*S)->second;
+            DEBUG(if (DebugPairSelection) dbgs() << "\tweight {"
+                   << *S->first << " <-> " << *S->second << "} = " <<
+                   ESContrib << "\n");
+            EffSize += ESContrib;
+          }
+
+          // The edge weights contribute in a negative sense: they represent
+          // the cost of shuffles.
+          VPPIteratorPair IP = ConnectedPairDeps.equal_range(*S);
+          if (IP.first != ConnectedPairDeps.end()) {
+            unsigned NumDepsDirect = 0, NumDepsSwap = 0;
+            for (std::multimap<ValuePair, ValuePair>::iterator Q = IP.first;
+                 Q != IP.second; ++Q) {
+              if (!PrunedTree.count(Q->second))
+                continue;
+              DenseMap<VPPair, unsigned>::iterator R =
+                PairConnectionTypes.find(VPPair(Q->second, Q->first));
+              assert(R != PairConnectionTypes.end() &&
+                     "Cannot find pair connection type");
+              if (R->second == PairConnectionDirect)
+                ++NumDepsDirect;
+              else if (R->second == PairConnectionSwap)
+                ++NumDepsSwap;
+            }
+
+            // If there are more swaps than direct connections, then
+            // the pair order will be flipped during fusion. So the real
+            // number of swaps is the minimum number.
+            FlipOrder = !FixedOrderPairs.count(*S) &&
+              ((NumDepsSwap > NumDepsDirect) ||
+                FixedOrderPairs.count(ValuePair(S->second, S->first)));
+
+            for (std::multimap<ValuePair, ValuePair>::iterator Q = IP.first;
+                 Q != IP.second; ++Q) {
+              if (!PrunedTree.count(Q->second))
+                continue;
+              DenseMap<VPPair, unsigned>::iterator R =
+                PairConnectionTypes.find(VPPair(Q->second, Q->first));
+              assert(R != PairConnectionTypes.end() &&
+                     "Cannot find pair connection type");
+              Type *Ty1 = Q->second.first->getType(),
+                   *Ty2 = Q->second.second->getType();
+              Type *VTy = getVecTypeForPair(Ty1, Ty2);
+              if ((R->second == PairConnectionDirect && FlipOrder) ||
+                  (R->second == PairConnectionSwap && !FlipOrder)  ||
+                  R->second == PairConnectionSplat) {
+                int ESContrib = (int) getInstrCost(Instruction::ShuffleVector,
+                                                   VTy, VTy);
+                DEBUG(if (DebugPairSelection) dbgs() << "\tcost {" <<
+                  *Q->second.first << " <-> " << *Q->second.second <<
+                    "} -> {" <<
+                  *S->first << " <-> " << *S->second << "} = " <<
+                   ESContrib << "\n");
+                EffSize -= ESContrib;
+              }
+            }
+          }
+
+          // Compute the cost of outgoing edges. We assume that edges outgoing
+          // to shuffles, inserts or extracts can be merged, and so contribute
+          // no additional cost.
+          if (!S->first->getType()->isVoidTy()) {
+            Type *Ty1 = S->first->getType(),
+                 *Ty2 = S->second->getType();
+            Type *VTy = getVecTypeForPair(Ty1, Ty2);
+
+            bool NeedsExtraction = false;
+            for (Value::use_iterator I = S->first->use_begin(),
+                 IE = S->first->use_end(); I != IE; ++I) {
+              if (ShuffleVectorInst *SI = dyn_cast<ShuffleVectorInst>(*I)) {
+                // Shuffle can be folded if it has no other input
+                if (isa<UndefValue>(SI->getOperand(1)))
+                  continue;
+              }
+              if (isa<ExtractElementInst>(*I))
+                continue;
+              if (PrunedTreeInstrs.count(*I))
+                continue;
+              NeedsExtraction = true;
+              break;
+            }
+
+            if (NeedsExtraction) {
+              int ESContrib;
+              if (Ty1->isVectorTy())
+                ESContrib = (int) getInstrCost(Instruction::ShuffleVector,
+                                               Ty1, VTy);
+              else
+                ESContrib = (int) VTTI->getVectorInstrCost(
+                                    Instruction::ExtractElement, VTy, 0);
+
+              DEBUG(if (DebugPairSelection) dbgs() << "\tcost {" <<
+                *S->first << "} = " << ESContrib << "\n");
+              EffSize -= ESContrib;
+            }
+
+            NeedsExtraction = false;
+            for (Value::use_iterator I = S->second->use_begin(),
+                 IE = S->second->use_end(); I != IE; ++I) {
+              if (ShuffleVectorInst *SI = dyn_cast<ShuffleVectorInst>(*I)) {
+                // Shuffle can be folded if it has no other input
+                if (isa<UndefValue>(SI->getOperand(1)))
+                  continue;
+              }
+              if (isa<ExtractElementInst>(*I))
+                continue;
+              if (PrunedTreeInstrs.count(*I))
+                continue;
+              NeedsExtraction = true;
+              break;
+            }
+
+            if (NeedsExtraction) {
+              int ESContrib;
+              if (Ty2->isVectorTy())
+                ESContrib = (int) getInstrCost(Instruction::ShuffleVector,
+                                               Ty2, VTy);
+              else
+                ESContrib = (int) VTTI->getVectorInstrCost(
+                                    Instruction::ExtractElement, VTy, 1);
+              DEBUG(if (DebugPairSelection) dbgs() << "\tcost {" <<
+                *S->second << "} = " << ESContrib << "\n");
+              EffSize -= ESContrib;
+            }
+          }
+
+          // Compute the cost of incoming edges.
+          if (!isa<LoadInst>(S->first) && !isa<StoreInst>(S->first)) {
+            Instruction *S1 = cast<Instruction>(S->first),
+                        *S2 = cast<Instruction>(S->second);
+            for (unsigned o = 0; o < S1->getNumOperands(); ++o) {
+              Value *O1 = S1->getOperand(o), *O2 = S2->getOperand(o);
+
+              // Combining constants into vector constants (or small vector
+              // constants into larger ones are assumed free).
+              if (isa<Constant>(O1) && isa<Constant>(O2))
+                continue;
+
+              if (FlipOrder)
+                std::swap(O1, O2);
+
+              ValuePair VP  = ValuePair(O1, O2);
+              ValuePair VPR = ValuePair(O2, O1);
+
+              // Internal edges are not handled here.
+              if (PrunedTree.count(VP) || PrunedTree.count(VPR))
+                continue;
+
+              Type *Ty1 = O1->getType(),
+                   *Ty2 = O2->getType();
+              Type *VTy = getVecTypeForPair(Ty1, Ty2);
+
+              // Combining vector operations of the same type is also assumed
+              // folded with other operations.
+              if (Ty1 == Ty2) {
+                // If both are insert elements, then both can be widened.
+                InsertElementInst *IEO1 = dyn_cast<InsertElementInst>(O1),
+                                  *IEO2 = dyn_cast<InsertElementInst>(O2);
+                if (IEO1 && IEO2 && isPureIEChain(IEO1) && isPureIEChain(IEO2))
+                  continue;
+                // If both are extract elements, and both have the same input
+                // type, then they can be replaced with a shuffle
+                ExtractElementInst *EIO1 = dyn_cast<ExtractElementInst>(O1),
+                                   *EIO2 = dyn_cast<ExtractElementInst>(O2);
+                if (EIO1 && EIO2 &&
+                    EIO1->getOperand(0)->getType() ==
+                      EIO2->getOperand(0)->getType())
+                  continue;
+                // If both are a shuffle with equal operand types and only two
+                // unqiue operands, then they can be replaced with a single
+                // shuffle
+                ShuffleVectorInst *SIO1 = dyn_cast<ShuffleVectorInst>(O1),
+                                  *SIO2 = dyn_cast<ShuffleVectorInst>(O2);
+                if (SIO1 && SIO2 &&
+                    SIO1->getOperand(0)->getType() ==
+                      SIO2->getOperand(0)->getType()) {
+                  SmallSet<Value *, 4> SIOps;
+                  SIOps.insert(SIO1->getOperand(0));
+                  SIOps.insert(SIO1->getOperand(1));
+                  SIOps.insert(SIO2->getOperand(0));
+                  SIOps.insert(SIO2->getOperand(1));
+                  if (SIOps.size() <= 2)
+                    continue;
+                }
+              }
+
+              int ESContrib;
+              // This pair has already been formed.
+              if (IncomingPairs.count(VP)) {
+                continue;
+              } else if (IncomingPairs.count(VPR)) {
+                ESContrib = (int) getInstrCost(Instruction::ShuffleVector,
+                                               VTy, VTy);
+              } else if (!Ty1->isVectorTy() && !Ty2->isVectorTy()) {
+                ESContrib = (int) VTTI->getVectorInstrCost(
+                                    Instruction::InsertElement, VTy, 0);
+                ESContrib += (int) VTTI->getVectorInstrCost(
+                                     Instruction::InsertElement, VTy, 1);
+              } else if (!Ty1->isVectorTy()) {
+                // O1 needs to be inserted into a vector of size O2, and then
+                // both need to be shuffled together.
+                ESContrib = (int) VTTI->getVectorInstrCost(
+                                    Instruction::InsertElement, Ty2, 0);
+                ESContrib += (int) getInstrCost(Instruction::ShuffleVector,
+                                                VTy, Ty2);
+              } else if (!Ty2->isVectorTy()) {
+                // O2 needs to be inserted into a vector of size O1, and then
+                // both need to be shuffled together.
+                ESContrib = (int) VTTI->getVectorInstrCost(
+                                    Instruction::InsertElement, Ty1, 0);
+                ESContrib += (int) getInstrCost(Instruction::ShuffleVector,
+                                                VTy, Ty1);
+              } else {
+                Type *TyBig = Ty1, *TySmall = Ty2;
+                if (Ty2->getVectorNumElements() > Ty1->getVectorNumElements())
+                  std::swap(TyBig, TySmall);
+
+                ESContrib = (int) getInstrCost(Instruction::ShuffleVector,
+                                               VTy, TyBig);
+                if (TyBig != TySmall)
+                  ESContrib += (int) getInstrCost(Instruction::ShuffleVector,
+                                                  TyBig, TySmall);
+              }
+
+              DEBUG(if (DebugPairSelection) dbgs() << "\tcost {"
+                     << *O1 << " <-> " << *O2 << "} = " <<
+                     ESContrib << "\n");
+              EffSize -= ESContrib;
+              IncomingPairs.insert(VP);
+            }
+          }
+        }
+
+        if (!HasNontrivialInsts) {
+          DEBUG(if (DebugPairSelection) dbgs() <<
+                "\tNo non-trivial instructions in tree;"
+                " override to zero effective size\n");
+          EffSize = 0;
+        }
+      } else {
+        for (DenseSet<ValuePair>::iterator S = PrunedTree.begin(),
+             E = PrunedTree.end(); S != E; ++S)
+          EffSize += (int) getDepthFactor(S->first);
+      }
 
       DEBUG(if (DebugPairSelection)
              dbgs() << "BBV: found pruned Tree for pair {"
              << *J->first << " <-> " << *J->second << "} of depth " <<
              MaxDepth << " and size " << PrunedTree.size() <<
             " (effective size: " << EffSize << ")\n");
-      if (MaxDepth >= Config.ReqChainDepth && EffSize > BestEffSize) {
+      if (((VTTI && !UseChainDepthWithTI) ||
+            MaxDepth >= Config.ReqChainDepth) &&
+          EffSize > 0 && EffSize > BestEffSize) {
         BestMaxDepth = MaxDepth;
         BestEffSize = EffSize;
         BestTree = PrunedTree;
@@ -1419,8 +1984,12 @@
   // that will be fused into vector instructions.
   void BBVectorize::choosePairs(
                       std::multimap<Value *, Value *> &CandidatePairs,
+                      DenseMap<ValuePair, int> &CandidatePairCostSavings,
                       std::vector<Value *> &PairableInsts,
+                      DenseSet<ValuePair> &FixedOrderPairs,
+                      DenseMap<VPPair, unsigned> &PairConnectionTypes,
                       std::multimap<ValuePair, ValuePair> &ConnectedPairs,
+                      std::multimap<ValuePair, ValuePair> &ConnectedPairDeps,
                       DenseSet<ValuePair> &PairableInstUsers,
                       DenseMap<Value *, Value *>& ChosenPairs) {
     bool UseCycleCheck =
@@ -1435,9 +2004,12 @@
       VPIteratorPair ChoiceRange = CandidatePairs.equal_range(*I);
 
       // The best pair to choose and its tree:
-      size_t BestMaxDepth = 0, BestEffSize = 0;
+      size_t BestMaxDepth = 0;
+      int BestEffSize = 0;
       DenseSet<ValuePair> BestTree;
-      findBestTreeFor(CandidatePairs, PairableInsts, ConnectedPairs,
+      findBestTreeFor(CandidatePairs, CandidatePairCostSavings,
+                      PairableInsts, FixedOrderPairs, PairConnectionTypes,
+                      ConnectedPairs, ConnectedPairDeps,
                       PairableInstUsers, PairableInstUserMap, ChosenPairs,
                       BestTree, BestMaxDepth, BestEffSize, ChoiceRange,
                       UseCycleCheck);
@@ -1490,24 +2062,19 @@
   // Returns the value that is to be used as the pointer input to the vector
   // instruction that fuses I with J.
   Value *BBVectorize::getReplacementPointerInput(LLVMContext& Context,
-                     Instruction *I, Instruction *J, unsigned o,
-                     bool FlipMemInputs) {
+                     Instruction *I, Instruction *J, unsigned o) {
     Value *IPtr, *JPtr;
-    unsigned IAlignment, JAlignment;
+    unsigned IAlignment, JAlignment, IAddressSpace, JAddressSpace;
     int64_t OffsetInElmts;
 
-    // Note: the analysis might fail here, that is why FlipMemInputs has
+    // Note: the analysis might fail here, that is why the pair order has
     // been precomputed (OffsetInElmts must be unused here).
     (void) getPairPtrInfo(I, J, IPtr, JPtr, IAlignment, JAlignment,
-                          OffsetInElmts);
+                          IAddressSpace, JAddressSpace,
+                          OffsetInElmts, false);
 
     // The pointer value is taken to be the one with the lowest offset.
-    Value *VPtr;
-    if (!FlipMemInputs) {
-      VPtr = IPtr;
-    } else {
-      VPtr = JPtr;
-    }
+    Value *VPtr = IPtr;
 
     Type *ArgTypeI = cast<PointerType>(IPtr->getType())->getElementType();
     Type *ArgTypeJ = cast<PointerType>(JPtr->getType())->getElementType();
@@ -1515,7 +2082,7 @@
     Type *VArgPtrType = PointerType::get(VArgType,
       cast<PointerType>(IPtr->getType())->getAddressSpace());
     return new BitCastInst(VPtr, VArgPtrType, getReplacementName(I, true, o),
-                        /* insert before */ FlipMemInputs ? J : I);
+                        /* insert before */ I);
   }
 
   void BBVectorize::fillNewShuffleMask(LLVMContext& Context, Instruction *J,
@@ -1585,23 +2152,12 @@
                                   Instruction *J, unsigned o, Value *&LOp,
                                   unsigned numElemL,
                                   Type *ArgTypeL, Type *ArgTypeH,
-                                  unsigned IdxOff) {
+                                  bool IBeforeJ, unsigned IdxOff) {
     bool ExpandedIEChain = false;
     if (InsertElementInst *LIE = dyn_cast<InsertElementInst>(LOp)) {
       // If we have a pure insertelement chain, then this can be rewritten
       // into a chain that directly builds the larger type.
-      bool PureChain = true;
-      InsertElementInst *LIENext = LIE;
-      do {
-        if (!isa<UndefValue>(LIENext->getOperand(0)) &&
-            !isa<InsertElementInst>(LIENext->getOperand(0))) {
-          PureChain = false;
-          break;
-        }
-      } while ((LIENext =
-                 dyn_cast<InsertElementInst>(LIENext->getOperand(0))));
-
-      if (PureChain) {
+      if (isPureIEChain(LIE)) {
         SmallVector<Value *, 8> VectElemts(numElemL,
           UndefValue::get(ArgTypeL->getScalarType()));
         InsertElementInst *LIENext = LIE;
@@ -1619,8 +2175,9 @@
           LIENext = InsertElementInst::Create(LIEPrev, VectElemts[i],
                              ConstantInt::get(Type::getInt32Ty(Context),
                                               i + IdxOff),
-                             getReplacementName(I, true, o, i+1));
-          LIENext->insertBefore(J);
+                             getReplacementName(IBeforeJ ? I : J,
+                                                true, o, i+1));
+          LIENext->insertBefore(IBeforeJ ? J : I);
           LIEPrev = LIENext;
         }
 
@@ -1635,7 +2192,7 @@
   // Returns the value to be used as the specified operand of the vector
   // instruction that fuses I with J.
   Value *BBVectorize::getReplacementInput(LLVMContext& Context, Instruction *I,
-                     Instruction *J, unsigned o, bool FlipMemInputs) {
+                     Instruction *J, unsigned o, bool IBeforeJ) {
     Value *CV0 = ConstantInt::get(Type::getInt32Ty(Context), 0);
     Value *CV1 = ConstantInt::get(Type::getInt32Ty(Context), 1);
 
@@ -1646,12 +2203,6 @@
 
     Instruction *L = I, *H = J;
     Type *ArgTypeL = ArgTypeI, *ArgTypeH = ArgTypeJ;
-    if (FlipMemInputs) {
-      L = J;
-      H = I;
-      ArgTypeL = ArgTypeJ;
-      ArgTypeH = ArgTypeI;
-    }
 
     unsigned numElemL;
     if (ArgTypeL->isVectorTy())
@@ -1804,8 +2355,9 @@
           Instruction *S =
             new ShuffleVectorInst(I1, UndefValue::get(I1T),
                                   ConstantVector::get(Mask),
-                                  getReplacementName(I, true, o));
-          S->insertBefore(J);
+                                  getReplacementName(IBeforeJ ? I : J,
+                                                     true, o));
+          S->insertBefore(IBeforeJ ? J : I);
           return S;
         }
 
@@ -1826,8 +2378,9 @@
           Instruction *NewI1 =
             new ShuffleVectorInst(I1, UndefValue::get(I1T),
                                   ConstantVector::get(Mask),
-                                  getReplacementName(I, true, o, 1));
-          NewI1->insertBefore(J);
+                                  getReplacementName(IBeforeJ ? I : J,
+                                                     true, o, 1));
+          NewI1->insertBefore(IBeforeJ ? J : I);
           I1 = NewI1;
           I1T = I2T;
           I1Elem = I2Elem;
@@ -1842,8 +2395,9 @@
           Instruction *NewI2 =
             new ShuffleVectorInst(I2, UndefValue::get(I2T),
                                   ConstantVector::get(Mask),
-                                  getReplacementName(I, true, o, 1));
-          NewI2->insertBefore(J);
+                                  getReplacementName(IBeforeJ ? I : J,
+                                                     true, o, 1));
+          NewI2->insertBefore(IBeforeJ ? J : I);
           I2 = NewI2;
           I2T = I1T;
           I2Elem = I1Elem;
@@ -1863,8 +2417,8 @@
 
         Instruction *NewOp =
           new ShuffleVectorInst(I1, I2, ConstantVector::get(Mask),
-                                getReplacementName(I, true, o));
-        NewOp->insertBefore(J);
+                                getReplacementName(IBeforeJ ? I : J, true, o));
+        NewOp->insertBefore(IBeforeJ ? J : I);
         return NewOp;
       }
     }
@@ -1872,17 +2426,17 @@
     Type *ArgType = ArgTypeL;
     if (numElemL < numElemH) {
       if (numElemL == 1 && expandIEChain(Context, I, J, o, HOp, numElemH,
-                                         ArgTypeL, VArgType, 1)) {
+                                         ArgTypeL, VArgType, IBeforeJ, 1)) {
         // This is another short-circuit case: we're combining a scalar into
         // a vector that is formed by an IE chain. We've just expanded the IE
         // chain, now insert the scalar and we're done.
 
         Instruction *S = InsertElementInst::Create(HOp, LOp, CV0,
-                                               getReplacementName(I, true, o));
-        S->insertBefore(J);
+                           getReplacementName(IBeforeJ ? I : J, true, o));
+        S->insertBefore(IBeforeJ ? J : I);
         return S;
       } else if (!expandIEChain(Context, I, J, o, LOp, numElemL, ArgTypeL,
-                                ArgTypeH)) {
+                                ArgTypeH, IBeforeJ)) {
         // The two vector inputs to the shuffle must be the same length,
         // so extend the smaller vector to be the same length as the larger one.
         Instruction *NLOp;
@@ -1897,29 +2451,32 @@
     
           NLOp = new ShuffleVectorInst(LOp, UndefValue::get(ArgTypeL),
                                        ConstantVector::get(Mask),
-                                       getReplacementName(I, true, o, 1));
+                                       getReplacementName(IBeforeJ ? I : J,
+                                                          true, o, 1));
         } else {
           NLOp = InsertElementInst::Create(UndefValue::get(ArgTypeH), LOp, CV0,
-                                           getReplacementName(I, true, o, 1));
+                                           getReplacementName(IBeforeJ ? I : J,
+                                                              true, o, 1));
         }
   
-        NLOp->insertBefore(J);
+        NLOp->insertBefore(IBeforeJ ? J : I);
         LOp = NLOp;
       }
 
       ArgType = ArgTypeH;
     } else if (numElemL > numElemH) {
       if (numElemH == 1 && expandIEChain(Context, I, J, o, LOp, numElemL,
-                                         ArgTypeH, VArgType)) {
+                                         ArgTypeH, VArgType, IBeforeJ)) {
         Instruction *S =
           InsertElementInst::Create(LOp, HOp, 
                                     ConstantInt::get(Type::getInt32Ty(Context),
                                                      numElemL),
-                                    getReplacementName(I, true, o));
-        S->insertBefore(J);
+                                    getReplacementName(IBeforeJ ? I : J,
+                                                       true, o));
+        S->insertBefore(IBeforeJ ? J : I);
         return S;
       } else if (!expandIEChain(Context, I, J, o, HOp, numElemH, ArgTypeH,
-                                ArgTypeL)) {
+                                ArgTypeL, IBeforeJ)) {
         Instruction *NHOp;
         if (numElemH > 1) {
           std::vector<Constant *> Mask(numElemL);
@@ -1931,13 +2488,15 @@
     
           NHOp = new ShuffleVectorInst(HOp, UndefValue::get(ArgTypeH),
                                        ConstantVector::get(Mask),
-                                       getReplacementName(I, true, o, 1));
+                                       getReplacementName(IBeforeJ ? I : J,
+                                                          true, o, 1));
         } else {
           NHOp = InsertElementInst::Create(UndefValue::get(ArgTypeL), HOp, CV0,
-                                           getReplacementName(I, true, o, 1));
+                                           getReplacementName(IBeforeJ ? I : J,
+                                                              true, o, 1));
         }
   
-        NHOp->insertBefore(J);
+        NHOp->insertBefore(IBeforeJ ? J : I);
         HOp = NHOp;
       }
     }
@@ -1955,19 +2514,21 @@
       }
 
       Instruction *BV = new ShuffleVectorInst(LOp, HOp,
-                                              ConstantVector::get(Mask),
-                                              getReplacementName(I, true, o));
-      BV->insertBefore(J);
+                          ConstantVector::get(Mask),
+                          getReplacementName(IBeforeJ ? I : J, true, o));
+      BV->insertBefore(IBeforeJ ? J : I);
       return BV;
     }
 
     Instruction *BV1 = InsertElementInst::Create(
                                           UndefValue::get(VArgType), LOp, CV0,
-                                          getReplacementName(I, true, o, 1));
-    BV1->insertBefore(I);
+                                          getReplacementName(IBeforeJ ? I : J,
+                                                             true, o, 1));
+    BV1->insertBefore(IBeforeJ ? J : I);
     Instruction *BV2 = InsertElementInst::Create(BV1, HOp, CV1,
-                                          getReplacementName(I, true, o, 2));
-    BV2->insertBefore(J);
+                                          getReplacementName(IBeforeJ ? I : J,
+                                                             true, o, 2));
+    BV2->insertBefore(IBeforeJ ? J : I);
     return BV2;
   }
 
@@ -1976,7 +2537,7 @@
   void BBVectorize::getReplacementInputsForPair(LLVMContext& Context,
                      Instruction *I, Instruction *J,
                      SmallVector<Value *, 3> &ReplacedOperands,
-                     bool FlipMemInputs) {
+                     bool IBeforeJ) {
     unsigned NumOperands = I->getNumOperands();
 
     for (unsigned p = 0, o = NumOperands-1; p < NumOperands; ++p, --o) {
@@ -1985,8 +2546,7 @@
 
       if (isa<LoadInst>(I) || (o == 1 && isa<StoreInst>(I))) {
         // This is the pointer for a load/store instruction.
-        ReplacedOperands[o] = getReplacementPointerInput(Context, I, J, o,
-                                FlipMemInputs);
+        ReplacedOperands[o] = getReplacementPointerInput(Context, I, J, o);
         continue;
       } else if (isa<CallInst>(I)) {
         Function *F = cast<CallInst>(I)->getCalledFunction();
@@ -2014,8 +2574,7 @@
         continue;
       }
 
-      ReplacedOperands[o] =
-        getReplacementInput(Context, I, J, o, FlipMemInputs);
+      ReplacedOperands[o] = getReplacementInput(Context, I, J, o, IBeforeJ);
     }
   }
 
@@ -2026,8 +2585,7 @@
   void BBVectorize::replaceOutputsOfPair(LLVMContext& Context, Instruction *I,
                      Instruction *J, Instruction *K,
                      Instruction *&InsertionPt,
-                     Instruction *&K1, Instruction *&K2,
-                     bool FlipMemInputs) {
+                     Instruction *&K1, Instruction *&K2) {
     if (isa<StoreInst>(I)) {
       AA->replaceWithNewValue(I, K);
       AA->replaceWithNewValue(J, K);
@@ -2057,13 +2615,11 @@
         }
 
         K1 = new ShuffleVectorInst(K, UndefValue::get(VType),
-                                   ConstantVector::get(
-                                     FlipMemInputs ? Mask2 : Mask1),
+                                   ConstantVector::get( Mask1),
                                    getReplacementName(K, false, 1));
       } else {
         Value *CV0 = ConstantInt::get(Type::getInt32Ty(Context), 0);
-        Value *CV1 = ConstantInt::get(Type::getInt32Ty(Context), numElem-1);
-        K1 = ExtractElementInst::Create(K, FlipMemInputs ? CV1 : CV0,
+        K1 = ExtractElementInst::Create(K, CV0,
                                           getReplacementName(K, false, 1));
       }
 
@@ -2075,13 +2631,11 @@
         }
 
         K2 = new ShuffleVectorInst(K, UndefValue::get(VType),
-                                   ConstantVector::get(
-                                     FlipMemInputs ? Mask1 : Mask2),
+                                   ConstantVector::get( Mask2),
                                    getReplacementName(K, false, 2));
       } else {
-        Value *CV0 = ConstantInt::get(Type::getInt32Ty(Context), 0);
         Value *CV1 = ConstantInt::get(Type::getInt32Ty(Context), numElem-1);
-        K2 = ExtractElementInst::Create(K, FlipMemInputs ? CV0 : CV1,
+        K2 = ExtractElementInst::Create(K, CV1,
                                           getReplacementName(K, false, 2));
       }
 
@@ -2181,36 +2735,6 @@
     }
   }
 
-  // As with the aliasing information, SCEV can also change because of
-  // vectorization. This information is used to compute relative pointer
-  // offsets; the necessary information will be cached here prior to
-  // fusion.
-  void BBVectorize::collectPtrInfo(std::vector<Value *> &PairableInsts,
-                                   DenseMap<Value *, Value *> &ChosenPairs,
-                                   DenseSet<Value *> &LowPtrInsts) {
-    for (std::vector<Value *>::iterator PI = PairableInsts.begin(),
-      PIE = PairableInsts.end(); PI != PIE; ++PI) {
-      DenseMap<Value *, Value *>::iterator P = ChosenPairs.find(*PI);
-      if (P == ChosenPairs.end()) continue;
-
-      Instruction *I = cast<Instruction>(P->first);
-      Instruction *J = cast<Instruction>(P->second);
-
-      if (!isa<LoadInst>(I) && !isa<StoreInst>(I))
-        continue;
-
-      Value *IPtr, *JPtr;
-      unsigned IAlignment, JAlignment;
-      int64_t OffsetInElmts;
-      if (!getPairPtrInfo(I, J, IPtr, JPtr, IAlignment, JAlignment,
-                          OffsetInElmts) || abs64(OffsetInElmts) != 1)
-        llvm_unreachable("Pre-fusion pointer analysis failed");
-
-      Value *LowPI = (OffsetInElmts > 0) ? I : J;
-      LowPtrInsts.insert(LowPI);
-    }
-  }
-
   // When the first instruction in each pair is cloned, it will inherit its
   // parent's metadata. This metadata must be combined with that of the other
   // instruction in a safe way.
@@ -2244,27 +2768,27 @@
   // second member).
   void BBVectorize::fuseChosenPairs(BasicBlock &BB,
                      std::vector<Value *> &PairableInsts,
-                     DenseMap<Value *, Value *> &ChosenPairs) {
+                     DenseMap<Value *, Value *> &ChosenPairs,
+                     DenseSet<ValuePair> &FixedOrderPairs,
+                     DenseMap<VPPair, unsigned> &PairConnectionTypes,
+                     std::multimap<ValuePair, ValuePair> &ConnectedPairs,
+                     std::multimap<ValuePair, ValuePair> &ConnectedPairDeps) {
     LLVMContext& Context = BB.getContext();
 
     // During the vectorization process, the order of the pairs to be fused
     // could be flipped. So we'll add each pair, flipped, into the ChosenPairs
     // list. After a pair is fused, the flipped pair is removed from the list.
-    std::vector<ValuePair> FlippedPairs;
-    FlippedPairs.reserve(ChosenPairs.size());
+    DenseSet<ValuePair> FlippedPairs;
     for (DenseMap<Value *, Value *>::iterator P = ChosenPairs.begin(),
          E = ChosenPairs.end(); P != E; ++P)
-      FlippedPairs.push_back(ValuePair(P->second, P->first));
-    for (std::vector<ValuePair>::iterator P = FlippedPairs.begin(),
+      FlippedPairs.insert(ValuePair(P->second, P->first));
+    for (DenseSet<ValuePair>::iterator P = FlippedPairs.begin(),
          E = FlippedPairs.end(); P != E; ++P)
       ChosenPairs.insert(*P);
 
     std::multimap<Value *, Value *> LoadMoveSet;
     collectLoadMoveSet(BB, PairableInsts, ChosenPairs, LoadMoveSet);
 
-    DenseSet<Value *> LowPtrInsts;
-    collectPtrInfo(PairableInsts, ChosenPairs, LowPtrInsts);
-
     DEBUG(dbgs() << "BBV: initial: \n" << BB << "\n");
 
     for (BasicBlock::iterator PI = BB.getFirstInsertionPt(); PI != BB.end();) {
@@ -2304,44 +2828,92 @@
         continue;
       }
 
-      bool FlipMemInputs = false;
-      if (isa<LoadInst>(I) || isa<StoreInst>(I))
-        FlipMemInputs = (LowPtrInsts.find(I) == LowPtrInsts.end());
+      // If the pair must have the other order, then flip it.
+      bool FlipPairOrder = FixedOrderPairs.count(ValuePair(J, I));
+      if (!FlipPairOrder && !FixedOrderPairs.count(ValuePair(I, J))) {
+        // This pair does not have a fixed order, and so we might want to
+        // flip it if that will yield fewer shuffles. We count the number
+        // of dependencies connected via swaps, and those directly connected,
+        // and flip the order if the number of swaps is greater.
+        bool OrigOrder = true;
+        VPPIteratorPair IP = ConnectedPairDeps.equal_range(ValuePair(I, J));
+        if (IP.first == ConnectedPairDeps.end()) {
+          IP = ConnectedPairDeps.equal_range(ValuePair(J, I));
+          OrigOrder = false;
+        }
+
+        if (IP.first != ConnectedPairDeps.end()) {
+          unsigned NumDepsDirect = 0, NumDepsSwap = 0;
+          for (std::multimap<ValuePair, ValuePair>::iterator Q = IP.first;
+               Q != IP.second; ++Q) {
+            DenseMap<VPPair, unsigned>::iterator R =
+              PairConnectionTypes.find(VPPair(Q->second, Q->first));
+            assert(R != PairConnectionTypes.end() &&
+                   "Cannot find pair connection type");
+            if (R->second == PairConnectionDirect)
+              ++NumDepsDirect;
+            else if (R->second == PairConnectionSwap)
+              ++NumDepsSwap;
+          }
+
+          if (!OrigOrder)
+            std::swap(NumDepsDirect, NumDepsSwap);
 
+          if (NumDepsSwap > NumDepsDirect) {
+            FlipPairOrder = true;
+            DEBUG(dbgs() << "BBV: reordering pair: " << *I <<
+                            " <-> " << *J << "\n");
+          }
+        }
+      }
+
+      Instruction *L = I, *H = J;
+      if (FlipPairOrder)
+        std::swap(H, L);
+
+      // If the pair being fused uses the opposite order from that in the pair
+      // connection map, then we need to flip the types.
+      VPPIteratorPair IP = ConnectedPairs.equal_range(ValuePair(H, L));
+      for (std::multimap<ValuePair, ValuePair>::iterator Q = IP.first;
+           Q != IP.second; ++Q) {
+        DenseMap<VPPair, unsigned>::iterator R = PairConnectionTypes.find(*Q);
+        assert(R != PairConnectionTypes.end() &&
+               "Cannot find pair connection type");
+        if (R->second == PairConnectionDirect)
+          R->second = PairConnectionSwap;
+        else if (R->second == PairConnectionSwap)
+          R->second = PairConnectionDirect;
+      }
+
+      bool LBeforeH = !FlipPairOrder;
       unsigned NumOperands = I->getNumOperands();
       SmallVector<Value *, 3> ReplacedOperands(NumOperands);
-      getReplacementInputsForPair(Context, I, J, ReplacedOperands,
-        FlipMemInputs);
+      getReplacementInputsForPair(Context, L, H, ReplacedOperands,
+                                  LBeforeH);
 
       // Make a copy of the original operation, change its type to the vector
       // type and replace its operands with the vector operands.
-      Instruction *K = I->clone();
-      if (I->hasName()) K->takeName(I);
+      Instruction *K = L->clone();
+      if (L->hasName())
+        K->takeName(L);
+      else if (H->hasName())
+        K->takeName(H);
 
       if (!isa<StoreInst>(K))
-        K->mutateType(getVecTypeForPair(I->getType(), J->getType()));
+        K->mutateType(getVecTypeForPair(L->getType(), H->getType()));
 
-      combineMetadata(K, J);
+      combineMetadata(K, H);
+      K->intersectOptionalDataWith(H);
 
       for (unsigned o = 0; o < NumOperands; ++o)
         K->setOperand(o, ReplacedOperands[o]);
 
-      // If we've flipped the memory inputs, make sure that we take the correct
-      // alignment.
-      if (FlipMemInputs) {
-        if (isa<StoreInst>(K))
-          cast<StoreInst>(K)->setAlignment(cast<StoreInst>(J)->getAlignment());
-        else
-          cast<LoadInst>(K)->setAlignment(cast<LoadInst>(J)->getAlignment());
-      }
-
       K->insertAfter(J);
 
       // Instruction insertion point:
       Instruction *InsertionPt = K;
       Instruction *K1 = 0, *K2 = 0;
-      replaceOutputsOfPair(Context, I, J, K, InsertionPt, K1, K2,
-        FlipMemInputs);
+      replaceOutputsOfPair(Context, L, H, K, InsertionPt, K1, K2);
 
       // The use tree of the first original instruction must be moved to after
       // the location of the second instruction. The entire use tree of the
@@ -2351,10 +2923,10 @@
       moveUsesOfIAfterJ(BB, LoadMoveSet, InsertionPt, I, J);
 
       if (!isa<StoreInst>(I)) {
-        I->replaceAllUsesWith(K1);
-        J->replaceAllUsesWith(K2);
-        AA->replaceWithNewValue(I, K1);
-        AA->replaceWithNewValue(J, K2);
+        L->replaceAllUsesWith(K1);
+        H->replaceAllUsesWith(K2);
+        AA->replaceWithNewValue(L, K1);
+        AA->replaceWithNewValue(H, K2);
       }
 
       // Instructions that may read from memory may be in the load move set.
@@ -2387,6 +2959,9 @@
       SE->forgetValue(J);
       I->eraseFromParent();
       J->eraseFromParent();
+
+      DEBUG(if (PrintAfterEveryPair) dbgs() << "BBV: block is now: \n" <<
+                                               BB << "\n");
     }
 
     DEBUG(dbgs() << "BBV: final: \n" << BB << "\n");
@@ -2397,6 +2972,7 @@
 static const char bb_vectorize_name[] = "Basic-Block Vectorization";
 INITIALIZE_PASS_BEGIN(BBVectorize, BBV_NAME, bb_vectorize_name, false, false)
 INITIALIZE_AG_DEPENDENCY(AliasAnalysis)
+INITIALIZE_PASS_DEPENDENCY(DominatorTree)
 INITIALIZE_PASS_DEPENDENCY(ScalarEvolution)
 INITIALIZE_PASS_END(BBVectorize, BBV_NAME, bb_vectorize_name, false, false)
 

Modified: llvm/branches/AMDILBackend/lib/Transforms/Vectorize/CMakeLists.txt
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Vectorize/CMakeLists.txt?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Vectorize/CMakeLists.txt (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Vectorize/CMakeLists.txt Tue Jan 15 11:16:16 2013
@@ -1,6 +1,7 @@
 add_llvm_library(LLVMVectorize
   BBVectorize.cpp
   Vectorize.cpp
+  LoopVectorize.cpp
   )
 
 add_dependencies(LLVMVectorize intrinsics_gen)

Modified: llvm/branches/AMDILBackend/lib/Transforms/Vectorize/Vectorize.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/Transforms/Vectorize/Vectorize.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/Transforms/Vectorize/Vectorize.cpp (original)
+++ llvm/branches/AMDILBackend/lib/Transforms/Vectorize/Vectorize.cpp Tue Jan 15 11:16:16 2013
@@ -7,7 +7,7 @@
 //
 //===----------------------------------------------------------------------===//
 //
-// This file implements common infrastructure for libLLVMVectorizeOpts.a, which 
+// This file implements common infrastructure for libLLVMVectorizeOpts.a, which
 // implements several vectorization transformations over the LLVM intermediate
 // representation, including the C bindings for that library.
 //
@@ -23,10 +23,11 @@
 
 using namespace llvm;
 
-/// initializeVectorizationPasses - Initialize all passes linked into the 
+/// initializeVectorizationPasses - Initialize all passes linked into the
 /// Vectorization library.
 void llvm::initializeVectorization(PassRegistry &Registry) {
   initializeBBVectorizePass(Registry);
+  initializeLoopVectorizePass(Registry);
 }
 
 void LLVMInitializeVectorization(LLVMPassRegistryRef R) {
@@ -37,3 +38,6 @@
   unwrap(PM)->add(createBBVectorizePass());
 }
 
+void LLVMAddLoopVectorizePass(LLVMPassManagerRef PM) {
+  unwrap(PM)->add(createLoopVectorizePass());
+}

Modified: llvm/branches/AMDILBackend/lib/VMCore/AsmWriter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/VMCore/AsmWriter.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/VMCore/AsmWriter.cpp (original)
+++ llvm/branches/AMDILBackend/lib/VMCore/AsmWriter.cpp Tue Jan 15 11:16:16 2013
@@ -66,6 +66,25 @@
   return 0;
 }
 
+static void PrintCallingConv(unsigned cc, raw_ostream &Out)
+{
+  switch (cc) {
+    case CallingConv::Fast:         Out << "fastcc"; break;
+    case CallingConv::Cold:         Out << "coldcc"; break;
+    case CallingConv::X86_StdCall:  Out << "x86_stdcallcc"; break;
+    case CallingConv::X86_FastCall: Out << "x86_fastcallcc"; break;
+    case CallingConv::X86_ThisCall: Out << "x86_thiscallcc"; break;
+    case CallingConv::Intel_OCL_BI: Out << "intel_ocl_bicc"; break;
+    case CallingConv::ARM_APCS:     Out << "arm_apcscc"; break;
+    case CallingConv::ARM_AAPCS:    Out << "arm_aapcscc"; break;
+    case CallingConv::ARM_AAPCS_VFP:Out << "arm_aapcs_vfpcc"; break;
+    case CallingConv::MSP430_INTR:  Out << "msp430_intrcc"; break;
+    case CallingConv::PTX_Kernel:   Out << "ptx_kernel"; break;
+    case CallingConv::PTX_Device:   Out << "ptx_device"; break;
+    default:                        Out << "cc" << cc; break;
+  }
+}
+ 
 // PrintEscapedString - Print each character of the specified string, escaping
 // it if it is not printable or if it is an escape char.
 static void PrintEscapedString(StringRef Name, raw_ostream &Out) {
@@ -141,8 +160,8 @@
 /// TypePrinting - Type printing machinery.
 namespace {
 class TypePrinting {
-  TypePrinting(const TypePrinting &);   // DO NOT IMPLEMENT
-  void operator=(const TypePrinting&);  // DO NOT IMPLEMENT
+  TypePrinting(const TypePrinting &) LLVM_DELETED_FUNCTION;
+  void operator=(const TypePrinting&) LLVM_DELETED_FUNCTION;
 public:
 
   /// NamedTypes - The named types that are used by the current module.
@@ -380,8 +399,8 @@
   /// Add all of the functions arguments, basic blocks, and instructions.
   void processFunction();
 
-  SlotTracker(const SlotTracker &);  // DO NOT IMPLEMENT
-  void operator=(const SlotTracker &);  // DO NOT IMPLEMENT
+  SlotTracker(const SlotTracker &) LLVM_DELETED_FUNCTION;
+  void operator=(const SlotTracker &) LLVM_DELETED_FUNCTION;
 };
 
 }  // end anonymous namespace
@@ -1029,6 +1048,9 @@
       Out << "sideeffect ";
     if (IA->isAlignStack())
       Out << "alignstack ";
+    // We don't emit the AD_ATT dialect as it's the assumed default.
+    if (IA->getDialect() == InlineAsm::AD_Intel)
+      Out << "inteldialect ";
     Out << '"';
     PrintEscapedString(IA->getAsmString(), Out);
     Out << "\", \"";
@@ -1222,8 +1244,8 @@
   // Print the type
   TypePrinter.print(Operand->getType(), Out);
   // Print parameter attributes list
-  if (Attrs != Attribute::None)
-    Out << ' ' << Attribute::getAsString(Attrs);
+  if (Attrs.hasAttributes())
+    Out << ' ' << Attrs.getAsString();
   Out << ' ';
   // Print the operand
   WriteAsOperandInternal(Out, Operand, &TypePrinter, &Machine, TheModule);
@@ -1285,8 +1307,9 @@
   // Output all globals.
   if (!M->global_empty()) Out << '\n';
   for (Module::const_global_iterator I = M->global_begin(), E = M->global_end();
-       I != E; ++I)
-    printGlobal(I);
+       I != E; ++I) {
+    printGlobal(I); Out << '\n';
+  }
 
   // Output all aliases.
   if (!M->alias_empty()) Out << "\n";
@@ -1353,12 +1376,12 @@
   case GlobalValue::LinkerPrivateWeakLinkage:
     Out << "linker_private_weak ";
     break;
-  case GlobalValue::LinkerPrivateWeakDefAutoLinkage:
-    Out << "linker_private_weak_def_auto ";
-    break;
   case GlobalValue::InternalLinkage:      Out << "internal ";       break;
   case GlobalValue::LinkOnceAnyLinkage:   Out << "linkonce ";       break;
   case GlobalValue::LinkOnceODRLinkage:   Out << "linkonce_odr ";   break;
+  case GlobalValue::LinkOnceODRAutoHideLinkage:
+    Out << "linkonce_odr_auto_hide ";
+    break;
   case GlobalValue::WeakAnyLinkage:       Out << "weak ";           break;
   case GlobalValue::WeakODRLinkage:       Out << "weak_odr ";       break;
   case GlobalValue::CommonLinkage:        Out << "common ";         break;
@@ -1436,7 +1459,6 @@
     Out << ", align " << GV->getAlignment();
 
   printInfoComment(*GV);
-  Out << '\n';
 }
 
 void AssemblyWriter::printAlias(const GlobalAlias *GA) {
@@ -1527,27 +1549,16 @@
   PrintVisibility(F->getVisibility(), Out);
 
   // Print the calling convention.
-  switch (F->getCallingConv()) {
-  case CallingConv::C: break;   // default
-  case CallingConv::Fast:         Out << "fastcc "; break;
-  case CallingConv::Cold:         Out << "coldcc "; break;
-  case CallingConv::X86_StdCall:  Out << "x86_stdcallcc "; break;
-  case CallingConv::X86_FastCall: Out << "x86_fastcallcc "; break;
-  case CallingConv::X86_ThisCall: Out << "x86_thiscallcc "; break;
-  case CallingConv::ARM_APCS:     Out << "arm_apcscc "; break;
-  case CallingConv::ARM_AAPCS:    Out << "arm_aapcscc "; break;
-  case CallingConv::ARM_AAPCS_VFP:Out << "arm_aapcs_vfpcc "; break;
-  case CallingConv::MSP430_INTR:  Out << "msp430_intrcc "; break;
-  case CallingConv::PTX_Kernel:   Out << "ptx_kernel "; break;
-  case CallingConv::PTX_Device:   Out << "ptx_device "; break;
-  default: Out << "cc" << F->getCallingConv() << " "; break;
+  if (F->getCallingConv() != CallingConv::C) {
+    PrintCallingConv(F->getCallingConv(), Out);
+    Out << " ";
   }
 
   FunctionType *FT = F->getFunctionType();
   const AttrListPtr &Attrs = F->getAttributes();
   Attributes RetAttrs = Attrs.getRetAttributes();
-  if (RetAttrs != Attribute::None)
-    Out <<  Attribute::getAsString(Attrs.getRetAttributes()) << ' ';
+  if (RetAttrs.hasAttributes())
+    Out <<  Attrs.getRetAttributes().getAsString() << ' ';
   TypePrinter.print(F->getReturnType(), Out);
   Out << ' ';
   WriteAsOperandInternal(Out, F, &TypePrinter, &Machine, F->getParent());
@@ -1576,8 +1587,8 @@
       TypePrinter.print(FT->getParamType(i), Out);
 
       Attributes ArgAttrs = Attrs.getParamAttributes(i+1);
-      if (ArgAttrs != Attribute::None)
-        Out << ' ' << Attribute::getAsString(ArgAttrs);
+      if (ArgAttrs.hasAttributes())
+        Out << ' ' << ArgAttrs.getAsString();
     }
   }
 
@@ -1590,8 +1601,8 @@
   if (F->hasUnnamedAddr())
     Out << " unnamed_addr";
   Attributes FnAttrs = Attrs.getFnAttributes();
-  if (FnAttrs != Attribute::None)
-    Out << ' ' << Attribute::getAsString(Attrs.getFnAttributes());
+  if (FnAttrs.hasAttributes())
+    Out << ' ' << Attrs.getFnAttributes().getAsString();
   if (F->hasSection()) {
     Out << " section \"";
     PrintEscapedString(F->getSection(), Out);
@@ -1624,8 +1635,8 @@
   TypePrinter.print(Arg->getType(), Out);
 
   // Output parameter attributes list
-  if (Attrs != Attribute::None)
-    Out << ' ' << Attribute::getAsString(Attrs);
+  if (Attrs.hasAttributes())
+    Out << ' ' << Attrs.getAsString();
 
   // Output name, if available...
   if (Arg->hasName()) {
@@ -1828,20 +1839,9 @@
     Out << " void";
   } else if (const CallInst *CI = dyn_cast<CallInst>(&I)) {
     // Print the calling convention being used.
-    switch (CI->getCallingConv()) {
-    case CallingConv::C: break;   // default
-    case CallingConv::Fast:  Out << " fastcc"; break;
-    case CallingConv::Cold:  Out << " coldcc"; break;
-    case CallingConv::X86_StdCall:  Out << " x86_stdcallcc"; break;
-    case CallingConv::X86_FastCall: Out << " x86_fastcallcc"; break;
-    case CallingConv::X86_ThisCall: Out << " x86_thiscallcc"; break;
-    case CallingConv::ARM_APCS:     Out << " arm_apcscc "; break;
-    case CallingConv::ARM_AAPCS:    Out << " arm_aapcscc "; break;
-    case CallingConv::ARM_AAPCS_VFP:Out << " arm_aapcs_vfpcc "; break;
-    case CallingConv::MSP430_INTR:  Out << " msp430_intrcc "; break;
-    case CallingConv::PTX_Kernel:   Out << " ptx_kernel"; break;
-    case CallingConv::PTX_Device:   Out << " ptx_device"; break;
-    default: Out << " cc" << CI->getCallingConv(); break;
+    if (CI->getCallingConv() != CallingConv::C) {
+      Out << " ";
+      PrintCallingConv(CI->getCallingConv(), Out);
     }
 
     Operand = CI->getCalledValue();
@@ -1850,8 +1850,8 @@
     Type *RetTy = FTy->getReturnType();
     const AttrListPtr &PAL = CI->getAttributes();
 
-    if (PAL.getRetAttributes() != Attribute::None)
-      Out << ' ' << Attribute::getAsString(PAL.getRetAttributes());
+    if (PAL.getRetAttributes().hasAttributes())
+      Out << ' ' << PAL.getRetAttributes().getAsString();
 
     // If possible, print out the short form of the call instruction.  We can
     // only do this if the first argument is a pointer to a nonvararg function,
@@ -1874,8 +1874,8 @@
       writeParamOperand(CI->getArgOperand(op), PAL.getParamAttributes(op + 1));
     }
     Out << ')';
-    if (PAL.getFnAttributes() != Attribute::None)
-      Out << ' ' << Attribute::getAsString(PAL.getFnAttributes());
+    if (PAL.getFnAttributes().hasAttributes())
+      Out << ' ' << PAL.getFnAttributes().getAsString();
   } else if (const InvokeInst *II = dyn_cast<InvokeInst>(&I)) {
     Operand = II->getCalledValue();
     PointerType *PTy = cast<PointerType>(Operand->getType());
@@ -1884,24 +1884,13 @@
     const AttrListPtr &PAL = II->getAttributes();
 
     // Print the calling convention being used.
-    switch (II->getCallingConv()) {
-    case CallingConv::C: break;   // default
-    case CallingConv::Fast:  Out << " fastcc"; break;
-    case CallingConv::Cold:  Out << " coldcc"; break;
-    case CallingConv::X86_StdCall:  Out << " x86_stdcallcc"; break;
-    case CallingConv::X86_FastCall: Out << " x86_fastcallcc"; break;
-    case CallingConv::X86_ThisCall: Out << " x86_thiscallcc"; break;
-    case CallingConv::ARM_APCS:     Out << " arm_apcscc "; break;
-    case CallingConv::ARM_AAPCS:    Out << " arm_aapcscc "; break;
-    case CallingConv::ARM_AAPCS_VFP:Out << " arm_aapcs_vfpcc "; break;
-    case CallingConv::MSP430_INTR:  Out << " msp430_intrcc "; break;
-    case CallingConv::PTX_Kernel:   Out << " ptx_kernel"; break;
-    case CallingConv::PTX_Device:   Out << " ptx_device"; break;
-    default: Out << " cc" << II->getCallingConv(); break;
+    if (II->getCallingConv() != CallingConv::C) {
+      Out << " ";
+      PrintCallingConv(II->getCallingConv(), Out);
     }
 
-    if (PAL.getRetAttributes() != Attribute::None)
-      Out << ' ' << Attribute::getAsString(PAL.getRetAttributes());
+    if (PAL.getRetAttributes().hasAttributes())
+      Out << ' ' << PAL.getRetAttributes().getAsString();
 
     // If possible, print out the short form of the invoke instruction. We can
     // only do this if the first argument is a pointer to a nonvararg function,
@@ -1925,8 +1914,8 @@
     }
 
     Out << ')';
-    if (PAL.getFnAttributes() != Attribute::None)
-      Out << ' ' << Attribute::getAsString(PAL.getFnAttributes());
+    if (PAL.getFnAttributes().hasAttributes())
+      Out << ' ' << PAL.getFnAttributes().getAsString();
 
     Out << "\n          to ";
     writeOperand(II->getNormalDest(), true);

Modified: llvm/branches/AMDILBackend/lib/VMCore/Attributes.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/VMCore/Attributes.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/VMCore/Attributes.cpp (original)
+++ llvm/branches/AMDILBackend/lib/VMCore/Attributes.cpp Tue Jan 15 11:16:16 2013
@@ -7,11 +7,14 @@
 //
 //===----------------------------------------------------------------------===//
 //
-// This file implements the AttributesList class and Attribute utilities.
+// This file implements the Attributes, AttributeImpl, AttrBuilder,
+// AttributeListImpl, and AttrListPtr classes.
 //
 //===----------------------------------------------------------------------===//
 
 #include "llvm/Attributes.h"
+#include "AttributesImpl.h"
+#include "LLVMContextImpl.h"
 #include "llvm/Type.h"
 #include "llvm/ADT/StringExtras.h"
 #include "llvm/ADT/FoldingSet.h"
@@ -23,215 +26,382 @@
 using namespace llvm;
 
 //===----------------------------------------------------------------------===//
-// Attribute Function Definitions
+// Attributes Implementation
 //===----------------------------------------------------------------------===//
 
-std::string Attribute::getAsString(Attributes Attrs) {
+Attributes Attributes::get(LLVMContext &Context, ArrayRef<AttrVal> Vals) {
+  AttrBuilder B;
+  for (ArrayRef<AttrVal>::iterator I = Vals.begin(), E = Vals.end();
+       I != E; ++I)
+    B.addAttribute(*I);
+  return Attributes::get(Context, B);
+}
+
+Attributes Attributes::get(LLVMContext &Context, AttrBuilder &B) {
+  // If there are no attributes, return an empty Attributes class.
+  if (!B.hasAttributes())
+    return Attributes();
+
+  // Otherwise, build a key to look up the existing attributes.
+  LLVMContextImpl *pImpl = Context.pImpl;
+  FoldingSetNodeID ID;
+  ID.AddInteger(B.Raw());
+
+  void *InsertPoint;
+  AttributesImpl *PA = pImpl->AttrsSet.FindNodeOrInsertPos(ID, InsertPoint);
+
+  if (!PA) {
+    // If we didn't find any existing attributes of the same shape then create a
+    // new one and insert it.
+    PA = new AttributesImpl(B.Raw());
+    pImpl->AttrsSet.InsertNode(PA, InsertPoint);
+  }
+
+  // Return the AttributesList that we found or created.
+  return Attributes(PA);
+}
+
+bool Attributes::hasAttribute(AttrVal Val) const {
+  return Attrs && Attrs->hasAttribute(Val);
+}
+
+bool Attributes::hasAttributes() const {
+  return Attrs && Attrs->hasAttributes();
+}
+
+bool Attributes::hasAttributes(const Attributes &A) const {
+  return Attrs && Attrs->hasAttributes(A);
+}
+
+/// This returns the alignment field of an attribute as a byte alignment value.
+unsigned Attributes::getAlignment() const {
+  if (!hasAttribute(Attributes::Alignment))
+    return 0;
+  return 1U << ((Attrs->getAlignment() >> 16) - 1);
+}
+
+/// This returns the stack alignment field of an attribute as a byte alignment
+/// value.
+unsigned Attributes::getStackAlignment() const {
+  if (!hasAttribute(Attributes::StackAlignment))
+    return 0;
+  return 1U << ((Attrs->getStackAlignment() >> 26) - 1);
+}
+
+uint64_t Attributes::Raw() const {
+  return Attrs ? Attrs->Raw() : 0;
+}
+
+Attributes Attributes::typeIncompatible(Type *Ty) {
+  AttrBuilder Incompatible;
+
+  if (!Ty->isIntegerTy())
+    // Attributes that only apply to integers.
+    Incompatible.addAttribute(Attributes::SExt)
+      .addAttribute(Attributes::ZExt);
+
+  if (!Ty->isPointerTy())
+    // Attributes that only apply to pointers.
+    Incompatible.addAttribute(Attributes::ByVal)
+      .addAttribute(Attributes::Nest)
+      .addAttribute(Attributes::NoAlias)
+      .addAttribute(Attributes::NoCapture)
+      .addAttribute(Attributes::StructRet);
+
+  return Attributes::get(Ty->getContext(), Incompatible);
+}
+
+/// encodeLLVMAttributesForBitcode - This returns an integer containing an
+/// encoding of all the LLVM attributes found in the given attribute bitset.
+/// Any change to this encoding is a breaking change to bitcode compatibility.
+uint64_t Attributes::encodeLLVMAttributesForBitcode(Attributes Attrs) {
+  // FIXME: It doesn't make sense to store the alignment information as an
+  // expanded out value, we should store it as a log2 value.  However, we can't
+  // just change that here without breaking bitcode compatibility.  If this ever
+  // becomes a problem in practice, we should introduce new tag numbers in the
+  // bitcode file and have those tags use a more efficiently encoded alignment
+  // field.
+
+  // Store the alignment in the bitcode as a 16-bit raw value instead of a 5-bit
+  // log2 encoded value. Shift the bits above the alignment up by 11 bits.
+  uint64_t EncodedAttrs = Attrs.Raw() & 0xffff;
+  if (Attrs.hasAttribute(Attributes::Alignment))
+    EncodedAttrs |= Attrs.getAlignment() << 16;
+  EncodedAttrs |= (Attrs.Raw() & (0xffffULL << 21)) << 11;
+  return EncodedAttrs;
+}
+
+/// decodeLLVMAttributesForBitcode - This returns an attribute bitset containing
+/// the LLVM attributes that have been decoded from the given integer.  This
+/// function must stay in sync with 'encodeLLVMAttributesForBitcode'.
+Attributes Attributes::decodeLLVMAttributesForBitcode(LLVMContext &C,
+                                                      uint64_t EncodedAttrs) {
+  // The alignment is stored as a 16-bit raw value from bits 31--16.  We shift
+  // the bits above 31 down by 11 bits.
+  unsigned Alignment = (EncodedAttrs & (0xffffULL << 16)) >> 16;
+  assert((!Alignment || isPowerOf2_32(Alignment)) &&
+         "Alignment must be a power of two.");
+
+  AttrBuilder B(EncodedAttrs & 0xffff);
+  if (Alignment)
+    B.addAlignmentAttr(Alignment);
+  B.addRawValue((EncodedAttrs & (0xffffULL << 32)) >> 11);
+  return Attributes::get(C, B);
+}
+
+std::string Attributes::getAsString() const {
   std::string Result;
-  if (Attrs & Attribute::ZExt)
+  if (hasAttribute(Attributes::ZExt))
     Result += "zeroext ";
-  if (Attrs & Attribute::SExt)
+  if (hasAttribute(Attributes::SExt))
     Result += "signext ";
-  if (Attrs & Attribute::NoReturn)
+  if (hasAttribute(Attributes::NoReturn))
     Result += "noreturn ";
-  if (Attrs & Attribute::NoUnwind)
+  if (hasAttribute(Attributes::NoUnwind))
     Result += "nounwind ";
-  if (Attrs & Attribute::UWTable)
+  if (hasAttribute(Attributes::UWTable))
     Result += "uwtable ";
-  if (Attrs & Attribute::ReturnsTwice)
+  if (hasAttribute(Attributes::ReturnsTwice))
     Result += "returns_twice ";
-  if (Attrs & Attribute::InReg)
+  if (hasAttribute(Attributes::InReg))
     Result += "inreg ";
-  if (Attrs & Attribute::NoAlias)
+  if (hasAttribute(Attributes::NoAlias))
     Result += "noalias ";
-  if (Attrs & Attribute::NoCapture)
+  if (hasAttribute(Attributes::NoCapture))
     Result += "nocapture ";
-  if (Attrs & Attribute::StructRet)
+  if (hasAttribute(Attributes::StructRet))
     Result += "sret ";
-  if (Attrs & Attribute::ByVal)
+  if (hasAttribute(Attributes::ByVal))
     Result += "byval ";
-  if (Attrs & Attribute::Nest)
+  if (hasAttribute(Attributes::Nest))
     Result += "nest ";
-  if (Attrs & Attribute::ReadNone)
+  if (hasAttribute(Attributes::ReadNone))
     Result += "readnone ";
-  if (Attrs & Attribute::ReadOnly)
+  if (hasAttribute(Attributes::ReadOnly))
     Result += "readonly ";
-  if (Attrs & Attribute::OptimizeForSize)
+  if (hasAttribute(Attributes::OptimizeForSize))
     Result += "optsize ";
-  if (Attrs & Attribute::NoInline)
+  if (hasAttribute(Attributes::NoInline))
     Result += "noinline ";
-  if (Attrs & Attribute::InlineHint)
+  if (hasAttribute(Attributes::InlineHint))
     Result += "inlinehint ";
-  if (Attrs & Attribute::AlwaysInline)
+  if (hasAttribute(Attributes::AlwaysInline))
     Result += "alwaysinline ";
-  if (Attrs & Attribute::StackProtect)
+  if (hasAttribute(Attributes::StackProtect))
     Result += "ssp ";
-  if (Attrs & Attribute::StackProtectReq)
+  if (hasAttribute(Attributes::StackProtectReq))
     Result += "sspreq ";
-  if (Attrs & Attribute::NoRedZone)
+  if (hasAttribute(Attributes::NoRedZone))
     Result += "noredzone ";
-  if (Attrs & Attribute::NoImplicitFloat)
+  if (hasAttribute(Attributes::NoImplicitFloat))
     Result += "noimplicitfloat ";
-  if (Attrs & Attribute::Naked)
+  if (hasAttribute(Attributes::Naked))
     Result += "naked ";
-  if (Attrs & Attribute::NonLazyBind)
+  if (hasAttribute(Attributes::NonLazyBind))
     Result += "nonlazybind ";
-  if (Attrs & Attribute::AddressSafety)
+  if (hasAttribute(Attributes::AddressSafety))
     Result += "address_safety ";
-  if (Attrs & Attribute::StackAlignment) {
+  if (hasAttribute(Attributes::MinSize))
+    Result += "minsize ";
+  if (hasAttribute(Attributes::StackAlignment)) {
     Result += "alignstack(";
-    Result += utostr(Attribute::getStackAlignmentFromAttrs(Attrs));
+    Result += utostr(getStackAlignment());
     Result += ") ";
   }
-  if (Attrs & Attribute::Alignment) {
+  if (hasAttribute(Attributes::Alignment)) {
     Result += "align ";
-    Result += utostr(Attribute::getAlignmentFromAttrs(Attrs));
+    Result += utostr(getAlignment());
     Result += " ";
   }
-  if (Attrs & Attribute::IANSDialect)
-    Result += "ia_nsdialect ";
-
   // Trim the trailing space.
   assert(!Result.empty() && "Unknown attribute!");
   Result.erase(Result.end()-1);
   return Result;
 }
 
-Attributes Attribute::typeIncompatible(Type *Ty) {
-  Attributes Incompatible = None;
-  
-  if (!Ty->isIntegerTy())
-    // Attributes that only apply to integers.
-    Incompatible |= SExt | ZExt;
-  
-  if (!Ty->isPointerTy())
-    // Attributes that only apply to pointers.
-    Incompatible |= ByVal | Nest | NoAlias | StructRet | NoCapture;
-  
-  return Incompatible;
-}
-
 //===----------------------------------------------------------------------===//
-// AttributeListImpl Definition
+// AttrBuilder Implementation
 //===----------------------------------------------------------------------===//
 
-namespace llvm {
-  class AttributeListImpl;
+AttrBuilder &AttrBuilder::addAttribute(Attributes::AttrVal Val){
+  Bits |= AttributesImpl::getAttrMask(Val);
+  return *this;
 }
 
-static ManagedStatic<FoldingSet<AttributeListImpl> > AttributesLists;
+AttrBuilder &AttrBuilder::addRawValue(uint64_t Val) {
+  Bits |= Val;
+  return *this;
+}
 
-namespace llvm {
-static ManagedStatic<sys::SmartMutex<true> > ALMutex;
+AttrBuilder &AttrBuilder::addAlignmentAttr(unsigned Align) {
+  if (Align == 0) return *this;
+  assert(isPowerOf2_32(Align) && "Alignment must be a power of two.");
+  assert(Align <= 0x40000000 && "Alignment too large.");
+  Bits |= (Log2_32(Align) + 1) << 16;
+  return *this;
+}
+AttrBuilder &AttrBuilder::addStackAlignmentAttr(unsigned Align){
+  // Default alignment, allow the target to define how to align it.
+  if (Align == 0) return *this;
+  assert(isPowerOf2_32(Align) && "Alignment must be a power of two.");
+  assert(Align <= 0x100 && "Alignment too large.");
+  Bits |= (Log2_32(Align) + 1) << 26;
+  return *this;
+}
 
-class AttributeListImpl : public FoldingSetNode {
-  sys::cas_flag RefCount;
-  
-  // AttributesList is uniqued, these should not be publicly available.
-  void operator=(const AttributeListImpl &); // Do not implement
-  AttributeListImpl(const AttributeListImpl &); // Do not implement
-  ~AttributeListImpl();                        // Private implementation
-public:
-  SmallVector<AttributeWithIndex, 4> Attrs;
-  
-  AttributeListImpl(ArrayRef<AttributeWithIndex> attrs)
-    : Attrs(attrs.begin(), attrs.end()) {
-    RefCount = 0;
-  }
-  
-  void AddRef() {
-    sys::SmartScopedLock<true> Lock(*ALMutex);
-    ++RefCount;
-  }
-  void DropRef() {
-    sys::SmartScopedLock<true> Lock(*ALMutex);
-    if (!AttributesLists.isConstructed())
-      return;
-    sys::cas_flag new_val = --RefCount;
-    if (new_val == 0)
-      delete this;
-  }
-  
-  void Profile(FoldingSetNodeID &ID) const {
-    Profile(ID, Attrs);
-  }
-  static void Profile(FoldingSetNodeID &ID, ArrayRef<AttributeWithIndex> Attrs){
-    for (unsigned i = 0, e = Attrs.size(); i != e; ++i) {
-      ID.AddInteger(Attrs[i].Attrs.Raw());
-      ID.AddInteger(Attrs[i].Index);
-    }
-  }
-};
+AttrBuilder &AttrBuilder::removeAttribute(Attributes::AttrVal Val) {
+  Bits &= ~AttributesImpl::getAttrMask(Val);
+  return *this;
 }
 
-AttributeListImpl::~AttributeListImpl() {
-  // NOTE: Lock must be acquired by caller.
-  AttributesLists->RemoveNode(this);
+AttrBuilder &AttrBuilder::addAttributes(const Attributes &A) {
+  Bits |= A.Raw();
+  return *this;
 }
 
+AttrBuilder &AttrBuilder::removeAttributes(const Attributes &A){
+  Bits &= ~A.Raw();
+  return *this;
+}
 
-AttrListPtr AttrListPtr::get(ArrayRef<AttributeWithIndex> Attrs) {
+bool AttrBuilder::hasAttribute(Attributes::AttrVal A) const {
+  return Bits & AttributesImpl::getAttrMask(A);
+}
+
+bool AttrBuilder::hasAttributes() const {
+  return Bits != 0;
+}
+bool AttrBuilder::hasAttributes(const Attributes &A) const {
+  return Bits & A.Raw();
+}
+bool AttrBuilder::hasAlignmentAttr() const {
+  return Bits & AttributesImpl::getAttrMask(Attributes::Alignment);
+}
+
+uint64_t AttrBuilder::getAlignment() const {
+  if (!hasAlignmentAttr())
+    return 0;
+  return 1U <<
+    (((Bits & AttributesImpl::getAttrMask(Attributes::Alignment)) >> 16) - 1);
+}
+
+uint64_t AttrBuilder::getStackAlignment() const {
+  if (!hasAlignmentAttr())
+    return 0;
+  return 1U <<
+    (((Bits & AttributesImpl::getAttrMask(Attributes::StackAlignment))>>26)-1);
+}
+
+//===----------------------------------------------------------------------===//
+// AttributeImpl Definition
+//===----------------------------------------------------------------------===//
+
+uint64_t AttributesImpl::getAttrMask(uint64_t Val) {
+  switch (Val) {
+  case Attributes::None:            return 0;
+  case Attributes::ZExt:            return 1 << 0;
+  case Attributes::SExt:            return 1 << 1;
+  case Attributes::NoReturn:        return 1 << 2;
+  case Attributes::InReg:           return 1 << 3;
+  case Attributes::StructRet:       return 1 << 4;
+  case Attributes::NoUnwind:        return 1 << 5;
+  case Attributes::NoAlias:         return 1 << 6;
+  case Attributes::ByVal:           return 1 << 7;
+  case Attributes::Nest:            return 1 << 8;
+  case Attributes::ReadNone:        return 1 << 9;
+  case Attributes::ReadOnly:        return 1 << 10;
+  case Attributes::NoInline:        return 1 << 11;
+  case Attributes::AlwaysInline:    return 1 << 12;
+  case Attributes::OptimizeForSize: return 1 << 13;
+  case Attributes::StackProtect:    return 1 << 14;
+  case Attributes::StackProtectReq: return 1 << 15;
+  case Attributes::Alignment:       return 31 << 16;
+  case Attributes::NoCapture:       return 1 << 21;
+  case Attributes::NoRedZone:       return 1 << 22;
+  case Attributes::NoImplicitFloat: return 1 << 23;
+  case Attributes::Naked:           return 1 << 24;
+  case Attributes::InlineHint:      return 1 << 25;
+  case Attributes::StackAlignment:  return 7 << 26;
+  case Attributes::ReturnsTwice:    return 1 << 29;
+  case Attributes::UWTable:         return 1 << 30;
+  case Attributes::NonLazyBind:     return 1U << 31;
+  case Attributes::AddressSafety:   return 1ULL << 32;
+  case Attributes::MinSize:         return 1ULL << 33;
+  }
+  llvm_unreachable("Unsupported attribute type");
+}
+
+bool AttributesImpl::hasAttribute(uint64_t A) const {
+  return (Bits & getAttrMask(A)) != 0;
+}
+
+bool AttributesImpl::hasAttributes() const {
+  return Bits != 0;
+}
+
+bool AttributesImpl::hasAttributes(const Attributes &A) const {
+  return Bits & A.Raw();        // FIXME: Raw() won't work here in the future.
+}
+
+uint64_t AttributesImpl::getAlignment() const {
+  return Bits & getAttrMask(Attributes::Alignment);
+}
+
+uint64_t AttributesImpl::getStackAlignment() const {
+  return Bits & getAttrMask(Attributes::StackAlignment);
+}
+
+//===----------------------------------------------------------------------===//
+// AttributeListImpl Definition
+//===----------------------------------------------------------------------===//
+
+AttrListPtr AttrListPtr::get(LLVMContext &C,
+                             ArrayRef<AttributeWithIndex> Attrs) {
   // If there are no attributes then return a null AttributesList pointer.
   if (Attrs.empty())
     return AttrListPtr();
-  
+
 #ifndef NDEBUG
   for (unsigned i = 0, e = Attrs.size(); i != e; ++i) {
-    assert(Attrs[i].Attrs != Attribute::None && 
+    assert(Attrs[i].Attrs.hasAttributes() &&
            "Pointless attribute!");
     assert((!i || Attrs[i-1].Index < Attrs[i].Index) &&
            "Misordered AttributesList!");
   }
 #endif
-  
+
   // Otherwise, build a key to look up the existing attributes.
+  LLVMContextImpl *pImpl = C.pImpl;
   FoldingSetNodeID ID;
   AttributeListImpl::Profile(ID, Attrs);
-  void *InsertPos;
-  
-  sys::SmartScopedLock<true> Lock(*ALMutex);
-  
-  AttributeListImpl *PAL =
-    AttributesLists->FindNodeOrInsertPos(ID, InsertPos);
-  
+
+  void *InsertPoint;
+  AttributeListImpl *PA = pImpl->AttrsLists.FindNodeOrInsertPos(ID,
+                                                                InsertPoint);
+
   // If we didn't find any existing attributes of the same shape then
   // create a new one and insert it.
-  if (!PAL) {
-    PAL = new AttributeListImpl(Attrs);
-    AttributesLists->InsertNode(PAL, InsertPos);
+  if (!PA) {
+    PA = new AttributeListImpl(Attrs);
+    pImpl->AttrsLists.InsertNode(PA, InsertPoint);
   }
-  
+
   // Return the AttributesList that we found or created.
-  return AttrListPtr(PAL);
+  return AttrListPtr(PA);
 }
 
-
 //===----------------------------------------------------------------------===//
 // AttrListPtr Method Implementations
 //===----------------------------------------------------------------------===//
 
-AttrListPtr::AttrListPtr(AttributeListImpl *LI) : AttrList(LI) {
-  if (LI) LI->AddRef();
-}
-
-AttrListPtr::AttrListPtr(const AttrListPtr &P) : AttrList(P.AttrList) {
-  if (AttrList) AttrList->AddRef();  
-}
-
 const AttrListPtr &AttrListPtr::operator=(const AttrListPtr &RHS) {
-  sys::SmartScopedLock<true> Lock(*ALMutex);
   if (AttrList == RHS.AttrList) return *this;
-  if (AttrList) AttrList->DropRef();
+
   AttrList = RHS.AttrList;
-  if (AttrList) AttrList->AddRef();
   return *this;
 }
 
-AttrListPtr::~AttrListPtr() {
-  if (AttrList) AttrList->DropRef();
-}
-
-/// getNumSlots - Return the number of slots used in this attribute list. 
+/// getNumSlots - Return the number of slots used in this attribute list.
 /// This is the number of arguments that have an attribute set on them
 /// (including the function itself).
 unsigned AttrListPtr::getNumSlots() const {
@@ -245,48 +415,60 @@
   return AttrList->Attrs[Slot];
 }
 
-
-/// getAttributes - The attributes for the specified index are
-/// returned.  Attributes for the result are denoted with Idx = 0.
-/// Function notes are denoted with idx = ~0.
+/// getAttributes - The attributes for the specified index are returned.
+/// Attributes for the result are denoted with Idx = 0.  Function notes are
+/// denoted with idx = ~0.
 Attributes AttrListPtr::getAttributes(unsigned Idx) const {
-  if (AttrList == 0) return Attribute::None;
-  
+  if (AttrList == 0) return Attributes();
+
   const SmallVector<AttributeWithIndex, 4> &Attrs = AttrList->Attrs;
   for (unsigned i = 0, e = Attrs.size(); i != e && Attrs[i].Index <= Idx; ++i)
     if (Attrs[i].Index == Idx)
       return Attrs[i].Attrs;
-  return Attribute::None;
+
+  return Attributes();
 }
 
 /// hasAttrSomewhere - Return true if the specified attribute is set for at
 /// least one parameter or for the return value.
-bool AttrListPtr::hasAttrSomewhere(Attributes Attr) const {
+bool AttrListPtr::hasAttrSomewhere(Attributes::AttrVal Attr) const {
   if (AttrList == 0) return false;
-  
+
   const SmallVector<AttributeWithIndex, 4> &Attrs = AttrList->Attrs;
   for (unsigned i = 0, e = Attrs.size(); i != e; ++i)
-    if (Attrs[i].Attrs & Attr)
+    if (Attrs[i].Attrs.hasAttribute(Attr))
       return true;
+
   return false;
 }
 
+unsigned AttrListPtr::getNumAttrs() const {
+  return AttrList ? AttrList->Attrs.size() : 0;
+}
+
+Attributes &AttrListPtr::getAttributesAtIndex(unsigned i) const {
+  assert(AttrList && "Trying to get an attribute from an empty list!");
+  assert(i < AttrList->Attrs.size() && "Index out of range!");
+  return AttrList->Attrs[i].Attrs;
+}
 
-AttrListPtr AttrListPtr::addAttr(unsigned Idx, Attributes Attrs) const {
+AttrListPtr AttrListPtr::addAttr(LLVMContext &C, unsigned Idx,
+                                 Attributes Attrs) const {
   Attributes OldAttrs = getAttributes(Idx);
 #ifndef NDEBUG
   // FIXME it is not obvious how this should work for alignment.
   // For now, say we can't change a known alignment.
-  Attributes OldAlign = OldAttrs & Attribute::Alignment;
-  Attributes NewAlign = Attrs & Attribute::Alignment;
+  unsigned OldAlign = OldAttrs.getAlignment();
+  unsigned NewAlign = Attrs.getAlignment();
   assert((!OldAlign || !NewAlign || OldAlign == NewAlign) &&
          "Attempt to change alignment!");
 #endif
-  
-  Attributes NewAttrs = OldAttrs | Attrs;
-  if (NewAttrs == OldAttrs)
+
+  AttrBuilder NewAttrs =
+    AttrBuilder(OldAttrs).addAttributes(Attrs);
+  if (NewAttrs == AttrBuilder(OldAttrs))
     return *this;
-  
+
   SmallVector<AttributeWithIndex, 8> NewAttrList;
   if (AttrList == 0)
     NewAttrList.push_back(AttributeWithIndex::get(Idx, Attrs));
@@ -299,61 +481,67 @@
 
     // If there are attributes already at this index, merge them in.
     if (i != e && OldAttrList[i].Index == Idx) {
-      Attrs |= OldAttrList[i].Attrs;
+      Attrs =
+        Attributes::get(C, AttrBuilder(Attrs).
+                        addAttributes(OldAttrList[i].Attrs));
       ++i;
     }
-    
+
     NewAttrList.push_back(AttributeWithIndex::get(Idx, Attrs));
-    
+
     // Copy attributes for arguments after this one.
-    NewAttrList.insert(NewAttrList.end(), 
+    NewAttrList.insert(NewAttrList.end(),
                        OldAttrList.begin()+i, OldAttrList.end());
   }
-  
-  return get(NewAttrList);
+
+  return get(C, NewAttrList);
 }
 
-AttrListPtr AttrListPtr::removeAttr(unsigned Idx, Attributes Attrs) const {
+AttrListPtr AttrListPtr::removeAttr(LLVMContext &C, unsigned Idx,
+                                    Attributes Attrs) const {
 #ifndef NDEBUG
   // FIXME it is not obvious how this should work for alignment.
   // For now, say we can't pass in alignment, which no current use does.
-  assert(!(Attrs & Attribute::Alignment) && "Attempt to exclude alignment!");
+  assert(!Attrs.hasAttribute(Attributes::Alignment) &&
+         "Attempt to exclude alignment!");
 #endif
   if (AttrList == 0) return AttrListPtr();
-  
+
   Attributes OldAttrs = getAttributes(Idx);
-  Attributes NewAttrs = OldAttrs & ~Attrs;
-  if (NewAttrs == OldAttrs)
+  AttrBuilder NewAttrs =
+    AttrBuilder(OldAttrs).removeAttributes(Attrs);
+  if (NewAttrs == AttrBuilder(OldAttrs))
     return *this;
 
   SmallVector<AttributeWithIndex, 8> NewAttrList;
   const SmallVector<AttributeWithIndex, 4> &OldAttrList = AttrList->Attrs;
   unsigned i = 0, e = OldAttrList.size();
-  
+
   // Copy attributes for arguments before this one.
   for (; i != e && OldAttrList[i].Index < Idx; ++i)
     NewAttrList.push_back(OldAttrList[i]);
-  
+
   // If there are attributes already at this index, merge them in.
   assert(OldAttrList[i].Index == Idx && "Attribute isn't set?");
-  Attrs = OldAttrList[i].Attrs & ~Attrs;
+  Attrs = Attributes::get(C, AttrBuilder(OldAttrList[i].Attrs).
+                          removeAttributes(Attrs));
   ++i;
-  if (Attrs)  // If any attributes left for this parameter, add them.
+  if (Attrs.hasAttributes()) // If any attributes left for this param, add them.
     NewAttrList.push_back(AttributeWithIndex::get(Idx, Attrs));
-  
+
   // Copy attributes for arguments after this one.
-  NewAttrList.insert(NewAttrList.end(), 
+  NewAttrList.insert(NewAttrList.end(),
                      OldAttrList.begin()+i, OldAttrList.end());
-  
-  return get(NewAttrList);
+
+  return get(C, NewAttrList);
 }
 
 void AttrListPtr::dump() const {
   dbgs() << "PAL[ ";
   for (unsigned i = 0; i < getNumSlots(); ++i) {
     const AttributeWithIndex &PAWI = getSlot(i);
-    dbgs() << "{" << PAWI.Index << "," << PAWI.Attrs << "} ";
+    dbgs() << "{" << PAWI.Index << "," << PAWI.Attrs.getAsString() << "} ";
   }
-  
+
   dbgs() << "]\n";
 }

Copied: llvm/branches/AMDILBackend/lib/VMCore/AttributesImpl.h (from r167701, llvm/trunk/lib/VMCore/AttributesImpl.h)
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/VMCore/AttributesImpl.h?p2=llvm/branches/AMDILBackend/lib/VMCore/AttributesImpl.h&p1=llvm/trunk/lib/VMCore/AttributesImpl.h&r1=167701&r2=172541&rev=172541&view=diff
==============================================================================
--- llvm/trunk/lib/VMCore/AttributesImpl.h (original)
+++ llvm/branches/AMDILBackend/lib/VMCore/AttributesImpl.h Tue Jan 15 11:16:16 2013
@@ -15,12 +15,11 @@
 #ifndef LLVM_ATTRIBUTESIMPL_H
 #define LLVM_ATTRIBUTESIMPL_H
 
+#include "llvm/Attributes.h"
 #include "llvm/ADT/FoldingSet.h"
 
 namespace llvm {
 
-class Attributes;
-
 class AttributesImpl : public FoldingSetNode {
   uint64_t Bits;                // FIXME: We will be expanding this.
 public:
@@ -46,6 +45,27 @@
   }
 };
 
+class AttributeListImpl : public FoldingSetNode {
+  // AttributesList is uniqued, these should not be publicly available.
+  void operator=(const AttributeListImpl &) LLVM_DELETED_FUNCTION;
+  AttributeListImpl(const AttributeListImpl &) LLVM_DELETED_FUNCTION;
+public:
+  SmallVector<AttributeWithIndex, 4> Attrs;
+
+  AttributeListImpl(ArrayRef<AttributeWithIndex> attrs)
+    : Attrs(attrs.begin(), attrs.end()) {}
+
+  void Profile(FoldingSetNodeID &ID) const {
+    Profile(ID, Attrs);
+  }
+  static void Profile(FoldingSetNodeID &ID, ArrayRef<AttributeWithIndex> Attrs){
+    for (unsigned i = 0, e = Attrs.size(); i != e; ++i) {
+      ID.AddInteger(Attrs[i].Attrs.Raw());
+      ID.AddInteger(Attrs[i].Index);
+    }
+  }
+};
+
 } // end llvm namespace
 
 #endif

Modified: llvm/branches/AMDILBackend/lib/VMCore/AutoUpgrade.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/VMCore/AutoUpgrade.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/VMCore/AutoUpgrade.cpp (original)
+++ llvm/branches/AMDILBackend/lib/VMCore/AutoUpgrade.cpp Tue Jan 15 11:16:16 2013
@@ -148,7 +148,8 @@
   if (NewFn)
     F = NewFn;
   if (unsigned id = F->getIntrinsicID())
-    F->setAttributes(Intrinsic::getAttributes((Intrinsic::ID)id));
+    F->setAttributes(Intrinsic::getAttributes(F->getContext(),
+                                              (Intrinsic::ID)id));
   return Upgraded;
 }
 

Modified: llvm/branches/AMDILBackend/lib/VMCore/CMakeLists.txt
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/VMCore/CMakeLists.txt?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/VMCore/CMakeLists.txt (original)
+++ llvm/branches/AMDILBackend/lib/VMCore/CMakeLists.txt Tue Jan 15 11:16:16 2013
@@ -1,5 +1,3 @@
-set(LLVM_REQUIRES_RTTI 1)
-
 add_llvm_library(LLVMCore
   AsmWriter.cpp
   Attributes.cpp
@@ -8,6 +6,7 @@
   ConstantFold.cpp
   Constants.cpp
   Core.cpp
+  DataLayout.cpp
   DebugInfo.cpp
   DebugLoc.cpp
   DIBuilder.cpp
@@ -32,6 +31,7 @@
   PrintModulePass.cpp
   Type.cpp
   TypeFinder.cpp
+  TargetTransformInfo.cpp
   Use.cpp
   User.cpp
   Value.cpp
@@ -42,7 +42,7 @@
 
 # Workaround: It takes over 20 minutes to compile with msvc10.
 # FIXME: Suppressing optimizations to core libraries would not be good thing.
-if( MSVC_VERSION EQUAL 1600 )
+if( MSVC_VERSION LESS 1700 )
 set_property(
   SOURCE Function.cpp
   PROPERTY COMPILE_FLAGS "/Og-"

Modified: llvm/branches/AMDILBackend/lib/VMCore/ConstantFold.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/VMCore/ConstantFold.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/VMCore/ConstantFold.cpp (original)
+++ llvm/branches/AMDILBackend/lib/VMCore/ConstantFold.cpp Tue Jan 15 11:16:16 2013
@@ -12,7 +12,7 @@
 // ConstantExpr::get* methods to automatically fold constants when possible.
 //
 // The current constant folding implementation is implemented in two pieces: the
-// pieces that don't need TargetData, and the pieces that do. This is to avoid
+// pieces that don't need DataLayout, and the pieces that do. This is to avoid
 // a dependence in VMCore on Target.
 //
 //===----------------------------------------------------------------------===//
@@ -87,9 +87,13 @@
   Instruction::CastOps firstOp = Instruction::CastOps(Op->getOpcode());
   Instruction::CastOps secondOp = Instruction::CastOps(opc);
 
+  // Assume that pointers are never more than 64 bits wide.
+  IntegerType *FakeIntPtrTy = Type::getInt64Ty(DstTy->getContext());
+
   // Let CastInst::isEliminableCastPair do the heavy lifting.
   return CastInst::isEliminableCastPair(firstOp, secondOp, SrcTy, MidTy, DstTy,
-                                        Type::getInt64Ty(DstTy->getContext()));
+                                        FakeIntPtrTy, FakeIntPtrTy,
+                                        FakeIntPtrTy);
 }
 
 static Constant *FoldBitCast(Constant *V, Type *DestTy) {
@@ -514,10 +518,6 @@
     return UndefValue::get(DestTy);
   }
 
-  // No compile-time operations on this type yet.
-  if (V->getType()->isPPC_FP128Ty() || DestTy->isPPC_FP128Ty())
-    return 0;
-
   if (V->isNullValue() && !DestTy->isX86_MMXTy())
     return Constant::getNullValue(DestTy);
 
@@ -576,6 +576,7 @@
                   DestTy->isDoubleTy() ? APFloat::IEEEdouble :
                   DestTy->isX86_FP80Ty() ? APFloat::x87DoubleExtended :
                   DestTy->isFP128Ty() ? APFloat::IEEEquad :
+                  DestTy->isPPC_FP128Ty() ? APFloat::PPCDoubleDouble :
                   APFloat::Bogus,
                   APFloat::rmNearestTiesToEven, &ignored);
       return ConstantFP::get(V->getContext(), Val);
@@ -646,7 +647,8 @@
   case Instruction::SIToFP:
     if (ConstantInt *CI = dyn_cast<ConstantInt>(V)) {
       APInt api = CI->getValue();
-      APFloat apf(APInt::getNullValue(DestTy->getPrimitiveSizeInBits()), true);
+      APFloat apf(APInt::getNullValue(DestTy->getPrimitiveSizeInBits()),
+                  !DestTy->isPPC_FP128Ty() /* isEEEE */);
       (void)apf.convertFromAPInt(api, 
                                  opc==Instruction::SIToFP,
                                  APFloat::rmNearestTiesToEven);
@@ -867,10 +869,6 @@
 
 Constant *llvm::ConstantFoldBinaryInstruction(unsigned Opcode,
                                               Constant *C1, Constant *C2) {
-  // No compile-time operations on this type yet.
-  if (C1->getType()->isPPC_FP128Ty())
-    return 0;
-
   // Handle UndefValue up front.
   if (isa<UndefValue>(C1) || isa<UndefValue>(C2)) {
     switch (Opcode) {
@@ -1273,10 +1271,6 @@
   assert(V1->getType() == V2->getType() &&
          "Cannot compare values of different types!");
 
-  // No compile-time operations on this type yet.
-  if (V1->getType()->isPPC_FP128Ty())
-    return FCmpInst::BAD_FCMP_PREDICATE;
-
   // Handle degenerate case quickly
   if (V1 == V2) return FCmpInst::FCMP_OEQ;
 
@@ -1602,10 +1596,6 @@
     return ConstantInt::get(ResultTy, CmpInst::isTrueWhenEqual(pred));
   }
 
-  // No compile-time operations on this type yet.
-  if (C1->getType()->isPPC_FP128Ty())
-    return 0;
-
   // icmp eq/ne(null,GV) -> false/true
   if (C1->isNullValue()) {
     if (const GlobalValue *GV = dyn_cast<GlobalValue>(C2))

Modified: llvm/branches/AMDILBackend/lib/VMCore/Constants.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/VMCore/Constants.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/VMCore/Constants.cpp (original)
+++ llvm/branches/AMDILBackend/lib/VMCore/Constants.cpp Tue Jan 15 11:16:16 2013
@@ -245,6 +245,33 @@
   }
 }
 
+/// isThreadDependent - Return true if the value can vary between threads.
+bool Constant::isThreadDependent() const {
+  SmallPtrSet<const Constant*, 64> Visited;
+  SmallVector<const Constant*, 64> WorkList;
+  WorkList.push_back(this);
+  Visited.insert(this);
+
+  while (!WorkList.empty()) {
+    const Constant *C = WorkList.pop_back_val();
+
+    if (const GlobalVariable *GV = dyn_cast<GlobalVariable>(C)) {
+      if (GV->isThreadLocal())
+        return true;
+    }
+
+    for (unsigned I = 0, E = C->getNumOperands(); I != E; ++I) {
+      const Constant *D = dyn_cast<Constant>(C->getOperand(I));
+      if (!D)
+        continue;
+      if (Visited.insert(D))
+        WorkList.push_back(D);
+    }
+  }
+
+  return false;
+}
+
 /// isConstantUsed - Return true if the constant has users other than constant
 /// exprs and other dangling things.
 bool Constant::isConstantUsed() const {

Modified: llvm/branches/AMDILBackend/lib/VMCore/ConstantsContext.h
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/VMCore/ConstantsContext.h?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/VMCore/ConstantsContext.h (original)
+++ llvm/branches/AMDILBackend/lib/VMCore/ConstantsContext.h Tue Jan 15 11:16:16 2013
@@ -33,7 +33,7 @@
 /// behind the scenes to implement unary constant exprs.
 class UnaryConstantExpr : public ConstantExpr {
   virtual void anchor();
-  void *operator new(size_t, unsigned);  // DO NOT IMPLEMENT
+  void *operator new(size_t, unsigned) LLVM_DELETED_FUNCTION;
 public:
   // allocate space for exactly one operand
   void *operator new(size_t s) {
@@ -50,7 +50,7 @@
 /// behind the scenes to implement binary constant exprs.
 class BinaryConstantExpr : public ConstantExpr {
   virtual void anchor();
-  void *operator new(size_t, unsigned);  // DO NOT IMPLEMENT
+  void *operator new(size_t, unsigned) LLVM_DELETED_FUNCTION;
 public:
   // allocate space for exactly two operands
   void *operator new(size_t s) {
@@ -71,7 +71,7 @@
 /// behind the scenes to implement select constant exprs.
 class SelectConstantExpr : public ConstantExpr {
   virtual void anchor();
-  void *operator new(size_t, unsigned);  // DO NOT IMPLEMENT
+  void *operator new(size_t, unsigned) LLVM_DELETED_FUNCTION;
 public:
   // allocate space for exactly three operands
   void *operator new(size_t s) {
@@ -92,7 +92,7 @@
 /// extractelement constant exprs.
 class ExtractElementConstantExpr : public ConstantExpr {
   virtual void anchor();
-  void *operator new(size_t, unsigned);  // DO NOT IMPLEMENT
+  void *operator new(size_t, unsigned) LLVM_DELETED_FUNCTION;
 public:
   // allocate space for exactly two operands
   void *operator new(size_t s) {
@@ -113,7 +113,7 @@
 /// insertelement constant exprs.
 class InsertElementConstantExpr : public ConstantExpr {
   virtual void anchor();
-  void *operator new(size_t, unsigned);  // DO NOT IMPLEMENT
+  void *operator new(size_t, unsigned) LLVM_DELETED_FUNCTION;
 public:
   // allocate space for exactly three operands
   void *operator new(size_t s) {
@@ -135,7 +135,7 @@
 /// shufflevector constant exprs.
 class ShuffleVectorConstantExpr : public ConstantExpr {
   virtual void anchor();
-  void *operator new(size_t, unsigned);  // DO NOT IMPLEMENT
+  void *operator new(size_t, unsigned) LLVM_DELETED_FUNCTION;
 public:
   // allocate space for exactly three operands
   void *operator new(size_t s) {
@@ -160,7 +160,7 @@
 /// extractvalue constant exprs.
 class ExtractValueConstantExpr : public ConstantExpr {
   virtual void anchor();
-  void *operator new(size_t, unsigned);  // DO NOT IMPLEMENT
+  void *operator new(size_t, unsigned) LLVM_DELETED_FUNCTION;
 public:
   // allocate space for exactly one operand
   void *operator new(size_t s) {
@@ -186,7 +186,7 @@
 /// insertvalue constant exprs.
 class InsertValueConstantExpr : public ConstantExpr {
   virtual void anchor();
-  void *operator new(size_t, unsigned);  // DO NOT IMPLEMENT
+  void *operator new(size_t, unsigned) LLVM_DELETED_FUNCTION;
 public:
   // allocate space for exactly one operand
   void *operator new(size_t s) {
@@ -234,7 +234,7 @@
 // needed in order to store the predicate value for these instructions.
 class CompareConstantExpr : public ConstantExpr {
   virtual void anchor();
-  void *operator new(size_t, unsigned);  // DO NOT IMPLEMENT
+  void *operator new(size_t, unsigned) LLVM_DELETED_FUNCTION;
 public:
   // allocate space for exactly two operands
   void *operator new(size_t s) {
@@ -352,18 +352,21 @@
 struct InlineAsmKeyType {
   InlineAsmKeyType(StringRef AsmString,
                    StringRef Constraints, bool hasSideEffects,
-                   bool isAlignStack)
+                   bool isAlignStack, InlineAsm::AsmDialect asmDialect)
     : asm_string(AsmString), constraints(Constraints),
-      has_side_effects(hasSideEffects), is_align_stack(isAlignStack) {}
+      has_side_effects(hasSideEffects), is_align_stack(isAlignStack),
+      asm_dialect(asmDialect) {}
   std::string asm_string;
   std::string constraints;
   bool has_side_effects;
   bool is_align_stack;
+  InlineAsm::AsmDialect asm_dialect;
   bool operator==(const InlineAsmKeyType& that) const {
     return this->asm_string == that.asm_string &&
            this->constraints == that.constraints &&
            this->has_side_effects == that.has_side_effects &&
-           this->is_align_stack == that.is_align_stack;
+           this->is_align_stack == that.is_align_stack &&
+           this->asm_dialect == that.asm_dialect;
   }
   bool operator<(const InlineAsmKeyType& that) const {
     if (this->asm_string != that.asm_string)
@@ -374,6 +377,8 @@
       return this->has_side_effects < that.has_side_effects;
     if (this->is_align_stack != that.is_align_stack)
       return this->is_align_stack < that.is_align_stack;
+    if (this->asm_dialect != that.asm_dialect)
+      return this->asm_dialect < that.asm_dialect;
     return false;
   }
 
@@ -490,7 +495,8 @@
 struct ConstantCreator<InlineAsm, PointerType, InlineAsmKeyType> {
   static InlineAsm *create(PointerType *Ty, const InlineAsmKeyType &Key) {
     return new InlineAsm(Ty, Key.asm_string, Key.constraints,
-                         Key.has_side_effects, Key.is_align_stack);
+                         Key.has_side_effects, Key.is_align_stack,
+                         Key.asm_dialect);
   }
 };
 
@@ -499,7 +505,8 @@
   typedef InlineAsmKeyType ValType;
   static ValType getValType(InlineAsm *Asm) {
     return InlineAsmKeyType(Asm->getAsmString(), Asm->getConstraintString(),
-                            Asm->hasSideEffects(), Asm->isAlignStack());
+                            Asm->hasSideEffects(), Asm->isAlignStack(),
+                            Asm->getDialect());
   }
 };
 

Modified: llvm/branches/AMDILBackend/lib/VMCore/Core.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/VMCore/Core.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/VMCore/Core.cpp (original)
+++ llvm/branches/AMDILBackend/lib/VMCore/Core.cpp Tue Jan 15 11:16:16 2013
@@ -568,6 +568,19 @@
   return 0;
 }
 
+unsigned LLVMGetMDNodeNumOperands(LLVMValueRef V)
+{
+  return cast<MDNode>(unwrap(V))->getNumOperands();
+}
+
+void LLVMGetMDNodeOperands(LLVMValueRef V, LLVMValueRef *Dest)
+{
+  const MDNode *N = cast<MDNode>(unwrap(V));
+  const unsigned numOperands = N->getNumOperands();
+  for (unsigned i = 0; i < numOperands; i++)
+    Dest[i] = wrap(N->getOperand(i));
+}
+
 unsigned LLVMGetNamedMetadataNumOperands(LLVMModuleRef M, const char* name)
 {
   if (NamedMDNode *N = unwrap(M)->getNamedMetadata(name)) {
@@ -1084,6 +1097,8 @@
     return LLVMLinkOnceAnyLinkage;
   case GlobalValue::LinkOnceODRLinkage:
     return LLVMLinkOnceODRLinkage;
+  case GlobalValue::LinkOnceODRAutoHideLinkage:
+    return LLVMLinkOnceODRAutoHideLinkage;
   case GlobalValue::WeakAnyLinkage:
     return LLVMWeakAnyLinkage;
   case GlobalValue::WeakODRLinkage:
@@ -1098,8 +1113,6 @@
     return LLVMLinkerPrivateLinkage;
   case GlobalValue::LinkerPrivateWeakLinkage:
     return LLVMLinkerPrivateWeakLinkage;
-  case GlobalValue::LinkerPrivateWeakDefAutoLinkage:
-    return LLVMLinkerPrivateWeakDefAutoLinkage;
   case GlobalValue::DLLImportLinkage:
     return LLVMDLLImportLinkage;
   case GlobalValue::DLLExportLinkage:
@@ -1129,6 +1142,9 @@
   case LLVMLinkOnceODRLinkage:
     GV->setLinkage(GlobalValue::LinkOnceODRLinkage);
     break;
+  case LLVMLinkOnceODRAutoHideLinkage:
+    GV->setLinkage(GlobalValue::LinkOnceODRAutoHideLinkage);
+    break;
   case LLVMWeakAnyLinkage:
     GV->setLinkage(GlobalValue::WeakAnyLinkage);
     break;
@@ -1150,9 +1166,6 @@
   case LLVMLinkerPrivateWeakLinkage:
     GV->setLinkage(GlobalValue::LinkerPrivateWeakLinkage);
     break;
-  case LLVMLinkerPrivateWeakDefAutoLinkage:
-    GV->setLinkage(GlobalValue::LinkerPrivateWeakDefAutoLinkage);
-    break;
   case LLVMDLLImportLinkage:
     GV->setLinkage(GlobalValue::DLLImportLinkage);
     break;
@@ -1368,14 +1381,20 @@
 void LLVMAddFunctionAttr(LLVMValueRef Fn, LLVMAttribute PA) {
   Function *Func = unwrap<Function>(Fn);
   const AttrListPtr PAL = Func->getAttributes();
-  const AttrListPtr PALnew = PAL.addAttr(~0U, Attributes(PA));
+  AttrBuilder B(PA);
+  const AttrListPtr PALnew =
+    PAL.addAttr(Func->getContext(), AttrListPtr::FunctionIndex,
+                Attributes::get(Func->getContext(), B));
   Func->setAttributes(PALnew);
 }
 
 void LLVMRemoveFunctionAttr(LLVMValueRef Fn, LLVMAttribute PA) {
   Function *Func = unwrap<Function>(Fn);
   const AttrListPtr PAL = Func->getAttributes();
-  const AttrListPtr PALnew = PAL.removeAttr(~0U, Attributes(PA));
+  AttrBuilder B(PA);
+  const AttrListPtr PALnew =
+    PAL.removeAttr(Func->getContext(), AttrListPtr::FunctionIndex,
+                   Attributes::get(Func->getContext(), B));
   Func->setAttributes(PALnew);
 }
 
@@ -1445,11 +1464,15 @@
 }
 
 void LLVMAddAttribute(LLVMValueRef Arg, LLVMAttribute PA) {
-  unwrap<Argument>(Arg)->addAttr(Attributes(PA));
+  Argument *A = unwrap<Argument>(Arg);
+  AttrBuilder B(PA);
+  A->addAttr(Attributes::get(A->getContext(), B));
 }
 
 void LLVMRemoveAttribute(LLVMValueRef Arg, LLVMAttribute PA) {
-  unwrap<Argument>(Arg)->removeAttr(Attributes(PA));
+  Argument *A = unwrap<Argument>(Arg);
+  AttrBuilder B(PA);
+  A->removeAttr(Attributes::get(A->getContext(), B));
 }
 
 LLVMAttribute LLVMGetAttribute(LLVMValueRef Arg) {
@@ -1461,8 +1484,10 @@
   
 
 void LLVMSetParamAlignment(LLVMValueRef Arg, unsigned align) {
-  unwrap<Argument>(Arg)->addAttr(
-          Attribute::constructAlignmentFromInt(align));
+  AttrBuilder B;
+  B.addAlignmentAttr(align);
+  unwrap<Argument>(Arg)->addAttr(Attributes::
+                                 get(unwrap<Argument>(Arg)->getContext(), B));
 }
 
 /*--.. Operations on basic blocks ..........................................--*/
@@ -1651,23 +1676,28 @@
 void LLVMAddInstrAttribute(LLVMValueRef Instr, unsigned index, 
                            LLVMAttribute PA) {
   CallSite Call = CallSite(unwrap<Instruction>(Instr));
+  AttrBuilder B(PA);
   Call.setAttributes(
-    Call.getAttributes().addAttr(index, Attributes(PA)));
+    Call.getAttributes().addAttr(Call->getContext(), index,
+                                 Attributes::get(Call->getContext(), B)));
 }
 
 void LLVMRemoveInstrAttribute(LLVMValueRef Instr, unsigned index, 
                               LLVMAttribute PA) {
   CallSite Call = CallSite(unwrap<Instruction>(Instr));
+  AttrBuilder B(PA);
   Call.setAttributes(
-    Call.getAttributes().removeAttr(index, Attributes(PA)));
+    Call.getAttributes().removeAttr(Call->getContext(), index,
+                                    Attributes::get(Call->getContext(), B)));
 }
 
 void LLVMSetInstrParamAlignment(LLVMValueRef Instr, unsigned index, 
                                 unsigned align) {
   CallSite Call = CallSite(unwrap<Instruction>(Instr));
-  Call.setAttributes(
-    Call.getAttributes().addAttr(index, 
-        Attribute::constructAlignmentFromInt(align)));
+  AttrBuilder B;
+  B.addAlignmentAttr(align);
+  Call.setAttributes(Call.getAttributes().addAttr(Call->getContext(), index,
+                                       Attributes::get(Call->getContext(), B)));
 }
 
 /*--.. Operations on call instructions (only) ..............................--*/

Modified: llvm/branches/AMDILBackend/lib/VMCore/DIBuilder.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/VMCore/DIBuilder.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/VMCore/DIBuilder.cpp (original)
+++ llvm/branches/AMDILBackend/lib/VMCore/DIBuilder.cpp Tue Jan 15 11:16:16 2013
@@ -492,7 +492,8 @@
     NULL,
     Elements,
     ConstantInt::get(Type::getInt32Ty(VMContext), RunTimeLang),
-    Constant::getNullValue(Type::getInt32Ty(VMContext))
+    ConstantInt::get(Type::getInt32Ty(VMContext), 0),
+    ConstantInt::get(Type::getInt32Ty(VMContext), 0),
   };
   return DIType(MDNode::get(VMContext, Elts));
 }
@@ -550,7 +551,7 @@
                                         uint64_t SizeInBits,
                                         uint64_t AlignInBits,
                                         DIArray Elements,
-                                        DIType ClassType, unsigned Flags) {
+                                        DIType ClassType) {
   // TAG_enumeration_type is encoded in DICompositeType format.
   Value *Elts[] = {
     GetTagConstant(VMContext, dwarf::DW_TAG_enumeration_type),
@@ -561,7 +562,7 @@
     ConstantInt::get(Type::getInt64Ty(VMContext), SizeInBits),
     ConstantInt::get(Type::getInt64Ty(VMContext), AlignInBits),
     ConstantInt::get(Type::getInt32Ty(VMContext), 0),
-    ConstantInt::get(Type::getInt32Ty(VMContext), Flags),
+    ConstantInt::get(Type::getInt32Ty(VMContext), 0),
     ClassType,
     Elements,
     ConstantInt::get(Type::getInt32Ty(VMContext), 0),
@@ -640,6 +641,30 @@
   return DIType(MDNode::get(VMContext, Elts));
 }
 
+/// createArtificialType - Create a new DIType with "artificial" flag set.
+DIType DIBuilder::createObjectPointerType(DIType Ty) {
+  if (Ty.isObjectPointer())
+    return Ty;
+
+  SmallVector<Value *, 9> Elts;
+  MDNode *N = Ty;
+  assert (N && "Unexpected input DIType!");
+  for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i) {
+    if (Value *V = N->getOperand(i))
+      Elts.push_back(V);
+    else
+      Elts.push_back(Constant::getNullValue(Type::getInt32Ty(VMContext)));
+  }
+
+  unsigned CurFlags = Ty.getFlags();
+  CurFlags = CurFlags | (DIType::FlagObjectPointer | DIType::FlagArtificial);
+
+  // Flags are stored at this slot.
+  Elts[8] = ConstantInt::get(Type::getInt32Ty(VMContext), CurFlags);
+
+  return DIType(MDNode::get(VMContext, Elts));
+}
+
 /// retainType - Retain DIType in a module even if it is not referenced
 /// through debug info anchors.
 void DIBuilder::retainType(DIType T) {
@@ -682,7 +707,9 @@
 /// can be RAUW'd if the full type is seen.
 DIType DIBuilder::createForwardDecl(unsigned Tag, StringRef Name,
                                     DIDescriptor Scope, DIFile F,
-                                    unsigned Line, unsigned RuntimeLang) {
+                                    unsigned Line, unsigned RuntimeLang,
+                                    uint64_t SizeInBits,
+                                    uint64_t AlignInBits) {
   // Create a temporary MDNode.
   Value *Elts[] = {
     GetTagConstant(VMContext, Tag),
@@ -690,9 +717,8 @@
     MDString::get(VMContext, Name),
     F,
     ConstantInt::get(Type::getInt32Ty(VMContext), Line),
-    // To ease transition include sizes etc of 0.
-    ConstantInt::get(Type::getInt32Ty(VMContext), 0),
-    ConstantInt::get(Type::getInt32Ty(VMContext), 0),
+    ConstantInt::get(Type::getInt64Ty(VMContext), SizeInBits),
+    ConstantInt::get(Type::getInt64Ty(VMContext), AlignInBits),
     ConstantInt::get(Type::getInt32Ty(VMContext), 0),
     ConstantInt::get(Type::getInt32Ty(VMContext),
                      DIDescriptor::FlagFwdDecl),

Modified: llvm/branches/AMDILBackend/lib/VMCore/DebugInfo.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/VMCore/DebugInfo.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/VMCore/DebugInfo.cpp (original)
+++ llvm/branches/AMDILBackend/lib/VMCore/DebugInfo.cpp Tue Jan 15 11:16:16 2013
@@ -111,6 +111,16 @@
   return 0;
 }
 
+void DIDescriptor::replaceFunctionField(unsigned Elt, Function *F) {
+  if (DbgNode == 0)
+    return;
+
+  if (Elt < DbgNode->getNumOperands()) {
+    MDNode *Node = const_cast<MDNode*>(DbgNode);
+    Node->replaceOperandWith(Elt, F);
+  }
+}
+
 unsigned DIVariable::getNumAddrElements() const {
   if (getVersion() <= LLVMDebugVersion8)
     return DbgNode->getNumOperands()-6;

Modified: llvm/branches/AMDILBackend/lib/VMCore/Dominators.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/VMCore/Dominators.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/VMCore/Dominators.cpp (original)
+++ llvm/branches/AMDILBackend/lib/VMCore/Dominators.cpp Tue Jan 15 11:16:16 2013
@@ -39,20 +39,17 @@
 VerifyDomInfoX("verify-dom-info", cl::location(VerifyDomInfo),
                cl::desc("Verify dominator info (time consuming)"));
 
-namespace llvm {
-  class BasicBlockEdge {
-    const BasicBlock *Start;
-    const BasicBlock *End;
-  public:
-    BasicBlockEdge(const BasicBlock *Start_, const BasicBlock *End_) :
-      Start(Start_), End(End_) { }
-    const BasicBlock *getStart() const {
-      return Start;
-    }
-    const BasicBlock *getEnd() const {
-      return End;
-    }
-  };
+bool BasicBlockEdge::isSingleEdge() const {
+  const TerminatorInst *TI = Start->getTerminator();
+  unsigned NumEdgesToEnd = 0;
+  for (unsigned int i = 0, n = TI->getNumSuccessors(); i < n; ++i) {
+    if (TI->getSuccessor(i) == End)
+      ++NumEdgesToEnd;
+    if (NumEdgesToEnd >= 2)
+      return false;
+  }
+  assert(NumEdgesToEnd == 1);
+  return true;
 }
 
 //===----------------------------------------------------------------------===//
@@ -164,6 +161,11 @@
 
 bool DominatorTree::dominates(const BasicBlockEdge &BBE,
                               const BasicBlock *UseBB) const {
+  // Assert that we have a single edge. We could handle them by simply
+  // returning false, but since isSingleEdge is linear on the number of
+  // edges, the callers can normally handle them more efficiently.
+  assert(BBE.isSingleEdge());
+
   // If the BB the edge ends in doesn't dominate the use BB, then the
   // edge also doesn't.
   const BasicBlock *Start = BBE.getStart();
@@ -210,6 +212,11 @@
 
 bool DominatorTree::dominates(const BasicBlockEdge &BBE,
                               const Use &U) const {
+  // Assert that we have a single edge. We could handle them by simply
+  // returning false, but since isSingleEdge is linear on the number of
+  // edges, the callers can normally handle them more efficiently.
+  assert(BBE.isSingleEdge());
+
   Instruction *UserInst = cast<Instruction>(U.getUser());
   // A PHI in the end of the edge is dominated by it.
   PHINode *PN = dyn_cast<PHINode>(UserInst);

Modified: llvm/branches/AMDILBackend/lib/VMCore/Function.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/VMCore/Function.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/VMCore/Function.cpp (original)
+++ llvm/branches/AMDILBackend/lib/VMCore/Function.cpp Tue Jan 15 11:16:16 2013
@@ -78,7 +78,8 @@
 /// in its containing function.
 bool Argument::hasByValAttr() const {
   if (!getType()->isPointerTy()) return false;
-  return getParent()->paramHasAttr(getArgNo()+1, Attribute::ByVal);
+  return getParent()->getParamAttributes(getArgNo()+1).
+    hasAttribute(Attributes::ByVal);
 }
 
 unsigned Argument::getParamAlignment() const {
@@ -91,21 +92,24 @@
 /// it in its containing function.
 bool Argument::hasNestAttr() const {
   if (!getType()->isPointerTy()) return false;
-  return getParent()->paramHasAttr(getArgNo()+1, Attribute::Nest);
+  return getParent()->getParamAttributes(getArgNo()+1).
+    hasAttribute(Attributes::Nest);
 }
 
 /// hasNoAliasAttr - Return true if this argument has the noalias attribute on
 /// it in its containing function.
 bool Argument::hasNoAliasAttr() const {
   if (!getType()->isPointerTy()) return false;
-  return getParent()->paramHasAttr(getArgNo()+1, Attribute::NoAlias);
+  return getParent()->getParamAttributes(getArgNo()+1).
+    hasAttribute(Attributes::NoAlias);
 }
 
 /// hasNoCaptureAttr - Return true if this argument has the nocapture attribute
 /// on it in its containing function.
 bool Argument::hasNoCaptureAttr() const {
   if (!getType()->isPointerTy()) return false;
-  return getParent()->paramHasAttr(getArgNo()+1, Attribute::NoCapture);
+  return getParent()->getParamAttributes(getArgNo()+1).
+    hasAttribute(Attributes::NoCapture);
 }
 
 /// hasSRetAttr - Return true if this argument has the sret attribute on
@@ -114,7 +118,8 @@
   if (!getType()->isPointerTy()) return false;
   if (this != getParent()->arg_begin())
     return false; // StructRet param must be first param
-  return getParent()->paramHasAttr(1, Attribute::StructRet);
+  return getParent()->getParamAttributes(1).
+    hasAttribute(Attributes::StructRet);
 }
 
 /// addAttr - Add a Attribute to an argument
@@ -180,7 +185,7 @@
 
   // Ensure intrinsics have the right parameter attributes.
   if (unsigned IID = getIntrinsicID())
-    setAttributes(Intrinsic::getAttributes(Intrinsic::ID(IID)));
+    setAttributes(Intrinsic::getAttributes(getContext(), Intrinsic::ID(IID)));
 
 }
 
@@ -244,13 +249,13 @@
 
 void Function::addAttribute(unsigned i, Attributes attr) {
   AttrListPtr PAL = getAttributes();
-  PAL = PAL.addAttr(i, attr);
+  PAL = PAL.addAttr(getContext(), i, attr);
   setAttributes(PAL);
 }
 
 void Function::removeAttribute(unsigned i, Attributes attr) {
   AttrListPtr PAL = getAttributes();
-  PAL = PAL.removeAttr(i, attr);
+  PAL = PAL.removeAttr(getContext(), i, attr);
   setAttributes(PAL);
 }
 

Modified: llvm/branches/AMDILBackend/lib/VMCore/GCOV.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/VMCore/GCOV.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/VMCore/GCOV.cpp (original)
+++ llvm/branches/AMDILBackend/lib/VMCore/GCOV.cpp Tue Jan 15 11:16:16 2013
@@ -28,19 +28,19 @@
 }
 
 /// isGCDAFile - Return true if Format identifies a .gcda file.
-static bool isGCDAFile(GCOVFormat Format) {
-  return Format == GCDA_402 || Format == GCDA_404;
+static bool isGCDAFile(GCOV::GCOVFormat Format) {
+  return Format == GCOV::GCDA_402 || Format == GCOV::GCDA_404;
 }
 
 /// isGCNOFile - Return true if Format identifies a .gcno file.
-static bool isGCNOFile(GCOVFormat Format) {
-  return Format == GCNO_402 || Format == GCNO_404;
+static bool isGCNOFile(GCOV::GCOVFormat Format) {
+  return Format == GCOV::GCNO_402 || Format == GCOV::GCNO_404;
 }
 
 /// read - Read GCOV buffer.
 bool GCOVFile::read(GCOVBuffer &Buffer) {
-  GCOVFormat Format = Buffer.readGCOVFormat();
-  if (Format == InvalidGCOV)
+  GCOV::GCOVFormat Format = Buffer.readGCOVFormat();
+  if (Format == GCOV::InvalidGCOV)
     return false;
 
   unsigned i = 0;
@@ -48,7 +48,7 @@
     GCOVFunction *GFun = NULL;
     if (isGCDAFile(Format)) {
       // Use existing function while reading .gcda file.
-      assert (i < Functions.size() && ".gcda data does not match .gcno data");
+      assert(i < Functions.size() && ".gcda data does not match .gcno data");
       GFun = Functions[i];
     } else if (isGCNOFile(Format)){
       GFun = new GCOVFunction();
@@ -87,21 +87,21 @@
 
 /// read - Read a aunction from the buffer. Return false if buffer cursor
 /// does not point to a function tag.
-bool GCOVFunction::read(GCOVBuffer &Buff, GCOVFormat Format) {
+bool GCOVFunction::read(GCOVBuffer &Buff, GCOV::GCOVFormat Format) {
   if (!Buff.readFunctionTag())
     return false;
 
   Buff.readInt(); // Function header length
   Ident = Buff.readInt(); 
   Buff.readInt(); // Checksum #1
-  if (Format != GCNO_402)
+  if (Format != GCOV::GCNO_402)
     Buff.readInt(); // Checksum #2
 
   Name = Buff.readString();
-  if (Format == GCNO_402 || Format == GCNO_404)
+  if (Format == GCOV::GCNO_402 || Format == GCOV::GCNO_404)
     Filename = Buff.readString();
 
-  if (Format == GCDA_402 || Format == GCDA_404) {
+  if (Format == GCOV::GCDA_402 || Format == GCOV::GCDA_404) {
     Buff.readArcTag();
     uint32_t Count = Buff.readInt() / 2;
     for (unsigned i = 0, e = Count; i != e; ++i) {
@@ -113,7 +113,9 @@
   LineNumber = Buff.readInt();
 
   // read blocks.
-  assert (Buff.readBlockTag() && "Block Tag not found!");
+  bool BlockTagFound = Buff.readBlockTag();
+  (void)BlockTagFound;
+  assert(BlockTagFound && "Block Tag not found!");
   uint32_t BlockCount = Buff.readInt();
   for (int i = 0, e = BlockCount; i != e; ++i) {
     Buff.readInt(); // Block flags;
@@ -124,7 +126,7 @@
   while (Buff.readEdgeTag()) {
     uint32_t EdgeCount = (Buff.readInt() - 1) / 2;
     uint32_t BlockNo = Buff.readInt();
-    assert (BlockNo < BlockCount && "Unexpected Block number!");
+    assert(BlockNo < BlockCount && "Unexpected Block number!");
     for (int i = 0, e = EdgeCount; i != e; ++i) {
       Blocks[BlockNo]->addEdge(Buff.readInt());
       Buff.readInt(); // Edge flag
@@ -136,7 +138,7 @@
     uint32_t LineTableLength = Buff.readInt();
     uint32_t Size = Buff.getCursor() + LineTableLength*4;
     uint32_t BlockNo = Buff.readInt();
-    assert (BlockNo < BlockCount && "Unexpected Block number!");
+    assert(BlockNo < BlockCount && "Unexpected Block number!");
     GCOVBlock *Block = Blocks[BlockNo];
     Buff.readInt(); // flag
     while (Buff.getCursor() != (Size - 4)) {

Modified: llvm/branches/AMDILBackend/lib/VMCore/IRBuilder.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/VMCore/IRBuilder.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/VMCore/IRBuilder.cpp (original)
+++ llvm/branches/AMDILBackend/lib/VMCore/IRBuilder.cpp Tue Jan 15 11:16:16 2013
@@ -80,7 +80,7 @@
 
 CallInst *IRBuilderBase::
 CreateMemCpy(Value *Dst, Value *Src, Value *Size, unsigned Align,
-             bool isVolatile, MDNode *TBAATag) {
+             bool isVolatile, MDNode *TBAATag, MDNode *TBAAStructTag) {
   Dst = getCastedInt8PtrValue(Dst);
   Src = getCastedInt8PtrValue(Src);
 
@@ -94,6 +94,10 @@
   // Set the TBAA info if present.
   if (TBAATag)
     CI->setMetadata(LLVMContext::MD_tbaa, TBAATag);
+
+  // Set the TBAA Struct info if present.
+  if (TBAAStructTag)
+    CI->setMetadata(LLVMContext::MD_tbaa_struct, TBAAStructTag);
   
   return CI;  
 }

Modified: llvm/branches/AMDILBackend/lib/VMCore/InlineAsm.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/VMCore/InlineAsm.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/VMCore/InlineAsm.cpp (original)
+++ llvm/branches/AMDILBackend/lib/VMCore/InlineAsm.cpp Tue Jan 15 11:16:16 2013
@@ -27,19 +27,20 @@
 
 InlineAsm *InlineAsm::get(FunctionType *Ty, StringRef AsmString,
                           StringRef Constraints, bool hasSideEffects,
-                          bool isAlignStack) {
-  InlineAsmKeyType Key(AsmString, Constraints, hasSideEffects, isAlignStack);
+                          bool isAlignStack, AsmDialect asmDialect) {
+  InlineAsmKeyType Key(AsmString, Constraints, hasSideEffects, isAlignStack,
+                       asmDialect);
   LLVMContextImpl *pImpl = Ty->getContext().pImpl;
   return pImpl->InlineAsms.getOrCreate(PointerType::getUnqual(Ty), Key);
 }
 
 InlineAsm::InlineAsm(PointerType *Ty, const std::string &asmString,
                      const std::string &constraints, bool hasSideEffects,
-                     bool isAlignStack)
+                     bool isAlignStack, AsmDialect asmDialect)
   : Value(Ty, Value::InlineAsmVal),
-    AsmString(asmString), 
-    Constraints(constraints), HasSideEffects(hasSideEffects), 
-    IsAlignStack(isAlignStack) {
+    AsmString(asmString), Constraints(constraints),
+    HasSideEffects(hasSideEffects), IsAlignStack(isAlignStack),
+    Dialect(asmDialect) {
 
   // Do various checks on the constraint string and type.
   assert(Verify(getFunctionType(), constraints) &&

Modified: llvm/branches/AMDILBackend/lib/VMCore/Instructions.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/VMCore/Instructions.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/VMCore/Instructions.cpp (original)
+++ llvm/branches/AMDILBackend/lib/VMCore/Instructions.cpp Tue Jan 15 11:16:16 2013
@@ -332,21 +332,30 @@
 
 void CallInst::addAttribute(unsigned i, Attributes attr) {
   AttrListPtr PAL = getAttributes();
-  PAL = PAL.addAttr(i, attr);
+  PAL = PAL.addAttr(getContext(), i, attr);
   setAttributes(PAL);
 }
 
 void CallInst::removeAttribute(unsigned i, Attributes attr) {
   AttrListPtr PAL = getAttributes();
-  PAL = PAL.removeAttr(i, attr);
+  PAL = PAL.removeAttr(getContext(), i, attr);
   setAttributes(PAL);
 }
 
-bool CallInst::paramHasAttr(unsigned i, Attributes attr) const {
-  if (AttributeList.paramHasAttr(i, attr))
+bool CallInst::hasFnAttr(Attributes::AttrVal A) const {
+  if (AttributeList.getParamAttributes(AttrListPtr::FunctionIndex)
+      .hasAttribute(A))
     return true;
   if (const Function *F = getCalledFunction())
-    return F->paramHasAttr(i, attr);
+    return F->getParamAttributes(AttrListPtr::FunctionIndex).hasAttribute(A);
+  return false;
+}
+
+bool CallInst::paramHasAttr(unsigned i, Attributes::AttrVal A) const {
+  if (AttributeList.getParamAttributes(i).hasAttribute(A))
+    return true;
+  if (const Function *F = getCalledFunction())
+    return F->getParamAttributes(i).hasAttribute(A);
   return false;
 }
 
@@ -562,23 +571,32 @@
   return setSuccessor(idx, B);
 }
 
-bool InvokeInst::paramHasAttr(unsigned i, Attributes attr) const {
-  if (AttributeList.paramHasAttr(i, attr))
+bool InvokeInst::hasFnAttr(Attributes::AttrVal A) const {
+  if (AttributeList.getParamAttributes(AttrListPtr::FunctionIndex).
+      hasAttribute(A))
     return true;
   if (const Function *F = getCalledFunction())
-    return F->paramHasAttr(i, attr);
+    return F->getParamAttributes(AttrListPtr::FunctionIndex).hasAttribute(A);
+  return false;
+}
+
+bool InvokeInst::paramHasAttr(unsigned i, Attributes::AttrVal A) const {
+  if (AttributeList.getParamAttributes(i).hasAttribute(A))
+    return true;
+  if (const Function *F = getCalledFunction())
+    return F->getParamAttributes(i).hasAttribute(A);
   return false;
 }
 
 void InvokeInst::addAttribute(unsigned i, Attributes attr) {
   AttrListPtr PAL = getAttributes();
-  PAL = PAL.addAttr(i, attr);
+  PAL = PAL.addAttr(getContext(), i, attr);
   setAttributes(PAL);
 }
 
 void InvokeInst::removeAttribute(unsigned i, Attributes attr) {
   AttrListPtr PAL = getAttributes();
-  PAL = PAL.removeAttr(i, attr);
+  PAL = PAL.removeAttr(getContext(), i, attr);
   setAttributes(PAL);
 }
 
@@ -1381,18 +1399,6 @@
   return getIndexedTypeInternal(Ptr, IdxList);
 }
 
-unsigned GetElementPtrInst::getAddressSpace(Value *Ptr) {
-  Type *Ty = Ptr->getType();
-
-  if (VectorType *VTy = dyn_cast<VectorType>(Ty))
-    Ty = VTy->getElementType();
-
-  if (PointerType *PTy = dyn_cast<PointerType>(Ty))
-    return PTy->getAddressSpace();
-
-  llvm_unreachable("Invalid GEP pointer type");
-}
-
 /// hasAllZeroIndices - Return true if all of the indices of this GEP are
 /// zeros.  If so, the result pointer and the first operand have the same
 /// value, just potentially different types.
@@ -2112,7 +2118,8 @@
 /// If no such cast is permited, the function returns 0.
 unsigned CastInst::isEliminableCastPair(
   Instruction::CastOps firstOp, Instruction::CastOps secondOp,
-  Type *SrcTy, Type *MidTy, Type *DstTy, Type *IntPtrTy) {
+  Type *SrcTy, Type *MidTy, Type *DstTy, Type *SrcIntPtrTy, Type *MidIntPtrTy,
+  Type *DstIntPtrTy) {
   // Define the 144 possibilities for these two cast instructions. The values
   // in this matrix determine what to do in a given situation and select the
   // case in the switch below.  The rows correspond to firstOp, the columns 
@@ -2215,9 +2222,9 @@
       return 0;
     case 7: { 
       // ptrtoint, inttoptr -> bitcast (ptr -> ptr) if int size is >= ptr size
-      if (!IntPtrTy)
+      if (!SrcIntPtrTy || DstIntPtrTy != SrcIntPtrTy)
         return 0;
-      unsigned PtrSize = IntPtrTy->getScalarSizeInBits();
+      unsigned PtrSize = SrcIntPtrTy->getScalarSizeInBits();
       unsigned MidSize = MidTy->getScalarSizeInBits();
       if (MidSize >= PtrSize)
         return Instruction::BitCast;
@@ -2256,9 +2263,9 @@
       return 0;
     case 13: {
       // inttoptr, ptrtoint -> bitcast if SrcSize<=PtrSize and SrcSize==DstSize
-      if (!IntPtrTy)
+      if (!MidIntPtrTy)
         return 0;
-      unsigned PtrSize = IntPtrTy->getScalarSizeInBits();
+      unsigned PtrSize = MidIntPtrTy->getScalarSizeInBits();
       unsigned SrcSize = SrcTy->getScalarSizeInBits();
       unsigned DstSize = DstTy->getScalarSizeInBits();
       if (SrcSize <= PtrSize && SrcSize == DstSize)
@@ -2836,7 +2843,7 @@
 //                               CmpInst Classes
 //===----------------------------------------------------------------------===//
 
-void CmpInst::Anchor() const {}
+void CmpInst::anchor() {}
 
 CmpInst::CmpInst(Type *ty, OtherOps op, unsigned short predicate,
                  Value *LHS, Value *RHS, const Twine &Name,

Modified: llvm/branches/AMDILBackend/lib/VMCore/LLVMContext.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/VMCore/LLVMContext.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/VMCore/LLVMContext.cpp (original)
+++ llvm/branches/AMDILBackend/lib/VMCore/LLVMContext.cpp Tue Jan 15 11:16:16 2013
@@ -53,6 +53,11 @@
   unsigned RangeID = getMDKindID("range");
   assert(RangeID == MD_range && "range kind id drifted");
   (void)RangeID;
+
+  // Create the 'tbaa.struct' metadata kind.
+  unsigned TBAAStructID = getMDKindID("tbaa.struct");
+  assert(TBAAStructID == MD_tbaa_struct && "tbaa.struct kind id drifted");
+  (void)TBAAStructID;
 }
 LLVMContext::~LLVMContext() { delete pImpl; }
 

Modified: llvm/branches/AMDILBackend/lib/VMCore/LLVMContextImpl.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/VMCore/LLVMContextImpl.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/VMCore/LLVMContextImpl.cpp (original)
+++ llvm/branches/AMDILBackend/lib/VMCore/LLVMContextImpl.cpp Tue Jan 15 11:16:16 2013
@@ -12,6 +12,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "LLVMContextImpl.h"
+#include "llvm/Attributes.h"
 #include "llvm/Module.h"
 #include "llvm/ADT/STLExtras.h"
 #include <algorithm>
@@ -93,7 +94,21 @@
        E = CDSConstants.end(); I != E; ++I)
     delete I->second;
   CDSConstants.clear();
-  
+
+  // Destroy attributes.
+  for (FoldingSetIterator<AttributesImpl> I = AttrsSet.begin(),
+         E = AttrsSet.end(); I != E; ) {
+    FoldingSetIterator<AttributesImpl> Elem = I++;
+    delete &*Elem;
+  }
+
+  // Destroy attribute lists.
+  for (FoldingSetIterator<AttributeListImpl> I = AttrsLists.begin(),
+         E = AttrsLists.end(); I != E; ) {
+    FoldingSetIterator<AttributeListImpl> Elem = I++;
+    delete &*Elem;
+  }
+
   // Destroy MDNodes.  ~MDNode can move and remove nodes between the MDNodeSet
   // and the NonUniquedMDNodes sets, so copy the values out first.
   SmallVector<MDNode*, 8> MDNodes;
@@ -107,6 +122,7 @@
     (*I)->destroy();
   assert(MDNodeSet.empty() && NonUniquedMDNodes.empty() &&
          "Destroying all MDNodes didn't empty the Context's sets.");
+
   // Destroy MDStrings.
   DeleteContainerSeconds(MDStringCache);
 }

Modified: llvm/branches/AMDILBackend/lib/VMCore/LLVMContextImpl.h
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/VMCore/LLVMContextImpl.h?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/VMCore/LLVMContextImpl.h (original)
+++ llvm/branches/AMDILBackend/lib/VMCore/LLVMContextImpl.h Tue Jan 15 11:16:16 2013
@@ -16,6 +16,7 @@
 #define LLVM_LLVMCONTEXT_IMPL_H
 
 #include "llvm/LLVMContext.h"
+#include "AttributesImpl.h"
 #include "ConstantsContext.h"
 #include "LeaksContext.h"
 #include "llvm/Constants.h"
@@ -253,10 +254,14 @@
   typedef DenseMap<DenseMapAPFloatKeyInfo::KeyTy, ConstantFP*, 
                          DenseMapAPFloatKeyInfo> FPMapTy;
   FPMapTy FPConstants;
-  
+
+  FoldingSet<AttributesImpl> AttrsSet;
+  FoldingSet<AttributeListImpl> AttrsLists;
+
   StringMap<Value*> MDStringCache;
-  
+
   FoldingSet<MDNode> MDNodeSet;
+
   // MDNodes may be uniqued or not uniqued.  When they're not uniqued, they
   // aren't in the MDNodeSet, but they're still shared between objects, so no
   // one object can destroy them.  This set allows us to at least destroy them

Modified: llvm/branches/AMDILBackend/lib/VMCore/Makefile
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/VMCore/Makefile?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/VMCore/Makefile (original)
+++ llvm/branches/AMDILBackend/lib/VMCore/Makefile Tue Jan 15 11:16:16 2013
@@ -9,7 +9,6 @@
 LEVEL = ../..
 LIBRARYNAME = LLVMCore
 BUILD_ARCHIVE = 1
-REQUIRES_RTTI = 1
 
 BUILT_SOURCES = $(PROJ_OBJ_ROOT)/include/llvm/Intrinsics.gen
 

Modified: llvm/branches/AMDILBackend/lib/VMCore/Metadata.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/VMCore/Metadata.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/VMCore/Metadata.cpp (original)
+++ llvm/branches/AMDILBackend/lib/VMCore/Metadata.cpp Tue Jan 15 11:16:16 2013
@@ -200,7 +200,7 @@
 // destroy - Delete this node.  Only when there are no uses.
 void MDNode::destroy() {
   setValueSubclassData(getSubclassDataFromValue() | DestroyFlag);
-  // Placement delete, the free the memory.
+  // Placement delete, then free the memory.
   this->~MDNode();
   free(this);
 }

Modified: llvm/branches/AMDILBackend/lib/VMCore/PassManager.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/VMCore/PassManager.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/VMCore/PassManager.cpp (original)
+++ llvm/branches/AMDILBackend/lib/VMCore/PassManager.cpp Tue Jan 15 11:16:16 2013
@@ -1189,7 +1189,7 @@
   assert(PassDebugging >= Details);
   if (Set.empty())
     return;
-  dbgs() << (void*)P << std::string(getDepth()*2+3, ' ') << Msg << " Analyses:";
+  dbgs() << (const void*)P << std::string(getDepth()*2+3, ' ') << Msg << " Analyses:";
   for (unsigned i = 0; i != Set.size(); ++i) {
     if (i) dbgs() << ',';
     const PassInfo *PInf = PassRegistry::getPassRegistry()->getPassInfo(Set[i]);

Modified: llvm/branches/AMDILBackend/lib/VMCore/Type.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/VMCore/Type.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/VMCore/Type.cpp (original)
+++ llvm/branches/AMDILBackend/lib/VMCore/Type.cpp Tue Jan 15 11:16:16 2013
@@ -47,35 +47,17 @@
   return this;
 }
 
+const Type *Type::getScalarType() const {
+  if (const VectorType *VTy = dyn_cast<VectorType>(this))
+    return VTy->getElementType();
+  return this;
+}
+
 /// isIntegerTy - Return true if this is an IntegerType of the specified width.
 bool Type::isIntegerTy(unsigned Bitwidth) const {
   return isIntegerTy() && cast<IntegerType>(this)->getBitWidth() == Bitwidth;
 }
 
-/// isIntOrIntVectorTy - Return true if this is an integer type or a vector of
-/// integer types.
-///
-bool Type::isIntOrIntVectorTy() const {
-  if (isIntegerTy())
-    return true;
-  if (getTypeID() != Type::VectorTyID) return false;
-  
-  return cast<VectorType>(this)->getElementType()->isIntegerTy();
-}
-
-/// isFPOrFPVectorTy - Return true if this is a FP type or a vector of FP types.
-///
-bool Type::isFPOrFPVectorTy() const {
-  if (getTypeID() == Type::HalfTyID || getTypeID() == Type::FloatTyID ||
-      getTypeID() == Type::DoubleTyID ||
-      getTypeID() == Type::FP128TyID || getTypeID() == Type::X86_FP80TyID || 
-      getTypeID() == Type::PPC_FP128TyID)
-    return true;
-  if (getTypeID() != Type::VectorTyID) return false;
-  
-  return cast<VectorType>(this)->getElementType()->isFloatingPointTy();
-}
-
 // canLosslesslyBitCastTo - Return true if this type can be converted to
 // 'Ty' without any reinterpretation of bits.  For example, i8* to i32*.
 //
@@ -220,8 +202,6 @@
   return cast<StructType>(this)->getElementType(N);
 }
 
-
-
 Type *Type::getSequentialElementType() const {
   return cast<SequentialType>(this)->getElementType();
 }
@@ -235,12 +215,10 @@
 }
 
 unsigned Type::getPointerAddressSpace() const {
-  return cast<PointerType>(this)->getAddressSpace();
+  return cast<PointerType>(getScalarType())->getAddressSpace();
 }
 
 
-
-
 //===----------------------------------------------------------------------===//
 //                          Primitive 'Type' data
 //===----------------------------------------------------------------------===//
@@ -400,12 +378,10 @@
   return FT;
 }
 
-
 FunctionType *FunctionType::get(Type *Result, bool isVarArg) {
   return get(Result, ArrayRef<Type *>(), isVarArg);
 }
 
-
 /// isValidReturnType - Return true if the specified type is valid as a return
 /// type.
 bool FunctionType::isValidReturnType(Type *RetTy) {
@@ -553,7 +529,6 @@
   return create(Context, StringRef());
 }
 
-
 StructType *StructType::create(ArrayRef<Type*> Elements, StringRef Name,
                                bool isPacked) {
   assert(!Elements.empty() &&
@@ -637,7 +612,6 @@
   return std::equal(element_begin(), element_end(), Other->element_begin());
 }
 
-
 /// getTypeByName - Return the type with the specified name, or null if there
 /// is none by that name.
 StructType *Module::getTypeByName(StringRef Name) const {
@@ -700,7 +674,6 @@
   NumElements = NumEl;
 }
 
-
 ArrayType *ArrayType::get(Type *elementType, uint64_t NumElements) {
   Type *ElementType = const_cast<Type*>(elementType);
   assert(isValidElementType(ElementType) && "Invalid type for array element!");

Modified: llvm/branches/AMDILBackend/lib/VMCore/User.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/VMCore/User.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/VMCore/User.cpp (original)
+++ llvm/branches/AMDILBackend/lib/VMCore/User.cpp Tue Jan 15 11:16:16 2013
@@ -10,6 +10,7 @@
 #include "llvm/Constant.h"
 #include "llvm/GlobalValue.h"
 #include "llvm/User.h"
+#include "llvm/Operator.h"
 
 namespace llvm {
 
@@ -78,4 +79,12 @@
   ::operator delete(Storage);
 }
 
+//===----------------------------------------------------------------------===//
+//                             Operator Class
+//===----------------------------------------------------------------------===//
+
+Operator::~Operator() {
+  llvm_unreachable("should never destroy an Operator");
+}
+
 } // End llvm namespace

Modified: llvm/branches/AMDILBackend/lib/VMCore/Value.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/VMCore/Value.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/VMCore/Value.cpp (original)
+++ llvm/branches/AMDILBackend/lib/VMCore/Value.cpp Tue Jan 15 11:16:16 2013
@@ -394,7 +394,7 @@
   // It's also not always safe to follow a bitcast, for example:
   //   bitcast i8* (alloca i8) to i32*
   // would result in a 4-byte load from a 1-byte alloca. Some cases could
-  // be handled using TargetData to check sizes and alignments though.
+  // be handled using DataLayout to check sizes and alignments though.
 
   // These are obviously ok.
   if (isa<AllocaInst>(V)) return true;

Modified: llvm/branches/AMDILBackend/lib/VMCore/ValueTypes.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/VMCore/ValueTypes.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/VMCore/ValueTypes.cpp (original)
+++ llvm/branches/AMDILBackend/lib/VMCore/ValueTypes.cpp Tue Jan 15 11:16:16 2013
@@ -55,24 +55,32 @@
   return LLVMTy->isVectorTy();
 }
 
+bool EVT::isExtended16BitVector() const {
+  return isExtendedVector() && getExtendedSizeInBits() == 16;
+}
+
+bool EVT::isExtended32BitVector() const {
+  return isExtendedVector() && getExtendedSizeInBits() == 32;
+}
+
 bool EVT::isExtended64BitVector() const {
-  return isExtendedVector() && getSizeInBits() == 64;
+  return isExtendedVector() && getExtendedSizeInBits() == 64;
 }
 
 bool EVT::isExtended128BitVector() const {
-  return isExtendedVector() && getSizeInBits() == 128;
+  return isExtendedVector() && getExtendedSizeInBits() == 128;
 }
 
 bool EVT::isExtended256BitVector() const {
-  return isExtendedVector() && getSizeInBits() == 256;
+  return isExtendedVector() && getExtendedSizeInBits() == 256;
 }
 
 bool EVT::isExtended512BitVector() const {
-  return isExtendedVector() && getSizeInBits() == 512;
+  return isExtendedVector() && getExtendedSizeInBits() == 512;
 }
 
 bool EVT::isExtended1024BitVector() const {
-  return isExtendedVector() && getSizeInBits() == 1024;
+  return isExtendedVector() && getExtendedSizeInBits() == 1024;
 }
 
 EVT EVT::getExtendedVectorElementType() const {
@@ -120,15 +128,21 @@
   case MVT::Other:   return "ch";
   case MVT::Glue:    return "glue";
   case MVT::x86mmx:  return "x86mmx";
+  case MVT::v2i1:    return "v2i1";
+  case MVT::v4i1:    return "v4i1";
+  case MVT::v8i1:    return "v8i1";
+  case MVT::v16i1:   return "v16i1";
   case MVT::v2i8:    return "v2i8";
   case MVT::v4i8:    return "v4i8";
   case MVT::v8i8:    return "v8i8";
   case MVT::v16i8:   return "v16i8";
   case MVT::v32i8:   return "v32i8";
+  case MVT::v1i16:   return "v1i16";
   case MVT::v2i16:   return "v2i16";
   case MVT::v4i16:   return "v4i16";
   case MVT::v8i16:   return "v8i16";
   case MVT::v16i16:  return "v16i16";
+  case MVT::v1i32:   return "v1i32";
   case MVT::v2i32:   return "v2i32";
   case MVT::v4i32:   return "v4i32";
   case MVT::v8i32:   return "v8i32";
@@ -171,15 +185,21 @@
   case MVT::f128:    return Type::getFP128Ty(Context);
   case MVT::ppcf128: return Type::getPPC_FP128Ty(Context);
   case MVT::x86mmx:  return Type::getX86_MMXTy(Context);
+  case MVT::v2i1:    return VectorType::get(Type::getInt1Ty(Context), 2);
+  case MVT::v4i1:    return VectorType::get(Type::getInt1Ty(Context), 4);
+  case MVT::v8i1:    return VectorType::get(Type::getInt1Ty(Context), 8);
+  case MVT::v16i1:   return VectorType::get(Type::getInt1Ty(Context), 16);
   case MVT::v2i8:    return VectorType::get(Type::getInt8Ty(Context), 2);
   case MVT::v4i8:    return VectorType::get(Type::getInt8Ty(Context), 4);
   case MVT::v8i8:    return VectorType::get(Type::getInt8Ty(Context), 8);
   case MVT::v16i8:   return VectorType::get(Type::getInt8Ty(Context), 16);
   case MVT::v32i8:   return VectorType::get(Type::getInt8Ty(Context), 32);
+  case MVT::v1i16:   return VectorType::get(Type::getInt16Ty(Context), 1);
   case MVT::v2i16:   return VectorType::get(Type::getInt16Ty(Context), 2);
   case MVT::v4i16:   return VectorType::get(Type::getInt16Ty(Context), 4);
   case MVT::v8i16:   return VectorType::get(Type::getInt16Ty(Context), 8);
   case MVT::v16i16:  return VectorType::get(Type::getInt16Ty(Context), 16);
+  case MVT::v1i32:   return VectorType::get(Type::getInt32Ty(Context), 1);
   case MVT::v2i32:   return VectorType::get(Type::getInt32Ty(Context), 2);
   case MVT::v4i32:   return VectorType::get(Type::getInt32Ty(Context), 4);
   case MVT::v8i32:   return VectorType::get(Type::getInt32Ty(Context), 8);

Modified: llvm/branches/AMDILBackend/lib/VMCore/Verifier.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/lib/VMCore/Verifier.cpp?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/lib/VMCore/Verifier.cpp (original)
+++ llvm/branches/AMDILBackend/lib/VMCore/Verifier.cpp Tue Jan 15 11:16:16 2013
@@ -400,8 +400,8 @@
             "Only global arrays can have appending linkage!", GVar);
   }
 
-  Assert1(!GV.hasLinkerPrivateWeakDefAutoLinkage() || GV.hasDefaultVisibility(),
-          "linker_private_weak_def_auto can only have default visibility!",
+  Assert1(!GV.hasLinkOnceODRAutoHideLinkage() || GV.hasDefaultVisibility(),
+          "linkonce_odr_auto_hide can only have default visibility!",
           &GV);
 }
 
@@ -526,40 +526,60 @@
 // value of the specified type.  The value V is printed in error messages.
 void Verifier::VerifyParameterAttrs(Attributes Attrs, Type *Ty,
                                     bool isReturnValue, const Value *V) {
-  if (Attrs == Attribute::None)
+  if (!Attrs.hasAttributes())
     return;
 
-  Attributes FnCheckAttr = Attrs & Attribute::FunctionOnly;
-  Assert1(!FnCheckAttr, "Attribute " + Attribute::getAsString(FnCheckAttr) +
-          " only applies to the function!", V);
-
-  if (isReturnValue) {
-    Attributes RetI = Attrs & Attribute::ParameterOnly;
-    Assert1(!RetI, "Attribute " + Attribute::getAsString(RetI) +
-            " does not apply to return values!", V);
-  }
-
-  for (unsigned i = 0;
-       i < array_lengthof(Attribute::MutuallyIncompatible); ++i) {
-    Attributes MutI = Attrs & Attribute::MutuallyIncompatible[i];
-    Assert1(MutI.isEmptyOrSingleton(), "Attributes " +
-            Attribute::getAsString(MutI) + " are incompatible!", V);
-  }
-
-  Attributes TypeI = Attrs & Attribute::typeIncompatible(Ty);
-  Assert1(!TypeI, "Wrong type for attribute " +
-          Attribute::getAsString(TypeI), V);
-
-  Attributes ByValI = Attrs & Attribute::ByVal;
-  if (PointerType *PTy = dyn_cast<PointerType>(Ty)) {
-    Assert1(!ByValI || PTy->getElementType()->isSized(),
-            "Attribute " + Attribute::getAsString(ByValI) +
-            " does not support unsized types!", V);
-  } else {
-    Assert1(!ByValI,
-            "Attribute " + Attribute::getAsString(ByValI) +
-            " only applies to parameters with pointer type!", V);
-  }
+  Assert1(!Attrs.hasFunctionOnlyAttrs(),
+          "Some attributes in '" + Attrs.getAsString() +
+          "' only apply to functions!", V);
+
+  if (isReturnValue)
+    Assert1(!Attrs.hasParameterOnlyAttrs(),
+            "Attributes 'byval', 'nest', 'sret', and 'nocapture' "
+            "do not apply to return values!", V);
+
+  // Check for mutually incompatible attributes.
+  Assert1(!((Attrs.hasAttribute(Attributes::ByVal) &&
+             Attrs.hasAttribute(Attributes::Nest)) ||
+            (Attrs.hasAttribute(Attributes::ByVal) &&
+             Attrs.hasAttribute(Attributes::StructRet)) ||
+            (Attrs.hasAttribute(Attributes::Nest) &&
+             Attrs.hasAttribute(Attributes::StructRet))), "Attributes "
+          "'byval, nest, and sret' are incompatible!", V);
+
+  Assert1(!((Attrs.hasAttribute(Attributes::ByVal) &&
+             Attrs.hasAttribute(Attributes::Nest)) ||
+            (Attrs.hasAttribute(Attributes::ByVal) &&
+             Attrs.hasAttribute(Attributes::InReg)) ||
+            (Attrs.hasAttribute(Attributes::Nest) &&
+             Attrs.hasAttribute(Attributes::InReg))), "Attributes "
+          "'byval, nest, and inreg' are incompatible!", V);
+
+  Assert1(!(Attrs.hasAttribute(Attributes::ZExt) &&
+            Attrs.hasAttribute(Attributes::SExt)), "Attributes "
+          "'zeroext and signext' are incompatible!", V);
+
+  Assert1(!(Attrs.hasAttribute(Attributes::ReadNone) &&
+            Attrs.hasAttribute(Attributes::ReadOnly)), "Attributes "
+          "'readnone and readonly' are incompatible!", V);
+
+  Assert1(!(Attrs.hasAttribute(Attributes::NoInline) &&
+            Attrs.hasAttribute(Attributes::AlwaysInline)), "Attributes "
+          "'noinline and alwaysinline' are incompatible!", V);
+
+  Assert1(!AttrBuilder(Attrs).
+            hasAttributes(Attributes::typeIncompatible(Ty)),
+          "Wrong types for attribute: " +
+          Attributes::typeIncompatible(Ty).getAsString(), V);
+
+  if (PointerType *PTy = dyn_cast<PointerType>(Ty))
+    Assert1(!Attrs.hasAttribute(Attributes::ByVal) ||
+            PTy->getElementType()->isSized(),
+            "Attribute 'byval' does not support unsized types!", V);
+  else
+    Assert1(!Attrs.hasAttribute(Attributes::ByVal),
+            "Attribute 'byval' only applies to parameters with pointer type!",
+            V);
 }
 
 // VerifyFunctionAttrs - Check parameter attributes against a function type.
@@ -585,26 +605,50 @@
 
     VerifyParameterAttrs(Attr.Attrs, Ty, Attr.Index == 0, V);
 
-    if (Attr.Attrs & Attribute::Nest) {
+    if (Attr.Attrs.hasAttribute(Attributes::Nest)) {
       Assert1(!SawNest, "More than one parameter has attribute nest!", V);
       SawNest = true;
     }
 
-    if (Attr.Attrs & Attribute::StructRet)
+    if (Attr.Attrs.hasAttribute(Attributes::StructRet))
       Assert1(Attr.Index == 1, "Attribute sret not on first parameter!", V);
   }
 
   Attributes FAttrs = Attrs.getFnAttributes();
-  Attributes NotFn = FAttrs & (~Attribute::FunctionOnly);
-  Assert1(!NotFn, "Attribute " + Attribute::getAsString(NotFn) +
-          " does not apply to the function!", V);
-
-  for (unsigned i = 0;
-       i < array_lengthof(Attribute::MutuallyIncompatible); ++i) {
-    Attributes MutI = FAttrs & Attribute::MutuallyIncompatible[i];
-    Assert1(MutI.isEmptyOrSingleton(), "Attributes " +
-            Attribute::getAsString(MutI) + " are incompatible!", V);
-  }
+  AttrBuilder NotFn(FAttrs);
+  NotFn.removeFunctionOnlyAttrs();
+  Assert1(!NotFn.hasAttributes(), "Attributes '" +
+          Attributes::get(V->getContext(), NotFn).getAsString() +
+          "' do not apply to the function!", V);
+
+  // Check for mutually incompatible attributes.
+  Assert1(!((FAttrs.hasAttribute(Attributes::ByVal) &&
+             FAttrs.hasAttribute(Attributes::Nest)) ||
+            (FAttrs.hasAttribute(Attributes::ByVal) &&
+             FAttrs.hasAttribute(Attributes::StructRet)) ||
+            (FAttrs.hasAttribute(Attributes::Nest) &&
+             FAttrs.hasAttribute(Attributes::StructRet))), "Attributes "
+          "'byval, nest, and sret' are incompatible!", V);
+
+  Assert1(!((FAttrs.hasAttribute(Attributes::ByVal) &&
+             FAttrs.hasAttribute(Attributes::Nest)) ||
+            (FAttrs.hasAttribute(Attributes::ByVal) &&
+             FAttrs.hasAttribute(Attributes::InReg)) ||
+            (FAttrs.hasAttribute(Attributes::Nest) &&
+             FAttrs.hasAttribute(Attributes::InReg))), "Attributes "
+          "'byval, nest, and inreg' are incompatible!", V);
+
+  Assert1(!(FAttrs.hasAttribute(Attributes::ZExt) &&
+            FAttrs.hasAttribute(Attributes::SExt)), "Attributes "
+          "'zeroext and signext' are incompatible!", V);
+
+  Assert1(!(FAttrs.hasAttribute(Attributes::ReadNone) &&
+            FAttrs.hasAttribute(Attributes::ReadOnly)), "Attributes "
+          "'readnone and readonly' are incompatible!", V);
+
+  Assert1(!(FAttrs.hasAttribute(Attributes::NoInline) &&
+            FAttrs.hasAttribute(Attributes::AlwaysInline)), "Attributes "
+          "'noinline and alwaysinline' are incompatible!", V);
 }
 
 static bool VerifyAttributeCount(const AttrListPtr &Attrs, unsigned Params) {
@@ -661,6 +705,7 @@
   case CallingConv::Cold:
   case CallingConv::X86_FastCall:
   case CallingConv::X86_ThisCall:
+  case CallingConv::Intel_OCL_BI:
   case CallingConv::PTX_Kernel:
   case CallingConv::PTX_Device:
     Assert1(!F.isVarArg(),
@@ -1093,7 +1138,7 @@
 
   // BitCast implies a no-op cast of type only. No bits change.
   // However, you can't cast pointers to anything but pointers.
-  Assert1(DestTy->isPointerTy() == DestTy->isPointerTy(),
+  Assert1(SrcTy->isPointerTy() == DestTy->isPointerTy(),
           "Bitcast requires both operands to be pointer or neither", &I);
   Assert1(SrcBitSize == DestBitSize, "Bitcast requires types of same width",&I);
 
@@ -1170,9 +1215,8 @@
 
       VerifyParameterAttrs(Attr, CS.getArgument(Idx-1)->getType(), false, I);
 
-      Attributes VArgI = Attr & Attribute::VarArgsIncompatible;
-      Assert1(!VArgI, "Attribute " + Attribute::getAsString(VArgI) +
-              " cannot be used for vararg call arguments!", I);
+      Assert1(!Attr.hasIncompatibleWithVarArgsAttrs(),
+              "Attribute 'sret' cannot be used for vararg call arguments!", I);
     }
 
   // Verify that there's no metadata unless it's a direct call to an intrinsic.
@@ -1378,6 +1422,15 @@
             "Load cannot have Release ordering", &LI);
     Assert1(LI.getAlignment() != 0,
             "Atomic load must specify explicit alignment", &LI);
+    if (!ElTy->isPointerTy()) {
+      Assert2(ElTy->isIntegerTy(),
+              "atomic store operand must have integer type!",
+              &LI, ElTy);
+      unsigned Size = ElTy->getPrimitiveSizeInBits();
+      Assert2(Size >= 8 && !(Size & (Size - 1)),
+              "atomic store operand must be power-of-two byte-sized integer",
+              &LI, ElTy);
+    }
   } else {
     Assert1(LI.getSynchScope() == CrossThread,
             "Non-atomic load cannot have SynchronizationScope specified", &LI);
@@ -1444,6 +1497,15 @@
             "Store cannot have Acquire ordering", &SI);
     Assert1(SI.getAlignment() != 0,
             "Atomic store must specify explicit alignment", &SI);
+    if (!ElTy->isPointerTy()) {
+      Assert2(ElTy->isIntegerTy(),
+              "atomic store operand must have integer type!",
+              &SI, ElTy);
+      unsigned Size = ElTy->getPrimitiveSizeInBits();
+      Assert2(Size >= 8 && !(Size & (Size - 1)),
+              "atomic store operand must be power-of-two byte-sized integer",
+              &SI, ElTy);
+    }
   } else {
     Assert1(SI.getSynchScope() == CrossThread,
             "Non-atomic store cannot have SynchronizationScope specified", &SI);
@@ -1471,6 +1533,13 @@
   PointerType *PTy = dyn_cast<PointerType>(CXI.getOperand(0)->getType());
   Assert1(PTy, "First cmpxchg operand must be a pointer.", &CXI);
   Type *ElTy = PTy->getElementType();
+  Assert2(ElTy->isIntegerTy(),
+          "cmpxchg operand must have integer type!",
+          &CXI, ElTy);
+  unsigned Size = ElTy->getPrimitiveSizeInBits();
+  Assert2(Size >= 8 && !(Size & (Size - 1)),
+          "cmpxchg operand must be power-of-two byte-sized integer",
+          &CXI, ElTy);
   Assert2(ElTy == CXI.getOperand(1)->getType(),
           "Expected value type does not match pointer operand type!",
           &CXI, ElTy);
@@ -1488,6 +1557,13 @@
   PointerType *PTy = dyn_cast<PointerType>(RMWI.getOperand(0)->getType());
   Assert1(PTy, "First atomicrmw operand must be a pointer.", &RMWI);
   Type *ElTy = PTy->getElementType();
+  Assert2(ElTy->isIntegerTy(),
+          "atomicrmw operand must have integer type!",
+          &RMWI, ElTy);
+  unsigned Size = ElTy->getPrimitiveSizeInBits();
+  Assert2(Size >= 8 && !(Size & (Size - 1)),
+          "atomicrmw operand must be power-of-two byte-sized integer",
+          &RMWI, ElTy);
   Assert2(ElTy == RMWI.getOperand(1)->getType(),
           "Argument value type does not match pointer operand type!",
           &RMWI, ElTy);
@@ -1536,7 +1612,7 @@
   // landing pad block may be branched to only by the unwind edge of an invoke.
   for (pred_iterator I = pred_begin(BB), E = pred_end(BB); I != E; ++I) {
     const InvokeInst *II = dyn_cast<InvokeInst>((*I)->getTerminator());
-    Assert1(II && II->getUnwindDest() == BB,
+    Assert1(II && II->getUnwindDest() == BB && II->getNormalDest() != BB,
             "Block containing LandingPadInst must be jumped to "
             "only by the unwind edge of an invoke.", &LPI);
   }
@@ -1575,6 +1651,13 @@
 
 void Verifier::verifyDominatesUse(Instruction &I, unsigned i) {
   Instruction *Op = cast<Instruction>(I.getOperand(i));
+  // If the we have an invalid invoke, don't try to compute the dominance.
+  // We already reject it in the invoke specific checks and the dominance
+  // computation doesn't handle multiple edges.
+  if (InvokeInst *II = dyn_cast<InvokeInst>(Op)) {
+    if (II->getNormalDest() == II->getUnwindDest())
+      return;
+  }
 
   const Use &U = I.getOperandUse(i);
   Assert2(InstsInThisBlock.count(Op) || DT->dominates(Op, U),

Modified: llvm/branches/AMDILBackend/projects/CMakeLists.txt
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/projects/CMakeLists.txt?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/projects/CMakeLists.txt (original)
+++ llvm/branches/AMDILBackend/projects/CMakeLists.txt Tue Jan 15 11:16:16 2013
@@ -10,3 +10,10 @@
     endif()
   endif()
 endforeach(entry)
+
+# Also add in the compiler-rt tree if present and we have a sufficiently
+# recent version of CMake.
+if(${CMAKE_VERSION} VERSION_GREATER 2.8.7 AND
+   ${LLVM_BUILD_RUNTIME})
+  add_llvm_external_project(compiler-rt)
+endif()

Modified: llvm/branches/AMDILBackend/projects/sample/Makefile.llvm.rules
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/projects/sample/Makefile.llvm.rules?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/projects/sample/Makefile.llvm.rules (original)
+++ llvm/branches/AMDILBackend/projects/sample/Makefile.llvm.rules Tue Jan 15 11:16:16 2013
@@ -1437,7 +1437,7 @@
 uninstall-local::
 	$(Echo) Uninstall circumvented with NO_INSTALL
 else
-DestTool = $(DESTDIR)$(PROJ_bindir)/$(TOOLEXENAME)
+DestTool = $(DESTDIR)$(PROJ_bindir)/$(program_prefix)$(TOOLEXENAME)
 
 install-local:: $(DestTool)
 
@@ -1451,7 +1451,7 @@
 
 # TOOLALIAS install.
 ifdef TOOLALIAS
-DestToolAlias = $(DESTDIR)$(PROJ_bindir)/$(TOOLALIAS)$(EXEEXT)
+DestToolAlias = $(DESTDIR)$(PROJ_bindir)/$(program_prefix)$(TOOLALIAS)$(EXEEXT)
 
 install-local:: $(DestToolAlias)
 

Modified: llvm/branches/AMDILBackend/projects/sample/autoconf/configure.ac
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/projects/sample/autoconf/configure.ac?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/projects/sample/autoconf/configure.ac (original)
+++ llvm/branches/AMDILBackend/projects/sample/autoconf/configure.ac Tue Jan 15 11:16:16 2013
@@ -304,8 +304,8 @@
   sparc*-*)               llvm_cv_target_arch="Sparc" ;;
   powerpc*-*)             llvm_cv_target_arch="PowerPC" ;;
   arm*-*)                 llvm_cv_target_arch="ARM" ;;
-  mips-*)                 llvm_cv_target_arch="Mips" ;;
-  mipsel-*)               llvm_cv_target_arch="Mips" ;;
+  mips-* | mips64-*)      llvm_cv_target_arch="Mips" ;;
+  mipsel-* | mips64el-*)  llvm_cv_target_arch="Mips" ;;
   xcore-*)                llvm_cv_target_arch="XCore" ;;
   msp430-*)               llvm_cv_target_arch="MSP430" ;;
   hexagon-*)              llvm_cv_target_arch="Hexagon" ;;

Modified: llvm/branches/AMDILBackend/projects/sample/configure
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/projects/sample/configure?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/projects/sample/configure (original)
+++ llvm/branches/AMDILBackend/projects/sample/configure Tue Jan 15 11:16:16 2013
@@ -3840,8 +3840,8 @@
   sparc*-*)               llvm_cv_target_arch="Sparc" ;;
   powerpc*-*)             llvm_cv_target_arch="PowerPC" ;;
   arm*-*)                 llvm_cv_target_arch="ARM" ;;
-  mips-*)                 llvm_cv_target_arch="Mips" ;;
-  mipsel-*)               llvm_cv_target_arch="Mips" ;;
+  mips-* | mips64-*)      llvm_cv_target_arch="Mips" ;;
+  mipsel-* | mips64el-*)  llvm_cv_target_arch="Mips" ;;
   xcore-*)                llvm_cv_target_arch="XCore" ;;
   msp430-*)               llvm_cv_target_arch="MSP430" ;;
   hexagon-*)              llvm_cv_target_arch="Hexagon" ;;

Modified: llvm/branches/AMDILBackend/runtime/libprofile/CMakeLists.txt
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/runtime/libprofile/CMakeLists.txt?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/runtime/libprofile/CMakeLists.txt (original)
+++ llvm/branches/AMDILBackend/runtime/libprofile/CMakeLists.txt Tue Jan 15 11:16:16 2013
@@ -13,7 +13,8 @@
   PROPERTIES
   OUTPUT_NAME "profile_rt" )
 
-add_llvm_loadable_module( profile_rt-shared ${SOURCES} )
+set(BUILD_SHARED_LIBS ON)
+add_llvm_library( profile_rt-shared ${SOURCES} )
 set_target_properties( profile_rt-shared
   PROPERTIES
   OUTPUT_NAME "profile_rt" )

Modified: llvm/branches/AMDILBackend/runtime/libprofile/CommonProfiling.c
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/runtime/libprofile/CommonProfiling.c?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/runtime/libprofile/CommonProfiling.c (original)
+++ llvm/branches/AMDILBackend/runtime/libprofile/CommonProfiling.c Tue Jan 15 11:16:16 2013
@@ -28,14 +28,35 @@
 
 static char *SavedArgs = 0;
 static unsigned SavedArgsLength = 0;
+static const char *SavedEnvVar = 0;
 
 static const char *OutputFilename = "llvmprof.out";
 
+/* check_environment_variable - Check to see if the LLVMPROF_OUTPUT environment
+ * variable is set.  If it is then save it and set OutputFilename.
+ */
+static void check_environment_variable(void) {
+  const char *EnvVar;
+  if (SavedEnvVar) return; /* Guarantee that we can't leak memory. */
+
+  if ((EnvVar = getenv("LLVMPROF_OUTPUT")) != NULL) {
+    /* The string that getenv returns is allowed to be statically allocated,
+     * which means it may be changed by future calls to getenv, so copy it.
+     */
+    SavedEnvVar = strdup(EnvVar);
+    OutputFilename = SavedEnvVar;
+  }
+}
+
 /* save_arguments - Save argc and argv as passed into the program for the file
  * we output.
+ * If either the LLVMPROF_OUTPUT environment variable or the -llvmprof-output
+ * command line argument are set then change OutputFilename to the provided
+ * value.  The command line argument value overrides the environment variable.
  */
 int save_arguments(int argc, const char **argv) {
   unsigned Length, i;
+  if (!SavedEnvVar && !SavedArgs) check_environment_variable();
   if (SavedArgs || !argv) return argc;  /* This can be called multiple times */
 
   /* Check to see if there are any arguments passed into the program for the
@@ -54,6 +75,7 @@
         puts("-llvmprof-output requires a filename argument!");
       else {
         OutputFilename = strdup(argv[1]);
+        if (SavedEnvVar) { free((void *)SavedEnvVar); SavedEnvVar = 0; }
         memmove((char**)&argv[1], &argv[2], (argc-1)*sizeof(char*));
         --argc;
       }

Modified: llvm/branches/AMDILBackend/runtime/libprofile/Makefile
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/runtime/libprofile/Makefile?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/runtime/libprofile/Makefile (original)
+++ llvm/branches/AMDILBackend/runtime/libprofile/Makefile Tue Jan 15 11:16:16 2013
@@ -44,8 +44,15 @@
     # command line.
     DARWIN_VERS := $(shell echo $(TARGET_TRIPLE) | sed 's/.*darwin\([0-9]*\).*/\1/')
     ifneq ($(DARWIN_VERS),8)
-       LLVMLibsOptions    := $(LLVMLibsOptions)  \
+       LLVMLibsOptions    := $(LLVMLibsOptions) \
                             -Wl,-install_name \
                             -Wl,"@executable_path/../lib/lib$(LIBRARYNAME)$(SHLIBEXT)"
     endif
+
+    # If we're doing an Apple-style build, add the LTO object path.
+    ifeq ($(RC_BUILDIT),YES)
+       TempFile           := $(shell mkdir -p ${OBJROOT}/dSYMs ; mktemp ${OBJROOT}/dSYMs/profile_rt-lto.XXXXXX)
+       LLVMLibsOptions    := $(LLVMLibsOptions) \
+                             -Wl,-object_path_lto -Wl,$(TempFile)
+    endif
 endif

Modified: llvm/branches/AMDILBackend/runtime/libprofile/Profiling.h
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/runtime/libprofile/Profiling.h?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/runtime/libprofile/Profiling.h (original)
+++ llvm/branches/AMDILBackend/runtime/libprofile/Profiling.h Tue Jan 15 11:16:16 2013
@@ -15,7 +15,7 @@
 #ifndef PROFILING_H
 #define PROFILING_H
 
-#include "llvm/Analysis/ProfileInfoTypes.h" /* for enum ProfilingType */
+#include "llvm/Analysis/ProfileDataTypes.h" /* for enum ProfilingType */
 
 /* save_arguments - Save argc and argv as passed into the program for the file
  * we output.

Modified: llvm/branches/AMDILBackend/test/Analysis/BasicAA/nocapture.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/Analysis/BasicAA/nocapture.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/Analysis/BasicAA/nocapture.ll (original)
+++ llvm/branches/AMDILBackend/test/Analysis/BasicAA/nocapture.ll Tue Jan 15 11:16:16 2013
@@ -13,3 +13,24 @@
        ret i32 %c
 }
 
+declare void @test3(i32** %p, i32* %q) nounwind
+
+define i32 @test4(i32* noalias nocapture %p) nounwind {
+; CHECK: call void @test3
+; CHECK: store i32 0, i32* %p
+; CHECK: store i32 1, i32* %x
+; CHECK: %y = load i32* %p
+; CHECK: ret i32 %y
+entry:
+       %q = alloca i32*
+       ; Here test3 might store %p to %q. This doesn't violate %p's nocapture
+       ; attribute since the copy doesn't outlive the function.
+       call void @test3(i32** %q, i32* %p) nounwind
+       store i32 0, i32* %p
+       %x = load i32** %q
+       ; This store might write to %p and so we can't eliminate the subsequent
+       ; load
+       store i32 1, i32* %x
+       %y = load i32* %p
+       ret i32 %y
+}

Modified: llvm/branches/AMDILBackend/test/Analysis/BranchProbabilityInfo/basic.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/Analysis/BranchProbabilityInfo/basic.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/Analysis/BranchProbabilityInfo/basic.ll (original)
+++ llvm/branches/AMDILBackend/test/Analysis/BranchProbabilityInfo/basic.ll Tue Jan 15 11:16:16 2013
@@ -88,3 +88,30 @@
 }
 
 !1 = metadata !{metadata !"branch_weights", i32 4, i32 4, i32 64, i32 4, i32 4}
+
+define i32 @test4(i32 %x) nounwind uwtable readnone ssp {
+; CHECK: Printing analysis {{.*}} for function 'test4'
+entry:
+  %conv = sext i32 %x to i64
+  switch i64 %conv, label %return [
+    i64 0, label %sw.bb
+    i64 1, label %sw.bb
+    i64 2, label %sw.bb
+    i64 5, label %sw.bb1
+  ], !prof !2
+; CHECK: edge entry -> return probability is 7 / 85
+; CHECK: edge entry -> sw.bb probability is 14 / 85
+; CHECK: edge entry -> sw.bb1 probability is 64 / 85
+
+sw.bb:
+  br label %return
+
+sw.bb1:
+  br label %return
+
+return:
+  %retval.0 = phi i32 [ 5, %sw.bb1 ], [ 1, %sw.bb ], [ 0, %entry ]
+  ret i32 %retval.0
+}
+
+!2 = metadata !{metadata !"branch_weights", i32 7, i32 6, i32 4, i32 4, i32 64}

Removed: llvm/branches/AMDILBackend/test/Analysis/LoopDependenceAnalysis/alias.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/Analysis/LoopDependenceAnalysis/alias.ll?rev=172540&view=auto
==============================================================================
--- llvm/branches/AMDILBackend/test/Analysis/LoopDependenceAnalysis/alias.ll (original)
+++ llvm/branches/AMDILBackend/test/Analysis/LoopDependenceAnalysis/alias.ll (removed)
@@ -1,44 +0,0 @@
-; RUN: opt < %s -analyze -basicaa -lda | FileCheck %s
-
-;; x[5] = x[6] // with x being a pointer passed as argument
-
-define void @f1(i32* nocapture %xptr) nounwind {
-entry:
-  %x.ld.addr = getelementptr i32* %xptr, i64 6
-  %x.st.addr = getelementptr i32* %xptr, i64 5
-  br label %for.body
-
-for.body:
-  %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
-  %x = load i32* %x.ld.addr
-  store i32 %x, i32* %x.st.addr
-; CHECK: 0,1: dep
-  %i.next = add i64 %i, 1
-  %exitcond = icmp eq i64 %i.next, 256
-  br i1 %exitcond, label %for.end, label %for.body
-
-for.end:
-  ret void
-}
-
-;; x[5] = x[6] // with x being an array on the stack
-
-define void @foo(...) nounwind {
-entry:
-  %xptr = alloca [256 x i32], align 4
-  %x.ld.addr = getelementptr [256 x i32]* %xptr, i64 0, i64 6
-  %x.st.addr = getelementptr [256 x i32]* %xptr, i64 0, i64 5
-  br label %for.body
-
-for.body:
-  %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
-  %x = load i32* %x.ld.addr
-  store i32 %x, i32* %x.st.addr
-; CHECK: 0,1: ind
-  %i.next = add i64 %i, 1
-  %exitcond = icmp eq i64 %i.next, 256
-  br i1 %exitcond, label %for.end, label %for.body
-
-for.end:
-  ret void
-}

Removed: llvm/branches/AMDILBackend/test/Analysis/LoopDependenceAnalysis/lit.local.cfg
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/Analysis/LoopDependenceAnalysis/lit.local.cfg?rev=172540&view=auto
==============================================================================
--- llvm/branches/AMDILBackend/test/Analysis/LoopDependenceAnalysis/lit.local.cfg (original)
+++ llvm/branches/AMDILBackend/test/Analysis/LoopDependenceAnalysis/lit.local.cfg (removed)
@@ -1 +0,0 @@
-config.suffixes = ['.ll', '.c', '.cpp']

Removed: llvm/branches/AMDILBackend/test/Analysis/LoopDependenceAnalysis/siv-strong.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/Analysis/LoopDependenceAnalysis/siv-strong.ll?rev=172540&view=auto
==============================================================================
--- llvm/branches/AMDILBackend/test/Analysis/LoopDependenceAnalysis/siv-strong.ll (original)
+++ llvm/branches/AMDILBackend/test/Analysis/LoopDependenceAnalysis/siv-strong.ll (removed)
@@ -1,110 +0,0 @@
-; RUN: opt < %s -analyze -basicaa -lda | FileCheck %s
-
- at x = common global [256 x i32] zeroinitializer, align 4
- at y = common global [256 x i32] zeroinitializer, align 4
-
-;; for (i = 0; i < 256; i++)
-;;   x[i] = x[i] + y[i]
-
-define void @f1(...) nounwind {
-entry:
-  br label %for.body
-
-for.body:
-  %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
-  %y.addr = getelementptr [256 x i32]* @y, i64 0, i64 %i
-  %x.addr = getelementptr [256 x i32]* @x, i64 0, i64 %i
-  %x = load i32* %x.addr      ; 0
-  %y = load i32* %y.addr      ; 1
-  %r = add i32 %y, %x
-  store i32 %r, i32* %x.addr  ; 2
-; CHECK: 0,2: dep
-; CHECK: 1,2: ind
-  %i.next = add i64 %i, 1
-  %exitcond = icmp eq i64 %i.next, 256
-  br i1 %exitcond, label %for.end, label %for.body
-
-for.end:
-  ret void
-}
-
-;; for (i = 0; i < 256; i++)
-;;   x[i+1] = x[i] + y[i]
-
-define void @f2(...) nounwind {
-entry:
-  br label %for.body
-
-for.body:
-  %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
-  %y.ld.addr = getelementptr [256 x i32]* @y, i64 0, i64 %i
-  %x.ld.addr = getelementptr [256 x i32]* @x, i64 0, i64 %i
-  %i.next = add i64 %i, 1
-  %x.st.addr = getelementptr [256 x i32]* @x, i64 0, i64 %i.next
-  %x = load i32* %x.ld.addr     ; 0
-  %y = load i32* %y.ld.addr     ; 1
-  %r = add i32 %y, %x
-  store i32 %r, i32* %x.st.addr ; 2
-; CHECK: 0,2: dep
-; CHECK: 1,2: ind
-  %exitcond = icmp eq i64 %i.next, 256
-  br i1 %exitcond, label %for.end, label %for.body
-
-for.end:
-  ret void
-}
-
-;; for (i = 0; i < 10; i++)
-;;   x[i+20] = x[i] + y[i]
-
-define void @f3(...) nounwind {
-entry:
-  br label %for.body
-
-for.body:
-  %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
-  %y.ld.addr = getelementptr [256 x i32]* @y, i64 0, i64 %i
-  %x.ld.addr = getelementptr [256 x i32]* @x, i64 0, i64 %i
-  %i.20 = add i64 %i, 20
-  %x.st.addr = getelementptr [256 x i32]* @x, i64 0, i64 %i.20
-  %x = load i32* %x.ld.addr     ; 0
-  %y = load i32* %y.ld.addr     ; 1
-  %r = add i32 %y, %x
-  store i32 %r, i32* %x.st.addr ; 2
-; CHECK: 0,2: dep
-; CHECK: 1,2: ind
-  %i.next = add i64 %i, 1
-  %exitcond = icmp eq i64 %i.next, 10
-  br i1 %exitcond, label %for.end, label %for.body
-
-for.end:
-  ret void
-}
-
-;; for (i = 0; i < 10; i++)
-;;   x[10*i+1] = x[10*i] + y[i]
-
-define void @f4(...) nounwind {
-entry:
-  br label %for.body
-
-for.body:
-  %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
-  %i.10 = mul i64 %i, 10
-  %y.ld.addr = getelementptr [256 x i32]* @y, i64 0, i64 %i.10
-  %x.ld.addr = getelementptr [256 x i32]* @x, i64 0, i64 %i.10
-  %i.10.1 = add i64 %i.10, 1
-  %x.st.addr = getelementptr [256 x i32]* @x, i64 0, i64 %i.10.1
-  %x = load i32* %x.ld.addr     ; 0
-  %y = load i32* %y.ld.addr     ; 1
-  %r = add i32 %y, %x
-  store i32 %r, i32* %x.st.addr ; 2
-; CHECK: 0,2: dep
-; CHECK: 1,2: ind
-  %i.next = add i64 %i, 1
-  %exitcond = icmp eq i64 %i.next, 10
-  br i1 %exitcond, label %for.end, label %for.body
-
-for.end:
-  ret void
-}

Removed: llvm/branches/AMDILBackend/test/Analysis/LoopDependenceAnalysis/siv-weak-crossing.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/Analysis/LoopDependenceAnalysis/siv-weak-crossing.ll?rev=172540&view=auto
==============================================================================
--- llvm/branches/AMDILBackend/test/Analysis/LoopDependenceAnalysis/siv-weak-crossing.ll (original)
+++ llvm/branches/AMDILBackend/test/Analysis/LoopDependenceAnalysis/siv-weak-crossing.ll (removed)
@@ -1,118 +0,0 @@
-; RUN: opt < %s -analyze -basicaa -lda | FileCheck %s
-
- at x = common global [256 x i32] zeroinitializer, align 4
- at y = common global [256 x i32] zeroinitializer, align 4
-
-;; for (i = 0; i < 256; i++)
-;;   x[i] = x[255 - i] + y[i]
-
-define void @f1(...) nounwind {
-entry:
-  br label %for.body
-
-for.body:
-  %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
-  %i.255 = sub i64 255, %i
-  %y.ld.addr = getelementptr [256 x i32]* @y, i64 0, i64 %i
-  %x.ld.addr = getelementptr [256 x i32]* @x, i64 0, i64 %i.255
-  %x.st.addr = getelementptr [256 x i32]* @x, i64 0, i64 %i
-  %x = load i32* %x.ld.addr     ; 0
-  %y = load i32* %y.ld.addr     ; 1
-  %r = add i32 %y, %x
-  store i32 %r, i32* %x.st.addr ; 2
-; CHECK: 0,2: dep
-; CHECK: 1,2: ind
-  %i.next = add i64 %i, 1
-  %exitcond = icmp eq i64 %i.next, 256
-  br i1 %exitcond, label %for.end, label %for.body
-
-for.end:
-  ret void
-}
-
-;; for (i = 0; i < 100; i++)
-;;   x[i] = x[255 - i] + y[i]
-
-define void @f2(...) nounwind {
-entry:
-  br label %for.body
-
-for.body:
-  %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
-  %i.255 = sub i64 255, %i
-  %y.ld.addr = getelementptr [256 x i32]* @y, i64 0, i64 %i
-  %x.ld.addr = getelementptr [256 x i32]* @x, i64 0, i64 %i.255
-  %x.st.addr = getelementptr [256 x i32]* @x, i64 0, i64 %i
-  %x = load i32* %x.ld.addr     ; 0
-  %y = load i32* %y.ld.addr     ; 1
-  %r = add i32 %y, %x
-  store i32 %r, i32* %x.st.addr ; 2
-; CHECK: 0,2: dep
-; CHECK: 1,2: ind
-  %i.next = add i64 %i, 1
-  %exitcond = icmp eq i64 %i.next, 100
-  br i1 %exitcond, label %for.end, label %for.body
-
-for.end:
-  ret void
-}
-
-;; // the first iteration (i=0) leads to an out-of-bounds access of x. as the
-;; // result of this access is undefined, _any_ dependence result is safe.
-;; for (i = 0; i < 256; i++)
-;;   x[i] = x[256 - i] + y[i]
-
-define void @f3(...) nounwind {
-entry:
-  br label %for.body
-
-for.body:
-  %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
-  %i.256 = sub i64 0, %i
-  %y.ld.addr = getelementptr [256 x i32]* @y, i64 0, i64 %i
-  %x.ld.addr = getelementptr [256 x i32]* @x, i64 1, i64 %i.256
-  %x.st.addr = getelementptr [256 x i32]* @x, i64 0, i64 %i
-  %x = load i32* %x.ld.addr     ; 0
-  %y = load i32* %y.ld.addr     ; 1
-  %r = add i32 %y, %x
-  store i32 %r, i32* %x.st.addr ; 2
-; CHECK: 0,2: dep
-; CHECK: 1,2:
-  %i.next = add i64 %i, 1
-  %exitcond = icmp eq i64 %i.next, 256
-  br i1 %exitcond, label %for.end, label %for.body
-
-for.end:
-  ret void
-}
-
-;; // slightly contrived but valid IR for the following loop, where all
-;; // accesses in all iterations are within bounds. while this example's first
-;; // (ZIV-)subscript is (0, 1), accesses are dependent.
-;; for (i = 1; i < 256; i++)
-;;   x[i] = x[256 - i] + y[i]
-
-define void @f4(...) nounwind {
-entry:
-  br label %for.body
-
-for.body:
-  %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
-  %i.1 = add i64 1, %i
-  %i.256 = sub i64 -1, %i
-  %y.ld.addr = getelementptr [256 x i32]* @y, i64 0, i64 %i.1
-  %x.ld.addr = getelementptr [256 x i32]* @x, i64 1, i64 %i.256
-  %x.st.addr = getelementptr [256 x i32]* @x, i64 0, i64 %i.1
-  %x = load i32* %x.ld.addr     ; 0
-  %y = load i32* %y.ld.addr     ; 1
-  %r = add i32 %y, %x
-  store i32 %r, i32* %x.st.addr ; 2
-; CHECK: 0,2: dep
-; CHECK: 1,2: ind
-  %i.next = add i64 %i, 1
-  %exitcond = icmp eq i64 %i.next, 256
-  br i1 %exitcond, label %for.end, label %for.body
-
-for.end:
-  ret void
-}

Removed: llvm/branches/AMDILBackend/test/Analysis/LoopDependenceAnalysis/siv-weak-zero.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/Analysis/LoopDependenceAnalysis/siv-weak-zero.ll?rev=172540&view=auto
==============================================================================
--- llvm/branches/AMDILBackend/test/Analysis/LoopDependenceAnalysis/siv-weak-zero.ll (original)
+++ llvm/branches/AMDILBackend/test/Analysis/LoopDependenceAnalysis/siv-weak-zero.ll (removed)
@@ -1,56 +0,0 @@
-; RUN: opt < %s -analyze -basicaa -lda | FileCheck %s
-
- at x = common global [256 x i32] zeroinitializer, align 4
- at y = common global [256 x i32] zeroinitializer, align 4
-
-;; for (i = 0; i < 256; i++)
-;;   x[i] = x[42] + y[i]
-
-define void @f1(...) nounwind {
-entry:
-  %x.ld.addr = getelementptr [256 x i32]* @x, i64 0, i64 42
-  br label %for.body
-
-for.body:
-  %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
-  %x.addr = getelementptr [256 x i32]* @x, i64 0, i64 %i
-  %y.addr = getelementptr [256 x i32]* @y, i64 0, i64 %i
-  %x = load i32* %x.ld.addr   ; 0
-  %y = load i32* %y.addr      ; 1
-  %r = add i32 %y, %x
-  store i32 %r, i32* %x.addr  ; 2
-; CHECK: 0,2: dep
-; CHECK: 1,2: ind
-  %i.next = add i64 %i, 1
-  %exitcond = icmp eq i64 %i.next, 256
-  br i1 %exitcond, label %for.end, label %for.body
-
-for.end:
-  ret void
-}
-
-;; for (i = 0; i < 250; i++)
-;;   x[i] = x[255] + y[i]
-
-define void @f2(...) nounwind {
-entry:
-  %x.ld.addr = getelementptr [256 x i32]* @x, i64 0, i64 255
-  br label %for.body
-
-for.body:
-  %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
-  %x.addr = getelementptr [256 x i32]* @x, i64 0, i64 %i
-  %y.addr = getelementptr [256 x i32]* @y, i64 0, i64 %i
-  %x = load i32* %x.ld.addr   ; 0
-  %y = load i32* %y.addr      ; 1
-  %r = add i32 %y, %x
-  store i32 %r, i32* %x.addr  ; 2
-; CHECK: 0,2: dep
-; CHECK: 1,2: ind
-  %i.next = add i64 %i, 1
-  %exitcond = icmp eq i64 %i.next, 250
-  br i1 %exitcond, label %for.end, label %for.body
-
-for.end:
-  ret void
-}

Removed: llvm/branches/AMDILBackend/test/Analysis/LoopDependenceAnalysis/ziv.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/Analysis/LoopDependenceAnalysis/ziv.ll?rev=172540&view=auto
==============================================================================
--- llvm/branches/AMDILBackend/test/Analysis/LoopDependenceAnalysis/ziv.ll (original)
+++ llvm/branches/AMDILBackend/test/Analysis/LoopDependenceAnalysis/ziv.ll (removed)
@@ -1,63 +0,0 @@
-; RUN: opt < %s -analyze -basicaa -lda | FileCheck %s
-
- at x = common global [256 x i32] zeroinitializer, align 4
-
-;; x[5] = x[6]
-
-define void @f1(...) nounwind {
-entry:
-  br label %for.body
-
-for.body:
-  %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
-  %x = load i32* getelementptr ([256 x i32]* @x, i32 0, i64 6)
-  store i32 %x, i32* getelementptr ([256 x i32]* @x, i32 0, i64 5)
-; CHECK: 0,1: ind
-  %i.next = add i64 %i, 1
-  %exitcond = icmp eq i64 %i.next, 256
-  br i1 %exitcond, label %for.end, label %for.body
-
-for.end:
-  ret void
-}
-
-;; x[c] = x[c+1] // with c being a loop-invariant constant
-
-define void @f2(i64 %c0) nounwind {
-entry:
-  %c1 = add i64 %c0, 1
-  %x.ld.addr = getelementptr [256 x i32]* @x, i64 0, i64 %c0
-  %x.st.addr = getelementptr [256 x i32]* @x, i64 0, i64 %c1
-  br label %for.body
-
-for.body:
-  %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
-  %x = load i32* %x.ld.addr
-  store i32 %x, i32* %x.st.addr
-; CHECK: 0,1: ind
-  %i.next = add i64 %i, 1
-  %exitcond = icmp eq i64 %i.next, 256
-  br i1 %exitcond, label %for.end, label %for.body
-
-for.end:
-  ret void
-}
-
-;; x[6] = x[6]
-
-define void @f3(...) nounwind {
-entry:
-  br label %for.body
-
-for.body:
-  %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
-  %x = load i32* getelementptr ([256 x i32]* @x, i32 0, i64 6)
-  store i32 %x, i32* getelementptr ([256 x i32]* @x, i32 0, i64 6)
-; CHECK: 0,1: dep
-  %i.next = add i64 %i, 1
-  %exitcond = icmp eq i64 %i.next, 256
-  br i1 %exitcond, label %for.end, label %for.body
-
-for.end:
-  ret void
-}

Modified: llvm/branches/AMDILBackend/test/Analysis/ScalarEvolution/2012-05-29-MulAddRec.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/Analysis/ScalarEvolution/2012-05-29-MulAddRec.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/Analysis/ScalarEvolution/2012-05-29-MulAddRec.ll (original)
+++ llvm/branches/AMDILBackend/test/Analysis/ScalarEvolution/2012-05-29-MulAddRec.ll Tue Jan 15 11:16:16 2013
@@ -16,7 +16,7 @@
 ; CHECK: for.body:
 ; CHECK: %inc.9 = add i8 %inc.8, 1
 ; CHECK: %0 = add i8 %inc1, 10
-; CHEKC: br label %for.cond
+; CHECK: br label %for.cond
 
 target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
 define void @func() noreturn nounwind uwtable ssp {

Removed: llvm/branches/AMDILBackend/test/Assembler/2003-06-17-InvokeDisassemble.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/Assembler/2003-06-17-InvokeDisassemble.ll?rev=172540&view=auto
==============================================================================
--- llvm/branches/AMDILBackend/test/Assembler/2003-06-17-InvokeDisassemble.ll (original)
+++ llvm/branches/AMDILBackend/test/Assembler/2003-06-17-InvokeDisassemble.ll (removed)
@@ -1,13 +0,0 @@
-; RUN: llvm-as < %s | llvm-dis
-
-define void @test() {
-  invoke void @test( )
-    to label %Next unwind label %Next
-
-Next:           ; preds = %0, %0
-  %lpad = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
-            cleanup
-  ret void
-}
-
-declare i32 @__gxx_personality_v0(...)

Modified: llvm/branches/AMDILBackend/test/Assembler/2008-09-02-FunctionNotes2.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/Assembler/2008-09-02-FunctionNotes2.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/Assembler/2008-09-02-FunctionNotes2.ll (original)
+++ llvm/branches/AMDILBackend/test/Assembler/2008-09-02-FunctionNotes2.ll Tue Jan 15 11:16:16 2013
@@ -1,5 +1,5 @@
 ; Test function notes
-; RUN: not llvm-as %s -o /dev/null 2>&1 | grep "Attributes noinline alwaysinline are incompatible"
+; RUN: not llvm-as %s -o /dev/null 2>&1 | grep "Attributes 'noinline and alwaysinline' are incompatible"
 define void @fn1() alwaysinline  noinline {
   ret void
 }

Modified: llvm/branches/AMDILBackend/test/Bindings/Ocaml/ipo_opts.ml
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/Bindings/Ocaml/ipo_opts.ml?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/Bindings/Ocaml/ipo_opts.ml (original)
+++ llvm/branches/AMDILBackend/test/Bindings/Ocaml/ipo_opts.ml Tue Jan 15 11:16:16 2013
@@ -43,10 +43,10 @@
       ignore (build_ret (build_call fn [| |] "" b) b);
   end;
 
-  let td = TargetData.create (target_triple m) in
+  let td = DataLayout.create (target_triple m) in
   
   ignore (PassManager.create ()
-           ++ TargetData.add td
+           ++ DataLayout.add td
            ++ add_argument_promotion
            ++ add_constant_merge
            ++ add_dead_arg_elimination
@@ -63,7 +63,7 @@
            ++ PassManager.run_module m
            ++ PassManager.dispose);
 
-  TargetData.dispose td
+  DataLayout.dispose td
 
 
 (*===-- Driver ------------------------------------------------------------===*)

Modified: llvm/branches/AMDILBackend/test/Bindings/Ocaml/scalar_opts.ml
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/Bindings/Ocaml/scalar_opts.ml?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/Bindings/Ocaml/scalar_opts.ml (original)
+++ llvm/branches/AMDILBackend/test/Bindings/Ocaml/scalar_opts.ml Tue Jan 15 11:16:16 2013
@@ -38,10 +38,10 @@
   let fn = define_function "fn" fty m in
   ignore (build_ret_void (builder_at_end context (entry_block fn)));
   
-  let td = TargetData.create (target_triple m) in
+  let td = DataLayout.create (target_triple m) in
   
   ignore (PassManager.create_function m
-           ++ TargetData.add td
+           ++ DataLayout.add td
            ++ add_verifier
            ++ add_constant_propagation
            ++ add_sccp
@@ -78,7 +78,7 @@
            ++ PassManager.finalize
            ++ PassManager.dispose);
   
-  TargetData.dispose td
+  DataLayout.dispose td
 
 
 (*===-- Driver ------------------------------------------------------------===*)

Modified: llvm/branches/AMDILBackend/test/Bindings/Ocaml/target.ml
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/Bindings/Ocaml/target.ml?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/Bindings/Ocaml/target.ml (original)
+++ llvm/branches/AMDILBackend/test/Bindings/Ocaml/target.ml Tue Jan 15 11:16:16 2013
@@ -33,10 +33,10 @@
 (*===-- Target Data -------------------------------------------------------===*)
 
 let test_target_data () =
-  let td = TargetData.create (target_triple m) in
+  let td = DataLayout.create (target_triple m) in
   let sty = struct_type context [| i32_type; i64_type |] in
   
-  ignore (TargetData.as_string td);
+  ignore (DataLayout.as_string td);
   ignore (byte_order td);
   ignore (pointer_size td);
   ignore (intptr_type td);
@@ -49,7 +49,7 @@
   ignore (element_at_offset td sty (Int64.of_int 1));
   ignore (offset_of_element td sty 1);
   
-  TargetData.dispose td
+  DataLayout.dispose td
 
 
 (*===-- Driver ------------------------------------------------------------===*)

Modified: llvm/branches/AMDILBackend/test/Bindings/Ocaml/vmcore.ml
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/Bindings/Ocaml/vmcore.ml?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/Bindings/Ocaml/vmcore.ml (original)
+++ llvm/branches/AMDILBackend/test/Bindings/Ocaml/vmcore.ml Tue Jan 15 11:16:16 2013
@@ -113,14 +113,14 @@
   ignore (define_global "const_int_string" c m);
   insist (i32_type = type_of c);
 
-  (* RUN: grep 'const_string.*"cruel\00world"' < %t.ll
+  (* RUN: grep 'const_string.*"cruel\\00world"' < %t.ll
    *)
   group "string";
   let c = const_string context "cruel\000world" in
   ignore (define_global "const_string" c m);
   insist ((array_type i8_type 11) = type_of c);
 
-  (* RUN: grep 'const_stringz.*"hi\00again\00"' < %t.ll
+  (* RUN: grep 'const_stringz.*"hi\\00again\\00"' < %t.ll
    *)
   group "stringz";
   let c = const_stringz context "hi\000again" in
@@ -187,7 +187,7 @@
   ignore (define_global "const_all_ones" c m);
 
   group "pointer null"; begin
-    (* RUN: grep "const_pointer_null = global i64* null" < %t.ll
+    (* RUN: grep "const_pointer_null = global i64\* null" < %t.ll
      *)
     let c = const_pointer_null (pointer_type i64_type) in
     ignore (define_global "const_pointer_null" c m);
@@ -542,7 +542,7 @@
 (*===-- Aliases -----------------------------------------------------------===*)
 
 let test_aliases () =
-  (* RUN: grep "@alias = alias i32* @aliasee" < %t.ll
+  (* RUN: grep "@alias = alias i32\* @aliasee" < %t.ll
    *)
   let v = declare_global i32_type "aliasee" m in
   ignore (add_alias m (pointer_type i32_type) v "alias")
@@ -554,7 +554,7 @@
   let ty = function_type i32_type [| i32_type; i64_type |] in
   let ty2 = function_type i8_type [| i8_type; i64_type |] in
   
-  (* RUN: grep "declare i32 @Fn1\(i32, i64\)" < %t.ll
+  (* RUN: grep 'declare i32 @Fn1(i32, i64)' < %t.ll
    *)
   begin group "declare";
     insist (None = lookup_function "Fn1" m);
@@ -935,7 +935,7 @@
 
   group "malloc/free"; begin
       (* RUN: grep "call.*@malloc(i32 ptrtoint" < %t.ll
-       * RUN: grep "call.*@free(i8*" < %t.ll
+       * RUN: grep "call.*@free(i8\*" < %t.ll
        * RUN: grep "call.*@malloc(i32 %" < %t.ll
        *)
       let bb1 = append_block context "MallocBlock1" fn in
@@ -947,7 +947,7 @@
   end;
 
   group "indirectbr"; begin
-    (* RUN: grep "indirectbr i8* blockaddress(@X7, %IBRBlock2), [label %IBRBlock2, label %IBRBlock3]" < %t.ll
+    (* RUN: grep "indirectbr i8\* blockaddress(@X7, %IBRBlock2), \[label %IBRBlock2, label %IBRBlock3\]" < %t.ll
      *)
     let bb1 = append_block context "IBRBlock1" fn in
 
@@ -1054,10 +1054,10 @@
 
     (* RUN: grep "%build_alloca = alloca i32" < %t.ll
      * RUN: grep "%build_array_alloca = alloca i32, i32 %P2" < %t.ll
-     * RUN: grep "%build_load = load i32* %build_array_alloca" < %t.ll
-     * RUN: grep "store i32 %P2, i32* %build_alloca" < %t.ll
-     * RUN: grep "%build_gep = getelementptr i32* %build_array_alloca, i32 %P2" < %t.ll
-     * RUN: grep "%build_in_bounds_gep = getelementptr inbounds i32* %build_array_alloca, i32 %P2" < %t.ll
+     * RUN: grep "%build_load = load i32\* %build_array_alloca" < %t.ll
+     * RUN: grep "store i32 %P2, i32\* %build_alloca" < %t.ll
+     * RUN: grep "%build_gep = getelementptr i32\* %build_array_alloca, i32 %P2" < %t.ll
+     * RUN: grep "%build_in_bounds_gep = getelementptr inbounds i32\* %build_array_alloca, i32 %P2" < %t.ll
      * RUN: grep "%build_struct_gep = getelementptr inbounds.*%build_alloca2, i32 0, i32 1" < %t.ll
      *)
     let alloca = build_alloca i32_type "build_alloca" b in
@@ -1106,14 +1106,14 @@
      * RUN: grep "%build_fptrunc2 = fptrunc double %build_sitofp to float" < %t.ll
      * RUN: grep "%build_fpext = fpext float %build_fptrunc to double" < %t.ll
      * RUN: grep "%build_fpext2 = fpext float %build_fptrunc to double" < %t.ll
-     * RUN: grep "%build_inttoptr = inttoptr i32 %P1 to i8*" < %t.ll
-     * RUN: grep "%build_ptrtoint = ptrtoint i8* %build_inttoptr to i64" < %t.ll
-     * RUN: grep "%build_ptrtoint2 = ptrtoint i8* %build_inttoptr to i64" < %t.ll
+     * RUN: grep "%build_inttoptr = inttoptr i32 %P1 to i8\*" < %t.ll
+     * RUN: grep "%build_ptrtoint = ptrtoint i8\* %build_inttoptr to i64" < %t.ll
+     * RUN: grep "%build_ptrtoint2 = ptrtoint i8\* %build_inttoptr to i64" < %t.ll
      * RUN: grep "%build_bitcast = bitcast i64 %build_ptrtoint to double" < %t.ll
      * RUN: grep "%build_bitcast2 = bitcast i64 %build_ptrtoint to double" < %t.ll
      * RUN: grep "%build_bitcast3 = bitcast i64 %build_ptrtoint to double" < %t.ll
      * RUN: grep "%build_bitcast4 = bitcast i64 %build_ptrtoint to double" < %t.ll
-     * RUN: grep "%build_pointercast = bitcast i8* %build_inttoptr to i16*" < %t.ll
+     * RUN: grep "%build_pointercast = bitcast i8\* %build_inttoptr to i16*" < %t.ll
      *)
     let inst28 = build_trunc p1 i8_type "build_trunc" atentry in
     let inst29 = build_zext inst28 i32_type "build_zext" atentry in
@@ -1148,7 +1148,7 @@
      * RUN: grep "%build_fcmp_false = fcmp false float %F1, %F2" < %t.ll
      * RUN: grep "%build_fcmp_true = fcmp true float %F2, %F1" < %t.ll
      * RUN: grep "%build_is_null.*= icmp eq.*%X0,.*null" < %t.ll
-     * RUN: grep "%build_is_not_null = icmp ne i8* %X1, null" < %t.ll
+     * RUN: grep "%build_is_not_null = icmp ne i8\* %X1, null" < %t.ll
      * RUN: grep "%build_ptrdiff" < %t.ll
      *)
     ignore (build_icmp Icmp.Ne    p1 p2 "build_icmp_ne" atentry);
@@ -1167,7 +1167,7 @@
   group "miscellaneous"; begin
     (* RUN: grep "%build_call = tail call cc63 i32 @.*(i32 signext %P2, i32 %P1)" < %t.ll
      * RUN: grep "%build_select = select i1 %build_icmp, i32 %P1, i32 %P2" < %t.ll
-     * RUN: grep "%build_va_arg = va_arg i8** null, i32" < %t.ll
+     * RUN: grep "%build_va_arg = va_arg i8\*\* null, i32" < %t.ll
      * RUN: grep "%build_extractelement = extractelement <4 x i32> %Vec1, i32 %P2" < %t.ll
      * RUN: grep "%build_insertelement = insertelement <4 x i32> %Vec1, i32 %P1, i32 %P2" < %t.ll
      * RUN: grep "%build_shufflevector = shufflevector <4 x i32> %Vec1, <4 x i32> %Vec2, <4 x i32> <i32 1, i32 1, i32 0, i32 0>" < %t.ll
@@ -1240,8 +1240,8 @@
   end;
 
   group "dbg"; begin
-    (* RUN: grep "%dbg = add i32 %P1, %P2, !dbg !1" < %t.ll
-     * RUN: grep "!1 = metadata !{i32 2, i32 3, metadata !2, metadata !2}" < %t.ll
+    (* RUN: grep '%dbg = add i32 %P1, %P2, !dbg !1' < %t.ll
+     * RUN: grep '!1 = metadata !{i32 2, i32 3, metadata !2, metadata !2}' < %t.ll
      *)
     insist ((current_debug_location atentry) = None);
 

Modified: llvm/branches/AMDILBackend/test/Bitcode/blockaddress.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/Bitcode/blockaddress.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/Bitcode/blockaddress.ll (original)
+++ llvm/branches/AMDILBackend/test/Bitcode/blockaddress.ll Tue Jan 15 11:16:16 2013
@@ -28,3 +28,18 @@
 end:
   ret void
 }
+
+; PR13895
+define void @doitagain(i8** nocapture %pptr) {
+; CHECK: define void @doitagain
+entry:
+  br label %here
+
+here:
+  store i8* blockaddress(@doit, %here), i8** %pptr, align 8
+; CHECK: blockaddress(@doit, %here)
+  br label %end
+
+end:
+  ret void
+}

Modified: llvm/branches/AMDILBackend/test/BugPoint/crash-narrowfunctiontest.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/BugPoint/crash-narrowfunctiontest.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/BugPoint/crash-narrowfunctiontest.ll (original)
+++ llvm/branches/AMDILBackend/test/BugPoint/crash-narrowfunctiontest.ll Tue Jan 15 11:16:16 2013
@@ -2,6 +2,7 @@
 ;
 ; RUN: bugpoint -load %llvmshlibdir/BugpointPasses%shlibext %s -output-prefix %t -bugpoint-crashcalls -silence-passes > /dev/null
 ; REQUIRES: loadable_module
+; XFAIL: lto_on_osx
 
 define i32 @foo() { ret i32 1 }
 

Modified: llvm/branches/AMDILBackend/test/BugPoint/metadata.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/BugPoint/metadata.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/BugPoint/metadata.ll (original)
+++ llvm/branches/AMDILBackend/test/BugPoint/metadata.ll Tue Jan 15 11:16:16 2013
@@ -1,6 +1,7 @@
 ; RUN: bugpoint -load %llvmshlibdir/BugpointPasses%shlibext %s -output-prefix %t -bugpoint-crashcalls -silence-passes > /dev/null
 ; RUN: llvm-dis %t-reduced-simplified.bc -o - | FileCheck %s
 ; REQUIRES: loadable_module
+; XFAIL: lto_on_osx
 
 ; Bugpoint should keep the call's metadata attached to the call.
 

Modified: llvm/branches/AMDILBackend/test/BugPoint/remove_arguments_test.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/BugPoint/remove_arguments_test.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/BugPoint/remove_arguments_test.ll (original)
+++ llvm/branches/AMDILBackend/test/BugPoint/remove_arguments_test.ll Tue Jan 15 11:16:16 2013
@@ -1,6 +1,7 @@
 ; RUN: bugpoint -load %llvmshlibdir/BugpointPasses%shlibext %s -output-prefix %t -bugpoint-crashcalls -silence-passes
 ; RUN: llvm-dis %t-reduced-simplified.bc -o - | FileCheck %s
 ; REQUIRES: loadable_module
+; XFAIL: lto_on_osx
 
 ; Test to make sure that arguments are removed from the function if they are 
 ; unnecessary. And clean up any types that that frees up too.

Modified: llvm/branches/AMDILBackend/test/CMakeLists.txt
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CMakeLists.txt?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CMakeLists.txt (original)
+++ llvm/branches/AMDILBackend/test/CMakeLists.txt Tue Jan 15 11:16:16 2013
@@ -7,6 +7,11 @@
   ${CMAKE_CURRENT_BINARY_DIR}/Unit/lit.site.cfg
   )
 
+# Don't include check-llvm into check-all without LLVM_BUILD_TOOLS.
+if(NOT LLVM_BUILD_TOOLS)
+  set(EXCLUDE_FROM_ALL ON)
+endif()
+
 add_lit_testsuite(check-llvm "Running the LLVM regression tests"
   ${CMAKE_CURRENT_BINARY_DIR}
   PARAMS llvm_site_config=${CMAKE_CURRENT_BINARY_DIR}/lit.site.cfg
@@ -14,10 +19,16 @@
   DEPENDS UnitTests
           BugpointPasses LLVMHello
           llc lli llvm-ar llvm-as
-          llvm-diff
+          llvm-bcanalyzer llvm-diff
           llvm-dis llvm-extract llvm-dwarfdump
-          llvm-link llvm-mc llvm-nm llvm-objdump llvm-readobj
+          llvm-link
+          llvm-mc
+          llvm-mcmarkup
+          llvm-nm
+          llvm-objdump
+          llvm-readobj
           macho-dump opt
+          profile_rt-shared
           FileCheck count not
           yaml2obj
   )

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/2010-10-19-mc-elf-objheader.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/2010-10-19-mc-elf-objheader.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/2010-10-19-mc-elf-objheader.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/2010-10-19-mc-elf-objheader.ll Tue Jan 15 11:16:16 2013
@@ -13,12 +13,12 @@
 ; BASIC-NEXT:         0x00000000
 ; BASIC-NEXT:         0x00000000
 ; BASIC-NEXT:         0x0000003c
-; BASIC-NEXT:         0x00000020
+; BASIC-NEXT:         0x00000022
 ; BASIC-NEXT:         0x00000000
 ; BASIC-NEXT:         0x00000000
 ; BASIC-NEXT:         0x00000001
 ; BASIC-NEXT:         0x00000000
-; BASIC-NEXT:         '411f0000 00616561 62690001 15000000 06020801 09011401 15011703 18011901'
+; BASIC-NEXT:         '41210000 00616561 62690001 17000000 060a0741 08010902 14011501 17031801 1901'
 
 ; CORTEXA8:        .ARM.attributes
 ; CORTEXA8-NEXT:         0x70000003

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/2010-12-07-PEIBug.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/2010-12-07-PEIBug.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/2010-12-07-PEIBug.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/2010-12-07-PEIBug.ll Tue Jan 15 11:16:16 2013
@@ -1,4 +1,4 @@
-; RUN: llc < %s -mtriple=thumbv7-apple-ios -mcpu=cortex-a8 | FileCheck %s
+; RUN: llc < %s -mtriple=thumbv7-apple-ios -mcpu=cortex-a9 | FileCheck %s
 ; rdar://8728956
 
 define hidden void @foo() nounwind ssp {

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/2011-03-15-LdStMultipleBug.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/2011-03-15-LdStMultipleBug.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/2011-03-15-LdStMultipleBug.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/2011-03-15-LdStMultipleBug.ll Tue Jan 15 11:16:16 2013
@@ -1,4 +1,4 @@
-; RUN: llc < %s -mtriple=thumbv7-apple-darwin10 -relocation-model=pic -disable-fp-elim -disable-cgp-delete-dead-blocks -mcpu=cortex-a8 | FileCheck %s
+; RUN: llc < %s -mtriple=thumbv7-apple-darwin10 -relocation-model=pic -disable-fp-elim -mcpu=cortex-a8 | FileCheck %s
 
 ; Do not form Thumb2 ldrd / strd if the offset is not multiple of 4.
 ; rdar://9133587
@@ -21,12 +21,6 @@
   %x = getelementptr %struct.Outer* @oStruct, i32 0, i32 1, i32 %i.022, i32 0
   %y = getelementptr %struct.Outer* @oStruct, i32 0, i32 1, i32 %i.022, i32 1
   %inc = add i32 %i.022, 1
-  br i1 %tmp3, label %_Z14printIsNotZeroi.exit, label %if.then.i
-
-if.then.i:                                        ; preds = %for.body
-  unreachable
-
-_Z14printIsNotZeroi.exit:                         ; preds = %for.body
   %tmp8 = load i32* %x, align 4, !tbaa !0
   %tmp11 = load i32* %y, align 4, !tbaa !0
   %mul = mul nsw i32 %tmp11, %tmp8
@@ -37,7 +31,7 @@
   unreachable
 
 _Z14printIsNotZeroi.exit17:                       ; preds = %_Z14printIsNotZeroi.exit
-  br i1 undef, label %_Z14printIsNotZeroi.exit17.for.body_crit_edge, label %for.end
+  br label %_Z14printIsNotZeroi.exit17.for.body_crit_edge
 
 _Z14printIsNotZeroi.exit17.for.body_crit_edge:    ; preds = %_Z14printIsNotZeroi.exit17
   %b.phi.trans.insert = getelementptr %struct.Outer* @oStruct, i32 0, i32 1, i32 %inc, i32 3

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/2011-06-16-TailCallByVal.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/2011-06-16-TailCallByVal.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/2011-06-16-TailCallByVal.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/2011-06-16-TailCallByVal.ll Tue Jan 15 11:16:16 2013
@@ -1,4 +1,9 @@
 ; RUN: llc < %s -arm-tail-calls=1 | FileCheck %s
+
+; tail call inside a function where byval argument is splitted between
+; registers and stack is currently unsupported.
+; XFAIL: *
+
 target datalayout = "e-p:32:32:32-i1:8:32-i8:8:32-i16:16:32-i32:32:32-i64:32:32-f32:32:32-f64:32:32-v64:32:64-v128:32:128-a0:0:32-n32"
 target triple = "thumbv7-apple-ios"
 

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/2011-10-26-memset-with-neon.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/2011-10-26-memset-with-neon.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/2011-10-26-memset-with-neon.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/2011-10-26-memset-with-neon.ll Tue Jan 15 11:16:16 2013
@@ -9,8 +9,8 @@
 }
 
 ; Trigger multiple NEON stores.
-; CHECK:      vstmia
-; CHECK-NEXT: vstmia
+; CHECK:      vst1.64
+; CHECK-NEXT: vst1.64
 define void @f_0_40(i8* nocapture %c) nounwind optsize {
 entry:
   call void @llvm.memset.p0i8.i64(i8* %c, i8 0, i64 40, i32 16, i1 false)

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/2011-11-29-128bitArithmetics.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/2011-11-29-128bitArithmetics.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/2011-11-29-128bitArithmetics.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/2011-11-29-128bitArithmetics.ll Tue Jan 15 11:16:16 2013
@@ -8,12 +8,12 @@
 
 ; CHECK:      movw    r1, :lower16:{{.*}}
 ; CHECK:      movt    r1, :upper16:{{.*}}
-; CHECK:      vldmia  r1
+; CHECK:      vld1.64 {{.*}}, [r1, :128]
 ; CHECK:      vsqrt.f32       {{s[0-9]+}}, {{s[0-9]+}}
 ; CHECK:      vsqrt.f32       {{s[0-9]+}}, {{s[0-9]+}}
 ; CHECK:      vsqrt.f32       {{s[0-9]+}}, {{s[0-9]+}}
 ; CHECK:      vsqrt.f32       {{s[0-9]+}}, {{s[0-9]+}}
-; CHECK:      vstmia  {{.*}}
+; CHECK:      vst1.64  {{.*}}
 
 L.entry:
   %0 = load <4 x float>* @A, align 16
@@ -31,21 +31,21 @@
 
 ; CHECK:      movw  [[reg0:r[0-9]+]], :lower16:{{.*}}
 ; CHECK:      movt  [[reg0]], :upper16:{{.*}}
-; CHECK:      vldmia r{{[0-9][0-9]?}}, {{.*}}
+; CHECK:      vld1.64
 
-; CHECK:      {{[v]?mov}}  r0, {{[r|s][0-9]+}}
+; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}cosf
 
-; CHECK:      {{[v]?mov}}  r0, {{[r|s][0-9]+}}
+; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}cosf
 
-; CHECK:      {{[v]?mov}}  r0, {{[r|s][0-9]+}}
+; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}cosf
 
-; CHECK:      {{[v]?mov}}  r0, {{[r|s][0-9]+}}
+; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}cosf
 
-; CHECK:      vstmia  {{.*}}
+; CHECK:      vst1.64
 
 L.entry:
   %0 = load <4 x float>* @A, align 16
@@ -62,21 +62,21 @@
 
 ; CHECK:      movw  [[reg0:r[0-9]+]], :lower16:{{.*}}
 ; CHECK:      movt  [[reg0]], :upper16:{{.*}}
-; CHECK:      vldmia r{{[0-9][0-9]?}}, {{.*}}
+; CHECK:      vld1.64
 
-; CHECK:      {{[v]?mov}}  r0, {{[r|s][0-9]+}}
+; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}expf
 
-; CHECK:      {{[v]?mov}}  r0, {{[r|s][0-9]+}}
+; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}expf
 
-; CHECK:      {{[v]?mov}}  r0, {{[r|s][0-9]+}}
+; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}expf
 
-; CHECK:      {{[v]?mov}}  r0, {{[r|s][0-9]+}}
+; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}expf
 
-; CHECK:      vstmia  {{.*}}
+; CHECK:      vst1.64
 
 L.entry:
   %0 = load <4 x float>* @A, align 16
@@ -93,21 +93,21 @@
 
 ; CHECK:      movw  [[reg0:r[0-9]+]], :lower16:{{.*}}
 ; CHECK:      movt  [[reg0]], :upper16:{{.*}}
-; CHECK:      vldmia r{{[0-9][0-9]?}}, {{.*}}
+; CHECK:      vld1.64
 
-; CHECK:      {{[v]?mov}}  r0, {{[r|s][0-9]+}}
+; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}exp2f
 
-; CHECK:      {{[v]?mov}}  r0, {{[r|s][0-9]+}}
+; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}exp2f
 
-; CHECK:      {{[v]?mov}}  r0, {{[r|s][0-9]+}}
+; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}exp2f
 
-; CHECK:      {{[v]?mov}}  r0, {{[r|s][0-9]+}}
+; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}exp2f
 
-; CHECK:      vstmia  {{.*}}
+; CHECK:      vst1.64
 
 L.entry:
   %0 = load <4 x float>* @A, align 16
@@ -124,21 +124,21 @@
 
 ; CHECK:      movw  [[reg0:r[0-9]+]], :lower16:{{.*}}
 ; CHECK:      movt  [[reg0]], :upper16:{{.*}}
-; CHECK:      vldmia r{{[0-9][0-9]?}}, {{.*}}
+; CHECK:      vld1.64
 
-; CHECK:      {{[v]?mov}}  r0, {{[r|s][0-9]+}}
+; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}log10f
 
-; CHECK:      {{[v]?mov}}  r0, {{[r|s][0-9]+}}
+; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}log10f
 
-; CHECK:      {{[v]?mov}}  r0, {{[r|s][0-9]+}}
+; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}log10f
 
-; CHECK:      {{[v]?mov}}  r0, {{[r|s][0-9]+}}
+; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}log10f
 
-; CHECK:      vstmia  {{.*}}
+; CHECK:      vst1.64
 
 L.entry:
   %0 = load <4 x float>* @A, align 16
@@ -155,21 +155,21 @@
 
 ; CHECK:      movw  [[reg0:r[0-9]+]], :lower16:{{.*}}
 ; CHECK:      movt  [[reg0]], :upper16:{{.*}}
-; CHECK:      vldmia r{{[0-9][0-9]?}}, {{.*}}
+; CHECK:      vld1.64
 
-; CHECK:      {{[v]?mov}}  r0, {{[r|s][0-9]+}}
+; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}logf
 
-; CHECK:      {{[v]?mov}}  r0, {{[r|s][0-9]+}}
+; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}logf
 
-; CHECK:      {{[v]?mov}}  r0, {{[r|s][0-9]+}}
+; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}logf
 
-; CHECK:      {{[v]?mov}}  r0, {{[r|s][0-9]+}}
+; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}logf
 
-; CHECK:      vstmia  {{.*}}
+; CHECK:      vst1.64
 
 L.entry:
   %0 = load <4 x float>* @A, align 16
@@ -186,21 +186,21 @@
 
 ; CHECK:      movw  [[reg0:r[0-9]+]], :lower16:{{.*}}
 ; CHECK:      movt  [[reg0]], :upper16:{{.*}}
-; CHECK:      vldmia r{{[0-9][0-9]?}}, {{.*}}
+; CHECK:      vld1.64
 
-; CHECK:      {{[v]?mov}}  r0, {{[r|s][0-9]+}}
+; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}log2f
 
-; CHECK:      {{[v]?mov}}  r0, {{[r|s][0-9]+}}
+; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}log2f
 
-; CHECK:      {{[v]?mov}}  r0, {{[r|s][0-9]+}}
+; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}log2f
 
-; CHECK:      {{[v]?mov}}  r0, {{[r|s][0-9]+}}
+; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}log2f
 
-; CHECK:      vstmia  {{.*}}
+; CHECK:      vst1.64
 
 L.entry:
   %0 = load <4 x float>* @A, align 16
@@ -218,21 +218,21 @@
 
 ; CHECK:      movw  [[reg0:r[0-9]+]], :lower16:{{.*}}
 ; CHECK:      movt  [[reg0]], :upper16:{{.*}}
-; CHECK:      vldmia r{{[0-9][0-9]?}}, {{.*}}
+; CHECK:      vld1.64
 
-; CHECK:      {{[v]?mov}}  r0, {{[r|s][0-9]+}}
+; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}powf
 
-; CHECK:      {{[v]?mov}}  r0, {{[r|s][0-9]+}}
+; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}powf
 
-; CHECK:      {{[v]?mov}}  r0, {{[r|s][0-9]+}}
+; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}powf
 
-; CHECK:      {{[v]?mov}}  r0, {{[r|s][0-9]+}}
+; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}powf
 
-; CHECK:      vstmia  {{.*}}
+; CHECK:      vst1.64
 
 L.entry:
 
@@ -252,10 +252,10 @@
 
 ; CHECK:       movw  [[reg0:r[0-9]+]], :lower16:{{.*}}
 ; CHECK:       movt  [[reg0]], :upper16:{{.*}}
-; CHECK:       vldmia  [[reg0]], {{.*}}
+; CHECK:       vld1.64 {{.*}}, :128
 ; CHECK:       vmul.f32 {{.*}}
 
-; CHECK:      vstmia  {{.*}}
+; CHECK:      vst1.64
 
 L.entry:
 
@@ -275,21 +275,21 @@
 
 ; CHECK:      movw  [[reg0:r[0-9]+]], :lower16:{{.*}}
 ; CHECK:      movt  [[reg0]], :upper16:{{.*}}
-; CHECK:      vldmia r{{[0-9][0-9]?}}, {{.*}}
+; CHECK:      vld1.64
 
-; CHECK:      {{[v]?mov}}  r0, {{[r|s][0-9]+}}
+; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}sinf
 
-; CHECK:      {{[v]?mov}}  r0, {{[r|s][0-9]+}}
+; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}sinf
 
-; CHECK:      {{[v]?mov}}  r0, {{[r|s][0-9]+}}
+; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}sinf
 
-; CHECK:      {{[v]?mov}}  r0, {{[r|s][0-9]+}}
+; CHECK:      {{v?mov(.32)?}}  r0,
 ; CHECK:      bl  {{.*}}sinf
 
-; CHECK:      vstmia  {{.*}}
+; CHECK:      vst1.64
 
 L.entry:
   %0 = load <4 x float>* @A, align 16
@@ -300,3 +300,34 @@
 
 declare <4 x float> @llvm.sin.v4f32(<4 x float>) nounwind readonly
 
+define void @test_floor(<4 x float>* %X) nounwind {
+
+; CHECK: test_floor:
+
+; CHECK:      movw  [[reg0:r[0-9]+]], :lower16:{{.*}}
+; CHECK:      movt  [[reg0]], :upper16:{{.*}}
+; CHECK:      vld1.64
+
+; CHECK:      {{v?mov(.32)?}}  r0,
+; CHECK:      bl  {{.*}}floorf
+
+; CHECK:      {{v?mov(.32)?}}  r0,
+; CHECK:      bl  {{.*}}floorf
+
+; CHECK:      {{v?mov(.32)?}}  r0,
+; CHECK:      bl  {{.*}}floorf
+
+; CHECK:      {{v?mov(.32)?}}  r0,
+; CHECK:      bl  {{.*}}floorf
+
+; CHECK:      vst1.64
+
+L.entry:
+  %0 = load <4 x float>* @A, align 16
+  %1 = call <4 x float> @llvm.floor.v4f32(<4 x float> %0)
+  store <4 x float> %1, <4 x float>* %X, align 16
+  ret void
+}
+
+declare <4 x float> @llvm.floor.v4f32(<4 x float>) nounwind readonly
+

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/arm-modifier.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/arm-modifier.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/arm-modifier.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/arm-modifier.ll Tue Jan 15 11:16:16 2013
@@ -57,3 +57,12 @@
 store i64 %1, i64* @f3_var, align 4
 ret void
 }
+
+define i64 @f4(i64* %val) nounwind {
+entry:
+  ;CHECK: f4
+  ;CHECK: ldrexd [[REG1:(r[0-9]?[02468])]], {{r[0-9]?[13579]}}, [r0]
+  ;CHECK: mov r0, [[REG1]]
+  %0 = tail call i64 asm sideeffect "ldrexd $0, ${0:H}, [$1]", "=&r,r,*Qo"(i64* %val, i64* %val) nounwind
+  ret i64 %0
+}

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/atomic-op.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/atomic-op.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/atomic-op.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/atomic-op.ll Tue Jan 15 11:16:16 2013
@@ -159,3 +159,13 @@
   store i8 %3, i8* %old
   ret void
 }
+
+; CHECK: func4
+; This function should not need to use callee-saved registers.
+; rdar://problem/12203728
+; CHECK-NOT: r4
+define i32 @func4(i32* %p) nounwind optsize ssp {
+entry:
+  %0 = atomicrmw add i32* %p, i32 1 monotonic
+  ret i32 %0
+}

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/avoid-cpsr-rmw.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/avoid-cpsr-rmw.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/avoid-cpsr-rmw.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/avoid-cpsr-rmw.ll Tue Jan 15 11:16:16 2013
@@ -1,4 +1,5 @@
 ; RUN: llc < %s -mtriple=thumbv7-apple-darwin -mcpu=cortex-a9 | FileCheck %s
+; RUN: llc < %s -mtriple=thumbv7-apple-darwin -mcpu=swift     | FileCheck %s
 ; Avoid some 's' 16-bit instruction which partially update CPSR (and add false
 ; dependency) when it isn't dependent on last CPSR defining instruction.
 ; rdar://8928208

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/carry.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/carry.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/carry.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/carry.ll Tue Jan 15 11:16:16 2013
@@ -45,3 +45,16 @@
   %0 = sub nsw i64 0, %x
   ret i64 %0
 }
+
+; rdar://12559385
+define i64 @f5(i32 %vi) {
+entry:
+; CHECK: f5:
+; CHECK: movw [[REG:r[0-9]+]], #36102
+; CHECK: sbc r{{[0-9]+}}, r{{[0-9]+}}, [[REG]]
+    %v0 = zext i32 %vi to i64
+    %v1 = xor i64 %v0, -155057456198619
+    %v4 = add i64 %v1, 155057456198619
+    %v5 = add i64 %v4, %v1
+    ret i64 %v5
+}

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/coalesce-subregs.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/coalesce-subregs.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/coalesce-subregs.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/coalesce-subregs.ll Tue Jan 15 11:16:16 2013
@@ -1,4 +1,4 @@
-; RUN: llc < %s -mcpu=cortex-a9 | FileCheck %s
+; RUN: llc < %s -mcpu=cortex-a9 -verify-coalescing -verify-machineinstrs | FileCheck %s
 target datalayout = "e-p:32:32:32-i1:8:32-i8:8:32-i16:16:32-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:32:64-v128:32:128-a0:0:32-n32-S32"
 target triple = "thumbv7-apple-ios0.0.0"
 
@@ -66,3 +66,295 @@
 
 declare { <4 x float>, <4 x float> } @llvm.arm.neon.vld2.v4f32(i8*, i32) nounwind readonly
 declare void @llvm.arm.neon.vst2.v4f32(i8*, <4 x float>, <4 x float>, i32) nounwind
+
+; CHECK: f3
+; This function has lane insertions that span basic blocks.
+; The trivial REG_SEQUENCE lowering can't handle that, but the coalescer can.
+;
+; void f3(float *p, float *q) {
+;   float32x2_t x;
+;   x[1] = p[3];
+;   if (q)
+;     x[0] = q[0] + q[1];
+;   else
+;     x[0] = p[2];
+;   vst1_f32(p+4, x);
+; }
+;
+; CHECK-NOT: vmov
+; CHECK-NOT: vorr
+define void @f3(float* %p, float* %q) nounwind ssp {
+entry:
+  %arrayidx = getelementptr inbounds float* %p, i32 3
+  %0 = load float* %arrayidx, align 4
+  %vecins = insertelement <2 x float> undef, float %0, i32 1
+  %tobool = icmp eq float* %q, null
+  br i1 %tobool, label %if.else, label %if.then
+
+if.then:                                          ; preds = %entry
+  %1 = load float* %q, align 4
+  %arrayidx2 = getelementptr inbounds float* %q, i32 1
+  %2 = load float* %arrayidx2, align 4
+  %add = fadd float %1, %2
+  %vecins3 = insertelement <2 x float> %vecins, float %add, i32 0
+  br label %if.end
+
+if.else:                                          ; preds = %entry
+  %arrayidx4 = getelementptr inbounds float* %p, i32 2
+  %3 = load float* %arrayidx4, align 4
+  %vecins5 = insertelement <2 x float> %vecins, float %3, i32 0
+  br label %if.end
+
+if.end:                                           ; preds = %if.else, %if.then
+  %x.0 = phi <2 x float> [ %vecins3, %if.then ], [ %vecins5, %if.else ]
+  %add.ptr = getelementptr inbounds float* %p, i32 4
+  %4 = bitcast float* %add.ptr to i8*
+  tail call void @llvm.arm.neon.vst1.v2f32(i8* %4, <2 x float> %x.0, i32 4)
+  ret void
+}
+
+declare void @llvm.arm.neon.vst1.v2f32(i8*, <2 x float>, i32) nounwind
+declare <2 x float> @llvm.arm.neon.vld1.v2f32(i8*, i32) nounwind readonly
+
+; CHECK: f4
+; This function inserts a lane into a fully defined vector.
+; The destination lane isn't read, so the subregs can coalesce.
+; CHECK-NOT: vmov
+; CHECK-NOT: vorr
+define void @f4(float* %p, float* %q) nounwind ssp {
+entry:
+  %0 = bitcast float* %p to i8*
+  %vld1 = tail call <2 x float> @llvm.arm.neon.vld1.v2f32(i8* %0, i32 4)
+  %tobool = icmp eq float* %q, null
+  br i1 %tobool, label %if.end, label %if.then
+
+if.then:                                          ; preds = %entry
+  %1 = load float* %q, align 4
+  %arrayidx1 = getelementptr inbounds float* %q, i32 1
+  %2 = load float* %arrayidx1, align 4
+  %add = fadd float %1, %2
+  %vecins = insertelement <2 x float> %vld1, float %add, i32 1
+  br label %if.end
+
+if.end:                                           ; preds = %entry, %if.then
+  %x.0 = phi <2 x float> [ %vecins, %if.then ], [ %vld1, %entry ]
+  tail call void @llvm.arm.neon.vst1.v2f32(i8* %0, <2 x float> %x.0, i32 4)
+  ret void
+}
+
+; CHECK: f5
+; Coalesce vector lanes through phis.
+; CHECK: vmov.f32 {{.*}}, #1.0
+; CHECK-NOT: vmov
+; CHECK-NOT: vorr
+; CHECK: %if.end
+; We may leave the last insertelement in the if.end block.
+; It is inserting the %add value into a dead lane, but %add causes interference
+; in the entry block, and we don't do dead lane checks across basic blocks.
+define void @f5(float* %p, float* %q) nounwind ssp {
+entry:
+  %0 = bitcast float* %p to i8*
+  %vld1 = tail call <4 x float> @llvm.arm.neon.vld1.v4f32(i8* %0, i32 4)
+  %vecext = extractelement <4 x float> %vld1, i32 0
+  %vecext1 = extractelement <4 x float> %vld1, i32 1
+  %vecext2 = extractelement <4 x float> %vld1, i32 2
+  %vecext3 = extractelement <4 x float> %vld1, i32 3
+  %add = fadd float %vecext3, 1.000000e+00
+  %tobool = icmp eq float* %q, null
+  br i1 %tobool, label %if.end, label %if.then
+
+if.then:                                          ; preds = %entry
+  %arrayidx = getelementptr inbounds float* %q, i32 1
+  %1 = load float* %arrayidx, align 4
+  %add4 = fadd float %vecext, %1
+  %2 = load float* %q, align 4
+  %add6 = fadd float %vecext1, %2
+  %arrayidx7 = getelementptr inbounds float* %q, i32 2
+  %3 = load float* %arrayidx7, align 4
+  %add8 = fadd float %vecext2, %3
+  br label %if.end
+
+if.end:                                           ; preds = %entry, %if.then
+  %a.0 = phi float [ %add4, %if.then ], [ %vecext, %entry ]
+  %b.0 = phi float [ %add6, %if.then ], [ %vecext1, %entry ]
+  %c.0 = phi float [ %add8, %if.then ], [ %vecext2, %entry ]
+  %vecinit = insertelement <4 x float> undef, float %a.0, i32 0
+  %vecinit9 = insertelement <4 x float> %vecinit, float %b.0, i32 1
+  %vecinit10 = insertelement <4 x float> %vecinit9, float %c.0, i32 2
+  %vecinit11 = insertelement <4 x float> %vecinit10, float %add, i32 3
+  tail call void @llvm.arm.neon.vst1.v4f32(i8* %0, <4 x float> %vecinit11, i32 4)
+  ret void
+}
+
+declare <4 x float> @llvm.arm.neon.vld1.v4f32(i8*, i32) nounwind readonly
+
+declare void @llvm.arm.neon.vst1.v4f32(i8*, <4 x float>, i32) nounwind
+
+; CHECK: pr13999
+define void @pr13999() nounwind readonly {
+entry:
+ br i1 true, label %outer_loop, label %loop.end
+
+outer_loop:
+ %d = phi double [ 0.0, %entry ], [ %add, %after_inner_loop ]
+ %0 = insertelement <2 x double> <double 0.0, double 0.0>, double %d, i32 0
+ br i1 undef, label %after_inner_loop, label %inner_loop
+
+inner_loop:
+ br i1 true, label %after_inner_loop, label %inner_loop
+
+after_inner_loop:
+ %1 = phi <2 x double> [ %0, %outer_loop ], [ <double 0.0, double 0.0>,
+%inner_loop ]
+ %2 = extractelement <2 x double> %1, i32 1
+ %add = fadd double 1.0, %2
+ br i1 false, label %loop.end, label %outer_loop
+
+loop.end:
+ %d.end = phi double [ 0.0, %entry ], [ %add, %after_inner_loop ]
+ ret void
+}
+
+; CHECK: pr14078
+define arm_aapcs_vfpcc i32 @pr14078(i8* nocapture %arg, i8* nocapture %arg1, i32 %arg2) nounwind uwtable readonly {
+bb:
+  br i1 undef, label %bb31, label %bb3
+
+bb3:                                              ; preds = %bb12, %bb
+  %tmp = shufflevector <2 x i64> undef, <2 x i64> undef, <1 x i32> zeroinitializer
+  %tmp4 = bitcast <1 x i64> %tmp to <2 x float>
+  %tmp5 = shufflevector <2 x float> %tmp4, <2 x float> undef, <4 x i32> zeroinitializer
+  %tmp6 = bitcast <4 x float> %tmp5 to <2 x i64>
+  %tmp7 = shufflevector <2 x i64> %tmp6, <2 x i64> undef, <1 x i32> zeroinitializer
+  %tmp8 = bitcast <1 x i64> %tmp7 to <2 x float>
+  %tmp9 = tail call <2 x float> @baz(<2 x float> <float 0xFFFFFFFFE0000000, float 0.000000e+00>, <2 x float> %tmp8, <2 x float> zeroinitializer) nounwind
+  br i1 undef, label %bb10, label %bb12
+
+bb10:                                             ; preds = %bb3
+  %tmp11 = load <4 x float>* undef, align 8
+  br label %bb12
+
+bb12:                                             ; preds = %bb10, %bb3
+  %tmp13 = shufflevector <2 x float> %tmp9, <2 x float> zeroinitializer, <2 x i32> <i32 0, i32 2>
+  %tmp14 = bitcast <2 x float> %tmp13 to <1 x i64>
+  %tmp15 = shufflevector <1 x i64> %tmp14, <1 x i64> zeroinitializer, <2 x i32> <i32 0, i32 1>
+  %tmp16 = bitcast <2 x i64> %tmp15 to <4 x float>
+  %tmp17 = fmul <4 x float> zeroinitializer, %tmp16
+  %tmp18 = bitcast <4 x float> %tmp17 to <2 x i64>
+  %tmp19 = shufflevector <2 x i64> %tmp18, <2 x i64> undef, <1 x i32> zeroinitializer
+  %tmp20 = bitcast <1 x i64> %tmp19 to <2 x float>
+  %tmp21 = tail call <2 x float> @baz67(<2 x float> %tmp20, <2 x float> undef) nounwind
+  %tmp22 = tail call <2 x float> @baz67(<2 x float> %tmp21, <2 x float> %tmp21) nounwind
+  %tmp23 = shufflevector <2 x float> %tmp22, <2 x float> undef, <4 x i32> zeroinitializer
+  %tmp24 = bitcast <4 x float> %tmp23 to <2 x i64>
+  %tmp25 = shufflevector <2 x i64> %tmp24, <2 x i64> undef, <1 x i32> zeroinitializer
+  %tmp26 = bitcast <1 x i64> %tmp25 to <2 x float>
+  %tmp27 = extractelement <2 x float> %tmp26, i32 0
+  %tmp28 = fcmp olt float %tmp27, 0.000000e+00
+  %tmp29 = select i1 %tmp28, i32 0, i32 undef
+  %tmp30 = icmp ult i32 undef, %arg2
+  br i1 %tmp30, label %bb3, label %bb31
+
+bb31:                                             ; preds = %bb12, %bb
+  %tmp32 = phi i32 [ 1, %bb ], [ %tmp29, %bb12 ]
+  ret i32 %tmp32
+}
+
+declare <2 x float> @baz(<2 x float>, <2 x float>, <2 x float>) nounwind readnone
+
+declare <2 x float> @baz67(<2 x float>, <2 x float>) nounwind readnone
+
+%struct.wombat.5 = type { %struct.quux, %struct.quux, %struct.quux, %struct.quux }
+%struct.quux = type { <4 x float> }
+
+; CHECK: pr14079
+define linkonce_odr arm_aapcs_vfpcc %struct.wombat.5 @pr14079(i8* nocapture %arg, i8* nocapture %arg1, i8* nocapture %arg2) nounwind uwtable inlinehint {
+bb:
+  %tmp = shufflevector <2 x i64> zeroinitializer, <2 x i64> undef, <1 x i32> zeroinitializer
+  %tmp3 = bitcast <1 x i64> %tmp to <2 x float>
+  %tmp4 = shufflevector <2 x float> %tmp3, <2 x float> zeroinitializer, <2 x i32> <i32 1, i32 3>
+  %tmp5 = shufflevector <2 x float> %tmp4, <2 x float> undef, <2 x i32> <i32 1, i32 3>
+  %tmp6 = bitcast <2 x float> %tmp5 to <1 x i64>
+  %tmp7 = shufflevector <1 x i64> undef, <1 x i64> %tmp6, <2 x i32> <i32 0, i32 1>
+  %tmp8 = bitcast <2 x i64> %tmp7 to <4 x float>
+  %tmp9 = shufflevector <2 x i64> zeroinitializer, <2 x i64> undef, <1 x i32> <i32 1>
+  %tmp10 = bitcast <1 x i64> %tmp9 to <2 x float>
+  %tmp11 = shufflevector <2 x float> %tmp10, <2 x float> undef, <2 x i32> <i32 0, i32 2>
+  %tmp12 = shufflevector <2 x float> %tmp11, <2 x float> undef, <2 x i32> <i32 0, i32 2>
+  %tmp13 = bitcast <2 x float> %tmp12 to <1 x i64>
+  %tmp14 = shufflevector <1 x i64> %tmp13, <1 x i64> undef, <2 x i32> <i32 0, i32 1>
+  %tmp15 = bitcast <2 x i64> %tmp14 to <4 x float>
+  %tmp16 = insertvalue %struct.wombat.5 undef, <4 x float> %tmp8, 1, 0
+  %tmp17 = insertvalue %struct.wombat.5 %tmp16, <4 x float> %tmp15, 2, 0
+  %tmp18 = insertvalue %struct.wombat.5 %tmp17, <4 x float> undef, 3, 0
+  ret %struct.wombat.5 %tmp18
+}
+
+; CHECK: adjustCopiesBackFrom
+; The shuffle in if.else3 must be preserved even though adjustCopiesBackFrom
+; is tempted to remove it.
+; CHECK: %if.else3
+; CHECK: vorr d
+define internal void @adjustCopiesBackFrom(<2 x i64>* noalias nocapture sret %agg.result, <2 x i64> %in) {
+entry:
+  %0 = extractelement <2 x i64> %in, i32 0
+  %cmp = icmp slt i64 %0, 1
+  %.in = select i1 %cmp, <2 x i64> <i64 0, i64 undef>, <2 x i64> %in
+  %1 = extractelement <2 x i64> %in, i32 1
+  %cmp1 = icmp slt i64 %1, 1
+  br i1 %cmp1, label %if.then2, label %if.else3
+
+if.then2:                                         ; preds = %entry
+  %2 = insertelement <2 x i64> %.in, i64 0, i32 1
+  br label %if.end4
+
+if.else3:                                         ; preds = %entry
+  %3 = shufflevector <2 x i64> %.in, <2 x i64> %in, <2 x i32> <i32 0, i32 3>
+  br label %if.end4
+
+if.end4:                                          ; preds = %if.else3, %if.then2
+  %result.2 = phi <2 x i64> [ %2, %if.then2 ], [ %3, %if.else3 ]
+  store <2 x i64> %result.2, <2 x i64>* %agg.result, align 128
+  ret void
+}
+
+; <rdar://problem/12758887>
+; RegisterCoalescer::updateRegDefsUses() could visit an instruction more than
+; once under rare circumstances. When widening a register from QPR to DTriple
+; with the original virtual register in dsub_1_dsub_2, the double rewrite would
+; produce an invalid sub-register.
+;
+; This is because dsub_1_dsub_2 is not an idempotent sub-register index.
+; It will translate %vr:dsub_0 -> %vr:dsub_1.
+define hidden fastcc void @radar12758887() nounwind optsize ssp {
+entry:
+  br i1 undef, label %for.body, label %for.end70
+
+for.body:                                         ; preds = %for.end, %entry
+  br i1 undef, label %for.body29, label %for.end
+
+for.body29:                                       ; preds = %for.body29, %for.body
+  %0 = load <2 x double>* null, align 1
+  %splat40 = shufflevector <2 x double> %0, <2 x double> undef, <2 x i32> zeroinitializer
+  %mul41 = fmul <2 x double> undef, %splat40
+  %add42 = fadd <2 x double> undef, %mul41
+  %splat44 = shufflevector <2 x double> %0, <2 x double> undef, <2 x i32> <i32 1, i32 1>
+  %mul45 = fmul <2 x double> undef, %splat44
+  %add46 = fadd <2 x double> undef, %mul45
+  br i1 undef, label %for.end, label %for.body29
+
+for.end:                                          ; preds = %for.body29, %for.body
+  %accumR2.0.lcssa = phi <2 x double> [ zeroinitializer, %for.body ], [ %add42, %for.body29 ]
+  %accumI2.0.lcssa = phi <2 x double> [ zeroinitializer, %for.body ], [ %add46, %for.body29 ]
+  %1 = shufflevector <2 x double> %accumI2.0.lcssa, <2 x double> undef, <2 x i32> <i32 1, i32 0>
+  %add58 = fadd <2 x double> undef, %1
+  %mul61 = fmul <2 x double> %add58, undef
+  %add63 = fadd <2 x double> undef, %mul61
+  %add64 = fadd <2 x double> undef, %add63
+  %add67 = fadd <2 x double> undef, %add64
+  store <2 x double> %add67, <2 x double>* undef, align 1
+  br i1 undef, label %for.end70, label %for.body
+
+for.end70:                                        ; preds = %for.end, %entry
+  ret void
+}

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/constants.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/constants.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/constants.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/constants.ll Tue Jan 15 11:16:16 2013
@@ -1,4 +1,4 @@
-; RUN: llc < %s -mtriple=armv4t-unknown-linux-gnueabi -disable-cgp-branch-opts | FileCheck %s
+; RUN: llc < %s -mtriple=armv4t-unknown-linux-gnueabi -disable-cgp-branch-opts -verify-machineinstrs | FileCheck %s
 
 define i32 @f1() {
 ; CHECK: f1
@@ -45,6 +45,16 @@
         ret void
 }
 
+define i32 @f8() nounwind {
+; Check that constant propagation through (i32)-1 => (float)Nan => (i32)-1
+; gives expected result
+; CHECK: f8
+; CHECK: mvn r0, #0
+        %tmp0 = bitcast i32 -1 to float
+        %tmp1 = bitcast float %tmp0 to i32
+        ret i32 %tmp1
+}
+
 %t1 = type { <3 x float>, <3 x float> }
 
 @const1 = global %t1 { <3 x float> zeroinitializer,

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/debug-info-branch-folding.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/debug-info-branch-folding.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/debug-info-branch-folding.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/debug-info-branch-folding.ll Tue Jan 15 11:16:16 2013
@@ -3,16 +3,17 @@
 target triple = "thumbv7-apple-macosx10.6.7"
 
 ;CHECK: 	vadd.f32	q4, q8, q8
-;CHECK-NEXT: Ltmp
-;CHECK-NEXT: 	@DEBUG_VALUE: y <- Q4+0
-;CHECK-NEXT:    @DEBUG_VALUE: x <- Q4+0
+;CHECK-NEXT: Ltmp1
+
+;CHECK:@DEBUG_VALUE: x <- Q4+0
+;CHECK-NEXT:@DEBUG_VALUE: y <- Q4+0
 
 
 @.str = external constant [13 x i8]
 
 declare <4 x float> @test0001(float) nounwind readnone ssp
 
-define i32 @main(i32 %argc, i8** nocapture %argv) nounwind ssp {
+define i32 @main(i32 %argc, i8** nocapture %argv, i1 %cond) nounwind ssp {
 entry:
   br label %for.body9
 
@@ -21,7 +22,7 @@
   tail call void @llvm.dbg.value(metadata !{<4 x float> %add19}, i64 0, metadata !27), !dbg !39
   %add20 = fadd <4 x float> undef, <float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 1.000000e+00>, !dbg !39
   tail call void @llvm.dbg.value(metadata !{<4 x float> %add20}, i64 0, metadata !28), !dbg !39
-  br i1 undef, label %for.end54, label %for.body9, !dbg !44
+  br i1 %cond, label %for.end54, label %for.body9, !dbg !44
 
 for.end54:                                        ; preds = %for.body9
   %tmp115 = extractelement <4 x float> %add19, i32 1
@@ -52,7 +53,7 @@
 !7 = metadata !{i32 589860, metadata !2, metadata !"float", null, i32 0, i64 32, i64 32, i64 0, i32 0, i32 4} ; [ DW_TAG_base_type ]
 !8 = metadata !{metadata !9}
 !9 = metadata !{i32 589857, i64 0, i64 3}         ; [ DW_TAG_subrange_type ]
-!10 = metadata !{i32 589870, i32 0, metadata !1, metadata !"main", metadata !"main", metadata !"", metadata !1, i32 59, metadata !11, i1 false, i1 true, i32 0, i32 0, i32 0, i32 256, i1 true, i32 (i32, i8**)* @main, null} ; [ DW_TAG_subprogram ]
+!10 = metadata !{i32 589870, i32 0, metadata !1, metadata !"main", metadata !"main", metadata !"", metadata !1, i32 59, metadata !11, i1 false, i1 true, i32 0, i32 0, i32 0, i32 256, i1 true, i32 (i32, i8**, i1)* @main, null} ; [ DW_TAG_subprogram ]
 !11 = metadata !{i32 589845, metadata !1, metadata !"", metadata !1, i32 0, i64 0, i64 0, i32 0, i32 0, i32 0, metadata !12, i32 0, i32 0} ; [ DW_TAG_subroutine_type ]
 !12 = metadata !{metadata !13}
 !13 = metadata !{i32 589860, metadata !2, metadata !"int", null, i32 0, i64 32, i64 32, i64 0, i32 0, i32 5} ; [ DW_TAG_base_type ]

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/div.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/div.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/div.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/div.ll Tue Jan 15 11:16:16 2013
@@ -1,9 +1,13 @@
-; RUN: llc < %s -mtriple=arm-apple-darwin | FileCheck %s -check-prefix=CHECK-ARM
+; RUN: llc < %s -mtriple=arm-apple-ios -mcpu=cortex-a8 | FileCheck %s -check-prefix=CHECK-ARM
+; RUN: llc < %s -mtriple=arm-apple-ios -mcpu=swift     | FileCheck %s -check-prefix=CHECK-SWIFT
 
 define i32 @f1(i32 %a, i32 %b) {
 entry:
 ; CHECK-ARM: f1
 ; CHECK-ARM: __divsi3
+
+; CHECK-SWIFT: f1
+; CHECK-SWIFT: sdiv
         %tmp1 = sdiv i32 %a, %b         ; <i32> [#uses=1]
         ret i32 %tmp1
 }
@@ -12,6 +16,9 @@
 entry:
 ; CHECK-ARM: f2
 ; CHECK-ARM: __udivsi3
+
+; CHECK-SWIFT: f2
+; CHECK-SWIFT: udiv
         %tmp1 = udiv i32 %a, %b         ; <i32> [#uses=1]
         ret i32 %tmp1
 }
@@ -20,6 +27,10 @@
 entry:
 ; CHECK-ARM: f3
 ; CHECK-ARM: __modsi3
+
+; CHECK-SWIFT: f3
+; CHECK-SWIFT: sdiv
+; CHECK-SWIFT: mls
         %tmp1 = srem i32 %a, %b         ; <i32> [#uses=1]
         ret i32 %tmp1
 }
@@ -28,6 +39,10 @@
 entry:
 ; CHECK-ARM: f4
 ; CHECK-ARM: __umodsi3
+
+; CHECK-SWIFT: f4
+; CHECK-SWIFT: udiv
+; CHECK-SWIFT: mls
         %tmp1 = urem i32 %a, %b         ; <i32> [#uses=1]
         ret i32 %tmp1
 }

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/divmod.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/divmod.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/divmod.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/divmod.ll Tue Jan 15 11:16:16 2013
@@ -1,10 +1,18 @@
-; RUN: llc < %s -mtriple=arm-apple-ios5.0 -mcpu=cortex-a8 | FileCheck %s
+; RUN: llc < %s -mtriple=arm-apple-ios5.0 -mcpu=cortex-a8 | FileCheck %s -check-prefix=A8
+; RUN: llc < %s -mtriple=arm-apple-ios5.0 -mcpu=swift     | FileCheck %s -check-prefix=SWIFT
+
+; rdar://12481395
 
 define void @foo(i32 %x, i32 %y, i32* nocapture %P) nounwind ssp {
 entry:
-; CHECK: foo:
-; CHECK: bl ___divmodsi4
-; CHECK-NOT: bl ___divmodsi4
+; A8: foo:
+; A8: bl ___divmodsi4
+; A8-NOT: bl ___divmodsi4
+
+; SWIFT: foo:
+; SWIFT: sdiv
+; SWIFT: mls
+; SWIFT-NOT: bl __divmodsi4
   %div = sdiv i32 %x, %y
   store i32 %div, i32* %P, align 4
   %rem = srem i32 %x, %y
@@ -15,9 +23,14 @@
 
 define void @bar(i32 %x, i32 %y, i32* nocapture %P) nounwind ssp {
 entry:
-; CHECK: bar:
-; CHECK: bl ___udivmodsi4
-; CHECK-NOT: bl ___udivmodsi4
+; A8: bar:
+; A8: bl ___udivmodsi4
+; A8-NOT: bl ___udivmodsi4
+
+; SWIFT: bar:
+; SWIFT: udiv
+; SWIFT: mls
+; SWIFT-NOT: bl __udivmodsi4
   %div = udiv i32 %x, %y
   store i32 %div, i32* %P, align 4
   %rem = urem i32 %x, %y
@@ -32,14 +45,18 @@
 
 define void @do_indent(i32 %cols) nounwind {
 entry:
-; CHECK: do_indent:
+; A8: do_indent:
+; SWIFT: do_indent:
   %0 = load i32* @flags, align 4
   %1 = and i32 %0, 67108864
   %2 = icmp eq i32 %1, 0
   br i1 %2, label %bb1, label %bb
 
 bb:
-; CHECK: bl ___divmodsi4
+; A8: bl ___divmodsi4
+; SWIFT: sdiv
+; SWIFT: mls
+; SWIFT-NOT: bl __divmodsi4
   %3 = load i32* @tabsize, align 4
   %4 = srem i32 %cols, %3
   %5 = sdiv i32 %cols, %3
@@ -60,9 +77,14 @@
 ; rdar://11714607
 define i32 @howmany(i32 %x, i32 %y) nounwind {
 entry:
-; CHECK: howmany:
-; CHECK: bl ___udivmodsi4
-; CHECK-NOT: ___udivsi3
+; A8: howmany:
+; A8: bl ___udivmodsi4
+; A8-NOT: ___udivsi3
+
+; SWIFT: howmany:
+; SWIFT: udiv
+; SWIFT: mls
+; SWIFT-NOT: bl __udivmodsi4
   %rem = urem i32 %x, %y
   %div = udiv i32 %x, %y
   %not.cmp = icmp ne i32 %rem, 0

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/fabss.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/fabss.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/fabss.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/fabss.ll Tue Jan 15 11:16:16 2013
@@ -14,12 +14,12 @@
 declare float @fabsf(float)
 
 ; VFP2: test:
-; VFP2: 	vabs.f32	s1, s1
+; VFP2: 	vabs.f32	s2, s2
 
 ; NFP1: test:
 ; NFP1: 	vabs.f32	d1, d1
 ; NFP0: test:
-; NFP0: 	vabs.f32	s1, s1
+; NFP0: 	vabs.f32	s2, s2
 
 ; CORTEXA8: test:
 ; CORTEXA8:     vadd.f32        [[D1:d[0-9]+]]

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/fadds.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/fadds.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/fadds.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/fadds.ll Tue Jan 15 11:16:16 2013
@@ -10,14 +10,14 @@
 }
 
 ; VFP2: test:
-; VFP2: 	vadd.f32	s0, s1, s0
+; VFP2: 	vadd.f32	s
 
 ; NFP1: test:
-; NFP1: 	vadd.f32	d0, d1, d0
+; NFP1: 	vadd.f32	d
 ; NFP0: test:
-; NFP0: 	vadd.f32	s0, s1, s0
+; NFP0: 	vadd.f32	s
 
 ; CORTEXA8: test:
-; CORTEXA8: 	vadd.f32	d0, d1, d0
+; CORTEXA8: 	vadd.f32	d
 ; CORTEXA9: test:
 ; CORTEXA9: 	vadd.f32	s{{.}}, s{{.}}, s{{.}}

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/fast-isel-call.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/fast-isel-call.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/fast-isel-call.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/fast-isel-call.ll Tue Jan 15 11:16:16 2013
@@ -2,6 +2,8 @@
 ; RUN: llc < %s -O0 -fast-isel-abort -relocation-model=dynamic-no-pic -mtriple=thumbv7-apple-ios | FileCheck %s --check-prefix=THUMB
 ; RUN: llc < %s -O0 -fast-isel-abort -relocation-model=dynamic-no-pic -mtriple=armv7-apple-ios -arm-long-calls | FileCheck %s --check-prefix=ARM-LONG
 ; RUN: llc < %s -O0 -fast-isel-abort -relocation-model=dynamic-no-pic -mtriple=thumbv7-apple-ios -arm-long-calls | FileCheck %s --check-prefix=THUMB-LONG
+; RUN: llc < %s -O0 -relocation-model=dynamic-no-pic -mtriple=armv7-apple-ios -mattr=-vfp2 | FileCheck %s --check-prefix=ARM-NOVFP
+; RUN: llc < %s -O0 -relocation-model=dynamic-no-pic -mtriple=thumbv7-apple-ios -mattr=-vfp2 | FileCheck %s --check-prefix=THUMB-NOVFP
 
 define i32 @t0(i1 zeroext %a) nounwind {
   %1 = zext i1 %a to i32
@@ -221,3 +223,67 @@
 }
 
 declare i32 @CallVariadic(i32, ...)
+
+; Test fastcc
+
+define fastcc void @fast_callee(float %i) ssp {
+entry:
+; ARM: fast_callee
+; ARM: vmov r0, s0
+; THUMB: fast_callee
+; THUMB: vmov r0, s0
+; ARM-NOVFP: fast_callee
+; ARM-NOVFP-NOT: s0
+; THUMB-NOVFP: fast_callee
+; THUMB-NOVFP-NOT: s0
+  call void @print(float %i)
+  ret void
+}
+
+define void @fast_caller() ssp {
+entry:
+; ARM: fast_caller
+; ARM: vldr s0,
+; THUMB: fast_caller
+; THUMB: vldr s0,
+; ARM-NOVFP: fast_caller
+; ARM-NOVFP: movw r0, #13107
+; ARM-NOVFP: movt r0, #16611
+; THUMB-NOVFP: fast_caller
+; THUMB-NOVFP: movw r0, #13107
+; THUMB-NOVFP: movt r0, #16611
+  call fastcc void @fast_callee(float 0x401C666660000000)
+  ret void
+}
+
+define void @no_fast_callee(float %i) ssp {
+entry:
+; ARM: no_fast_callee
+; ARM: vmov s0, r0
+; THUMB: no_fast_callee
+; THUMB: vmov s0, r0
+; ARM-NOVFP: no_fast_callee
+; ARM-NOVFP-NOT: s0
+; THUMB-NOVFP: no_fast_callee
+; THUMB-NOVFP-NOT: s0
+  call void @print(float %i)
+  ret void
+}
+
+define void @no_fast_caller() ssp {
+entry:
+; ARM: no_fast_caller
+; ARM: vmov r0, s0
+; THUMB: no_fast_caller
+; THUMB: vmov r0, s0
+; ARM-NOVFP: no_fast_caller
+; ARM-NOVFP: movw r0, #13107
+; ARM-NOVFP: movt r0, #16611
+; THUMB-NOVFP: no_fast_caller
+; THUMB-NOVFP: movw r0, #13107
+; THUMB-NOVFP: movt r0, #16611
+  call void @no_fast_callee(float 0x401C666660000000)
+  ret void
+}
+
+declare void @print(float)

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/fast-isel.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/fast-isel.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/fast-isel.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/fast-isel.ll Tue Jan 15 11:16:16 2013
@@ -1,5 +1,7 @@
 ; RUN: llc < %s -O0 -fast-isel-abort -relocation-model=dynamic-no-pic -mtriple=armv7-apple-ios | FileCheck %s --check-prefix=ARM
 ; RUN: llc < %s -O0 -fast-isel-abort -relocation-model=dynamic-no-pic -mtriple=thumbv7-apple-ios | FileCheck %s --check-prefix=THUMB
+; RUN: llc < %s -O0 -arm-strict-align -relocation-model=dynamic-no-pic -mtriple=armv7-apple-ios | FileCheck %s --check-prefix=ARM-STRICT-ALIGN
+; RUN: llc < %s -O0 -arm-strict-align -relocation-model=dynamic-no-pic -mtriple=thumbv7-apple-ios | FileCheck %s --check-prefix=THUMB-STRICT-ALIGN
 
 ; Very basic fast-isel functionality.
 define i32 @add(i32 %a, i32 %b) nounwind {
@@ -238,3 +240,67 @@
 }
 
 declare void @llvm.trap() nounwind
+
+define void @unaligned_i16_store(i16 %x, i16* %y) nounwind {
+entry:
+; ARM-STRICT-ALIGN: @unaligned_i16_store
+; ARM-STRICT-ALIGN: strb
+; ARM-STRICT-ALIGN: strb
+
+; THUMB-STRICT-ALIGN: @unaligned_i16_store
+; THUMB-STRICT-ALIGN: strb
+; THUMB-STRICT-ALIGN: strb
+
+  store i16 %x, i16* %y, align 1
+  ret void
+}
+
+define i16 @unaligned_i16_load(i16* %x) nounwind {
+entry:
+; ARM-STRICT-ALIGN: @unaligned_i16_load
+; ARM-STRICT-ALIGN: ldrb
+; ARM-STRICT-ALIGN: ldrb
+
+; THUMB-STRICT-ALIGN: @unaligned_i16_load
+; THUMB-STRICT-ALIGN: ldrb
+; THUMB-STRICT-ALIGN: ldrb
+
+  %0 = load i16* %x, align 1
+  ret i16 %0
+}
+
+define void @unaligned_i32_store(i32 %x, i32* %y) nounwind {
+entry:
+; ARM-STRICT-ALIGN: @unaligned_i32_store
+; ARM-STRICT-ALIGN: strb
+; ARM-STRICT-ALIGN: strb
+; ARM-STRICT-ALIGN: strb
+; ARM-STRICT-ALIGN: strb
+
+; THUMB-STRICT-ALIGN: @unaligned_i32_store
+; THUMB-STRICT-ALIGN: strb
+; THUMB-STRICT-ALIGN: strb
+; THUMB-STRICT-ALIGN: strb
+; THUMB-STRICT-ALIGN: strb
+
+  store i32 %x, i32* %y, align 1
+  ret void
+}
+
+define i32 @unaligned_i32_load(i32* %x) nounwind {
+entry:
+; ARM-STRICT-ALIGN: @unaligned_i32_load
+; ARM-STRICT-ALIGN: ldrb
+; ARM-STRICT-ALIGN: ldrb
+; ARM-STRICT-ALIGN: ldrb
+; ARM-STRICT-ALIGN: ldrb
+
+; THUMB-STRICT-ALIGN: @unaligned_i32_load
+; THUMB-STRICT-ALIGN: ldrb
+; THUMB-STRICT-ALIGN: ldrb
+; THUMB-STRICT-ALIGN: ldrb
+; THUMB-STRICT-ALIGN: ldrb
+
+  %0 = load i32* %x, align 1
+  ret i32 %0
+}

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/fdivs.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/fdivs.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/fdivs.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/fdivs.ll Tue Jan 15 11:16:16 2013
@@ -10,14 +10,14 @@
 }
 
 ; VFP2: test:
-; VFP2: 	vdiv.f32	s0, s1, s0
+; VFP2: 	vdiv.f32	s0, s2, s0
 
 ; NFP1: test:
-; NFP1: 	vdiv.f32	s0, s1, s0
+; NFP1: 	vdiv.f32	s0, s2, s0
 ; NFP0: test:
-; NFP0: 	vdiv.f32	s0, s1, s0
+; NFP0: 	vdiv.f32	s0, s2, s0
 
 ; CORTEXA8: test:
-; CORTEXA8: 	vdiv.f32	s0, s1, s0
+; CORTEXA8: 	vdiv.f32	s0, s2, s0
 ; CORTEXA9: test:
 ; CORTEXA9: 	vdiv.f32	s{{.}}, s{{.}}, s{{.}}

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/fmuls.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/fmuls.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/fmuls.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/fmuls.ll Tue Jan 15 11:16:16 2013
@@ -10,15 +10,15 @@
 }
 
 ; VFP2: test:
-; VFP2: 	vmul.f32	s0, s1, s0
+; VFP2: 	vmul.f32	s
 
 ; NFP1: test:
-; NFP1: 	vmul.f32	d0, d1, d0
+; NFP1: 	vmul.f32	d
 ; NFP0: test:
-; NFP0: 	vmul.f32	s0, s1, s0
+; NFP0: 	vmul.f32	s
 
 ; CORTEXA8: test:
-; CORTEXA8: 	vmul.f32	d0, d1, d0
+; CORTEXA8: 	vmul.f32	d
 ; CORTEXA9: test:
 ; CORTEXA9: 	vmul.f32	s{{.}}, s{{.}}, s{{.}}
 

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/fp16.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/fp16.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/fp16.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/fp16.ll Tue Jan 15 11:16:16 2013
@@ -15,14 +15,14 @@
   %1 = load i16* @y, align 2
   %2 = tail call float @llvm.convert.from.fp16(i16 %0)
 ; CHECK: __gnu_h2f_ieee
-; CHECK-FP16: vcvtb.f16.f32
+; CHECK-FP16: vcvtb.f32.f16
   %3 = tail call float @llvm.convert.from.fp16(i16 %1)
 ; CHECK: __gnu_h2f_ieee
-; CHECK-FP16: vcvtb.f16.f32
+; CHECK-FP16: vcvtb.f32.f16
   %4 = fadd float %2, %3
   %5 = tail call i16 @llvm.convert.to.fp16(float %4)
 ; CHECK: __gnu_f2h_ieee
-; CHECK-FP16: vcvtb.f32.f16
+; CHECK-FP16: vcvtb.f16.f32
   store i16 %5, i16* @x, align 2
   ret void
 }

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/fp_convert.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/fp_convert.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/fp_convert.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/fp_convert.ll Tue Jan 15 11:16:16 2013
@@ -31,7 +31,7 @@
 ; VFP2: test3:
 ; VFP2: vcvt.f32.u32 s{{.}}, s{{.}}
 ; NEON: test3:
-; NEON: vcvt.f32.u32 d0, d0
+; NEON: vcvt.f32.u32 d
 entry:
         %0 = add i32 %a, %b
         %1 = uitofp i32 %0 to float
@@ -42,7 +42,7 @@
 ; VFP2: test4:
 ; VFP2: vcvt.f32.s32 s{{.}}, s{{.}}
 ; NEON: test4:
-; NEON: vcvt.f32.s32 d0, d0
+; NEON: vcvt.f32.s32 d
 entry:
         %0 = add i32 %a, %b
         %1 = sitofp i32 %0 to float

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/fsubs.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/fsubs.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/fsubs.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/fsubs.ll Tue Jan 15 11:16:16 2013
@@ -8,6 +8,6 @@
 	ret float %0
 }
 
-; VFP2: vsub.f32	s0, s1, s0
-; NFP1: vsub.f32	d0, d1, d0
-; NFP0: vsub.f32	s0, s1, s0
+; VFP2: vsub.f32	s
+; NFP1: vsub.f32	d
+; NFP0: vsub.f32	s

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/ifcvt1.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/ifcvt1.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/ifcvt1.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/ifcvt1.ll Tue Jan 15 11:16:16 2013
@@ -1,17 +1,21 @@
-; RUN: llc < %s -march=arm -mcpu=cortex-a8 | FileCheck %s
+; RUN: llc < %s -march=arm -mcpu=cortex-a8 | FileCheck %s -check-prefix=A8
+; RUN: llc < %s -march=arm -mcpu=swift     | FileCheck %s -check-prefix=SWIFT
 
 define i32 @t1(i32 %a, i32 %b) {
-; CHECK: t1:
+; A8: t1:
+; SWIFT: t1:
 	%tmp2 = icmp eq i32 %a, 0
 	br i1 %tmp2, label %cond_false, label %cond_true
 
 cond_true:
-; CHECK: subeq r0, r1, #1
+; A8: subeq r0, r1, #1
+; SWIFT: sub r0, r1, #1
 	%tmp5 = add i32 %b, 1
 	ret i32 %tmp5
 
 cond_false:
-; CHECK: addne r0, r1, #1
+; A8: addne r0, r1, #1
+; SWIFT: addne r0, r1, #1
 	%tmp7 = add i32 %b, -1
 	ret i32 %tmp7
 }

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/ifcvt5.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/ifcvt5.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/ifcvt5.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/ifcvt5.ll Tue Jan 15 11:16:16 2013
@@ -1,4 +1,6 @@
-; RUN: llc < %s -mtriple=armv7-apple-ios | FileCheck %s
+; RUN: llc < %s -mtriple=armv7-apple-ios -mcpu=cortex-a8 | FileCheck %s -check-prefix=A8
+; RUN: llc < %s -mtriple=armv7-apple-ios -mcpu=swift     | FileCheck %s -check-prefix=SWIFT
+; rdar://8402126
 
 @x = external global i32*		; <i32**> [#uses=1]
 
@@ -10,8 +12,12 @@
 }
 
 define i32 @t1(i32 %a, i32 %b) {
-; CHECK: t1:
-; CHECK: poplt {r7, pc}
+; A8: t1:
+; A8: poplt {r7, pc}
+
+; SWIFT: t1:
+; SWIFT: pop {r7, pc}
+; SWIFT: pop {r7, pc}
 entry:
 	%tmp1 = icmp sgt i32 %a, 10		; <i1> [#uses=1]
 	br i1 %tmp1, label %cond_true, label %UnifiedReturnBlock

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/ldr_post.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/ldr_post.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/ldr_post.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/ldr_post.ll Tue Jan 15 11:16:16 2013
@@ -1,4 +1,5 @@
 ; RUN: llc < %s -march=arm | FileCheck %s
+; RUN: llc < %s -march=arm -mcpu=swift | FileCheck %s
 
 ; CHECK: test1:
 ; CHECK: ldr {{.*, \[.*]}}, -r2

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/ldr_pre.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/ldr_pre.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/ldr_pre.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/ldr_pre.ll Tue Jan 15 11:16:16 2013
@@ -1,4 +1,5 @@
 ; RUN: llc < %s -march=arm | FileCheck %s
+; RUN: llc < %s -march=arm -mcpu=swift | FileCheck %s
 
 ; CHECK: test1:
 ; CHECK: ldr {{.*!}}

Removed: llvm/branches/AMDILBackend/test/CodeGen/ARM/longMACt.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/longMACt.ll?rev=172540&view=auto
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/longMACt.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/longMACt.ll (removed)
@@ -1,44 +0,0 @@
-; RUN: llc < %s -march=thumb -mattr=+thumb2 | FileCheck %s
-; Check generated signed and unsigned multiply accumulate long.
-
-define i64 @MACLongTest1(i32 %a, i32 %b, i64 %c) {
-;CHECK: MACLongTest1:
-;CHECK: umlal
-  %conv = zext i32 %a to i64
-  %conv1 = zext i32 %b to i64
-  %mul = mul i64 %conv1, %conv
-  %add = add i64 %mul, %c
-  ret i64 %add
-}
-
-define i64 @MACLongTest2(i32 %a, i32 %b, i64 %c)  {
-;CHECK: MACLongTest2:
-;CHECK: smlal
-  %conv = sext i32 %a to i64
-  %conv1 = sext i32 %b to i64
-  %mul = mul nsw i64 %conv1, %conv
-  %add = add nsw i64 %mul, %c
-  ret i64 %add
-}
-
-define i64 @MACLongTest3(i32 %a, i32 %b, i32 %c) {
-;CHECK: MACLongTest3:
-;CHECK: umlal
-  %conv = zext i32 %b to i64
-  %conv1 = zext i32 %a to i64
-  %mul = mul i64 %conv, %conv1
-  %conv2 = zext i32 %c to i64
-  %add = add i64 %mul, %conv2
-  ret i64 %add
-}
-
-define i64 @MACLongTest4(i32 %a, i32 %b, i32 %c) {
-;CHECK: MACLongTest4:
-;CHECK: smlal
-  %conv = sext i32 %b to i64
-  %conv1 = sext i32 %a to i64
-  %mul = mul nsw i64 %conv, %conv1
-  %conv2 = sext i32 %c to i64
-  %add = add nsw i64 %mul, %conv2
-  ret i64 %add
-}

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/mls.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/mls.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/mls.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/mls.ll Tue Jan 15 11:16:16 2013
@@ -1,4 +1,5 @@
 ; RUN: llc < %s -march=arm -mattr=+v6t2 | FileCheck %s
+; RUN: llc < %s -march=arm -mattr=+v6t2 -arm-use-mulops=false | FileCheck %s -check-prefix=NO_MULOPS
 
 define i32 @f1(i32 %a, i32 %b, i32 %c) {
     %tmp1 = mul i32 %a, %b
@@ -13,4 +14,15 @@
     ret i32 %tmp2
 }
 
+; CHECK: f1:
 ; CHECK: mls	r0, r0, r1, r2
+; NO_MULOPS: f1:
+; NO_MULOPS: mul r0, r0, r1
+; NO_MULOPS-NEXT: sub r0, r2, r0
+
+; CHECK: f2:
+; CHECK: mul r0, r0, r1
+; CHECK-NEXT: sub r0, r0, r2
+; NO_MULOPS: f2:
+; NO_MULOPS: mul r0, r0, r1
+; NO_MULOPS-NEXT: sub r0, r0, r2

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/neon_ld2.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/neon_ld2.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/neon_ld2.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/neon_ld2.ll Tue Jan 15 11:16:16 2013
@@ -1,10 +1,16 @@
 ; RUN: llc < %s -march=arm -mattr=+neon | FileCheck %s
+; RUN: llc < %s -march=arm -mcpu=swift | FileCheck %s --check-prefix=SWIFT
 
 ; CHECK: t1
-; CHECK: vldmia
-; CHECK: vldmia
+; CHECK: vld1.64
+; CHECK: vld1.64
 ; CHECK: vadd.i64 q
-; CHECK: vstmia
+; CHECK: vst1.64
+; SWIFT: t1
+; SWIFT: vld1.64 {{.d[0-9]+, d[0-9]+}, \[r[0-9]+, :128\]}}
+; SWIFT: vld1.64 {{.d[0-9]+, d[0-9]+}, \[r[0-9]+, :128\]}}
+; SWIFT: vadd.i64 q
+; SWIFT: vst1.64 {{.d[0-9]+, d[0-9]+}, \[r[0-9]+, :128\]}}
 define void @t1(<4 x i32>* %r, <2 x i64>* %a, <2 x i64>* %b) nounwind {
 entry:
 	%0 = load <2 x i64>* %a, align 16		; <<2 x i64>> [#uses=1]
@@ -16,11 +22,17 @@
 }
 
 ; CHECK: t2
-; CHECK: vldmia
-; CHECK: vldmia
+; CHECK: vld1.64
+; CHECK: vld1.64
 ; CHECK: vsub.i64 q
 ; CHECK: vmov r0, r1, d
 ; CHECK: vmov r2, r3, d
+; SWIFT: t2
+; SWIFT: vld1.64 {{.d[0-9]+, d[0-9]+}, \[r[0-9]+, :128\]}}
+; SWIFT: vld1.64 {{.d[0-9]+, d[0-9]+}, \[r[0-9]+, :128\]}}
+; SWIFT: vsub.i64 q
+; SWIFT: vmov r0, r1, d
+; SWIFT: vmov r2, r3, d
 define <4 x i32> @t2(<2 x i64>* %a, <2 x i64>* %b) nounwind readonly {
 entry:
 	%0 = load <2 x i64>* %a, align 16		; <<2 x i64>> [#uses=1]
@@ -30,3 +42,18 @@
 	ret <4 x i32> %3
 }
 
+; Limited alignment.
+; SWIFT: t3
+; SWIFT: vld1.64 {{.d[0-9]+, d[0-9]+}, \[r[0-9]+}}
+; SWIFT: vld1.64 {{.d[0-9]+, d[0-9]+}, \[r[0-9]+}}
+; SWIFT: vadd.i64 q
+; SWIFT: vst1.64 {{.d[0-9]+, d[0-9]+}, \[r[0-9]+}}
+define void @t3(<4 x i32>* %r, <2 x i64>* %a, <2 x i64>* %b) nounwind {
+entry:
+	%0 = load <2 x i64>* %a, align 8
+	%1 = load <2 x i64>* %b, align 8
+	%2 = add <2 x i64> %0, %1
+	%3 = bitcast <2 x i64> %2 to <4 x i32>
+	store <4 x i32> %3, <4 x i32>* %r, align 8
+	ret void
+}

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/opt-shuff-tstore.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/opt-shuff-tstore.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/opt-shuff-tstore.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/opt-shuff-tstore.ll Tue Jan 15 11:16:16 2013
@@ -2,7 +2,7 @@
 
 ; CHECK: func_4_8
 ; CHECK: vst1.32
-; CHECK-NEXT: bx lr
+; CHECK: bx lr
 define void @func_4_8(<4 x i8> %param, <4 x i8>* %p) {
   %r = add <4 x i8> %param, <i8 1, i8 2, i8 3, i8 4>
   store <4 x i8> %r, <4 x i8>* %p
@@ -11,7 +11,7 @@
 
 ; CHECK: func_2_16
 ; CHECK: vst1.32
-; CHECK-NEXT: bx lr
+; CHECK: bx lr
 define void @func_2_16(<2 x i16> %param, <2 x i16>* %p) {
   %r = add <2 x i16> %param, <i16 1, i16 2>
   store <2 x i16> %r, <2 x i16>* %p

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/reg_sequence.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/reg_sequence.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/reg_sequence.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/reg_sequence.ll Tue Jan 15 11:16:16 2013
@@ -1,5 +1,5 @@
-; RUN: llc < %s -march=arm -mcpu=cortex-a8 | FileCheck %s
-; RUN: llc < %s -march=arm -mcpu=cortex-a8 -regalloc=basic | FileCheck %s
+; RUN: llc < %s -mtriple=arm-apple-ios -mcpu=cortex-a8 | FileCheck %s
+; RUN: llc < %s -mtriple=arm-apple-ios -mcpu=cortex-a8 -regalloc=basic | FileCheck %s
 ; Implementing vld / vst as REG_SEQUENCE eliminates the extra vmov's.
 
 %struct.int16x8_t = type { <8 x i16> }
@@ -124,7 +124,6 @@
 return2:
 ; CHECK:        %return2
 ; CHECK:        vadd.i32
-; CHECK:        vorr {{q[0-9]+}}, {{q[0-9]+}}
 ; CHECK-NOT:    vmov
 ; CHECK:        vst2.32 {d{{[0-9]+}}, d{{[0-9]+}}, d{{[0-9]+}}, d{{[0-9]+}}}
   %tmp100 = extractvalue %struct.__neon_int32x4x2_t %tmp2, 0 ; <<4 x i32>> [#uses=1]
@@ -137,7 +136,7 @@
 
 define <8 x i16> @t5(i16* %A, <8 x i16>* %B) nounwind {
 ; CHECK:        t5:
-; CHECK:        vldmia
+; CHECK:        vld1.32
 ; How can FileCheck match Q and D registers? We need a lisp interpreter.
 ; CHECK:        vorr {{q[0-9]+}}, {{q[0-9]+}}, {{q[0-9]+}}
 ; CHECK-NOT:    vmov
@@ -243,8 +242,8 @@
 ; CHECK:        vldr
 ; CHECK-NOT:    vmov d{{.*}}, d16
 ; CHECK:        vmov.i32 d17
-; CHECK-NEXT:   vstmia r0, {d16, d17}
-; CHECK-NEXT:   vstmia r0, {d16, d17}
+; CHECK-NEXT:   vst1.64 {d16, d17}, [r0, :128]
+; CHECK-NEXT:   vst1.64 {d16, d17}, [r0, :128]
   %3 = bitcast double 0.000000e+00 to <2 x float> ; <<2 x float>> [#uses=2]
   %4 = shufflevector <2 x float> %3, <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; <<4 x float>> [#uses=1]
   store <4 x float> %4, <4 x float>* undef, align 16

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/select.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/select.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/select.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/select.ll Tue Jan 15 11:16:16 2013
@@ -76,12 +76,11 @@
 ; block generated, odds are good that we have close to the ideal code for this:
 ;
 ; CHECK-NEON:      _f8:
+; CHECK-NEON:      movw    [[R3:r[0-9]+]], #1123
 ; CHECK-NEON:      adr     [[R2:r[0-9]+]], LCPI7_0
-; CHECK-NEON-NEXT: movw    [[R3:r[0-9]+]], #1123
-; CHECK-NEON-NEXT: adds    {{r.*}}, [[R2]], #4
 ; CHECK-NEON-NEXT: cmp     r0, [[R3]]
-; CHECK-NEON-NEXT: it      ne
-; CHECK-NEON-NEXT: movne   {{r.*}}, [[R2]]
+; CHECK-NEON-NEXT: it      eq
+; CHECK-NEON-NEXT: addeq{{.*}} [[R2]], #4
 ; CHECK-NEON-NEXT: ldr
 ; CHECK-NEON:      bx
 

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/select_xform.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/select_xform.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/select_xform.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/select_xform.ll Tue Jan 15 11:16:16 2013
@@ -4,13 +4,13 @@
 
 define i32 @t1(i32 %a, i32 %b, i32 %c) nounwind {
 ; ARM: t1:
-; ARM: sub r0, r1, #-2147483647
-; ARM: movgt r0, r1
+; ARM: suble r1, r1, #-2147483647
+; ARM: mov r0, r1
 
 ; T2: t1:
 ; T2: mvn r0, #-2147483648
-; T2: add r0, r1
-; T2: movgt r0, r1
+; T2: addle r1, r0
+; T2: mov r0, r1
   %tmp1 = icmp sgt i32 %c, 10
   %tmp2 = select i1 %tmp1, i32 0, i32 2147483647
   %tmp3 = add i32 %tmp2, %b
@@ -19,12 +19,12 @@
 
 define i32 @t2(i32 %a, i32 %b, i32 %c, i32 %d) nounwind {
 ; ARM: t2:
-; ARM: sub r0, r1, #10
-; ARM: movgt r0, r1
+; ARM: suble r1, r1, #10
+; ARM: mov r0, r1
 
 ; T2: t2:
-; T2: sub.w r0, r1, #10
-; T2: movgt r0, r1
+; T2: suble r1, #10
+; T2: mov r0, r1
   %tmp1 = icmp sgt i32 %c, 10
   %tmp2 = select i1 %tmp1, i32 0, i32 10
   %tmp3 = sub i32 %b, %tmp2
@@ -33,12 +33,12 @@
 
 define i32 @t3(i32 %a, i32 %b, i32 %x, i32 %y) nounwind {
 ; ARM: t3:
-; ARM: mvnlt r2, #0
-; ARM: and r0, r2, r3
+; ARM: andge r3, r3, r2
+; ARM: mov r0, r3
 
 ; T2: t3:
-; T2: movlt.w r2, #-1
-; T2: and.w r0, r2, r3
+; T2: andge r3, r2
+; T2: mov r0, r3
   %cond = icmp slt i32 %a, %b
   %z = select i1 %cond, i32 -1, i32 %x
   %s = and i32 %z, %y
@@ -47,12 +47,12 @@
 
 define i32 @t4(i32 %a, i32 %b, i32 %x, i32 %y) nounwind {
 ; ARM: t4:
-; ARM: movlt r2, #0
-; ARM: orr r0, r2, r3
+; ARM: orrge r3, r3, r2
+; ARM: mov r0, r3
 
 ; T2: t4:
-; T2: movlt r2, #0
-; T2: orr.w r0, r2, r3
+; T2: orrge r3, r2
+; T2: mov r0, r3
   %cond = icmp slt i32 %a, %b
   %z = select i1 %cond, i32 0, i32 %x
   %s = or i32 %z, %y
@@ -81,7 +81,7 @@
 
 ; T2: t6:
 ; T2-NOT: movge
-; T2: eorlt.w r3, r3, r2
+; T2: eorlt r3, r2
   %cond = icmp slt i32 %a, %b
   %tmp1 = select i1 %cond, i32 %c, i32 0
   %tmp2 = xor i32 %tmp1, %d
@@ -104,3 +104,121 @@
   ret i32 %tmp3
 }
 
+; Fold ORRri into movcc.
+define i32 @t8(i32 %a, i32 %b) nounwind {
+; ARM: t8:
+; ARM: cmp r0, r1
+; ARM: orrge r0, r1, #1
+
+; T2: t8:
+; T2: cmp r0, r1
+; T2: orrge r0, r1, #1
+  %x = or i32 %b, 1
+  %cond = icmp slt i32 %a, %b
+  %tmp1 = select i1 %cond, i32 %a, i32 %x
+  ret i32 %tmp1
+}
+
+; Fold ANDrr into movcc.
+define i32 @t9(i32 %a, i32 %b, i32 %c) nounwind {
+; ARM: t9:
+; ARM: cmp r0, r1
+; ARM: andge r0, r1, r2
+
+; T2: t9:
+; T2: cmp r0, r1
+; T2: andge.w r0, r1, r2
+  %x = and i32 %b, %c
+  %cond = icmp slt i32 %a, %b
+  %tmp1 = select i1 %cond, i32 %a, i32 %x
+  ret i32 %tmp1
+}
+
+; Fold EORrs into movcc.
+define i32 @t10(i32 %a, i32 %b, i32 %c, i32 %d) nounwind {
+; ARM: t10:
+; ARM: cmp r0, r1
+; ARM: eorge r0, r1, r2, lsl #7
+
+; T2: t10:
+; T2: cmp r0, r1
+; T2: eorge.w r0, r1, r2, lsl #7
+  %s = shl i32 %c, 7
+  %x = xor i32 %b, %s
+  %cond = icmp slt i32 %a, %b
+  %tmp1 = select i1 %cond, i32 %a, i32 %x
+  ret i32 %tmp1
+}
+
+; Fold ORRri into movcc, reversing the condition.
+define i32 @t11(i32 %a, i32 %b) nounwind {
+; ARM: t11:
+; ARM: cmp r0, r1
+; ARM: orrlt r0, r1, #1
+
+; T2: t11:
+; T2: cmp r0, r1
+; T2: orrlt r0, r1, #1
+  %x = or i32 %b, 1
+  %cond = icmp slt i32 %a, %b
+  %tmp1 = select i1 %cond, i32 %x, i32 %a
+  ret i32 %tmp1
+}
+
+; Fold ADDri12 into movcc
+define i32 @t12(i32 %a, i32 %b) nounwind {
+; ARM: t12:
+; ARM: cmp r0, r1
+; ARM: addge r0, r1,
+
+; T2: t12:
+; T2: cmp r0, r1
+; T2: addwge r0, r1, #3000
+  %x = add i32 %b, 3000
+  %cond = icmp slt i32 %a, %b
+  %tmp1 = select i1 %cond, i32 %a, i32 %x
+  ret i32 %tmp1
+}
+
+; Handle frame index operands.
+define void @pr13628() nounwind uwtable align 2 {
+  %x3 = alloca i8, i32 256, align 8
+  %x4 = load i8* undef, align 1
+  %x5 = icmp ne i8 %x4, 0
+  %x6 = select i1 %x5, i8* %x3, i8* null
+  call void @bar(i8* %x6) nounwind
+  ret void
+}
+declare void @bar(i8*)
+
+; Fold zext i1 into predicated add
+define i32 @t13(i32 %c, i32 %a) nounwind readnone ssp {
+entry:
+; ARM: t13
+; ARM: cmp r1, #10
+; ARM: addgt r0, r0, #1
+
+; T2: t13
+; T2: cmp r1, #10
+; T2: addgt r0, #1
+  %cmp = icmp sgt i32 %a, 10
+  %conv = zext i1 %cmp to i32
+  %add = add i32 %conv, %c
+  ret i32 %add
+}
+
+; Fold sext i1 into predicated sub
+define i32 @t14(i32 %c, i32 %a) nounwind readnone ssp {
+entry:
+; ARM: t14
+; ARM: cmp r1, #10
+; ARM: subgt r0, r0, #1
+
+; T2: t14
+; T2: cmp r1, #10
+; T2: subgt r0, #1
+  %cmp = icmp sgt i32 %a, 10
+  %conv = sext i1 %cmp to i32
+  %add = add i32 %conv, %c
+  ret i32 %add
+}

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/struct_byval.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/struct_byval.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/struct_byval.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/struct_byval.ll Tue Jan 15 11:16:16 2013
@@ -44,3 +44,47 @@
 declare i32 @e1(%struct.SmallStruct* nocapture byval %in) nounwind
 declare i32 @e2(%struct.LargeStruct* nocapture byval %in) nounwind
 declare i32 @e3(%struct.LargeStruct* nocapture byval align 16 %in) nounwind
+
+; rdar://12442472
+; We can't do tail call since address of s is passed to the callee and part of
+; s is in caller's local frame.
+define void @f3(%struct.SmallStruct* nocapture byval %s) nounwind optsize {
+; CHECK: f3
+; CHECK: bl _consumestruct
+entry:
+  %0 = bitcast %struct.SmallStruct* %s to i8*
+  tail call void @consumestruct(i8* %0, i32 80) optsize
+  ret void
+}
+
+define void @f4(%struct.SmallStruct* nocapture byval %s) nounwind optsize {
+; CHECK: f4
+; CHECK: bl _consumestruct
+entry:
+  %addr = getelementptr inbounds %struct.SmallStruct* %s, i32 0, i32 0
+  %0 = bitcast i32* %addr to i8*
+  tail call void @consumestruct(i8* %0, i32 80) optsize
+  ret void
+}
+
+; We can do tail call here since s is in the incoming argument area.
+define void @f5(i32 %a, i32 %b, i32 %c, i32 %d, %struct.SmallStruct* nocapture byval %s) nounwind optsize {
+; CHECK: f5
+; CHECK: b _consumestruct
+entry:
+  %0 = bitcast %struct.SmallStruct* %s to i8*
+  tail call void @consumestruct(i8* %0, i32 80) optsize
+  ret void
+}
+
+define void @f6(i32 %a, i32 %b, i32 %c, i32 %d, %struct.SmallStruct* nocapture byval %s) nounwind optsize {
+; CHECK: f6
+; CHECK: b _consumestruct
+entry:
+  %addr = getelementptr inbounds %struct.SmallStruct* %s, i32 0, i32 0
+  %0 = bitcast i32* %addr to i8*
+  tail call void @consumestruct(i8* %0, i32 80) optsize
+  ret void
+}
+
+declare void @consumestruct(i8* nocapture %structp, i32 %structsize) nounwind

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/sub-cmp-peephole.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/sub-cmp-peephole.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/sub-cmp-peephole.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/sub-cmp-peephole.ll Tue Jan 15 11:16:16 2013
@@ -63,3 +63,24 @@
 if.else:
   ret i32 %sub
 }
+
+; If the sub/rsb instruction is predicated, we can't use the flags.
+; <rdar://problem/12263428>
+; Test case from MultiSource/Benchmarks/Ptrdist/bc/number.s
+; CHECK: bc_raise
+; CHECK: rsbeq
+; CHECK: cmp
+define i32 @bc_raise() nounwind ssp {
+entry:
+  %val.2.i = select i1 undef, i32 0, i32 undef
+  %sub.i = sub nsw i32 0, %val.2.i
+  %retval.0.i = select i1 undef, i32 %val.2.i, i32 %sub.i
+  %cmp1 = icmp eq i32 %retval.0.i, 0
+  br i1 %cmp1, label %land.lhs.true, label %if.end11
+
+land.lhs.true:                                    ; preds = %num2long.exit
+  ret i32 17
+
+if.end11:                                         ; preds = %num2long.exit
+  ret i32 23
+}

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/sub.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/sub.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/sub.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/sub.ll Tue Jan 15 11:16:16 2013
@@ -1,4 +1,4 @@
-; RUN: llc -march=arm < %s | FileCheck %s
+; RUN: llc -march=arm -mcpu=cortex-a8 < %s | FileCheck %s
 
 ; 171 = 0x000000ab
 define i64 @f1(i64 %a) {

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/subreg-remat.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/subreg-remat.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/subreg-remat.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/subreg-remat.ll Tue Jan 15 11:16:16 2013
@@ -4,14 +4,14 @@
 ;
 ; The vector %v2 is built like this:
 ;
-;   %vreg6:ssub_1<def> = VMOVSR %vreg0<kill>, pred:14, pred:%noreg, %vreg6<imp-def>; DPR_VFP2:%vreg6 GPR:%vreg0
+;   %vreg6:ssub_1<def> = ...
 ;   %vreg6:ssub_0<def> = VLDRS <cp#0>, 0, pred:14, pred:%noreg; mem:LD4[ConstantPool] DPR_VFP2:%vreg6
 ;
 ; When %vreg6 spills, the VLDRS constant pool load cannot be rematerialized
 ; since it implicitly reads the ssub_1 sub-register.
 ;
 ; CHECK: f1
-; CHECK: vmov    s1, r0
+; CHECK: vmov    d0, r0, r0
 ; CHECK: vldr s0, LCPI
 ; The vector must be spilled:
 ; CHECK: vstr d0,

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/trap.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/trap.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/trap.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/trap.ll Tue Jan 15 11:16:16 2013
@@ -14,4 +14,16 @@
   unreachable
 }
 
+define void @t2() nounwind {
+entry:
+; INSTR: t2:
+; INSTR: trap
+
+; FUNC: t2:
+; FUNC: bl __trap
+  call void @llvm.debugtrap()
+  unreachable
+}
+
 declare void @llvm.trap() nounwind
+declare void @llvm.debugtrap() nounwind

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/twoaddrinstr.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/twoaddrinstr.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/twoaddrinstr.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/twoaddrinstr.ll Tue Jan 15 11:16:16 2013
@@ -4,18 +4,18 @@
 define void @PR13378() nounwind {
 ; This was orriginally a crasher trying to schedule the instructions.
 ; CHECK:      PR13378:
-; CHECK:        vldmia
+; CHECK:        vld1.32
+; CHECK-NEXT:   vst1.32
+; CHECK-NEXT:   vst1.32
 ; CHECK-NEXT:   vmov.f32
-; CHECK-NEXT:   vstmia
-; CHECK-NEXT:   vstmia
 ; CHECK-NEXT:   vmov.f32
-; CHECK-NEXT:   vstmia
+; CHECK-NEXT:   vst1.32
 
 entry:
-  %0 = load <4 x float>* undef
-  store <4 x float> zeroinitializer, <4 x float>* undef
-  store <4 x float> %0, <4 x float>* undef
+  %0 = load <4 x float>* undef, align 4
+  store <4 x float> zeroinitializer, <4 x float>* undef, align 4
+  store <4 x float> %0, <4 x float>* undef, align 4
   %1 = insertelement <4 x float> %0, float 1.000000e+00, i32 3
-  store <4 x float> %1, <4 x float>* undef
+  store <4 x float> %1, <4 x float>* undef, align 4
   unreachable
 }

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/unaligned_load_store.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/unaligned_load_store.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/unaligned_load_store.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/unaligned_load_store.ll Tue Jan 15 11:16:16 2013
@@ -1,25 +1,25 @@
-; RUN: llc < %s -march=arm -pre-RA-sched=source | FileCheck %s -check-prefix=GENERIC
-; RUN: llc < %s -mtriple=armv6-apple-darwin | FileCheck %s -check-prefix=DARWIN_V6
-; RUN: llc < %s -mtriple=armv6-apple-darwin -arm-strict-align | FileCheck %s -check-prefix=GENERIC
-; RUN: llc < %s -mtriple=armv6-linux | FileCheck %s -check-prefix=GENERIC
+; RUN: llc < %s -march=arm -pre-RA-sched=source | FileCheck %s -check-prefix=EXPANDED
+; RUN: llc < %s -mtriple=armv6-apple-darwin -mcpu=cortex-a8 -mattr=-neon -arm-strict-align -pre-RA-sched=source | FileCheck %s -check-prefix=EXPANDED
+; RUN: llc < %s -mtriple=armv6-apple-darwin -mcpu=cortex-a8 | FileCheck %s -check-prefix=UNALIGNED
 
 ; rdar://7113725
+; rdar://12091029
 
 define void @t(i8* nocapture %a, i8* nocapture %b) nounwind {
 entry:
-; GENERIC: t:
-; GENERIC: ldrb [[R2:r[0-9]+]]
-; GENERIC: ldrb [[R3:r[0-9]+]]
-; GENERIC: ldrb [[R12:r[0-9]+]]
-; GENERIC: ldrb [[R1:r[0-9]+]]
-; GENERIC: strb [[R1]]
-; GENERIC: strb [[R12]]
-; GENERIC: strb [[R3]]
-; GENERIC: strb [[R2]]
-
-; DARWIN_V6: t:
-; DARWIN_V6: ldr r1
-; DARWIN_V6: str r1
+; EXPANDED: t:
+; EXPANDED: ldrb [[R2:r[0-9]+]]
+; EXPANDED: ldrb [[R3:r[0-9]+]]
+; EXPANDED: ldrb [[R12:r[0-9]+]]
+; EXPANDED: ldrb [[R1:r[0-9]+]]
+; EXPANDED: strb [[R1]]
+; EXPANDED: strb [[R12]]
+; EXPANDED: strb [[R3]]
+; EXPANDED: strb [[R2]]
+
+; UNALIGNED: t:
+; UNALIGNED: ldr r1
+; UNALIGNED: str r1
 
   %__src1.i = bitcast i8* %b to i32*              ; <i32*> [#uses=1]
   %__dest2.i = bitcast i8* %a to i32*             ; <i32*> [#uses=1]
@@ -27,3 +27,51 @@
   store i32 %tmp.i, i32* %__dest2.i, align 1
   ret void
 }
+
+define void @hword(double* %a, double* %b) nounwind {
+entry:
+; EXPANDED: hword:
+; EXPANDED-NOT: vld1
+; EXPANDED: ldrh
+; EXPANDED-NOT: str1
+; EXPANDED: strh
+
+; UNALIGNED: hword:
+; UNALIGNED: vld1.16
+; UNALIGNED: vst1.16
+  %tmp = load double* %a, align 2
+  store double %tmp, double* %b, align 2
+  ret void
+}
+
+define void @byte(double* %a, double* %b) nounwind {
+entry:
+; EXPANDED: byte:
+; EXPANDED-NOT: vld1
+; EXPANDED: ldrb
+; EXPANDED-NOT: str1
+; EXPANDED: strb
+
+; UNALIGNED: byte:
+; UNALIGNED: vld1.8
+; UNALIGNED: vst1.8
+  %tmp = load double* %a, align 1
+  store double %tmp, double* %b, align 1
+  ret void
+}
+
+define void @byte_word_ops(i32* %a, i32* %b) nounwind {
+entry:
+; EXPANDED: byte_word_ops:
+; EXPANDED: ldrb
+; EXPANDED: strb
+
+; UNALIGNED: byte_word_ops:
+; UNALIGNED-NOT: ldrb
+; UNALIGNED: ldr
+; UNALIGNED-NOT: strb
+; UNALIGNED: str
+  %tmp = load i32* %a, align 1
+  store i32 %tmp, i32* %b, align 1
+  ret void
+}

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/vbsl-constant.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/vbsl-constant.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/vbsl-constant.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/vbsl-constant.ll Tue Jan 15 11:16:16 2013
@@ -1,4 +1,4 @@
-; RUN: llc < %s -march=arm -mattr=+neon | FileCheck %s
+; RUN: llc < %s -mtriple=arm-apple-ios -mattr=+neon | FileCheck %s
 
 define <8 x i8> @v_bsli8(<8 x i8>* %A, <8 x i8>* %B, <8 x i8>* %C) nounwind {
 ;CHECK: v_bsli8:
@@ -59,8 +59,8 @@
 
 define <16 x i8> @v_bslQi8(<16 x i8>* %A, <16 x i8>* %B, <16 x i8>* %C) nounwind {
 ;CHECK: v_bslQi8:
-;CHECK: vldmia
-;CHECK: vldmia
+;CHECK: vld1.32
+;CHECK: vld1.32
 ;CHECK: vbsl
 	%tmp1 = load <16 x i8>* %A
 	%tmp2 = load <16 x i8>* %B
@@ -73,8 +73,8 @@
 
 define <8 x i16> @v_bslQi16(<8 x i16>* %A, <8 x i16>* %B, <8 x i16>* %C) nounwind {
 ;CHECK: v_bslQi16:
-;CHECK: vldmia
-;CHECK: vldmia
+;CHECK: vld1.32
+;CHECK: vld1.32
 ;CHECK: vbsl
 	%tmp1 = load <8 x i16>* %A
 	%tmp2 = load <8 x i16>* %B
@@ -87,8 +87,8 @@
 
 define <4 x i32> @v_bslQi32(<4 x i32>* %A, <4 x i32>* %B, <4 x i32>* %C) nounwind {
 ;CHECK: v_bslQi32:
-;CHECK: vldmia
-;CHECK: vldmia
+;CHECK: vld1.32
+;CHECK: vld1.32
 ;CHECK: vbsl
 	%tmp1 = load <4 x i32>* %A
 	%tmp2 = load <4 x i32>* %B
@@ -101,9 +101,9 @@
 
 define <2 x i64> @v_bslQi64(<2 x i64>* %A, <2 x i64>* %B, <2 x i64>* %C) nounwind {
 ;CHECK: v_bslQi64:
-;CHECK: vldmia
-;CHECK: vldmia
-;CHECK: vldmia
+;CHECK: vld1.32
+;CHECK: vld1.32
+;CHECK: vld1.64
 ;CHECK: vbsl
 	%tmp1 = load <2 x i64>* %A
 	%tmp2 = load <2 x i64>* %B

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/vbsl.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/vbsl.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/vbsl.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/vbsl.ll Tue Jan 15 11:16:16 2013
@@ -1,5 +1,7 @@
 ; RUN: llc < %s -march=arm -mattr=+neon | FileCheck %s
 
+; rdar://12471808
+
 define <8 x i8> @v_bsli8(<8 x i8>* %A, <8 x i8>* %B, <8 x i8>* %C) nounwind {
 ;CHECK: v_bsli8:
 ;CHECK: vbsl
@@ -103,3 +105,98 @@
 	%tmp7 = or <2 x i64> %tmp4, %tmp6
 	ret <2 x i64> %tmp7
 }
+
+define <8 x i8> @f1(<8 x i8> %a, <8 x i8> %b, <8 x i8> %c) nounwind readnone optsize ssp {
+; CHECK: f1:
+; CHECK: vbsl
+  %vbsl.i = tail call <8 x i8> @llvm.arm.neon.vbsl.v8i8(<8 x i8> %a, <8 x i8> %b, <8 x i8> %c) nounwind
+  ret <8 x i8> %vbsl.i
+}
+
+define <4 x i16> @f2(<4 x i16> %a, <4 x i16> %b, <4 x i16> %c) nounwind readnone optsize ssp {
+; CHECK: f2:
+; CHECK: vbsl
+  %vbsl3.i = tail call <4 x i16> @llvm.arm.neon.vbsl.v4i16(<4 x i16> %a, <4 x i16> %b, <4 x i16> %c) nounwind
+  ret <4 x i16> %vbsl3.i
+}
+
+define <2 x i32> @f3(<2 x i32> %a, <2 x i32> %b, <2 x i32> %c) nounwind readnone optsize ssp {
+; CHECK: f3:
+; CHECK: vbsl
+  %vbsl3.i = tail call <2 x i32> @llvm.arm.neon.vbsl.v2i32(<2 x i32> %a, <2 x i32> %b, <2 x i32> %c) nounwind
+  ret <2 x i32> %vbsl3.i
+}
+
+define <2 x float> @f4(<2 x float> %a, <2 x float> %b, <2 x float> %c) nounwind readnone optsize ssp {
+; CHECK: f4:
+; CHECK: vbsl
+  %vbsl4.i = tail call <2 x float> @llvm.arm.neon.vbsl.v2f32(<2 x float> %a, <2 x float> %b, <2 x float> %c) nounwind
+  ret <2 x float> %vbsl4.i
+}
+
+define <16 x i8> @g1(<16 x i8> %a, <16 x i8> %b, <16 x i8> %c) nounwind readnone optsize ssp {
+; CHECK: g1:
+; CHECK: vbsl
+  %vbsl.i = tail call <16 x i8> @llvm.arm.neon.vbsl.v16i8(<16 x i8> %a, <16 x i8> %b, <16 x i8> %c) nounwind
+  ret <16 x i8> %vbsl.i
+}
+
+define <8 x i16> @g2(<8 x i16> %a, <8 x i16> %b, <8 x i16> %c) nounwind readnone optsize ssp {
+; CHECK: g2:
+; CHECK: vbsl
+  %vbsl3.i = tail call <8 x i16> @llvm.arm.neon.vbsl.v8i16(<8 x i16> %a, <8 x i16> %b, <8 x i16> %c) nounwind
+  ret <8 x i16> %vbsl3.i
+}
+
+define <4 x i32> @g3(<4 x i32> %a, <4 x i32> %b, <4 x i32> %c) nounwind readnone optsize ssp {
+; CHECK: g3:
+; CHECK: vbsl
+  %vbsl3.i = tail call <4 x i32> @llvm.arm.neon.vbsl.v4i32(<4 x i32> %a, <4 x i32> %b, <4 x i32> %c) nounwind
+  ret <4 x i32> %vbsl3.i
+}
+
+define <4 x float> @g4(<4 x float> %a, <4 x float> %b, <4 x float> %c) nounwind readnone optsize ssp {
+; CHECK: g4:
+; CHECK: vbsl
+  %vbsl4.i = tail call <4 x float> @llvm.arm.neon.vbsl.v4f32(<4 x float> %a, <4 x float> %b, <4 x float> %c) nounwind
+  ret <4 x float> %vbsl4.i
+}
+
+define <1 x i64> @test_vbsl_s64(<1 x i64> %a, <1 x i64> %b, <1 x i64> %c) nounwind readnone optsize ssp {
+; CHECK: test_vbsl_s64:
+; CHECK: vbsl d
+  %vbsl3.i = tail call <1 x i64> @llvm.arm.neon.vbsl.v1i64(<1 x i64> %a, <1 x i64> %b, <1 x i64> %c) nounwind
+  ret <1 x i64> %vbsl3.i
+}
+
+define <1 x i64> @test_vbsl_u64(<1 x i64> %a, <1 x i64> %b, <1 x i64> %c) nounwind readnone optsize ssp {
+; CHECK: test_vbsl_u64:
+; CHECK: vbsl d
+  %vbsl3.i = tail call <1 x i64> @llvm.arm.neon.vbsl.v1i64(<1 x i64> %a, <1 x i64> %b, <1 x i64> %c) nounwind
+  ret <1 x i64> %vbsl3.i
+}
+
+define <2 x i64> @test_vbslq_s64(<2 x i64> %a, <2 x i64> %b, <2 x i64> %c) nounwind readnone optsize ssp {
+; CHECK: test_vbslq_s64:
+; CHECK: vbsl q
+  %vbsl3.i = tail call <2 x i64> @llvm.arm.neon.vbsl.v2i64(<2 x i64> %a, <2 x i64> %b, <2 x i64> %c) nounwind
+  ret <2 x i64> %vbsl3.i
+}
+
+define <2 x i64> @test_vbslq_u64(<2 x i64> %a, <2 x i64> %b, <2 x i64> %c) nounwind readnone optsize ssp {
+; CHECK: test_vbslq_u64:
+; CHECK: vbsl q
+  %vbsl3.i = tail call <2 x i64> @llvm.arm.neon.vbsl.v2i64(<2 x i64> %a, <2 x i64> %b, <2 x i64> %c) nounwind
+  ret <2 x i64> %vbsl3.i
+}
+
+declare <4 x i32> @llvm.arm.neon.vbsl.v4i32(<4 x i32>, <4 x i32>, <4 x i32>) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vbsl.v8i16(<8 x i16>, <8 x i16>, <8 x i16>) nounwind readnone
+declare <16 x i8> @llvm.arm.neon.vbsl.v16i8(<16 x i8>, <16 x i8>, <16 x i8>) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vbsl.v2i32(<2 x i32>, <2 x i32>, <2 x i32>) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vbsl.v4i16(<4 x i16>, <4 x i16>, <4 x i16>) nounwind readnone
+declare <8 x i8> @llvm.arm.neon.vbsl.v8i8(<8 x i8>, <8 x i8>, <8 x i8>) nounwind readnone
+declare <2 x float> @llvm.arm.neon.vbsl.v2f32(<2 x float>, <2 x float>, <2 x float>) nounwind readnone
+declare <4 x float> @llvm.arm.neon.vbsl.v4f32(<4 x float>, <4 x float>, <4 x float>) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vbsl.v2i64(<2 x i64>, <2 x i64>, <2 x i64>) nounwind readnone
+declare <1 x i64> @llvm.arm.neon.vbsl.v1i64(<1 x i64>, <1 x i64>, <1 x i64>) nounwind readnone

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/vdup.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/vdup.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/vdup.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/vdup.ll Tue Jan 15 11:16:16 2013
@@ -261,3 +261,73 @@
   store <8 x i8> %2, <8 x i8>* %ptr, align 8
   ret void
 }
+
+define <4 x i32> @tdupi(i32 %x, i32 %y) {
+;CHECK: tdupi
+;CHECK: vdup.32
+  %1 = insertelement <4 x i32> undef, i32 %x, i32 0
+  %2 = insertelement <4 x i32> %1, i32 %x, i32 1
+  %3 = insertelement <4 x i32> %2, i32 %x, i32 2
+  %4 = insertelement <4 x i32> %3, i32 %y, i32 3
+  ret <4 x i32> %4
+}
+
+define <4 x float> @tdupf(float %x, float %y) {
+;CHECK: tdupf
+;CHECK: vdup.32
+  %1 = insertelement <4 x float> undef, float %x, i32 0
+  %2 = insertelement <4 x float> %1, float %x, i32 1
+  %3 = insertelement <4 x float> %2, float %x, i32 2
+  %4 = insertelement <4 x float> %3, float %y, i32 3
+  ret <4 x float> %4
+}
+
+; This test checks that when splatting an element from a vector into another,
+; the value isn't moved out to GPRs first.
+define <4 x i32> @tduplane(<4 x i32> %invec) {
+;CHECK: tduplane
+;CHECK-NOT: vmov {{.*}}, d16[1]
+;CHECK: vdup.32 {{.*}}, d16[1]
+  %in = extractelement <4 x i32> %invec, i32 1
+  %1 = insertelement <4 x i32> undef, i32 %in, i32 0
+  %2 = insertelement <4 x i32> %1, i32 %in, i32 1
+  %3 = insertelement <4 x i32> %2, i32 %in, i32 2
+  %4 = insertelement <4 x i32> %3, i32 255, i32 3
+  ret <4 x i32> %4
+}
+
+define <2 x float> @check_f32(<4 x float> %v) nounwind {
+;CHECK: check_f32:
+;CHECK: vdup.32 {{.*}}, d{{..}}[1]
+  %x = extractelement <4 x float> %v, i32 3
+  %1 = insertelement  <2 x float> undef, float %x, i32 0
+  %2 = insertelement  <2 x float> %1, float %x, i32 1
+  ret <2 x float> %2
+}
+
+define <2 x i32> @check_i32(<4 x i32> %v) nounwind {
+;CHECK: check_i32:
+;CHECK: vdup.32 {{.*}}, d{{..}}[1]
+  %x = extractelement <4 x i32> %v, i32 3
+  %1 = insertelement  <2 x i32> undef, i32 %x, i32 0
+  %2 = insertelement  <2 x i32> %1, i32 %x, i32 1
+  ret <2 x i32> %2
+}
+
+define <4 x i16> @check_i16(<8 x i16> %v) nounwind {
+;CHECK: check_i16:
+;CHECK: vdup.16 {{.*}}, d{{..}}[3]
+  %x = extractelement <8 x i16> %v, i32 3
+  %1 = insertelement  <4 x i16> undef, i16 %x, i32 0
+  %2 = insertelement  <4 x i16> %1, i16 %x, i32 1
+  ret <4 x i16> %2
+}
+
+define <8 x i8> @check_i8(<16 x i8> %v) nounwind {
+;CHECK: check_i8:
+;CHECK: vdup.8 {{.*}}, d{{..}}[3]
+  %x = extractelement <16 x i8> %v, i32 3
+  %1 = insertelement  <8  x i8> undef, i8 %x, i32 0
+  %2 = insertelement  <8  x i8> %1, i8 %x, i32 1
+  ret <8 x i8> %2
+}

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/vector-extend-narrow.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/vector-extend-narrow.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/vector-extend-narrow.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/vector-extend-narrow.ll Tue Jan 15 11:16:16 2013
@@ -62,3 +62,14 @@
   %2 = sdiv <4 x i8> zeroinitializer, %1
   ret <4 x i8> %2
 }
+; CHECK: j:
+define <4 x i32> @j(<4 x i8>* %in) nounwind {
+  ; CHECK: vld1
+  ; CHECK: vmovl.u8
+  ; CHECK: vmovl.u16
+  ; CHECK-NOT: vand
+  %1 = load <4 x i8>* %in, align 4
+  %2 = zext <4 x i8> %1 to <4 x i32>
+  ret <4 x i32> %2
+}
+

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/vext.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/vext.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/vext.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/vext.ll Tue Jan 15 11:16:16 2013
@@ -74,6 +74,39 @@
 	ret <16 x i8> %tmp3
 }
 
+define <16 x i8> @test_vextq_undef_op2(<16 x i8> %a) nounwind {
+;CHECK: test_vextq_undef_op2:
+;CHECK: vext
+entry:
+  %tmp1 = shufflevector <16 x i8> %a, <16 x i8> undef, <16 x i32> <i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 0, i32 1>
+  ret <16 x i8> %tmp1
+}
+
+define <8 x i8> @test_vextd_undef_op2(<8 x i8> %a) nounwind {
+;CHECK: test_vextd_undef_op2:
+;CHECK: vext
+entry:
+  %tmp1 = shufflevector <8 x i8> %a, <8 x i8> undef, <8 x i32> <i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1>
+  ret <8 x i8> %tmp1
+}
+
+
+define <16 x i8> @test_vextq_undef_op2_undef(<16 x i8> %a) nounwind {
+;CHECK: test_vextq_undef_op2_undef:
+;CHECK: vext
+entry:
+  %tmp1 = shufflevector <16 x i8> %a, <16 x i8> undef, <16 x i32> <i32 2, i32 3, i32 4, i32 undef, i32 undef, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 0, i32 1>
+  ret <16 x i8> %tmp1
+}
+
+define <8 x i8> @test_vextd_undef_op2_undef(<8 x i8> %a) nounwind {
+;CHECK: test_vextd_undef_op2_undef:
+;CHECK: vext
+entry:
+  %tmp1 = shufflevector <8 x i8> %a, <8 x i8> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 6, i32 7, i32 undef, i32 1>
+  ret <8 x i8> %tmp1
+}
+
 ; Tests for ReconstructShuffle function. Indices have to be carefully
 ; chosen to reach lowering phase as a BUILD_VECTOR.
 

Modified: llvm/branches/AMDILBackend/test/CodeGen/ARM/vget_lane.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/ARM/vget_lane.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/ARM/vget_lane.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/ARM/vget_lane.ll Tue Jan 15 11:16:16 2013
@@ -200,7 +200,7 @@
 
 define <4 x i32> @vsetQ_lane32(<4 x i32>* %A, i32 %B) nounwind {
 ;CHECK: vsetQ_lane32:
-;CHECK: vmov.32
+;CHECK: vmov.32 d{{.*}}[1], r1
 	%tmp1 = load <4 x i32>* %A
 	%tmp2 = insertelement <4 x i32> %tmp1, i32 %B, i32 1
 	ret <4 x i32> %tmp2

Modified: llvm/branches/AMDILBackend/test/CodeGen/CellSPU/icmp16.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/CellSPU/icmp16.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/CellSPU/icmp16.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/CellSPU/icmp16.ll Tue Jan 15 11:16:16 2013
@@ -534,7 +534,7 @@
 define i16 @icmp_slt_immed04_i16(i16 %arg1, i16 %val1, i16 %val2) nounwind {
 ; CHECK:      icmp_slt_immed04_i16:
 ; CHECK:        lr
-; CHECK-NETX:   bi
+; CHECK-NEXT:   bi
 
 entry:
        %A = icmp slt i16 %arg1, 32768
@@ -559,7 +559,7 @@
 ; CHECK:        ilhu
 ; CHECK:        xorhi
 ; CHECK:        iohl
-; CHECK-NETX:   bi
+; CHECK:   bi
 
 entry:
        %A = icmp sle i16 %arg1, %arg2

Removed: llvm/branches/AMDILBackend/test/CodeGen/Generic/2009-06-03-UnreachableSplitPad.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/Generic/2009-06-03-UnreachableSplitPad.ll?rev=172540&view=auto
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/Generic/2009-06-03-UnreachableSplitPad.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/Generic/2009-06-03-UnreachableSplitPad.ll (removed)
@@ -1,19 +0,0 @@
-; RUN: llc < %s
-; PR4317
-
-declare i32 @b()
-
-define void @a() {
-entry:
-  ret void
-
-dummy:
-  invoke i32 @b() to label %reg unwind label %reg
-
-reg:
-  %lpad = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
-            catch i8* null
-  ret void
-}
-
-declare i32 @__gxx_personality_v0(...)

Modified: llvm/branches/AMDILBackend/test/CodeGen/Generic/donothing.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/Generic/donothing.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/Generic/donothing.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/Generic/donothing.ll Tue Jan 15 11:16:16 2013
@@ -7,7 +7,7 @@
 ; CHECK: f1
 define void @f1() nounwind uwtable ssp {
 entry:
-; CHECK-NOT donothing
+; CHECK-NOT: donothing
   invoke void @llvm.donothing()
   to label %invoke.cont unwind label %lpad
 
@@ -25,7 +25,7 @@
 ; CHECK: f2
 define void @f2() nounwind {
 entry:
-; CHECK-NOT donothing
+; CHECK-NOT: donothing
   call void @llvm.donothing()
   ret void
 }

Modified: llvm/branches/AMDILBackend/test/CodeGen/Hexagon/args.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/Hexagon/args.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/Hexagon/args.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/Hexagon/args.ll Tue Jan 15 11:16:16 2013
@@ -1,12 +1,12 @@
-; RUN: llc -march=hexagon -mcpu=hexagonv4 -disable-dfa-sched < %s | FileCheck %s
+; RUN: llc -march=hexagon -mcpu=hexagonv4 -disable-hexagon-misched < %s | FileCheck %s
 ; CHECK: r[[T0:[0-9]+]] = #7
 ; CHECK: memw(r29 + #0) = r[[T0]]
+; CHECK: r5 = #6
 ; CHECK: r0 = #1
 ; CHECK: r1 = #2
 ; CHECK: r2 = #3
 ; CHECK: r3 = #4
 ; CHECK: r4 = #5
-; CHECK: r5 = #6
 
 
 define void @foo() nounwind {

Modified: llvm/branches/AMDILBackend/test/CodeGen/Hexagon/newvaluestore.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/Hexagon/newvaluestore.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/Hexagon/newvaluestore.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/Hexagon/newvaluestore.ll Tue Jan 15 11:16:16 2013
@@ -1,4 +1,4 @@
-; RUN: llc -march=hexagon -mcpu=hexagonv4  < %s | FileCheck %s
+; RUN: llc -march=hexagon -mcpu=hexagonv4 -disable-hexagon-misched < %s | FileCheck %s
 ; Check that we generate new value store packet in V4
 
 @i = global i32 0, align 4

Modified: llvm/branches/AMDILBackend/test/CodeGen/Hexagon/static.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/Hexagon/static.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/Hexagon/static.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/Hexagon/static.ll Tue Jan 15 11:16:16 2013
@@ -1,4 +1,4 @@
-; RUN: llc -march=hexagon -mcpu=hexagonv4 -disable-dfa-sched < %s | FileCheck %s
+; RUN: llc -march=hexagon -mcpu=hexagonv4 -disable-dfa-sched -disable-hexagon-misched < %s | FileCheck %s
 
 @num = external global i32
 @acc = external global i32

Modified: llvm/branches/AMDILBackend/test/CodeGen/Mips/atomic.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/Mips/atomic.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/Mips/atomic.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/Mips/atomic.ll Tue Jan 15 11:16:16 2013
@@ -1,4 +1,4 @@
-; RUN: llc -march=mipsel < %s | FileCheck %s
+; RUN: llc -march=mipsel --disable-machine-licm < %s | FileCheck %s
 
 @x = common global i32 0, align 4
 
@@ -181,8 +181,9 @@
 
 ; CHECK:   $[[BB0:[A-Z_0-9]+]]:
 ; CHECK:   ll      $[[R10:[0-9]+]], 0($[[R2]])
+; CHECK:   and     $[[R18:[0-9]+]], $[[R9]], $[[R6]]
 ; CHECK:   and     $[[R13:[0-9]+]], $[[R10]], $[[R7]]
-; CHECK:   or      $[[R14:[0-9]+]], $[[R13]], $[[R9]]
+; CHECK:   or      $[[R14:[0-9]+]], $[[R13]], $[[R18]]
 ; CHECK:   sc      $[[R14]], 0($[[R2]])
 ; CHECK:   beq     $[[R14]], $zero, $[[BB0]]
 

Modified: llvm/branches/AMDILBackend/test/CodeGen/Mips/brdelayslot.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/Mips/brdelayslot.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/Mips/brdelayslot.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/Mips/brdelayslot.ll Tue Jan 15 11:16:16 2013
@@ -1,15 +1,37 @@
-; RUN: llc -march=mipsel  -enable-mips-delay-filler < %s | FileCheck %s
+; RUN: llc -march=mipsel -O0 < %s | FileCheck %s -check-prefix=None
+; RUN: llc -march=mipsel < %s | FileCheck %s -check-prefix=Default
 
 define void @foo1() nounwind {
 entry:
-; CHECK:      jalr 
-; CHECK-NOT:  nop 
-; CHECK:      jr 
-; CHECK-NOT:  nop
-; CHECK:      .end
+; Default:     jalr 
+; Default-NOT: nop 
+; Default:     jr 
+; Default-NOT: nop
+; Default:     .end
+; None: jalr 
+; None: nop 
+; None: jr 
+; None: nop
+; None: .end
 
   tail call void @foo2(i32 3) nounwind
   ret void
 }
 
 declare void @foo2(i32)
+
+; Check that cvt.d.w goes into jalr's delay slot.
+;
+define void @foo3(i32 %a) nounwind {
+entry:
+; Default:     foo3:
+; Default:     jalr
+; Default:     cvt.d.w
+
+  %conv = sitofp i32 %a to double
+  tail call void @foo4(double %conv) nounwind
+  ret void
+}
+
+declare void @foo4(double)
+

Modified: llvm/branches/AMDILBackend/test/CodeGen/Mips/helloworld.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/Mips/helloworld.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/Mips/helloworld.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/Mips/helloworld.ll Tue Jan 15 11:16:16 2013
@@ -24,10 +24,10 @@
 ; C1:	addiu	${{[0-9]+}}, %lo($.str)
 ; C2:	move	$25, ${{[0-9]+}}
 ; C1:	move 	$gp, ${{[0-9]+}}
-; C1:	jalr 	${{[0-9]+}}
+; C1:	jalrc 	${{[0-9]+}}
 ; SR:	restore 	$ra, [[FS]]
 ; PE:	li	$2, 0
-; PE:	jr 	$ra
+; PE:	jrc 	$ra
 
 }
 

Modified: llvm/branches/AMDILBackend/test/CodeGen/Mips/largeimm1.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/Mips/largeimm1.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/Mips/largeimm1.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/Mips/largeimm1.ll Tue Jan 15 11:16:16 2013
@@ -1,7 +1,7 @@
 ; RUN: llc -march=mipsel < %s | FileCheck %s
 
-; CHECK: lui $at, 49152
-; CHECK: lui $at, 16384
+; CHECK: lui ${{[0-9]+}}, 49152
+; CHECK: lui ${{[0-9]+}}, 16384
 define void @f() nounwind {
 entry:
   %a1 = alloca [1073741824 x i8], align 1

Modified: llvm/branches/AMDILBackend/test/CodeGen/Mips/largeimmprinting.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/Mips/largeimmprinting.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/Mips/largeimmprinting.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/Mips/largeimmprinting.ll Tue Jan 15 11:16:16 2013
@@ -1,4 +1,6 @@
-; RUN: llc -march=mipsel -mcpu=mips32r2 < %s | FileCheck %s
+; RUN: llc -march=mipsel < %s | FileCheck %s -check-prefix=32
+; RUN: llc -march=mips64el -mcpu=mips64 -mattr=n64 < %s | \
+; RUN: FileCheck %s -check-prefix=64
 
 %struct.S1 = type { [65536 x i8] }
 
@@ -6,9 +8,21 @@
 
 define void @f() nounwind {
 entry:
-; CHECK:  lui $at, 65535
-; CHECK:  addiu $at, $at, -16
-; CHECK:  addu  $sp, $sp, $at
+; 32:  lui $[[R0:[0-9]+]], 65535
+; 32:  addiu $[[R0]], $[[R0]], -24
+; 32:  addu $sp, $sp, $[[R0]]
+; 32:  lui $[[R1:[0-9]+]], 1
+; 32:  addu $[[R1]], $sp, $[[R1]]
+; 32:  sw $ra, 20($[[R1]])
+; 64:  daddiu  $[[R0:[0-9]+]], $zero, 1
+; 64:  dsll  $[[R0]], $[[R0]], 48
+; 64:  daddiu  $[[R0]], $[[R0]], -1
+; 64:  dsll  $[[R0]], $[[R0]], 16
+; 64:  daddiu  $[[R0]], $[[R0]], -48
+; 64:  daddu $sp, $sp, $[[R0]]
+; 64:  lui $[[R1:[0-9]+]], 1
+; 64:  daddu $[[R1]], $sp, $[[R1]]
+; 64:  sd  $ra, 40($[[R1]])
 
   %agg.tmp = alloca %struct.S1, align 1
   %tmp = getelementptr inbounds %struct.S1* %agg.tmp, i32 0, i32 0, i32 0

Modified: llvm/branches/AMDILBackend/test/CodeGen/Mips/longbranch.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/Mips/longbranch.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/Mips/longbranch.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/Mips/longbranch.ll Tue Jan 15 11:16:16 2013
@@ -6,9 +6,15 @@
 define void @foo1(i32 %s) nounwind {
 entry:
 ; O32: bal
+; O32: lui $1, 0
+; O32: addiu $1, $1, {{[0-9]+}} 
+; N64: lui $1, 0
+; N64: daddiu $1, $1, 0
+; N64: dsll $1, $1, 16
+; N64: daddiu $1, $1, 0
 ; N64: bal
-; N64: highest
-; N64: higher
+; N64: dsll $1, $1, 16
+; N64: daddiu $1, $1, {{[0-9]+}}  
 
   %tobool = icmp eq i32 %s, 0
   br i1 %tobool, label %if.end, label %if.then

Modified: llvm/branches/AMDILBackend/test/CodeGen/Mips/null.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/Mips/null.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/Mips/null.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/Mips/null.ll Tue Jan 15 11:16:16 2013
@@ -8,6 +8,6 @@
 ; 16: 	.set	mips16                  # @main
 
 
-; 16:	jr	$ra
+; 16:	jrc	$ra
 
 }

Modified: llvm/branches/AMDILBackend/test/CodeGen/Mips/o32_cc_byval.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/Mips/o32_cc_byval.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/Mips/o32_cc_byval.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/Mips/o32_cc_byval.ll Tue Jan 15 11:16:16 2013
@@ -119,6 +119,16 @@
   ret void
 }
 
+%struct.S4 = type { [4 x i32] }
+
+define void @f5(i64 %a0, %struct.S4* nocapture byval %a1) nounwind {
+entry:
+  tail call void @f6(%struct.S4* byval %a1, i64 %a0) nounwind
+  ret void
+}
+
+declare void @f6(%struct.S4* nocapture byval, i64)
+
 !0 = metadata !{metadata !"int", metadata !1}
 !1 = metadata !{metadata !"omnipotent char", metadata !2}
 !2 = metadata !{metadata !"Simple C/C++ TBAA", null}

Modified: llvm/branches/AMDILBackend/test/CodeGen/Mips/tls-alias.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/Mips/tls-alias.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/Mips/tls-alias.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/Mips/tls-alias.ll Tue Jan 15 11:16:16 2013
@@ -1,4 +1,4 @@
-; RUN: llc -march=mipsel -relocation-model=pic < %s | FileCheck %s
+; RUN: llc -march=mipsel -relocation-model=pic -disable-mips-delay-filler < %s | FileCheck %s
 
 @foo = thread_local global i32 42
 @bar = hidden alias i32* @foo

Modified: llvm/branches/AMDILBackend/test/CodeGen/Mips/tls.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/Mips/tls.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/Mips/tls.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/Mips/tls.ll Tue Jan 15 11:16:16 2013
@@ -1,8 +1,10 @@
-; RUN: llc -march=mipsel < %s | FileCheck %s -check-prefix=PIC
-; RUN: llc -march=mipsel -relocation-model=static < %s \
-; RUN:                             | FileCheck %s -check-prefix=STATIC
-; RUN: llc -march=mipsel -relocation-model=static < %s \
-; RUN:   -mips-fix-global-base-reg=false | FileCheck %s -check-prefix=STATICGP
+; RUN: llc -march=mipsel -disable-mips-delay-filler < %s | \
+; RUN:     FileCheck %s -check-prefix=PIC
+; RUN: llc -march=mipsel -relocation-model=static -disable-mips-delay-filler < \
+; RUN:     %s | FileCheck %s -check-prefix=STATIC
+; RUN: llc -march=mipsel -relocation-model=static -disable-mips-delay-filler \
+; RUN:     -mips-fix-global-base-reg=false < %s  | \
+; RUN:     FileCheck %s -check-prefix=STATICGP
 
 @t1 = thread_local global i32 0, align 4
 

Modified: llvm/branches/AMDILBackend/test/CodeGen/PowerPC/2010-03-09-indirect-call.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/PowerPC/2010-03-09-indirect-call.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/PowerPC/2010-03-09-indirect-call.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/PowerPC/2010-03-09-indirect-call.ll Tue Jan 15 11:16:16 2013
@@ -9,9 +9,8 @@
 
 define void @foo() nounwind ssp {
 entry:
-; Better: mtctr r12
-; CHECK: mr r12, [[REG:r[0-9]+]]
-; CHECK: mtctr [[REG]]
+; CHECK: mtctr r12
+; CHECK: bctrl
   %0 = load void (...)** @p, align 4              ; <void (...)*> [#uses=1]
   call void (...)* %0() nounwind
   br label %return

Modified: llvm/branches/AMDILBackend/test/CodeGen/PowerPC/big-endian-formal-args.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/PowerPC/big-endian-formal-args.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/PowerPC/big-endian-formal-args.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/PowerPC/big-endian-formal-args.ll Tue Jan 15 11:16:16 2013
@@ -2,10 +2,10 @@
 
 declare void @bar(i64 %x, i64 %y)
 
-; CHECK: li {{[53]}}, 0
+; CHECK: li 3, 0
 ; CHECK: li 4, 2
+; CHECK: li 5, 0
 ; CHECK: li 6, 3
-; CHECK: mr {{[53]}}, {{[53]}}
 
 define void @foo() {
   call void @bar(i64 2, i64 3)

Removed: llvm/branches/AMDILBackend/test/CodeGen/PowerPC/bl8_elf_nop.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/PowerPC/bl8_elf_nop.ll?rev=172540&view=auto
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/PowerPC/bl8_elf_nop.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/PowerPC/bl8_elf_nop.ll (removed)
@@ -1,16 +0,0 @@
-; RUN: llc < %s -mtriple=powerpc64-unknown-linux-gnu | FileCheck  %s
-target datalayout = "E-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v128:128:128-n32:64"
-target triple = "powerpc64-unknown-linux-gnu"
-
-declare i32 @clock() nounwind
-
-define i32 @func() {
-entry:
-  %call = call i32 @clock() nounwind
-  %call2 = add i32 %call, 7
-  ret i32 %call2
-}
-
-; CHECK: bl clock
-; CHECK-NEXT: nop
-

Modified: llvm/branches/AMDILBackend/test/CodeGen/PowerPC/coalesce-ext.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/PowerPC/coalesce-ext.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/PowerPC/coalesce-ext.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/PowerPC/coalesce-ext.ll Tue Jan 15 11:16:16 2013
@@ -13,5 +13,6 @@
   store volatile i32 %D, i32* %P
   ; Reuse low bits of extended register, don't extend live range of SUM.
   ; CHECK: stw [[EXT]]
-  ret i32 %D
+  %R = add i32 %D, %D
+  ret i32 %R
 }

Modified: llvm/branches/AMDILBackend/test/CodeGen/PowerPC/inlineasm-copy.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/PowerPC/inlineasm-copy.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/PowerPC/inlineasm-copy.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/PowerPC/inlineasm-copy.ll Tue Jan 15 11:16:16 2013
@@ -1,5 +1,6 @@
-; RUN: llc < %s -march=ppc32 | not grep mr
+; RUN: llc < %s -march=ppc32 -verify-machineinstrs | FileCheck %s
 
+; CHECK-NOT: mr
 define i32 @test(i32 %Y, i32 %X) {
 entry:
         %tmp = tail call i32 asm "foo $0", "=r"( )              ; <i32> [#uses=1]
@@ -12,3 +13,9 @@
         ret i32 %tmp1
 }
 
+; CHECK: test3
+define i32 @test3(i32 %Y, i32 %X) {
+entry:
+        %tmp1 = tail call { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } asm sideeffect "foo $0, $1", "=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19"( i32 %X, i32 %Y, i32 %X, i32 %Y, i32 %X, i32 %Y, i32 %X, i32 %Y, i32 %X, i32 %Y, i32 %X, i32 %Y, i32 %X, i32 %Y, i32 %X, i32 %Y, i32 %X, i32 %Y, i32 %X, i32 %Y )                ; <i32> [#uses=1]
+       ret i32 1
+}

Modified: llvm/branches/AMDILBackend/test/CodeGen/PowerPC/int-fp-conv-1.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/PowerPC/int-fp-conv-1.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/PowerPC/int-fp-conv-1.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/PowerPC/int-fp-conv-1.ll Tue Jan 15 11:16:16 2013
@@ -1,4 +1,5 @@
-; RUN: llc < %s -march=ppc64 | grep __floatditf
+; RUN: llc < %s -march=ppc64 | FileCheck %s
+; CHECK-NOT: __floatditf
 
 define i64 @__fixunstfdi(ppc_fp128 %a) nounwind  {
 entry:

Modified: llvm/branches/AMDILBackend/test/CodeGen/PowerPC/lbzux.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/PowerPC/lbzux.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/PowerPC/lbzux.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/PowerPC/lbzux.ll Tue Jan 15 11:16:16 2013
@@ -2,7 +2,7 @@
 target triple = "powerpc64-unknown-linux-gnu"
 ; RUN: llc < %s | FileCheck %s
 
-define fastcc void @allocateSpace() nounwind {
+define fastcc void @allocateSpace(i1 %cond1, i1 %cond2) nounwind {
 entry:
   %0 = load i8** undef, align 8, !tbaa !0
   br i1 undef, label %return, label %lor.lhs.false
@@ -20,10 +20,10 @@
   %idxprom17 = sext i32 0 to i64
   %arrayidx18 = getelementptr inbounds i8* %0, i64 %idxprom17
   %or = or i32 undef, undef
-  br i1 false, label %if.end71, label %while.body
+  br i1 %cond1, label %if.end71, label %while.body
 
 while.body:                                       ; preds = %while.cond
-  br i1 undef, label %while.cond, label %if.then45
+  br i1 %cond2, label %while.cond, label %if.then45
 
 if.then45:                                        ; preds = %while.body
   %idxprom48139 = zext i32 %or to i64

Removed: llvm/branches/AMDILBackend/test/CodeGen/PowerPC/ppc64-ind-call.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/PowerPC/ppc64-ind-call.ll?rev=172540&view=auto
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/PowerPC/ppc64-ind-call.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/PowerPC/ppc64-ind-call.ll (removed)
@@ -1,16 +0,0 @@
-; RUN: llc < %s -march=ppc64 | FileCheck %s
-target datalayout = "E-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v128:128:128-n32:64"
-target triple = "powerpc64-unknown-linux-gnu"
-
-define void @test1() {
-entry:
-  %call.i75 = call zeroext i8 undef(i8* undef, i8 zeroext 10)
-  unreachable
-}
-
-; CHECK: @test1
-; CHECK: ld 11, 0(3)
-; CHECK: ld 2, 8(3)
-; CHECK: bctrl
-; CHECK: ld 2, 40(1)
-

Modified: llvm/branches/AMDILBackend/test/CodeGen/PowerPC/ppc64-linux-func-size.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/PowerPC/ppc64-linux-func-size.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/PowerPC/ppc64-linux-func-size.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/PowerPC/ppc64-linux-func-size.ll Tue Jan 15 11:16:16 2013
@@ -5,6 +5,7 @@
 ; CHECK-NEXT:	.align 3
 ; CHECK-NEXT:	.quad .L.test1
 ; CHECK-NEXT:	.quad .TOC. at tocbase
+; CHECK-NEXT:   .quad 0
 ; CHECK-NEXT:	.text
 ; CHECK-NEXT: .L.test1:
 

Modified: llvm/branches/AMDILBackend/test/CodeGen/SPARC/2011-01-11-CC.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/SPARC/2011-01-11-CC.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/SPARC/2011-01-11-CC.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/SPARC/2011-01-11-CC.ll Tue Jan 15 11:16:16 2013
@@ -54,7 +54,7 @@
 ; V8: {{be|bne}}
 ; V9: test_select_dfp_icc
 ; V9: subcc
-; V9=NOT: {{be|bne}}
+; V9-NOT: {{be|bne}}
 ; V9: fmovd{{e|ne}} %icc
   %0 = icmp eq i32 %a, 0
   %1 = select i1 %0, double %f1, double %f2

Modified: llvm/branches/AMDILBackend/test/CodeGen/Thumb2/buildvector-crash.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/Thumb2/buildvector-crash.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/Thumb2/buildvector-crash.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/Thumb2/buildvector-crash.ll Tue Jan 15 11:16:16 2013
@@ -1,4 +1,4 @@
-; RUN: llc < %s -O3 -mtriple=thumbv7-apple-darwin10 -mcpu=cortex-a8 | FileCheck %s
+; RUN: llc < %s -O3 -mtriple=thumbv7-apple-ios -mcpu=cortex-a8 | FileCheck %s
 ; Formerly crashed, 3573915.
 
 define void @RotateStarsFP_Vec() nounwind {
@@ -13,5 +13,5 @@
   store <4 x float> %3, <4 x float>* undef, align 4
   br label %bb8
 ; CHECK: RotateStarsFP_Vec:
-; CHECK: vldmia
+; CHECK: vld1.64
 }

Modified: llvm/branches/AMDILBackend/test/CodeGen/Thumb2/carry.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/Thumb2/carry.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/Thumb2/carry.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/Thumb2/carry.ll Tue Jan 15 11:16:16 2013
@@ -20,3 +20,16 @@
 	%tmp2 = sub i64 %tmp1, %b
 	ret i64 %tmp2
 }
+
+; rdar://12559385
+define i64 @f3(i32 %vi) {
+entry:
+; CHECK: f3:
+; CHECK: movw [[REG:r[0-9]+]], #36102
+; CHECK: sbcs r{{[0-9]+}}, [[REG]]
+    %v0 = zext i32 %vi to i64
+    %v1 = xor i64 %v0, -155057456198619
+    %v4 = add i64 %v1, 155057456198619
+    %v5 = add i64 %v4, %v1
+    ret i64 %v5
+}

Modified: llvm/branches/AMDILBackend/test/CodeGen/Thumb2/cortex-fp.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/Thumb2/cortex-fp.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/Thumb2/cortex-fp.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/Thumb2/cortex-fp.ll Tue Jan 15 11:16:16 2013
@@ -7,8 +7,8 @@
 entry:
 ; CHECK: foo
 ; CORTEXM3: blx ___mulsf3
-; CORTEXM4: vmul.f32  s0, s1, s0
-; CORTEXA8: vmul.f32  d0, d1, d0
+; CORTEXM4: vmul.f32  s0, s2, s0
+; CORTEXA8: vmul.f32  d
   %0 = fmul float %a, %b
   ret float %0
 }
@@ -19,6 +19,6 @@
   %0 = fmul double %a, %b
 ; CORTEXM3: blx ___muldf3
 ; CORTEXM4: blx ___muldf3
-; CORTEXA8: vmul.f64  d16, d17, d16
+; CORTEXA8: vmul.f64  d
   ret double %0
 }

Modified: llvm/branches/AMDILBackend/test/CodeGen/Thumb2/div.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/Thumb2/div.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/Thumb2/div.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/Thumb2/div.ll Tue Jan 15 11:16:16 2013
@@ -2,6 +2,8 @@
 ; RUN:    | FileCheck %s -check-prefix=CHECK-THUMB
 ; RUN: llc < %s -march=thumb -mcpu=cortex-m3 -mattr=+thumb2 \
 ; RUN:    | FileCheck %s -check-prefix=CHECK-THUMBV7M
+; RUN: llc < %s -march=thumb -mcpu=swift \
+; RUN:    | FileCheck %s -check-prefix=CHECK-SWIFT-T2
 
 define i32 @f1(i32 %a, i32 %b) {
 entry:
@@ -9,6 +11,8 @@
 ; CHECK-THUMB: __divsi3
 ; CHECK-THUMBV7M: f1
 ; CHECK-THUMBV7M: sdiv
+; CHECK-SWIFT-T2: f1
+; CHECK-SWIFT-T2: sdiv
         %tmp1 = sdiv i32 %a, %b         ; <i32> [#uses=1]
         ret i32 %tmp1
 }
@@ -19,6 +23,8 @@
 ; CHECK-THUMB: __udivsi3
 ; CHECK-THUMBV7M: f2
 ; CHECK-THUMBV7M: udiv
+; CHECK-SWIFT-T2: f2
+; CHECK-SWIFT-T2: udiv
         %tmp1 = udiv i32 %a, %b         ; <i32> [#uses=1]
         ret i32 %tmp1
 }
@@ -29,6 +35,8 @@
 ; CHECK-THUMB: __modsi3
 ; CHECK-THUMBV7M: f3
 ; CHECK-THUMBV7M: sdiv
+; CHECK-SWIFT-T2: f3
+; CHECK-SWIFT-T2: sdiv
         %tmp1 = srem i32 %a, %b         ; <i32> [#uses=1]
         ret i32 %tmp1
 }
@@ -39,6 +47,8 @@
 ; CHECK-THUMB: __umodsi3
 ; CHECK-THUMBV7M: f4
 ; CHECK-THUMBV7M: udiv
+; CHECK-SWIFT-T2: f4
+; CHECK-SWIFT-T2: udiv
         %tmp1 = urem i32 %a, %b         ; <i32> [#uses=1]
         ret i32 %tmp1
 }

Modified: llvm/branches/AMDILBackend/test/CodeGen/Thumb2/machine-licm.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/Thumb2/machine-licm.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/Thumb2/machine-licm.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/Thumb2/machine-licm.ll Tue Jan 15 11:16:16 2013
@@ -95,7 +95,7 @@
 bb:                                               ; preds = %bb, %bb.nph
 ; CHECK: bb
 ; CHECK: eor.w
-; CHECK: eor.w {{(r[0-9])|(lr)}}, {{(r[0-9])|(lr)}}, [[REGISTER]]
+; CHECK: eorne.w {{(r[0-9])|(lr)}}, {{(r[0-9])|(lr)}}, [[REGISTER]]
 ; CHECK-NOT: eor
 ; CHECK: and
   %data_addr.013 = phi i8 [ %data, %bb.nph ], [ %8, %bb ] ; <i8> [#uses=2]

Modified: llvm/branches/AMDILBackend/test/CodeGen/Thumb2/thumb2-mla.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/Thumb2/thumb2-mla.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/Thumb2/thumb2-mla.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/Thumb2/thumb2-mla.ll Tue Jan 15 11:16:16 2013
@@ -1,4 +1,5 @@
 ; RUN: llc < %s -march=thumb -mattr=+thumb2 | FileCheck %s
+; RUN: llc < %s -march=thumb -mattr=+thumb2 -arm-use-mulops=false | FileCheck %s -check-prefix=NO_MULOPS
 
 define i32 @f1(i32 %a, i32 %b, i32 %c) {
     %tmp1 = mul i32 %a, %b
@@ -7,6 +8,9 @@
 }
 ; CHECK: f1:
 ; CHECK: 	mla	r0, r0, r1, r2
+; NO_MULOPS: f1:
+; NO_MULOPS: muls r0, r1, r0
+; NO_MULOPS-NEXT: add r0, r2
 
 define i32 @f2(i32 %a, i32 %b, i32 %c) {
     %tmp1 = mul i32 %a, %b
@@ -15,3 +19,6 @@
 }
 ; CHECK: f2:
 ; CHECK: 	mla	r0, r0, r1, r2
+; NO_MULOPS: f2:
+; NO_MULOPS: muls r0, r1, r0
+; NO_MULOPS-NEXT: add r0, r2

Modified: llvm/branches/AMDILBackend/test/CodeGen/Thumb2/thumb2-select_xform.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/Thumb2/thumb2-select_xform.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/Thumb2/thumb2-select_xform.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/Thumb2/thumb2-select_xform.ll Tue Jan 15 11:16:16 2013
@@ -4,9 +4,9 @@
 ; CHECK: t1
 ; CHECK: mvn r0, #-2147483648
 ; CHECK: cmp r2, #10
-; CHECK: add r0, r1
-; CHECK: it  gt
-; CHECK: movgt r0, r1
+; CHECK: it  le
+; CHECK: addle r1, r0
+; CHECK: mov r0, r1
         %tmp1 = icmp sgt i32 %c, 10
         %tmp2 = select i1 %tmp1, i32 0, i32 2147483647
         %tmp3 = add i32 %tmp2, %b
@@ -15,10 +15,10 @@
 
 define i32 @t2(i32 %a, i32 %b, i32 %c) nounwind {
 ; CHECK: t2
-; CHECK: add.w r0, r1, #-2147483648
 ; CHECK: cmp r2, #10
-; CHECK: it  gt
-; CHECK: movgt r0, r1
+; CHECK: it  le
+; CHECK: addle.w r1, r1, #-2147483648
+; CHECK: mov r0, r1
 
         %tmp1 = icmp sgt i32 %c, 10
         %tmp2 = select i1 %tmp1, i32 0, i32 2147483648
@@ -28,10 +28,10 @@
 
 define i32 @t3(i32 %a, i32 %b, i32 %c, i32 %d) nounwind {
 ; CHECK: t3
-; CHECK: sub.w r0, r1, #10
 ; CHECK: cmp r2, #10
-; CHECK: it  gt
-; CHECK: movgt r0, r1
+; CHECK: it  le
+; CHECK: suble r1, #10
+; CHECK: mov r0, r1
         %tmp1 = icmp sgt i32 %c, 10
         %tmp2 = select i1 %tmp1, i32 0, i32 10
         %tmp3 = sub i32 %b, %tmp2

Modified: llvm/branches/AMDILBackend/test/CodeGen/Thumb2/thumb2-smla.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/Thumb2/thumb2-smla.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/Thumb2/thumb2-smla.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/Thumb2/thumb2-smla.ll Tue Jan 15 11:16:16 2013
@@ -1,8 +1,12 @@
 ; RUN: llc < %s -march=thumb -mattr=+thumb2,+t2xtpk,+t2dsp | FileCheck %s
+; RUN: llc < %s -march=thumb -mattr=+thumb2,+t2xtpk,+t2dsp -arm-use-mulops=false | FileCheck %s -check-prefix=NO_MULOPS
 
 define i32 @f3(i32 %a, i16 %x, i32 %y) {
 ; CHECK: f3
 ; CHECK: smlabt r0, r1, r2, r0
+; NO_MULOPS: f3
+; NO_MULOPS: smultb r1, r2, r1
+; NO_MULOPS-NEXT: add r0, r1
         %tmp = sext i16 %x to i32               ; <i32> [#uses=1]
         %tmp2 = ashr i32 %y, 16         ; <i32> [#uses=1]
         %tmp3 = mul i32 %tmp2, %tmp             ; <i32> [#uses=1]

Modified: llvm/branches/AMDILBackend/test/CodeGen/Thumb2/thumb2-uxtb.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/Thumb2/thumb2-uxtb.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/Thumb2/thumb2-uxtb.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/Thumb2/thumb2-uxtb.ll Tue Jan 15 11:16:16 2013
@@ -128,9 +128,9 @@
 
 ; ARMv7M: test10
 ; ARMv7M: mov.w r1, #16253176
-; ARMv7M: mov.w r2, #458759
 ; ARMv7M: and.w r0, r1, r0, lsr #7
-; ARMv7M: and.w r1, r2, r0, lsr #5
+; ARMv7M: mov.w r1, #458759
+; ARMv7M: and.w r1, r1, r0, lsr #5
 ; ARMv7M: orrs r0, r1
 	%tmp1 = lshr i32 %p0, 7		; <i32> [#uses=1]
 	%tmp2 = and i32 %tmp1, 16253176		; <i32> [#uses=2]

Modified: llvm/branches/AMDILBackend/test/CodeGen/X86/2006-11-12-CSRetCC.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/X86/2006-11-12-CSRetCC.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/X86/2006-11-12-CSRetCC.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/X86/2006-11-12-CSRetCC.ll Tue Jan 15 11:16:16 2013
@@ -52,8 +52,8 @@
         %tmp21 = load double* %tmp20            ; <double> [#uses=1]
         %tmp.upgrd.6 = getelementptr [9 x i8]* @str, i32 0, i64 0               ; <i8*> [#uses=1]
         %tmp.upgrd.7 = call i32 (i8*, ...)* @printf( i8* %tmp.upgrd.6, double %tmp21, double %tmp19 )           ; <i32> [#uses=0]
-        br label %return
-return:         ; preds = %entry
+        br label %finish
+finish:
         %retval.upgrd.8 = load i32* %retval             ; <i32> [#uses=1]
         ret i32 %retval.upgrd.8
 }

Modified: llvm/branches/AMDILBackend/test/CodeGen/X86/2006-11-17-IllegalMove.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/X86/2006-11-17-IllegalMove.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/X86/2006-11-17-IllegalMove.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/X86/2006-11-17-IllegalMove.ll Tue Jan 15 11:16:16 2013
@@ -1,5 +1,5 @@
 ; RUN: llc < %s -march=x86-64 > %t
-; RUN: grep movb %t | count 2
+; RUN: grep movb %t | count 1
 ; RUN: grep "movzb[wl]" %t
 
 

Modified: llvm/branches/AMDILBackend/test/CodeGen/X86/2008-01-08-SchedulerCrash.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/X86/2008-01-08-SchedulerCrash.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/X86/2008-01-08-SchedulerCrash.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/X86/2008-01-08-SchedulerCrash.ll Tue Jan 15 11:16:16 2013
@@ -10,10 +10,10 @@
 
 	%struct.indexentry = type { i32, i8*, i8*, i8*, i8*, i8* }
 
-define i32 @_bfd_stab_section_find_nearest_line(i32 %offset) nounwind  {
+define i32 @_bfd_stab_section_find_nearest_line(i32 %offset, i1 %cond) nounwind  {
 entry:
 	%tmp910 = add i32 0, %offset		; <i32> [#uses=1]
-	br i1 true, label %bb951, label %bb917
+	br i1 %cond, label %bb951, label %bb917
 
 bb917:		; preds = %entry
 	ret i32 0
@@ -21,7 +21,7 @@
 bb951:		; preds = %bb986, %entry
 	%tmp955 = sdiv i32 0, 2		; <i32> [#uses=3]
 	%tmp961 = getelementptr %struct.indexentry* null, i32 %tmp955, i32 0		; <i32*> [#uses=1]
-	br i1 true, label %bb986, label %bb967
+	br i1 %cond, label %bb986, label %bb967
 
 bb967:		; preds = %bb951
 	ret i32 0

Modified: llvm/branches/AMDILBackend/test/CodeGen/X86/2010-01-08-Atomic64Bug.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/X86/2010-01-08-Atomic64Bug.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/X86/2010-01-08-Atomic64Bug.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/X86/2010-01-08-Atomic64Bug.ll Tue Jan 15 11:16:16 2013
@@ -1,4 +1,4 @@
-; RUN: llc < %s -mtriple=i386-apple-darwin | FileCheck %s
+; RUN: llc < %s -mtriple=i386-apple-darwin -mcpu=corei7 | FileCheck %s
 ; rdar://r7512579
 
 ; PHI defs in the atomic loop should be used by the add / adc
@@ -7,17 +7,16 @@
 define void @t(i64* nocapture %p) nounwind ssp {
 entry:
 ; CHECK: t:
-; CHECK: movl $1
-; CHECK: movl (%ebp), %eax
-; CHECK: movl 4(%ebp), %edx
+; CHECK: movl ([[REG:%[a-z]+]]), %eax
+; CHECK: movl 4([[REG]]), %edx
 ; CHECK: LBB0_1:
-; CHECK-NOT: movl $1
-; CHECK-NOT: movl $0
-; CHECK: addl
-; CHECK: adcl
+; CHECK: movl %eax, %ebx
+; CHECK: addl {{%[a-z]+}}, %ebx
+; CHECK: movl %edx, %ecx
+; CHECK: adcl {{%[a-z]+}}, %ecx
 ; CHECK: lock
-; CHECK: cmpxchg8b
-; CHECK: jne
+; CHECK-NEXT: cmpxchg8b ([[REG]])
+; CHECK-NEXT: jne
   %0 = atomicrmw add i64* %p, i64 1 seq_cst
   ret void
 }

Modified: llvm/branches/AMDILBackend/test/CodeGen/X86/2011-08-29-InitOrder.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/X86/2011-08-29-InitOrder.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/X86/2011-08-29-InitOrder.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/X86/2011-08-29-InitOrder.ll Tue Jan 15 11:16:16 2013
@@ -3,7 +3,7 @@
 ; PR5329
 
 @llvm.global_ctors = appending global [3 x { i32, void ()* }] [{ i32, void ()* } { i32 2000, void ()* @construct_2 }, { i32, void ()* } { i32 3000, void ()* @construct_3 }, { i32, void ()* } { i32 1000, void ()* @construct_1 }]
-; CHECK-DEFAULT  .section        .ctors.64535,"aw", at progbits
+; CHECK-DEFAULT: .section        .ctors.64535,"aw", at progbits
 ; CHECK-DEFAULT: .long construct_1
 ; CHECK-DEFAULT: .section        .ctors.63535,"aw", at progbits
 ; CHECK-DEFAULT: .long construct_2

Modified: llvm/branches/AMDILBackend/test/CodeGen/X86/2012-01-18-vbitcast.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/X86/2012-01-18-vbitcast.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/X86/2012-01-18-vbitcast.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/X86/2012-01-18-vbitcast.ll Tue Jan 15 11:16:16 2013
@@ -2,8 +2,8 @@
 
 ;CHECK: vcast
 define <2 x i32> @vcast(<2 x float> %a, <2 x float> %b) {
-;CHECK: pshufd
-;CHECK: pshufd
+;CHECK: pmovzxdq
+;CHECK: pmovzxdq
   %af = bitcast <2 x float> %a to <2 x i32>
   %bf = bitcast <2 x float> %b to <2 x i32>
   %x = sub <2 x i32> %af, %bf

Modified: llvm/branches/AMDILBackend/test/CodeGen/X86/2012-03-15-build_vector_wl.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/X86/2012-03-15-build_vector_wl.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/X86/2012-03-15-build_vector_wl.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/X86/2012-03-15-build_vector_wl.ll Tue Jan 15 11:16:16 2013
@@ -4,7 +4,7 @@
 define <4 x i8> @build_vector_again(<16 x i8> %in) nounwind readnone {
 entry:
   %out = shufflevector <16 x i8> %in, <16 x i8> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
-; CHECK: shufb
+; CHECK: pmovzxbd
   ret <4 x i8> %out
 ; CHECK: ret
 }

Modified: llvm/branches/AMDILBackend/test/CodeGen/X86/2012-04-26-sdglue.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/X86/2012-04-26-sdglue.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/X86/2012-04-26-sdglue.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/X86/2012-04-26-sdglue.ll Tue Jan 15 11:16:16 2013
@@ -5,7 +5,7 @@
 ; It's hard to test for the ISEL condition because CodeGen optimizes
 ; away the bugpointed code. Just ensure the basics are still there.
 ;CHECK: func:
-;CHECK: vpxor
+;CHECK: vxorps
 ;CHECK: vinsertf128
 ;CHECK: vpshufd
 ;CHECK: vpshufd

Modified: llvm/branches/AMDILBackend/test/CodeGen/X86/2012-07-10-extload64.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/X86/2012-07-10-extload64.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/X86/2012-07-10-extload64.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/X86/2012-07-10-extload64.ll Tue Jan 15 11:16:16 2013
@@ -3,7 +3,7 @@
 ; CHECK: load_store
 define void @load_store(<4 x i16>* %in) {
 entry:
-; CHECK: movsd
+; CHECK: pmovzxwd
   %A27 = load <4 x i16>* %in, align 4
   %A28 = add <4 x i16> %A27, %A27
 ; CHECK: movlpd
@@ -27,6 +27,6 @@
 BB:
   %t = load <2 x i32>* %ptr
   ret <2 x i32> %t
-;CHECK: movsd
+;CHECK: pmovzxdq
 ;CHECK: ret
 }

Modified: llvm/branches/AMDILBackend/test/CodeGen/X86/add-of-carry.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/X86/add-of-carry.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/X86/add-of-carry.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/X86/add-of-carry.ll Tue Jan 15 11:16:16 2013
@@ -30,4 +30,17 @@
   ret i32 %z.0
 }
 
+; <rdar://problem/12579915>
+define i32 @test3(i32 %x, i32 %y, i32 %res) nounwind uwtable readnone ssp {
+entry:
+  %cmp = icmp ugt i32 %x, %y
+  %dec = sext i1 %cmp to i32
+  %dec.res = add nsw i32 %dec, %res
+  ret i32 %dec.res
+; CHECK: test3:
+; CHECK: cmpl
+; CHECK: sbbl
+; CHECK: ret
+}
+
 declare { i32, i1 } @llvm.uadd.with.overflow.i32(i32, i32) nounwind readnone

Modified: llvm/branches/AMDILBackend/test/CodeGen/X86/alignment-2.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/X86/alignment-2.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/X86/alignment-2.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/X86/alignment-2.ll Tue Jan 15 11:16:16 2013
@@ -18,7 +18,9 @@
 define signext i8 @do_lo_list() nounwind optsize ssp {
 bb:
 ; CHECK:     do_lo_list
-; CHECK-NOT: movaps
+; Make sure we do not use movaps for the global variable.
+; It is okay to use movaps for writing the local variable on stack.
+; CHECK-NOT: movaps {{[0-9]*}}(%{{[a-z]*}}), {{%xmm[0-9]}}
   %myopt = alloca %struct.printQueryOpt, align 4
   %tmp = bitcast %struct.printQueryOpt* %myopt to i8*
   call void @llvm.memcpy.p0i8.p0i8.i32(i8* %tmp, i8* bitcast (%struct.printQueryOpt* getelementptr inbounds (%struct._psqlSettings* @pset, i32 0, i32 4) to i8*), i32 76, i32 4, i1 false)

Modified: llvm/branches/AMDILBackend/test/CodeGen/X86/atomic_add.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/X86/atomic_add.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/X86/atomic_add.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/X86/atomic_add.ll Tue Jan 15 11:16:16 2013
@@ -178,7 +178,8 @@
 define void @sub2(i16* nocapture %p, i32 %v) nounwind ssp {
 entry:
 ; CHECK: sub2:
-; CHECK: negl
+; CHECK-NOT: negl
+; CHECK: subw
 	%0 = trunc i32 %v to i16		; <i16> [#uses=1]
   %1 = atomicrmw sub i16* %p, i16 %0 monotonic
   ret void

Modified: llvm/branches/AMDILBackend/test/CodeGen/X86/atomic_op.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/X86/atomic_op.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/X86/atomic_op.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/X86/atomic_op.ll Tue Jan 15 11:16:16 2013
@@ -1,4 +1,4 @@
-; RUN: llc < %s -mcpu=generic -march=x86 | FileCheck %s
+; RUN: llc < %s -mcpu=generic -march=x86 -mattr=+cmov | FileCheck %s
 
 target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128"
 
@@ -107,13 +107,12 @@
         ; CHECK: cmpxchgl
   %17 = cmpxchg i32* %val2, i32 1976, i32 1 monotonic
 	store i32 %17, i32* %old
+        ; CHECK: movl  [[R17atomic:.*]], %eax
         ; CHECK: movl	$1401, %[[R17mask:[a-z]*]]
-        ; CHECK: movl	[[R17atomic:.*]], %eax
-        ; CHECK: movl	%eax, %[[R17newval:[a-z]*]]
-        ; CHECK: andl	%[[R17mask]], %[[R17newval]]
-        ; CHECK: notl	%[[R17newval]]
+        ; CHECK: andl	%eax, %[[R17mask]]
+        ; CHECK: notl	%[[R17mask]]
         ; CHECK: lock
-        ; CHECK: cmpxchgl	%[[R17newval]], [[R17atomic]]
+        ; CHECK: cmpxchgl	%[[R17mask]], [[R17atomic]]
         ; CHECK: jne
         ; CHECK: movl	%eax,
   %18 = atomicrmw nand i32* %val2, i32 1401 monotonic

Modified: llvm/branches/AMDILBackend/test/CodeGen/X86/avx-basic.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/X86/avx-basic.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/X86/avx-basic.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/X86/avx-basic.ll Tue Jan 15 11:16:16 2013
@@ -109,8 +109,8 @@
 ; rdar://10566486
 ; CHECK: fneg
 ; CHECK: vxorps
-define <16 x float> @fneg(<16 x float> addrspace(1)* nocapture %out) nounwind {
-  %1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, <float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00>
+define <16 x float> @fneg(<16 x float> %a) nounwind {
+  %1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a
   ret <16 x float> %1
 }
 

Modified: llvm/branches/AMDILBackend/test/CodeGen/X86/avx-intrinsics-x86.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/X86/avx-intrinsics-x86.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/X86/avx-intrinsics-x86.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/X86/avx-intrinsics-x86.ll Tue Jan 15 11:16:16 2013
@@ -1140,9 +1140,9 @@
 
 
 define i32 @test_x86_sse42_pcmpestri128(<16 x i8> %a0, <16 x i8> %a2) {
-  ; CHECK: movl
-  ; CHECK: movl
-  ; CHECK: vpcmpestri
+  ; CHECK: movl $7
+  ; CHECK: movl $7
+  ; CHECK: vpcmpestri $7
   ; CHECK: movl
   %res = call i32 @llvm.x86.sse42.pcmpestri128(<16 x i8> %a0, i32 7, <16 x i8> %a2, i32 7, i8 7) ; <i32> [#uses=1]
   ret i32 %res
@@ -1150,6 +1150,18 @@
 declare i32 @llvm.x86.sse42.pcmpestri128(<16 x i8>, i32, <16 x i8>, i32, i8) nounwind readnone
 
 
+define i32 @test_x86_sse42_pcmpestri128_load(<16 x i8>* %a0, <16 x i8>* %a2) {
+  ; CHECK: movl $7
+  ; CHECK: movl $7
+  ; CHECK: vpcmpestri $7, (
+  ; CHECK: movl
+  %1 = load <16 x i8>* %a0
+  %2 = load <16 x i8>* %a2
+  %res = call i32 @llvm.x86.sse42.pcmpestri128(<16 x i8> %1, i32 7, <16 x i8> %2, i32 7, i8 7) ; <i32> [#uses=1]
+  ret i32 %res
+}
+
+
 define i32 @test_x86_sse42_pcmpestria128(<16 x i8> %a0, <16 x i8> %a2) {
   ; CHECK: movl
   ; CHECK: movl
@@ -1216,8 +1228,19 @@
 declare <16 x i8> @llvm.x86.sse42.pcmpestrm128(<16 x i8>, i32, <16 x i8>, i32, i8) nounwind readnone
 
 
+define <16 x i8> @test_x86_sse42_pcmpestrm128_load(<16 x i8> %a0, <16 x i8>* %a2) {
+  ; CHECK: movl $7
+  ; CHECK: movl $7
+  ; CHECK: vpcmpestrm $7,
+  ; CHECK-NOT: vmov
+  %1 = load <16 x i8>* %a2
+  %res = call <16 x i8> @llvm.x86.sse42.pcmpestrm128(<16 x i8> %a0, i32 7, <16 x i8> %1, i32 7, i8 7) ; <<16 x i8>> [#uses=1]
+  ret <16 x i8> %res
+}
+
+
 define i32 @test_x86_sse42_pcmpistri128(<16 x i8> %a0, <16 x i8> %a1) {
-  ; CHECK: vpcmpistri
+  ; CHECK: vpcmpistri $7
   ; CHECK: movl
   %res = call i32 @llvm.x86.sse42.pcmpistri128(<16 x i8> %a0, <16 x i8> %a1, i8 7) ; <i32> [#uses=1]
   ret i32 %res
@@ -1225,6 +1248,16 @@
 declare i32 @llvm.x86.sse42.pcmpistri128(<16 x i8>, <16 x i8>, i8) nounwind readnone
 
 
+define i32 @test_x86_sse42_pcmpistri128_load(<16 x i8>* %a0, <16 x i8>* %a1) {
+  ; CHECK: vpcmpistri $7, (
+  ; CHECK: movl
+  %1 = load <16 x i8>* %a0
+  %2 = load <16 x i8>* %a1
+  %res = call i32 @llvm.x86.sse42.pcmpistri128(<16 x i8> %1, <16 x i8> %2, i8 7) ; <i32> [#uses=1]
+  ret i32 %res
+}
+
+
 define i32 @test_x86_sse42_pcmpistria128(<16 x i8> %a0, <16 x i8> %a1) {
   ; CHECK: vpcmpistri
   ; CHECK: seta
@@ -1271,7 +1304,7 @@
 
 
 define <16 x i8> @test_x86_sse42_pcmpistrm128(<16 x i8> %a0, <16 x i8> %a1) {
-  ; CHECK: vpcmpistrm
+  ; CHECK: vpcmpistrm $7
   ; CHECK-NOT: vmov
   %res = call <16 x i8> @llvm.x86.sse42.pcmpistrm128(<16 x i8> %a0, <16 x i8> %a1, i8 7) ; <<16 x i8>> [#uses=1]
   ret <16 x i8> %res
@@ -1279,6 +1312,15 @@
 declare <16 x i8> @llvm.x86.sse42.pcmpistrm128(<16 x i8>, <16 x i8>, i8) nounwind readnone
 
 
+define <16 x i8> @test_x86_sse42_pcmpistrm128_load(<16 x i8> %a0, <16 x i8>* %a1) {
+  ; CHECK: vpcmpistrm $7, (
+  ; CHECK-NOT: vmov
+  %1 = load <16 x i8>* %a1
+  %res = call <16 x i8> @llvm.x86.sse42.pcmpistrm128(<16 x i8> %a0, <16 x i8> %1, i8 7) ; <<16 x i8>> [#uses=1]
+  ret <16 x i8> %res
+}
+
+
 define <4 x float> @test_x86_sse_add_ss(<4 x float> %a0, <4 x float> %a1) {
   ; CHECK: vaddss
   %res = call <4 x float> @llvm.x86.sse.add.ss(<4 x float> %a0, <4 x float> %a1) ; <<4 x float>> [#uses=1]

Modified: llvm/branches/AMDILBackend/test/CodeGen/X86/avx-shuffle.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/X86/avx-shuffle.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/X86/avx-shuffle.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/X86/avx-shuffle.ll Tue Jan 15 11:16:16 2013
@@ -229,9 +229,8 @@
 }
 
 ; CHECK: test18
-; CHECK: vshufps
-; CHECK: vshufps
-; CHECK: vunpcklps
+; CHECK: vmovshdup
+; CHECK: vblendps
 ; CHECK: ret
 define <8 x float> @test18(<8 x float> %A, <8 x float>%B) nounwind {
   %S = shufflevector <8 x float> %A, <8 x float> %B, <8 x i32> <i32 1, i32 9, i32 3, i32 11, i32 5, i32 13, i32 7, i32 15>
@@ -239,9 +238,8 @@
 }
 
 ; CHECK: test19
-; CHECK: vshufps
-; CHECK: vshufps
-; CHECK: vunpcklps
+; CHECK: vmovsldup
+; CHECK: vblendps
 ; CHECK: ret
 define <8 x float> @test19(<8 x float> %A, <8 x float>%B) nounwind {
   %S = shufflevector <8 x float> %A, <8 x float> %B, <8 x i32> <i32 0, i32 8, i32 2, i32 10, i32 4, i32 12, i32 6, i32 14>

Modified: llvm/branches/AMDILBackend/test/CodeGen/X86/avx-vextractf128.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/X86/avx-vextractf128.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/X86/avx-vextractf128.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/X86/avx-vextractf128.ll Tue Jan 15 11:16:16 2013
@@ -19,12 +19,12 @@
 }
 
 ; CHECK: @t0
-; CHECK-NOT: vextractf128 $0, %ymm0, %xmm0
+; CHECK-NOT: vextractf128 $1, %ymm0, %xmm0
 ; CHECK-NOT: vmovaps %xmm0, (%rdi)
-; CHECK: vextractf128 $0, %ymm0, (%rdi)
+; CHECK: vextractf128 $1, %ymm0, (%rdi)
 define void @t0(float* nocapture %addr, <8 x float> %a) nounwind uwtable ssp {
 entry:
-  %0 = tail call <4 x float> @llvm.x86.avx.vextractf128.ps.256(<8 x float> %a, i8 0)
+  %0 = tail call <4 x float> @llvm.x86.avx.vextractf128.ps.256(<8 x float> %a, i8 1)
   %1 = bitcast float* %addr to <4 x float>*
   store <4 x float> %0, <4 x float>* %1, align 16
   ret void
@@ -32,27 +32,13 @@
 
 declare <4 x float> @llvm.x86.avx.vextractf128.ps.256(<8 x float>, i8) nounwind readnone
 
-; CHECK: @t1
-; CHECK-NOT: vextractf128 $0, %ymm0, %xmm0
-; CHECK-NOT: vmovups %xmm0, (%rdi)
-; CHECK: vextractf128 $0, %ymm0, (%rdi)
-define void @t1(float* %addr, <8 x float> %a) nounwind uwtable ssp {
-entry:
-  %0 = tail call <4 x float> @llvm.x86.avx.vextractf128.ps.256(<8 x float> %a, i8 0)
-  %1 = bitcast float* %addr to i8*
-  tail call void @llvm.x86.sse.storeu.ps(i8* %1, <4 x float> %0)
-  ret void
-}
-
-declare void @llvm.x86.sse.storeu.ps(i8*, <4 x float>) nounwind
-
 ; CHECK: @t2
-; CHECK-NOT: vextractf128 $0, %ymm0, %xmm0
+; CHECK-NOT: vextractf128 $1, %ymm0, %xmm0
 ; CHECK-NOT: vmovaps %xmm0, (%rdi)
-; CHECK: vextractf128 $0, %ymm0, (%rdi)
+; CHECK: vextractf128 $1, %ymm0, (%rdi)
 define void @t2(double* nocapture %addr, <4 x double> %a) nounwind uwtable ssp {
 entry:
-  %0 = tail call <2 x double> @llvm.x86.avx.vextractf128.pd.256(<4 x double> %a, i8 0)
+  %0 = tail call <2 x double> @llvm.x86.avx.vextractf128.pd.256(<4 x double> %a, i8 1)
   %1 = bitcast double* %addr to <2 x double>*
   store <2 x double> %0, <2 x double>* %1, align 16
   ret void
@@ -60,28 +46,14 @@
 
 declare <2 x double> @llvm.x86.avx.vextractf128.pd.256(<4 x double>, i8) nounwind readnone
 
-; CHECK: @t3
-; CHECK-NOT: vextractf128 $0, %ymm0, %xmm0
-; CHECK-NOT: vmovups %xmm0, (%rdi)
-; CHECK: vextractf128 $0, %ymm0, (%rdi)
-define void @t3(double* %addr, <4 x double> %a) nounwind uwtable ssp {
-entry:
-  %0 = tail call <2 x double> @llvm.x86.avx.vextractf128.pd.256(<4 x double> %a, i8 0)
-  %1 = bitcast double* %addr to i8*
-  tail call void @llvm.x86.sse2.storeu.pd(i8* %1, <2 x double> %0)
-  ret void
-}
-
-declare void @llvm.x86.sse2.storeu.pd(i8*, <2 x double>) nounwind
-
 ; CHECK: @t4
-; CHECK-NOT: vextractf128 $0, %ymm0, %xmm0
+; CHECK-NOT: vextractf128 $1, %ymm0, %xmm0
 ; CHECK-NOT: vmovaps %xmm0, (%rdi)
-; CHECK: vextractf128 $0, %ymm0, (%rdi)
+; CHECK: vextractf128 $1, %ymm0, (%rdi)
 define void @t4(<2 x i64>* nocapture %addr, <4 x i64> %a) nounwind uwtable ssp {
 entry:
   %0 = bitcast <4 x i64> %a to <8 x i32>
-  %1 = tail call <4 x i32> @llvm.x86.avx.vextractf128.si.256(<8 x i32> %0, i8 0)
+  %1 = tail call <4 x i32> @llvm.x86.avx.vextractf128.si.256(<8 x i32> %0, i8 1)
   %2 = bitcast <4 x i32> %1 to <2 x i64>
   store <2 x i64> %2, <2 x i64>* %addr, align 16
   ret void
@@ -90,17 +62,43 @@
 declare <4 x i32> @llvm.x86.avx.vextractf128.si.256(<8 x i32>, i8) nounwind readnone
 
 ; CHECK: @t5
-; CHECK-NOT: vextractf128 $0, %ymm0, %xmm0
-; CHECK-NOT: vmovdqu %xmm0, (%rdi)
-; CHECK: vextractf128 $0, %ymm0, (%rdi)
-define void @t5(<2 x i64>* %addr, <4 x i64> %a) nounwind uwtable ssp {
+; CHECK: vmovaps %xmm0, (%rdi)
+define void @t5(float* nocapture %addr, <8 x float> %a) nounwind uwtable ssp {
+entry:
+  %0 = tail call <4 x float> @llvm.x86.avx.vextractf128.ps.256(<8 x float> %a, i8 0)
+  %1 = bitcast float* %addr to <4 x float>*
+  store <4 x float> %0, <4 x float>* %1, align 16
+  ret void
+}
+
+; CHECK: @t6
+; CHECK: vmovaps %xmm0, (%rdi)
+define void @t6(double* nocapture %addr, <4 x double> %a) nounwind uwtable ssp {
+entry:
+  %0 = tail call <2 x double> @llvm.x86.avx.vextractf128.pd.256(<4 x double> %a, i8 0)
+  %1 = bitcast double* %addr to <2 x double>*
+  store <2 x double> %0, <2 x double>* %1, align 16
+  ret void
+}
+
+; CHECK: @t7
+; CHECK: vmovaps %xmm0, (%rdi)
+define void @t7(<2 x i64>* nocapture %addr, <4 x i64> %a) nounwind uwtable ssp {
 entry:
   %0 = bitcast <4 x i64> %a to <8 x i32>
   %1 = tail call <4 x i32> @llvm.x86.avx.vextractf128.si.256(<8 x i32> %0, i8 0)
-  %2 = bitcast <2 x i64>* %addr to i8*
-  %3 = bitcast <4 x i32> %1 to <16 x i8>
-  tail call void @llvm.x86.sse2.storeu.dq(i8* %2, <16 x i8> %3)
+  %2 = bitcast <4 x i32> %1 to <2 x i64>
+  store <2 x i64> %2, <2 x i64>* %addr, align 16
   ret void
 }
 
-declare void @llvm.x86.sse2.storeu.dq(i8*, <16 x i8>) nounwind
+; CHECK: @t8
+; CHECK: vmovups %xmm0, (%rdi)
+define void @t8(<2 x i64>* nocapture %addr, <4 x i64> %a) nounwind uwtable ssp {
+entry:
+  %0 = bitcast <4 x i64> %a to <8 x i32>
+  %1 = tail call <4 x i32> @llvm.x86.avx.vextractf128.si.256(<8 x i32> %0, i8 0)
+  %2 = bitcast <4 x i32> %1 to <2 x i64>
+  store <2 x i64> %2, <2 x i64>* %addr, align 1
+  ret void
+}

Modified: llvm/branches/AMDILBackend/test/CodeGen/X86/avx2-shuffle.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/X86/avx2-shuffle.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/X86/avx2-shuffle.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/X86/avx2-shuffle.ll Tue Jan 15 11:16:16 2013
@@ -26,3 +26,37 @@
   %shuffle.i = shufflevector <16 x i16> %src1, <16 x i16> %src1, <16 x i32> <i32 3, i32 undef, i32 1, i32 0, i32 4, i32 5, i32 6, i32 7, i32 11, i32 10, i32 9, i32 8, i32 12, i32 13, i32 14, i32 15>
   ret <16 x i16> %shuffle.i
 }
+
+; CHECK: vpshufb_test
+; CHECK: vpshufb {{.*\(%r.*}}, %ymm
+; CHECK: ret
+define <32 x i8> @vpshufb_test(<32 x i8> %a) nounwind {
+  %S = shufflevector <32 x i8> %a, <32 x i8> undef, <32 x i32> <i32 1, i32 9, i32 3, i32 11, i32 5, i32 13, i32 7, i32 15, 
+                                                                i32 1, i32 9, i32 3, i32 11, i32 5, i32 13, i32 7, i32 15,  
+                                                                i32 18, i32 19, i32 30, i32 16, i32 25, i32 23, i32 17, i32 25, 
+                                                                i32 20, i32 19, i32 31, i32 17, i32 23, i32 undef, i32 29, i32 18>
+  ret <32 x i8>%S
+}
+
+; CHECK: vpshufb1_test
+; CHECK: vpshufb {{.*\(%r.*}}, %ymm
+; CHECK: ret
+define <32 x i8> @vpshufb1_test(<32 x i8> %a) nounwind {
+  %S = shufflevector <32 x i8> %a, <32 x i8> zeroinitializer, <32 x i32> <i32 1, i32 9, i32 3, i32 11, i32 5, i32 13, i32 7, i32 15, 
+                                                                i32 1, i32 9, i32 36, i32 11, i32 5, i32 13, i32 7, i32 15,  
+                                                                i32 18, i32 49, i32 30, i32 16, i32 25, i32 23, i32 17, i32 25, 
+                                                                i32 20, i32 19, i32 31, i32 17, i32 23, i32 undef, i32 29, i32 18>
+  ret <32 x i8>%S
+}
+
+
+; CHECK: vpshufb2_test
+; CHECK: vpshufb {{.*\(%r.*}}, %ymm
+; CHECK: ret
+define <32 x i8> @vpshufb2_test(<32 x i8> %a) nounwind {
+  %S = shufflevector <32 x i8> zeroinitializer, <32 x i8> %a, <32 x i32> <i32 1, i32 9, i32 3, i32 11, i32 5, i32 13, i32 7, i32 15, 
+                                                                i32 1, i32 9, i32 36, i32 11, i32 5, i32 13, i32 7, i32 15,  
+                                                                i32 18, i32 49, i32 30, i32 16, i32 25, i32 23, i32 17, i32 25, 
+                                                                i32 20, i32 19, i32 31, i32 17, i32 23, i32 undef, i32 29, i32 18>
+  ret <32 x i8>%S
+}

Modified: llvm/branches/AMDILBackend/test/CodeGen/X86/crash.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/X86/crash.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/X86/crash.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/X86/crash.ll Tue Jan 15 11:16:16 2013
@@ -442,3 +442,150 @@
   ret void
 }
 declare void @_Z6PrintFz(...)
+
+ at a = external global i32, align 4
+ at fn1.g = private unnamed_addr constant [9 x i32*] [i32* null, i32* @a, i32* null, i32* null, i32* null, i32* null, i32* null, i32* null, i32* null], align 16
+ at e = external global i32, align 4
+
+define void @pr13943() nounwind uwtable ssp {
+entry:
+  %srcval = load i576* bitcast ([9 x i32*]* @fn1.g to i576*), align 16
+  br label %for.cond
+
+for.cond:                                         ; preds = %for.inc, %entry
+  %g.0 = phi i576 [ %srcval, %entry ], [ %ins, %for.inc ]
+  %0 = load i32* @e, align 4
+  %1 = lshr i576 %g.0, 64
+  %2 = trunc i576 %1 to i64
+  %3 = inttoptr i64 %2 to i32*
+  %cmp = icmp eq i32* undef, %3
+  %conv2 = zext i1 %cmp to i32
+  %and = and i32 %conv2, %0
+  tail call void (...)* @fn3(i32 %and) nounwind
+  %tobool = icmp eq i32 undef, 0
+  br i1 %tobool, label %for.inc, label %if.then
+
+if.then:                                          ; preds = %for.cond
+  ret void
+
+for.inc:                                          ; preds = %for.cond
+  %4 = shl i576 %1, 384
+  %mask = and i576 %g.0, -726838724295606890509921801691610055141362320587174446476410459910173841445449629921945328942266354949348255351381262292727973638307841
+  %5 = and i576 %4, 726838724295606890509921801691610055141362320587174446476410459910173841445449629921945328942266354949348255351381262292727973638307840
+  %ins = or i576 %5, %mask
+  br label %for.cond
+}
+
+declare void @fn3(...)
+
+; Check coalescing of IMPLICIT_DEF instructions:
+;
+; %vreg1 = IMPLICIT_DEF
+; %vreg2 = MOV32r0
+;
+; When coalescing %vreg1 and %vreg2, the IMPLICIT_DEF instruction should be
+; erased along with its value number.
+;
+define void @rdar12474033() nounwind ssp {
+bb:
+  br i1 undef, label %bb21, label %bb1
+
+bb1:                                              ; preds = %bb
+  switch i32 undef, label %bb10 [
+    i32 4, label %bb2
+    i32 1, label %bb9
+    i32 5, label %bb3
+    i32 6, label %bb3
+    i32 2, label %bb9
+  ]
+
+bb2:                                              ; preds = %bb1
+  unreachable
+
+bb3:                                              ; preds = %bb1, %bb1
+  br i1 undef, label %bb4, label %bb5
+
+bb4:                                              ; preds = %bb3
+  unreachable
+
+bb5:                                              ; preds = %bb3
+  %tmp = load <4 x float>* undef, align 1
+  %tmp6 = bitcast <4 x float> %tmp to i128
+  %tmp7 = load <4 x float>* undef, align 1
+  %tmp8 = bitcast <4 x float> %tmp7 to i128
+  br label %bb10
+
+bb9:                                              ; preds = %bb1, %bb1
+  unreachable
+
+bb10:                                             ; preds = %bb5, %bb1
+  %tmp11 = phi i128 [ undef, %bb1 ], [ %tmp6, %bb5 ]
+  %tmp12 = phi i128 [ 0, %bb1 ], [ %tmp8, %bb5 ]
+  switch i32 undef, label %bb21 [
+    i32 2, label %bb18
+    i32 3, label %bb13
+    i32 5, label %bb16
+    i32 6, label %bb17
+    i32 1, label %bb18
+  ]
+
+bb13:                                             ; preds = %bb10
+  br i1 undef, label %bb15, label %bb14
+
+bb14:                                             ; preds = %bb13
+  br label %bb21
+
+bb15:                                             ; preds = %bb13
+  unreachable
+
+bb16:                                             ; preds = %bb10
+  unreachable
+
+bb17:                                             ; preds = %bb10
+  unreachable
+
+bb18:                                             ; preds = %bb10, %bb10
+  %tmp19 = bitcast i128 %tmp11 to <4 x float>
+  %tmp20 = bitcast i128 %tmp12 to <4 x float>
+  br label %bb21
+
+bb21:                                             ; preds = %bb18, %bb14, %bb10, %bb
+  %tmp22 = phi <4 x float> [ undef, %bb ], [ undef, %bb10 ], [ undef, %bb14 ], [ %tmp20, %bb18 ]
+  %tmp23 = phi <4 x float> [ undef, %bb ], [ undef, %bb10 ], [ undef, %bb14 ], [ %tmp19, %bb18 ]
+  store <4 x float> %tmp23, <4 x float>* undef, align 16
+  store <4 x float> %tmp22, <4 x float>* undef, align 16
+  switch i32 undef, label %bb29 [
+    i32 5, label %bb27
+    i32 1, label %bb24
+    i32 2, label %bb25
+    i32 14, label %bb28
+    i32 4, label %bb26
+  ]
+
+bb24:                                             ; preds = %bb21
+  unreachable
+
+bb25:                                             ; preds = %bb21
+  br label %bb29
+
+bb26:                                             ; preds = %bb21
+  br label %bb29
+
+bb27:                                             ; preds = %bb21
+  unreachable
+
+bb28:                                             ; preds = %bb21
+  br label %bb29
+
+bb29:                                             ; preds = %bb28, %bb26, %bb25, %bb21
+  unreachable
+}
+
+define void @pr14194() nounwind uwtable {
+  %tmp = load i64* undef, align 16
+  %tmp1 = trunc i64 %tmp to i32
+  %tmp2 = lshr i64 %tmp, 32
+  %tmp3 = trunc i64 %tmp2 to i32
+  %tmp4 = call { i32, i32 } asm sideeffect "", "=&r,=&r,r,r,0,1,~{dirflag},~{fpsr},~{flags}"(i32 %tmp3, i32 undef, i32 %tmp3, i32 %tmp1) nounwind
+ ret void
+}

Modified: llvm/branches/AMDILBackend/test/CodeGen/X86/early-ifcvt.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/X86/early-ifcvt.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/X86/early-ifcvt.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/X86/early-ifcvt.ll Tue Jan 15 11:16:16 2013
@@ -1,4 +1,4 @@
-; RUN: llc < %s -enable-early-ifcvt -stress-early-ifcvt | FileCheck %s
+; RUN: llc < %s -x86-early-ifcvt -stress-early-ifcvt | FileCheck %s
 target triple = "x86_64-apple-macosx10.8.0"
 
 ; CHECK: mm2
@@ -37,3 +37,108 @@
   %sub = sub nsw i32 %max.1, %min.1
   ret i32 %sub
 }
+
+; CHECK: multipreds
+; Deal with alternative tail predecessors
+; CHECK-NOT: LBB
+; CHECK: cmov
+; CHECK-NOT: LBB
+; CHECK: cmov
+; CHECK-NOT: LBB
+; CHECK: fprintf
+
+define void @multipreds(i32 %sw) nounwind uwtable ssp {
+entry:
+  switch i32 %sw, label %if.then29 [
+    i32 0, label %if.then37
+    i32 127, label %if.end41
+  ]
+
+if.then29:
+  br label %if.end41
+
+if.then37:
+  br label %if.end41
+
+if.end41:
+  %exit_status.0 = phi i32 [ 2, %if.then29 ], [ 0, %if.then37 ], [ 66, %entry ]
+  call void (...)* @fprintf(i32 %exit_status.0) nounwind
+  unreachable
+}
+
+declare void @fprintf(...) nounwind
+
+; CHECK: BZ2_decompress
+; This test case contains irreducible control flow, so MachineLoopInfo doesn't
+; recognize the cycle in the CFG. This would confuse MachineTraceMetrics.
+define void @BZ2_decompress(i8* %s) nounwind ssp {
+entry:
+  switch i32 undef, label %sw.default [
+    i32 39, label %if.end.sw.bb2050_crit_edge
+    i32 36, label %sw.bb1788
+    i32 37, label %if.end.sw.bb1855_crit_edge
+    i32 40, label %sw.bb2409
+    i32 38, label %sw.bb1983
+    i32 44, label %if.end.sw.bb3058_crit_edge
+  ]
+
+if.end.sw.bb3058_crit_edge:                       ; preds = %entry
+  br label %save_state_and_return
+
+if.end.sw.bb1855_crit_edge:                       ; preds = %entry
+  br label %save_state_and_return
+
+if.end.sw.bb2050_crit_edge:                       ; preds = %entry
+  br label %sw.bb2050
+
+sw.bb1788:                                        ; preds = %entry
+  br label %save_state_and_return
+
+sw.bb1983:                                        ; preds = %entry
+  br i1 undef, label %save_state_and_return, label %if.then1990
+
+if.then1990:                                      ; preds = %sw.bb1983
+  br label %while.body2038
+
+while.body2038:                                   ; preds = %sw.bb2050, %if.then1990
+  %groupPos.8 = phi i32 [ 0, %if.then1990 ], [ %groupPos.9, %sw.bb2050 ]
+  br i1 undef, label %save_state_and_return, label %if.end2042
+
+if.end2042:                                       ; preds = %while.body2038
+  br i1 undef, label %if.end2048, label %while.end2104
+
+if.end2048:                                       ; preds = %if.end2042
+  %bsLive2054.pre = getelementptr inbounds i8* %s, i32 8
+  br label %sw.bb2050
+
+sw.bb2050:                                        ; preds = %if.end2048, %if.end.sw.bb2050_crit_edge
+  %groupPos.9 = phi i32 [ 0, %if.end.sw.bb2050_crit_edge ], [ %groupPos.8, %if.end2048 ]
+  %and2064 = and i32 undef, 1
+  br label %while.body2038
+
+while.end2104:                                    ; preds = %if.end2042
+  br i1 undef, label %save_state_and_return, label %if.end2117
+
+if.end2117:                                       ; preds = %while.end2104
+  br i1 undef, label %while.body2161.lr.ph, label %while.body2145.lr.ph
+
+while.body2145.lr.ph:                             ; preds = %if.end2117
+  br label %save_state_and_return
+
+while.body2161.lr.ph:                             ; preds = %if.end2117
+  br label %save_state_and_return
+
+sw.bb2409:                                        ; preds = %entry
+  br label %save_state_and_return
+
+sw.default:                                       ; preds = %entry
+  call void @BZ2_bz__AssertH__fail() nounwind
+  br label %save_state_and_return
+
+save_state_and_return:
+  %groupPos.14 = phi i32 [ 0, %sw.default ], [ %groupPos.8, %while.body2038 ], [ %groupPos.8, %while.end2104 ], [ 0, %if.end.sw.bb3058_crit_edge ], [ 0, %if.end.sw.bb1855_crit_edge ], [ %groupPos.8, %while.body2161.lr.ph ], [ %groupPos.8, %while.body2145.lr.ph ], [ 0, %sw.bb2409 ], [ 0, %sw.bb1788 ], [ 0, %sw.bb1983 ]
+  store i32 %groupPos.14, i32* undef, align 4
+  ret void
+}
+
+declare void @BZ2_bz__AssertH__fail()

Modified: llvm/branches/AMDILBackend/test/CodeGen/X86/fast-cc-callee-pops.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/X86/fast-cc-callee-pops.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/X86/fast-cc-callee-pops.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/X86/fast-cc-callee-pops.ll Tue Jan 15 11:16:16 2013
@@ -2,12 +2,12 @@
 
 ; Check that a fastcc function pops its stack variables before returning.
 
-define x86_fastcallcc void @func(i64 %X, i64 %Y, float %G, double %Z) nounwind {
+define x86_fastcallcc void @func(i64 inreg %X, i64 %Y, float %G, double %Z) nounwind {
         ret void
 ; CHECK: ret{{.*}}20
 }
 
-define x86_thiscallcc void @func2(i32 %X, i64 %Y, float %G, double %Z) nounwind {
+define x86_thiscallcc void @func2(i32 inreg %X, i64 %Y, float %G, double %Z) nounwind {
         ret void
 ; CHECK: ret{{.*}}20
 }

Modified: llvm/branches/AMDILBackend/test/CodeGen/X86/fast-cc-merge-stack-adj.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/X86/fast-cc-merge-stack-adj.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/X86/fast-cc-merge-stack-adj.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/X86/fast-cc-merge-stack-adj.ll Tue Jan 15 11:16:16 2013
@@ -3,7 +3,7 @@
 
 target triple = "i686-pc-linux-gnu"
 
-declare x86_fastcallcc void @func(i32*, i64)
+declare x86_fastcallcc void @func(i32*, i64 inreg)
 
 define x86_fastcallcc void @caller(i32, i64) {
         %X = alloca i32         ; <i32*> [#uses=1]

Modified: llvm/branches/AMDILBackend/test/CodeGen/X86/fast-cc-pass-in-regs.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/AMDILBackend/test/CodeGen/X86/fast-cc-pass-in-regs.ll?rev=172541&r1=172540&r2=172541&view=diff
==============================================================================
--- llvm/branches/AMDILBackend/test/CodeGen/X86/fast-cc-pass-in-regs.ll (original)
+++ llvm/branches/AMDILBackend/test/CodeGen/X86/fast-cc-pass-in-regs.ll Tue Jan 15 11:16:16 2013
@@ -1,7 +1,7 @@
 ; RUN: llc < %s -march=x86 -x86-asm-syntax=intel | FileCheck %s
 ; check that fastcc is pas