Fix of hang during Intel JIT profiling

[oota-llvm.git] / lib / Analysis / Loads.cpp

diff --git a/lib/Analysis/Loads.cpp b/lib/Analysis/Loads.cpp
index 2ea27fb62fcbb119b6a9cb75d28bb83e372ffe4b..873a27543dd62a09f9d88f12f80c4cd53a4a5f0d 100644 (file)
--- a/lib/Analysis/Loads.cpp
+++ b/lib/Analysis/Loads.cpp
@@ -17,6 +17,8 @@
 #include "llvm/GlobalAlias.h"
 #include "llvm/GlobalVariable.h"
 #include "llvm/IntrinsicInst.h"
+#include "llvm/LLVMContext.h"
+#include "llvm/Operator.h"
 using namespace llvm;
 
 /// AreEquivalentAddressValues - Test if A and B will obviously have the same
@@ -30,7 +32,7 @@ using namespace llvm;
 static bool AreEquivalentAddressValues(const Value *A, const Value *B) {
   // Test if the values are trivially equivalent.
   if (A == B) return true;
-  
+
   // Test if the values come from identical arithmetic instructions.
   // Use isIdenticalToWhenDefined instead of isIdenticalTo because
   // this function is only used when one address use dominates the
@@ -41,7 +43,7 @@ static bool AreEquivalentAddressValues(const Value *A, const Value *B) {
     if (const Instruction *BI = dyn_cast<Instruction>(B))
       if (cast<Instruction>(A)->isIdenticalToWhenDefined(BI))
         return true;
-  
+
   // Otherwise they may not be equivalent.
   return false;
 }
@@ -62,7 +64,7 @@ static Value *getUnderlyingObjectWithOffset(Value *V, const TargetData *TD,
         return V;
       SmallVector<Value*, 8> Indices(GEP->op_begin() + 1, GEP->op_end());
       ByteOffset += TD->getIndexedOffset(GEP->getPointerOperandType(),
-                                         &Indices[0], Indices.size());
+                                         Indices);
       V = GEP->getPointerOperand();
     } else if (Operator::getOpcode(V) == Instruction::BitCast) {
       V = cast<Operator>(V)->getOperand(0);
@@ -89,7 +91,7 @@ bool llvm::isSafeToLoadUnconditionally(Value *V, Instruction *ScanFrom,
   if (TD)
     Base = getUnderlyingObjectWithOffset(V, TD, ByteOffset);
 
-  const Type *BaseType = 0;
+  Type *BaseType = 0;
   unsigned BaseAlign = 0;
   if (const AllocaInst *AI = dyn_cast<AllocaInst>(Base)) {
     // An alloca is safe to load from as load as it is suitably aligned.
@@ -113,7 +115,7 @@ bool llvm::isSafeToLoadUnconditionally(Value *V, Instruction *ScanFrom,
         return true; // Loading directly from an alloca or global is OK.
 
       // Check if the load is within the bounds of the underlying object.
-      const PointerType *AddrTy = cast<PointerType>(V->getType());
+      PointerType *AddrTy = cast<PointerType>(V->getType());
       uint64_t LoadSize = TD->getTypeStoreSize(AddrTy->getElementType());
       if (ByteOffset + LoadSize <= TD->getTypeAllocSize(BaseType) &&
           (Align == 0 || (ByteOffset % Align) == 0))
@@ -159,16 +161,21 @@ bool llvm::isSafeToLoadUnconditionally(Value *V, Instruction *ScanFrom,
 /// MaxInstsToScan specifies the maximum instructions to scan in the block.  If
 /// it is set to 0, it will scan the whole block. You can also optionally
 /// specify an alias analysis implementation, which makes this more precise.
+///
+/// If TBAATag is non-null and a load or store is found, the TBAA tag from the
+/// load or store is recorded there.  If there is no TBAA tag or if no access
+/// is found, it is left unmodified.
 Value *llvm::FindAvailableLoadedValue(Value *Ptr, BasicBlock *ScanBB,
                                       BasicBlock::iterator &ScanFrom,
                                       unsigned MaxInstsToScan,
-                                      AliasAnalysis *AA) {
+                                      AliasAnalysis *AA,
+                                      MDNode **TBAATag) {
   if (MaxInstsToScan == 0) MaxInstsToScan = ~0U;
 
   // If we're using alias analysis to disambiguate get the size of *Ptr.
   uint64_t AccessSize = 0;
   if (AA) {
-    const Type *AccessTy = cast<PointerType>(Ptr->getType())->getElementType();
+    Type *AccessTy = cast<PointerType>(Ptr->getType())->getElementType();
     AccessSize = AA->getTypeStoreSize(AccessTy);
   }
   
@@ -187,14 +194,22 @@ Value *llvm::FindAvailableLoadedValue(Value *Ptr, BasicBlock *ScanBB,
     
     --ScanFrom;
     // If this is a load of Ptr, the loaded value is available.
+    // (This is true even if the load is volatile or atomic, although
+    // those cases are unlikely.)
     if (LoadInst *LI = dyn_cast<LoadInst>(Inst))
-      if (AreEquivalentAddressValues(LI->getOperand(0), Ptr))
+      if (AreEquivalentAddressValues(LI->getOperand(0), Ptr)) {
+        if (TBAATag) *TBAATag = LI->getMetadata(LLVMContext::MD_tbaa);
         return LI;
+      }
     
     if (StoreInst *SI = dyn_cast<StoreInst>(Inst)) {
       // If this is a store through Ptr, the value is available!
-      if (AreEquivalentAddressValues(SI->getOperand(1), Ptr))
+      // (This is true even if the store is volatile or atomic, although
+      // those cases are unlikely.)
+      if (AreEquivalentAddressValues(SI->getOperand(1), Ptr)) {
+        if (TBAATag) *TBAATag = SI->getMetadata(LLVMContext::MD_tbaa);
         return SI->getOperand(0);
+      }
       
       // If Ptr is an alloca and this is a store to a different alloca, ignore
       // the store.  This is a trivial form of alias analysis that is important