Address review comments

Author: david-arm

llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp

@@ -4397,8 +4397,14 @@ static bool shouldUnrollLoopWithInstruction(Instruction &I,
   return true;
 }
 
-static unsigned getLoopSize(Loop *L, AArch64TTIImpl &TTI,
-                            InstructionCost Budget) {
+// This function returns true if the loop:
+//  1. Contains only those instructions that should be unrolled,
+//  2. Has a valid cost,
+//  3. Has a cost within the supplied budget.
+// Otherwise it returns false.
+static bool isLoopSizeWithinBudget(Loop *L, AArch64TTIImpl &TTI,
+                                   InstructionCost Budget,
+                                   unsigned *FinalSize) {
   // Estimate the size of the loop.
   InstructionCost LoopCost = 0;
 
@@ -4424,7 +4430,9 @@ static unsigned getLoopSize(Loop *L, AArch64TTIImpl &TTI,
     }
   }
 
-  return *LoopCost.getValue();
+  if (FinalSize)
+    *FinalSize = *LoopCost.getValue();
+  return true;
 }
 
 static bool shouldUnrollMultiExitLoop(Loop *L, ScalarEvolution &SE,
@@ -4443,7 +4451,7 @@ static bool shouldUnrollMultiExitLoop(Loop *L, ScalarEvolution &SE,
     return false;
 
   // Estimate the size of the loop.
-  if (!getLoopSize(L, TTI, 5))
+  if (!isLoopSizeWithinBudget(L, TTI, 5, nullptr))
     return false;
 
   // Small search loops with multiple exits can be highly beneficial to unroll.
@@ -4485,6 +4493,9 @@ getAppleRuntimeUnrollPreferences(Loop *L, ScalarEvolution &SE,
        SE.getSmallConstantMaxTripCount(L) <= 32))
     return;
 
+  if (findStringMetadataForLoop(L, "llvm.loop.isvectorized"))
+    return;
+
   if (SE.getSymbolicMaxBackedgeTakenCount(L) != SE.getBackedgeTakenCount(L))
     return;
 
@@ -4496,8 +4507,8 @@ getAppleRuntimeUnrollPreferences(Loop *L, ScalarEvolution &SE,
   BasicBlock *Header = L->getHeader();
   if (Header == L->getLoopLatch()) {
     // Estimate the size of the loop.
-    unsigned Size = getLoopSize(L, TTI, 8);
-    if (!Size)
+    unsigned Size;
+    if (!isLoopSizeWithinBudget(L, TTI, 8, &Size))
       return;
 
     SmallPtrSet<Value *, 8> LoadedValues;
@@ -4609,6 +4620,11 @@ void AArch64TTIImpl::getUnrollingPreferences(Loop *L, ScalarEvolution &SE,
     if (EnableFalkorHWPFUnrollFix)
       getFalkorUnrollingPreferences(L, SE, UP);
     break;
+  case AArch64Subtarget::NeoverseV1:
+  case AArch64Subtarget::NeoverseV2:
+    if (!SmallMultiExitLoopUF.getNumOccurrences())
+      SmallMultiExitLoopUnrollFactor = 4;
+    break;
   default:
     break;
   }