[llvm] [LV] Separate runtime check cost from total overhead in profitability check (PR #176754)

[Luke Lau via llvm-commits]

================
@@ -9174,21 +9174,22 @@ static bool isOutsideLoopWorkProfitable(GeneratedRTChecks &Checks,
                                         VPCostContext &CostCtx, VPlan &Plan,
                                         ScalarEpilogueLowering SEL,
                                         std::optional<unsigned> VScale) {
-  InstructionCost TotalCost = Checks.getCost();
-  if (!TotalCost.isValid())
+  InstructionCost RtCost = Checks.getCost();
+  if (!RtCost.isValid())
     return false;
 
+  InstructionCost OverheadCost = RtCost;
   // Add on the cost of any work required in the vector early exit block, if
   // one exists.
-  TotalCost += calculateEarlyExitCost(CostCtx, Plan, VF.Width);
+  OverheadCost += calculateEarlyExitCost(CostCtx, Plan, VF.Width);
 
-  TotalCost += Plan.getMiddleBlock()->cost(VF.Width, CostCtx);
+  OverheadCost += Plan.getMiddleBlock()->cost(VF.Width, CostCtx);
 
   // When interleaving only scalar and vector cost will be equal, which in turn
   // would lead to a divide by 0. Fall back to hard threshold.
   if (VF.Width.isScalar()) {
     // TODO: Should we rename VectorizeMemoryCheckThreshold?
-    if (TotalCost > VectorizeMemoryCheckThreshold) {
+    if (OverheadCost > VectorizeMemoryCheckThreshold) {
----------------
lukel97 wrote:

VectorizeMemoryCheckThreshold does seem weird. Could we reuse the MinTC2 check instead for interleaving only? 

https://github.com/llvm/llvm-project/pull/176754