[llvm] LV: generalize profitability criterion over TC (PR #93300)

[David Green via llvm-commits]

================
@@ -4893,13 +4871,36 @@ bool LoopVectorizationPlanner::isMoreProfitable(
   // Assume vscale may be larger than 1 (or the value being tuned for),
   // so that scalable vectorization is slightly favorable over fixed-width
   // vectorization.
-  if (A.Width.isScalable() && !B.Width.isScalable())
-    return (CostA * B.Width.getFixedValue()) <= (CostB * EstimatedWidthA);
+  bool PreferScalable = A.Width.isScalable() && !B.Width.isScalable();
+  auto CmpFn = [PreferScalable](const InstructionCost &LHS,
+                                const InstructionCost &RHS) {
+    return PreferScalable ? LHS <= RHS : LHS < RHS;
+  };
 
   // To avoid the need for FP division:
-  //      (CostA / A.Width) < (CostB / B.Width)
-  // <=>  (CostA * B.Width) < (CostB * A.Width)
-  return (CostA * EstimatedWidthB) < (CostB * EstimatedWidthA);
+  //      (CostA / EstimatedWidthA) < (CostB / EstimatedWidthB)
+  // <=>  (CostA * EstimatedWidthB) < (CostB * EstimatedWidthA)
+  if (!MaxTripCount)
+    return CmpFn(CostA * EstimatedWidthB, CostB * EstimatedWidthA);
+
+  auto GetCost = [MaxTripCount, this](unsigned VF, InstructionCost VectorCost,
----------------
davemgreen wrote:

GetCostForTC was the suggested name from https://reviews.llvm.org/D147720#inline-1439570.

https://github.com/llvm/llvm-project/pull/93300