[llvm] Fix loop cache cost to avoid cost of zero for refgroups. (PR #88915)

[via llvm-commits]

================
@@ -299,7 +299,10 @@ CacheCostTy IndexedReference::computeRefCost(const Loop &L,
     Stride = SE.getNoopOrAnyExtend(Stride, WiderType);
     TripCount = SE.getNoopOrZeroExtend(TripCount, WiderType);
     const SCEV *Numerator = SE.getMulExpr(Stride, TripCount);
-    RefCost = SE.getUDivExpr(Numerator, CacheLineSize);
+    // Round the costs up to the nearest whole number i.e. when cost is
+    // calculated 12.5, it makes more sense to say 13 cache lines are used
+    // rather than 12 cache lines.
----------------
CongzheUalberta wrote:

// Round the cost up to the nearest integer number, e.g., when cost is
// calculated to be a number less than one, it makes more sense to say 
// one cache line is used rather than zero cache line is used.

https://github.com/llvm/llvm-project/pull/88915