[llvm] [SCEV] Unify and optimize constant folding (NFC) (PR #101473)

[Philip Reames via llvm-commits]

================
@@ -831,6 +831,49 @@ static bool hasHugeExpression(ArrayRef<const SCEV *> Ops) {
   });
 }
 
+/// Performs a number of common optimizations on the passed \p Ops. If the
+/// whole expression reduces down to a single operand, it will be returned.
+///
+/// The following optimizations are performed:
+///  * Fold constants using the \p Fold function.
+///  * Remove identity constants satisfying \p IsIdentity.
+///  * If a constant satisfies \p IsAbsorber, return it.
+///  * Sort operands by complexity.
+template <typename FoldT, typename IsIdentityT, typename IsAbsorberT>
+static const SCEV *
+constantFoldAndGroupOps(ScalarEvolution &SE, LoopInfo &LI, DominatorTree &DT,
+                        SmallVectorImpl<const SCEV *> &Ops, FoldT Fold,
+                        IsIdentityT IsIdentity, IsAbsorberT IsAbsorber) {
+  const SCEVConstant *Folded = nullptr;
+  for (unsigned Idx = 0; Idx < Ops.size();) {
+    const SCEV *Op = Ops[Idx];
+    if (const auto *C = dyn_cast<SCEVConstant>(Op)) {
+      if (!Folded)
+        Folded = C;
+      else
+        Folded = cast<SCEVConstant>(
----------------
preames wrote:

One thought - the indirection through lambdas may be relatively expensive here - unless they're reliability inlined.  You might want to either use template parameters for the callables (to force specialization).  

https://github.com/llvm/llvm-project/pull/101473