[llvm] [MachinePipeliner] Make Recurrence MII More Accurate (PR #105475)

[Ryotaro Kasuga via llvm-commits]

================
@@ -339,24 +341,58 @@ class NodeSet {
   using iterator = SetVector<SUnit *>::const_iterator;
 
   NodeSet() = default;
-  NodeSet(iterator S, iterator E) : Nodes(S, E), HasRecurrence(true) {
-    Latency = 0;
-    for (const SUnit *Node : Nodes) {
-      DenseMap<SUnit *, unsigned> SuccSUnitLatency;
-      for (const SDep &Succ : Node->Succs) {
-        auto SuccSUnit = Succ.getSUnit();
-        if (!Nodes.count(SuccSUnit))
+  NodeSet(iterator S, iterator E, const SwingSchedulerDAG *DAG)
+      : Nodes(S, E), HasRecurrence(true) {
+    // Calculate the latency of this node set.
+    // Example to demonstrate the calculation:
+    // Given: N0 -> N1 -> N2 -> N0
+    // Edges:
+    // (N0 -> N1, 3)
+    // (N0 -> N1, 5)
+    // (N1 -> N2, 2)
+    // (N2 -> N0, 1)
+    // The total latency which is a lower bound of the recurrence MII is the
+    // longest patch from N0 back to N0 given only the edges of this node set.
----------------
kasuga-fj wrote:

patch -> path? (typo?)

https://github.com/llvm/llvm-project/pull/105475