Fix reduce cycles and nodes in fork joins

• Fix reduce cycles and nodes in fork joins analyses (weren't previously considering nodes in reduce cycle but not dependent on thread ID).
• Clean up juno_matmul schedule.
• @xrouth2, this may have been causing the weird GCM issues you were running into?

hercules_ir/src/fork_join_analysis.rs

@@ -177,7 +177,7 @@ pub fn nodes_in_fork_joins(
         let mut set = HashSet::new();
         set.insert(*fork);
 
-        // Iterate uses of fork.
+        // Iterate uses of the fork.
         while let Some(item) = worklist.pop() {
             for u in def_use.get_users(item) {
                 let terminate = *u == *join
@@ -189,11 +189,15 @@ pub fn nodes_in_fork_joins(
                     worklist.push(*u);
                 }
                 set.insert(*u);
+            }
+        }
+        assert!(set.contains(join));
 
-                // Nodes in reduce cycles might not depend on the thread ID.
-                if terminate && let Some(cycle) = reduce_cycles.get(u) {
-                    set.extend(cycle);
-                }
+        // Add all the nodes in the reduce cycle. Some of these nodes may not
+        // use thread IDs of the fork, so do this explicitly.
+        for u in def_use.get_users(*join) {
+            if let Some(cycle) = reduce_cycles.get(u) {
+                set.extend(cycle);
             }
         }