More misc. rodinia opts

• Make GCM not spill ParallelReduce phis.
• Fix unification for inter-device communication.
• Make notion of parallel indices sets more flexible, and use in infer parallel reduce.
• Misc. schedule tweaks.

hercules_opt/src/gcm.rs

@@ -879,8 +879,8 @@ fn spill_clones(
                     || editor.func().nodes[a.idx()].is_reduce())
                 && !editor.func().nodes[a.idx()]
                     .try_reduce()
-                    .map(|(_, init, _)| {
-                        init == *b
+                    .map(|(_, init, reduct)| {
+                        (init == *b || reduct == *b)
                             && editor.func().schedules[a.idx()].contains(&Schedule::ParallelReduce)
                     })
                     .unwrap_or(false)