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/utils.rs

@@ -532,6 +532,24 @@ where
     let fork_thread_id_pairs = node_indices(indices).filter_map(|id| {
         if let Node::ThreadID { control, dimension } = nodes[id.idx()] {
             Some((control, dimension))
+        } else if let Node::Binary {
+            op: BinaryOperator::Add,
+            left: tid,
+            right: cons,
+        } = nodes[id.idx()]
+            && let Node::ThreadID { control, dimension } = nodes[tid.idx()]
+            && (nodes[cons.idx()].is_constant() || nodes[cons.idx()].is_dynamic_constant())
+        {
+            Some((control, dimension))
+        } else if let Node::Binary {
+            op: BinaryOperator::Add,
+            left: cons,
+            right: tid,
+        } = nodes[id.idx()]
+            && let Node::ThreadID { control, dimension } = nodes[tid.idx()]
+            && (nodes[cons.idx()].is_constant() || nodes[cons.idx()].is_dynamic_constant())
+        {
+            Some((control, dimension))
         } else {
             None
         }