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.

juno_samples/rodinia/bfs/src/gpu.sch

@@ -13,7 +13,7 @@ no-memset(bfs@cost);
 let cost_init = outline(bfs@cost_init);
 let loop1 = outline(bfs@loop1);
 let loop2 = outline(bfs@loop2);
-gpu(cost_init, loop1, loop2);
+gpu(loop1, loop2);
 
 simpl!(*);
 predication(*);
@@ -35,5 +35,5 @@ predication(*);
 reduce-slf(*);
 simpl!(*);
 
+unforkify(cost_init);
 gcm(*);
-xdot[true](*);