parallelize cost init loop

juno_samples/rodinia/bfs/src/bfs.jn

@@ -24,7 +24,7 @@ fn bfs<n, m: usize>(graph_nodes: Node[n], source: u32, edges: u32[m]) -> i32[n]
   @cost_init for i in 0..n {
     cost[i] = -1;
   }
-  cost[source as u64] = 0;
+  @cost_init cost[source as u64] = 0;
 
   // Nodes that were updated in the current iteration
   let updated: bool[n];

juno_samples/rodinia/bfs/src/cpu.sch

@@ -48,10 +48,9 @@ let collect_body = outline(inner);
 
 let init_body = init;
 // Following code seems to generate breaking RT code
-//fork-tile[32, 0, false, true](init);
-//let (outer, inner) = fork-reshape[[1], [0]](init);
-//let init_body = outline(inner);
-//inline(bfs@cost_init);
+fork-tile[32, 0, false, true](init);
+let (outer, inner) = fork-reshape[[1], [0]](init);
+let init_body = outline(inner);
 
 inline(bfs@loop1, bfs@loop2);
 delete-uncalled(*);