More optimizations

• A lot of misc. optimization work, notably parallelization of Rodinia CPU schedules.
• Support >32 tiles in GPU backend.
• Tweak some transformations.

juno_samples/rodinia/bfs/src/bfs.jn

 type Node = struct { edge_start: u32; num_edges: u32; };
+type StopProd = struct { stop: bool; };
+
+fn make_stop_prod() -> StopProd {
+  let ret : StopProd;
+  ret.stop = true;
+  return ret;
+}
 
 #[entry]
 fn bfs<n, m: usize>(graph_nodes: Node[n], source: u32, edges: u32[m]) -> i32[n] {
@@ -23,8 +30,6 @@ fn bfs<n, m: usize>(graph_nodes: Node[n], source: u32, edges: u32[m]) -> i32[n]
   let updated: bool[n];
 
   while !stop {
-    stop = true;
-
     @loop1 for i in 0..n {
       if mask[i] {
         mask[i] = false;
@@ -42,15 +47,16 @@ fn bfs<n, m: usize>(graph_nodes: Node[n], source: u32, edges: u32[m]) -> i32[n]
       }
     }
 
+    @make let stop_prod = make_stop_prod();
     @loop2 for i in 0..n {
-      stop = stop && !updated[i];
-    }
-
-    @loop3 for i in 0..n {
-      mask[i] = mask[i] || updated[i];
-      visited[i] = visited[i] || updated[i];
-      updated[i] = false;
+      if updated[i] {
+        mask[i] = true;
+        visited[i] = true;
+        updated[i] = false;
+	stop_prod.stop = updated[i];
+      }
     }
+    stop = stop_prod.stop;
   }
 
   return cost;