More optimizations

• Speedup backprop
• Fix gpu backend for different CUDA versions

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];