Exit early in getting most recent global_barrier if no barriers exist at all

loopy is creating a lot of temporary kernels when realizing
lots of reductions in sumpy kernels. Therefore calculating the
global barrier order is expensive.