Example 12: Nested user-defined parallelism

In this example, all operators from within the parallel regions are fused into two PEs corresponding to the two channels in the outer parallel region. The parallel transformation replicates operators B and C, and then replicates operators A and all replicas of B and C. This gives two replicas of operator A and four replicas of operators B and C. Fusion then places the replicas of operator A for each channel in their own PE along with the replicas of operators B and C for each inner local channel.

composite Comp12-2(input In; output C) {
  graph
    stream<Type> B = Functor(In) {}
    stream<Type> C = Functor(B) {}
}

composite Comp12-1(input In; output O) {
  graph
    stream<Type> A = Functor(In) {}
    @parallel(width=2)
    stream<Type> O = Comp12-2(A) {}
}

composite Main12 {
  graph
    stream<Type> Src = Source() {
      config placement: partitionIsolation;
    }

    @parallel(width=2)
    stream<Type> Out = Comp12-1(Src) {
      config placement: partitionColocation(byChannel());
    }
    () as Snk = Sink(Out) {
      config placement: partitionIsolation;
    }
}