Optimizing NEURON Simulation Environment Using Remote Memory Access with Recursive Doubling on Distributed Memory Systems

Algorithm 2

RMA_Allgather for spikes communication between processors.

Input parameters: sendbuf, count, datatype, MPI_Comm_World

Output Parameters: Recv_buf

    Start:

    Initialize local variables target, displacement, step, origin

    Calculate number of steps =

    if first step then

        Get communication size and proc_id

        Initialize recv_bufer

        Create window according to recv_bufer

    for to

        Calculate origin_addr as

        Origin_Addr ))

        Calculate target according to step and id (Algo 1)

        MPI Fence (Synchronization)

        MPI_Put values from origin to target

        Next

    End