collectives/br_tree_algorithms.hpp

#ifndef BR_TREE_ALGORITHMS_HPP_
#define BR_TREE_ALGORITHMS_HPP_

#include "libmpib_coll.h"
#include "mpib_coll.h"
#include "comm_tree.hpp"
#include <boost/graph/graphviz.hpp>

template <typename Builder>
int MPIB_Bcast_tree_algorithm(Builder builder, MPIB_child_traverse_order order,
    void* buffer, int count, MPI_Datatype datatype, int root, MPI_Comm comm)
{
    using namespace MPIB::comm_tree;
    // build graph
    int size;
    MPI_Comm_size(comm, &size);
    int rank;
    MPI_Comm_rank(comm, &rank);
    MPI_Aint extent;
    MPI_Type_extent(datatype, &extent);
    Graph graph;
    Vertex r, u, v;
    builder.build(size, root, rank, count * extent, graph, r, u, v);
    if ((rank == root) && mpib_coll_verbose)
        write_graphviz(cout, graph);
    // recv from parent
    if (rank != root)
        MPI_Recv(buffer, count, datatype, get(vertex_index, graph, u), 0, comm, MPI_STATUS_IGNORE);
    // send to children - start from the last
    MPI_Request* reqs = (MPI_Request*)malloc(sizeof(MPI_Request) * size);
    int child_counter;
    Adjacency_iterator ai, ai_end;
    tie(ai, ai_end) = adjacent_vertices(v, graph);
    for(child_counter = 0; ai != ai_end; child_counter++)
        MPI_Isend(buffer, count, datatype, get(vertex_index, graph, (order == R2L) ? *(--ai_end) : *(ai++)), 0, comm, &reqs[child_counter]);
    MPI_Waitall(child_counter, reqs, MPI_STATUSES_IGNORE);
    free(reqs);
    return MPI_SUCCESS;
}

template <typename Builder>
int MPIB_Reduce_tree_algorithm(Builder builder, MPIB_child_traverse_order order,
    void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype,
    MPI_Op op, int root, MPI_Comm comm)
{
    using namespace MPIB::comm_tree;
    // build graph
    int size;
    MPI_Comm_size(comm, &size);
    int rank;
    MPI_Comm_rank(comm, &rank);
    MPI_Aint extent;
    MPI_Type_extent(datatype, &extent);
    Graph graph;
    Vertex r, u, v;
    builder.build(size, root, rank, count * extent, graph, r, u, v);
    if ((rank == root) && mpib_coll_verbose)
        write_graphviz(cout, graph);
    // subproduct for MPI operation
    void* buffer = rank == root ? recvbuf : malloc(count * extent);
    // receive from children - start from the first
    MPI_Request* reqs = (MPI_Request*)malloc(sizeof(MPI_Request) * size);
    int child_counter;
    Adjacency_iterator ai, ai_end;
    tie(ai, ai_end) = adjacent_vertices(v, graph);
    for(child_counter = 0; ai != ai_end; child_counter++) {
        MPI_Irecv(buffer, count, datatype, get(vertex_index, graph, (order == R2L) ? *(--ai_end) : *(ai++)), 0, comm, &reqs[child_counter]);
        // MPI operation to be here: buffer op= sendbuf.
    }
    MPI_Waitall(child_counter, reqs, MPI_STATUSES_IGNORE);
    free(reqs);
    // send to parent
    if (rank != root) {
        MPI_Send(buffer, count, datatype, get(vertex_index, graph, u), 0, comm);
        free(buffer);
    }
    return MPI_SUCCESS;
}

#endif /*BR_TREE_ALGORITHMS_HPP_*/