/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil ; -*- */
/*
 *
 *  (C) 2017 by Argonne National Laboratory.
 *      See COPYRIGHT in top-level directory.
 */

#include "mpiimpl.h"

/* Intercommunicator Reduce_scatter_block
 *
 * We first do an intercommunicator reduce to rank 0 on left group,
 * then an intercommunicator reduce to rank 0 on right group, followed
 * by local intracommunicator scattervs in each group.
 */

#undef FUNCNAME
#define FUNCNAME MPIR_Reduce_scatter_block_inter_remote_reduce_local_scatter
#undef FCNAME
#define FCNAME MPL_QUOTE(FUNCNAME)
int MPIR_Reduce_scatter_block_inter_remote_reduce_local_scatter(const void *sendbuf,
                                                                void *recvbuf,
                                                                int recvcount,
                                                                MPI_Datatype datatype,
                                                                MPI_Op op,
                                                                MPIR_Comm * comm_ptr,
                                                                MPIR_Errflag_t * errflag)
{
    int rank, mpi_errno, root, local_size, total_count;
    int mpi_errno_ret = MPI_SUCCESS;
    MPI_Aint true_extent, true_lb = 0, extent;
    void *tmp_buf = NULL;
    MPIR_Comm *newcomm_ptr = NULL;
    MPIR_CHKLMEM_DECL(1);

    rank = comm_ptr->rank;
    local_size = comm_ptr->local_size;

    total_count = local_size * recvcount;

    if (rank == 0) {
        /* In each group, rank 0 allocates a temp. buffer for the
         * reduce */

        MPIR_Type_get_true_extent_impl(datatype, &true_lb, &true_extent);
        MPIR_Datatype_get_extent_macro(datatype, extent);

        MPIR_CHKLMEM_MALLOC(tmp_buf, void *, total_count * (MPL_MAX(extent, true_extent)),
                            mpi_errno, "tmp_buf", MPL_MEM_BUFFER);

        /* adjust for potential negative lower bound in datatype */
        tmp_buf = (void *) ((char *) tmp_buf - true_lb);
    }

    /* first do a reduce from right group to rank 0 in left group,
     * then from left group to rank 0 in right group */
    if (comm_ptr->is_low_group) {
        /* reduce from right group to rank 0 */
        root = (rank == 0) ? MPI_ROOT : MPI_PROC_NULL;
        mpi_errno = MPIR_Reduce_inter_auto(sendbuf, tmp_buf, total_count, datatype, op,
                                           root, comm_ptr, errflag);
        if (mpi_errno) {
            /* for communication errors, just record the error but continue */
            *errflag =
                MPIX_ERR_PROC_FAILED ==
                MPIR_ERR_GET_CLASS(mpi_errno) ? MPIR_ERR_PROC_FAILED : MPIR_ERR_OTHER;
            MPIR_ERR_SET(mpi_errno, *errflag, "**fail");
            MPIR_ERR_ADD(mpi_errno_ret, mpi_errno);
        }

        /* reduce to rank 0 of right group */
        root = 0;
        mpi_errno = MPIR_Reduce_inter_auto(sendbuf, tmp_buf, total_count, datatype, op,
                                           root, comm_ptr, errflag);
        if (mpi_errno) {
            /* for communication errors, just record the error but continue */
            *errflag =
                MPIX_ERR_PROC_FAILED ==
                MPIR_ERR_GET_CLASS(mpi_errno) ? MPIR_ERR_PROC_FAILED : MPIR_ERR_OTHER;
            MPIR_ERR_SET(mpi_errno, *errflag, "**fail");
            MPIR_ERR_ADD(mpi_errno_ret, mpi_errno);
        }
    } else {
        /* reduce to rank 0 of left group */
        root = 0;
        mpi_errno = MPIR_Reduce_inter_auto(sendbuf, tmp_buf, total_count, datatype, op,
                                           root, comm_ptr, errflag);
        if (mpi_errno) {
            /* for communication errors, just record the error but continue */
            *errflag =
                MPIX_ERR_PROC_FAILED ==
                MPIR_ERR_GET_CLASS(mpi_errno) ? MPIR_ERR_PROC_FAILED : MPIR_ERR_OTHER;
            MPIR_ERR_SET(mpi_errno, *errflag, "**fail");
            MPIR_ERR_ADD(mpi_errno_ret, mpi_errno);
        }

        /* reduce from right group to rank 0 */
        root = (rank == 0) ? MPI_ROOT : MPI_PROC_NULL;
        mpi_errno = MPIR_Reduce_inter_auto(sendbuf, tmp_buf, total_count, datatype, op,
                                           root, comm_ptr, errflag);
        if (mpi_errno) {
            /* for communication errors, just record the error but continue */
            *errflag =
                MPIX_ERR_PROC_FAILED ==
                MPIR_ERR_GET_CLASS(mpi_errno) ? MPIR_ERR_PROC_FAILED : MPIR_ERR_OTHER;
            MPIR_ERR_SET(mpi_errno, *errflag, "**fail");
            MPIR_ERR_ADD(mpi_errno_ret, mpi_errno);
        }
    }

    /* Get the local intracommunicator */
    if (!comm_ptr->local_comm)
        MPII_Setup_intercomm_localcomm(comm_ptr);

    newcomm_ptr = comm_ptr->local_comm;

    mpi_errno = MPIR_Scatter(tmp_buf, recvcount, datatype, recvbuf,
                             recvcount, datatype, 0, newcomm_ptr, errflag);
    if (mpi_errno) {
        /* for communication errors, just record the error but continue */
        *errflag =
            MPIX_ERR_PROC_FAILED ==
            MPIR_ERR_GET_CLASS(mpi_errno) ? MPIR_ERR_PROC_FAILED : MPIR_ERR_OTHER;
        MPIR_ERR_SET(mpi_errno, *errflag, "**fail");
        MPIR_ERR_ADD(mpi_errno_ret, mpi_errno);
    }

  fn_exit:
    MPIR_CHKLMEM_FREEALL();
    /* --BEGIN ERROR HANDLING-- */
    if (mpi_errno_ret)
        mpi_errno = mpi_errno_ret;
    else if (*errflag != MPIR_ERR_NONE)
        MPIR_ERR_SET(mpi_errno, *errflag, "**coll_fail");
    /* --END ERROR HANDLING-- */
    return mpi_errno;
  fn_fail:
    goto fn_exit;
}