/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil ; -*- */
/*
 *  (C) 2012 by Argonne National Laboratory.
 *      See COPYRIGHT in top-level directory.
 */

#include "mpiimpl.h"

/*
=== BEGIN_MPI_T_CVAR_INFO_BLOCK ===

cvars:
    - name        : MPIR_CVAR_INEIGHBOR_ALLTOALLW_INTRA_ALGORITHM
      category    : COLLECTIVE
      type        : string
      default     : auto
      class       : device
      verbosity   : MPI_T_VERBOSITY_USER_BASIC
      scope       : MPI_T_SCOPE_ALL_EQ
      description : |-
        Variable to select ineighbor_alltoallw algorithm
        auto   - Internal algorithm selection
        linear - Force linear algorithm

    - name        : MPIR_CVAR_INEIGHBOR_ALLTOALLW_INTER_ALGORITHM
      category    : COLLECTIVE
      type        : string
      default     : auto
      class       : device
      verbosity   : MPI_T_VERBOSITY_USER_BASIC
      scope       : MPI_T_SCOPE_ALL_EQ
      description : |-
        Variable to select ineighbor_alltoallw algorithm
        auto   - Internal algorithm selection
        linear - Force linear algorithm

    - name        : MPIR_CVAR_INEIGHBOR_ALLTOALLW_DEVICE_COLLECTIVE
      category    : COLLECTIVE
      type        : boolean
      default     : true
      class       : device
      verbosity   : MPI_T_VERBOSITY_USER_BASIC
      scope       : MPI_T_SCOPE_ALL_EQ
      description : >-
        If set to true, MPI_ineighbor_alltoallw will allow the device to override the
        MPIR-level collective algorithms. The device still has the
        option to call the MPIR-level algorithms manually.
        If set to false, the device-level ineighbor_alltoallw function will not be
        called.

=== END_MPI_T_CVAR_INFO_BLOCK ===
*/

/* -- Begin Profiling Symbol Block for routine MPI_Ineighbor_alltoallw */
#if defined(HAVE_PRAGMA_WEAK)
#pragma weak MPI_Ineighbor_alltoallw = PMPI_Ineighbor_alltoallw
#elif defined(HAVE_PRAGMA_HP_SEC_DEF)
#pragma _HP_SECONDARY_DEF PMPI_Ineighbor_alltoallw  MPI_Ineighbor_alltoallw
#elif defined(HAVE_PRAGMA_CRI_DUP)
#pragma _CRI duplicate MPI_Ineighbor_alltoallw as PMPI_Ineighbor_alltoallw
#elif defined(HAVE_WEAK_ATTRIBUTE)
int MPI_Ineighbor_alltoallw(const void *sendbuf, const int sendcounts[],
                            const MPI_Aint sdispls[], const MPI_Datatype sendtypes[],
                            void *recvbuf, const int recvcounts[], const MPI_Aint rdispls[],
                            const MPI_Datatype recvtypes[], MPI_Comm comm, MPI_Request * request)
    __attribute__ ((weak, alias("PMPI_Ineighbor_alltoallw")));
#endif
/* -- End Profiling Symbol Block */

/* Define MPICH_MPI_FROM_PMPI if weak symbols are not supported to build
   the MPI routines */
#ifndef MPICH_MPI_FROM_PMPI
#undef MPI_Ineighbor_alltoallw
#define MPI_Ineighbor_alltoallw PMPI_Ineighbor_alltoallw

#undef FUNCNAME
#define FUNCNAME MPIR_Ineighbor_alltoallw_sched_intra_auto
#undef FCNAME
#define FCNAME MPL_QUOTE(FUNCNAME)
int MPIR_Ineighbor_alltoallw_sched_intra_auto(const void *sendbuf, const int sendcounts[],
                                              const MPI_Aint sdispls[],
                                              const MPI_Datatype sendtypes[], void *recvbuf,
                                              const int recvcounts[], const MPI_Aint rdispls[],
                                              const MPI_Datatype recvtypes[], MPIR_Comm * comm_ptr,
                                              MPIR_Sched_t s)
{
    int mpi_errno = MPI_SUCCESS;

    mpi_errno =
        MPIR_Ineighbor_alltoallw_sched_allcomm_linear(sendbuf, sendcounts, sdispls, sendtypes,
                                                      recvbuf, recvcounts, rdispls, recvtypes,
                                                      comm_ptr, s);
    if (mpi_errno)
        MPIR_ERR_POP(mpi_errno);

  fn_exit:
    return mpi_errno;

  fn_fail:
    goto fn_exit;
}

#undef FUNCNAME
#define FUNCNAME MPIR_Ineighbor_alltoallw_sched_inter_auto
#undef FCNAME
#define FCNAME MPL_QUOTE(FUNCNAME)
int MPIR_Ineighbor_alltoallw_sched_inter_auto(const void *sendbuf, const int sendcounts[],
                                              const MPI_Aint sdispls[],
                                              const MPI_Datatype sendtypes[], void *recvbuf,
                                              const int recvcounts[], const MPI_Aint rdispls[],
                                              const MPI_Datatype recvtypes[], MPIR_Comm * comm_ptr,
                                              MPIR_Sched_t s)
{
    int mpi_errno = MPI_SUCCESS;

    mpi_errno =
        MPIR_Ineighbor_alltoallw_sched_allcomm_linear(sendbuf, sendcounts, sdispls, sendtypes,
                                                      recvbuf, recvcounts, rdispls, recvtypes,
                                                      comm_ptr, s);
    if (mpi_errno)
        MPIR_ERR_POP(mpi_errno);

  fn_exit:
    return mpi_errno;

  fn_fail:
    goto fn_exit;
}

#undef FUNCNAME
#define FUNCNAME MPIR_Ineighbor_alltoallw_sched_impl
#undef FCNAME
#define FCNAME MPL_QUOTE(FUNCNAME)
int MPIR_Ineighbor_alltoallw_sched_impl(const void *sendbuf, const int sendcounts[],
                                        const MPI_Aint sdispls[], const MPI_Datatype sendtypes[],
                                        void *recvbuf, const int recvcounts[],
                                        const MPI_Aint rdispls[], const MPI_Datatype recvtypes[],
                                        MPIR_Comm * comm_ptr, MPIR_Sched_t s)
{
    int mpi_errno = MPI_SUCCESS;

    if (comm_ptr->comm_kind == MPIR_COMM_KIND__INTRACOMM) {
        switch (MPIR_Ineighbor_alltoallw_intra_algo_choice) {
            case MPIR_INEIGHBOR_ALLTOALLW_INTRA_ALGO_LINEAR:
                mpi_errno =
                    MPIR_Ineighbor_alltoallw_sched_allcomm_linear(sendbuf, sendcounts, sdispls,
                                                                  sendtypes, recvbuf, recvcounts,
                                                                  rdispls, recvtypes, comm_ptr, s);
                break;
            case MPIR_INEIGHBOR_ALLTOALLW_INTRA_ALGO_AUTO:
                MPL_FALLTHROUGH;
            default:
                mpi_errno =
                    MPIR_Ineighbor_alltoallw_sched_intra_auto(sendbuf, sendcounts, sdispls,
                                                              sendtypes, recvbuf, recvcounts,
                                                              rdispls, recvtypes, comm_ptr, s);
                break;
        }
    } else {
        switch (MPIR_Ineighbor_alltoallw_inter_algo_choice) {
            case MPIR_INEIGHBOR_ALLTOALLW_INTER_ALGO_LINEAR:
                mpi_errno =
                    MPIR_Ineighbor_alltoallw_sched_allcomm_linear(sendbuf, sendcounts, sdispls,
                                                                  sendtypes, recvbuf, recvcounts,
                                                                  rdispls, recvtypes, comm_ptr, s);
                break;
            case MPIR_INEIGHBOR_ALLTOALLW_INTER_ALGO_AUTO:
                MPL_FALLTHROUGH;
            default:
                mpi_errno =
                    MPIR_Ineighbor_alltoallw_sched_inter_auto(sendbuf, sendcounts, sdispls,
                                                              sendtypes, recvbuf, recvcounts,
                                                              rdispls, recvtypes, comm_ptr, s);
                break;
        }
    }

    return mpi_errno;
}

#undef FUNCNAME
#define FUNCNAME MPIR_Ineighbor_alltoallw_sched
#undef FCNAME
#define FCNAME MPL_QUOTE(FUNCNAME)
int MPIR_Ineighbor_alltoallw_sched(const void *sendbuf, const int sendcounts[],
                                   const MPI_Aint sdispls[], const MPI_Datatype sendtypes[],
                                   void *recvbuf, const int recvcounts[],
                                   const MPI_Aint rdispls[], const MPI_Datatype recvtypes[],
                                   MPIR_Comm * comm_ptr, MPIR_Sched_t s)
{
    int mpi_errno = MPI_SUCCESS;

    if (MPIR_CVAR_BARRIER_DEVICE_COLLECTIVE && MPIR_CVAR_DEVICE_COLLECTIVES) {
        mpi_errno = MPID_Ineighbor_alltoallw_sched(sendbuf, sendcounts, sdispls, sendtypes, recvbuf,
                                                   recvcounts, rdispls, recvtypes, comm_ptr, s);
    } else {
        mpi_errno = MPIR_Ineighbor_alltoallw_sched_impl(sendbuf, sendcounts, sdispls, sendtypes,
                                                        recvbuf, recvcounts, rdispls, recvtypes,
                                                        comm_ptr, s);
    }

    return mpi_errno;
}

#undef FUNCNAME
#define FUNCNAME MPIR_Ineighbor_alltoallw_impl
#undef FCNAME
#define FCNAME MPL_QUOTE(FUNCNAME)
int MPIR_Ineighbor_alltoallw_impl(const void *sendbuf, const int sendcounts[],
                                  const MPI_Aint sdispls[],
                                  const MPI_Datatype sendtypes[],
                                  void *recvbuf, const int recvcounts[],
                                  const MPI_Aint rdispls[],
                                  const MPI_Datatype recvtypes[],
                                  MPIR_Comm * comm_ptr, MPIR_Request ** request)
{
    int mpi_errno = MPI_SUCCESS;
    int tag = -1;
    MPIR_Sched_t s = MPIR_SCHED_NULL;

    *request = NULL;

    mpi_errno = MPIR_Sched_next_tag(comm_ptr, &tag);
    if (mpi_errno)
        MPIR_ERR_POP(mpi_errno);
    mpi_errno = MPIR_Sched_create(&s);
    if (mpi_errno)
        MPIR_ERR_POP(mpi_errno);
    mpi_errno = MPIR_Ineighbor_alltoallw_sched(sendbuf, sendcounts, sdispls, sendtypes,
                                               recvbuf, recvcounts, rdispls, recvtypes,
                                               comm_ptr, s);
    if (mpi_errno)
        MPIR_ERR_POP(mpi_errno);

    mpi_errno = MPIR_Sched_start(&s, comm_ptr, tag, request);
    if (mpi_errno)
        MPIR_ERR_POP(mpi_errno);

  fn_exit:
    return mpi_errno;
  fn_fail:
    goto fn_exit;
}

#undef FUNCNAME
#define FUNCNAME MPIR_Ineighbor_alltoallw
#undef FCNAME
#define FCNAME MPL_QUOTE(FUNCNAME)
int MPIR_Ineighbor_alltoallw(const void *sendbuf, const int sendcounts[],
                             const MPI_Aint sdispls[],
                             const MPI_Datatype sendtypes[], void *recvbuf,
                             const int recvcounts[], const MPI_Aint rdispls[],
                             const MPI_Datatype recvtypes[],
                             MPIR_Comm * comm_ptr, MPIR_Request ** request)
{
    int mpi_errno = MPI_SUCCESS;

    if (MPIR_CVAR_BARRIER_DEVICE_COLLECTIVE && MPIR_CVAR_DEVICE_COLLECTIVES) {
        mpi_errno = MPID_Ineighbor_alltoallw(sendbuf, sendcounts, sdispls, sendtypes, recvbuf,
                                             recvcounts, rdispls, recvtypes, comm_ptr, request);
    } else {
        mpi_errno = MPIR_Ineighbor_alltoallw_impl(sendbuf, sendcounts, sdispls, sendtypes, recvbuf,
                                                  recvcounts, rdispls, recvtypes, comm_ptr,
                                                  request);
    }

    return mpi_errno;
}

#endif /* MPICH_MPI_FROM_PMPI */

#undef FUNCNAME
#define FUNCNAME MPI_Ineighbor_alltoallw
#undef FCNAME
#define FCNAME MPL_QUOTE(FUNCNAME)
/*@
MPI_Ineighbor_alltoallw - Nonblocking version of MPI_Neighbor_alltoallw.

Input Parameters:
+ sendbuf - starting address of the send buffer (choice)
. sendcounts - non-negative integer array (of length outdegree) specifying the number of elements to send to each neighbor
. sdispls - integer array (of length outdegree).  Entry j specifies the displacement in bytes (relative to sendbuf) from which to take the outgoing data destined for neighbor j (array of integers)
. sendtypes - array of datatypes (of length outdegree).  Entry j specifies the type of data to send to neighbor j (array of handles)
. recvcounts - non-negative integer array (of length indegree) specifying the number of elements that are received from each neighbor
. rdispls - integer array (of length indegree).  Entry i specifies the displacement in bytes (relative to recvbuf) at which to place the incoming data from neighbor i (array of integers).
. recvtypes - array of datatypes (of length indegree).  Entry i specifies the type of data received from neighbor i (array of handles).
- comm - communicator with topology structure (handle)

Output Parameters:
+ recvbuf - starting address of the receive buffer (choice)
- request - communication request (handle)

.N ThreadSafe

.N Fortran

.N Errors
@*/
int MPI_Ineighbor_alltoallw(const void *sendbuf, const int sendcounts[], const MPI_Aint sdispls[],
                            const MPI_Datatype sendtypes[], void *recvbuf, const int recvcounts[],
                            const MPI_Aint rdispls[], const MPI_Datatype recvtypes[], MPI_Comm comm,
                            MPI_Request * request)
{
    int mpi_errno = MPI_SUCCESS;
    MPIR_Comm *comm_ptr = NULL;
    MPIR_Request *request_ptr = NULL;
    MPIR_FUNC_TERSE_STATE_DECL(MPID_STATE_MPI_INEIGHBOR_ALLTOALLW);

    MPID_THREAD_CS_ENTER(GLOBAL, MPIR_THREAD_GLOBAL_ALLFUNC_MUTEX);
    MPIR_FUNC_TERSE_ENTER(MPID_STATE_MPI_INEIGHBOR_ALLTOALLW);

    /* Validate parameters, especially handles needing to be converted */
#ifdef HAVE_ERROR_CHECKING
    {
        MPID_BEGIN_ERROR_CHECKS;
        {
            MPIR_ERRTEST_COMM(comm, mpi_errno);

            /* TODO more checks may be appropriate */
        }
        MPID_END_ERROR_CHECKS;
    }
#endif /* HAVE_ERROR_CHECKING */

    /* Convert MPI object handles to object pointers */
    MPIR_Comm_get_ptr(comm, comm_ptr);
    MPIR_Assert(comm_ptr != NULL);

    /* Validate parameters and objects (post conversion) */
#ifdef HAVE_ERROR_CHECKING
    {
        MPID_BEGIN_ERROR_CHECKS;
        {
            MPIR_Comm_valid_ptr(comm_ptr, mpi_errno, FALSE);
            MPIR_ERRTEST_ARGNULL(request, "request", mpi_errno);
            /* TODO more checks may be appropriate (counts, in_place, buffer aliasing, etc) */
            if (mpi_errno != MPI_SUCCESS)
                goto fn_fail;
        }
        MPID_END_ERROR_CHECKS;
    }
#endif /* HAVE_ERROR_CHECKING */

    /* ... body of routine ...  */

    mpi_errno =
        MPIR_Ineighbor_alltoallw(sendbuf, sendcounts, sdispls, sendtypes, recvbuf, recvcounts,
                                 rdispls, recvtypes, comm_ptr, &request_ptr);
    if (mpi_errno)
        MPIR_ERR_POP(mpi_errno);

    /* create a complete request, if needed */
    if (!request_ptr)
        request_ptr = MPIR_Request_create_complete(MPIR_REQUEST_KIND__COLL);
    /* return the handle of the request to the user */
    *request = request_ptr->handle;

    /* ... end of body of routine ... */

  fn_exit:
    MPIR_FUNC_TERSE_EXIT(MPID_STATE_MPI_INEIGHBOR_ALLTOALLW);
    MPID_THREAD_CS_EXIT(GLOBAL, MPIR_THREAD_GLOBAL_ALLFUNC_MUTEX);
    return mpi_errno;

  fn_fail:
    /* --BEGIN ERROR HANDLING-- */
#ifdef HAVE_ERROR_CHECKING
    {
        mpi_errno =
            MPIR_Err_create_code(mpi_errno, MPIR_ERR_RECOVERABLE, FCNAME, __LINE__, MPI_ERR_OTHER,
                                 "**mpi_ineighbor_alltoallw",
                                 "**mpi_ineighbor_alltoallw %p %p %p %p %p %p %p %p %C %p", sendbuf,
                                 sendcounts, sdispls, sendtypes, recvbuf, recvcounts, rdispls,
                                 recvtypes, comm, request);
    }
#endif
    mpi_errno = MPIR_Err_return_comm(NULL, FCNAME, mpi_errno);
    goto fn_exit;
    /* --END ERROR HANDLING-- */
}