GsLib.hpp

Go to the documentation of this file.

///////////////////////////////////////////////////////////////////////////////
//
// File GsLib.hpp
//
// For more information, please see: http://www.nektar.info
//
// The MIT License
//
// Copyright (c) 2006 Division of Applied Mathematics, Brown University (USA),
// Department of Aeronautics, Imperial College London (UK), and Scientific
// Computing and Imaging Institute, University of Utah (USA).
//
// License for the specific language governing rights and limitations under
// Permission is hereby granted, free of charge, to any person obtaining a
// copy of this software and associated documentation files (the "Software"),
// to deal in the Software without restriction, including without limitation
// the rights to use, copy, modify, merge, publish, distribute, sublicense,
// and/or sell copies of the Software, and to permit persons to whom the
// Software is furnished to do so, subject to the following conditions:
//
// The above copyright notice and this permission notice shall be included
// in all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
// OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
// THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
// FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
// DEALINGS IN THE SOFTWARE.
//
// Description: wrapper of functions around GSLib routines
//
///////////////////////////////////////////////////////////////////////////////

#ifndef NEKTAR_LIB_UTILITIES_COMMUNICATION_GSLIB_HPP
#define NEKTAR_LIB_UTILITIES_COMMUNICATION_GSLIB_HPP

#include <iostream>

#include <LibUtilities/BasicConst/NektarUnivTypeDefs.hpp>
#include <LibUtilities/BasicUtils/SharedArray.hpp>
#ifdef NEKTAR_USE_MPI
#include <LibUtilities/Communication/CommMpi.h>
#endif

namespace Gs
{
using namespace std;
using namespace Nektar;
typedef enum { gs_double, gs_float, gs_int, gs_long, gs_dom_n } gs_dom;
typedef enum { gs_add, gs_mul, gs_min, gs_max, gs_bpr, gs_op_n } gs_op;
typedef enum { mode_plain, mode_vec, mode_many, mode_dry_run } gs_mode;

typedef struct
{
    void *ptr;
    size_t n, max;
} array;
typedef array buffer;
#ifdef NEKTAR_USE_MPI
typedef MPI_Comm comm_ext;
typedef MPI_Request comm_req;
#else
typedef int comm_ext;
typedef int comm_req;
#endif

struct comm
{
    unsigned int id;
    unsigned int np;
    comm_ext c;
};

typedef struct
{
    unsigned int n;     /* number of messages */
    unsigned int *p;    /* message source/dest proc */
    unsigned int *size; /* size of message */
    unsigned int total; /* sum of message sizes */
} pw_comm_data;

typedef struct
{
    pw_comm_data comm[2];
    const unsigned int *map[2];
    comm_req *req;
    unsigned int buffer_size;
} pw_data;

typedef struct
{
    const unsigned int *scatter_map, *gather_map;
    unsigned int size_r, size_r1, size_r2;
    unsigned int size_sk, size_s, size_total;
    unsigned int p1, p2;
    unsigned int nrecvn;
} cr_stage;

typedef struct
{
    cr_stage *stage[2];
    unsigned int nstages;
    unsigned int buffer_size, stage_buffer_size;
} cr_data;

typedef struct
{
    const unsigned int *map_to_buf[2], *map_from_buf[2];
    unsigned int buffer_size;
} allreduce_data;

typedef void exec_fun(void *data, gs_mode mode, unsigned vn, gs_dom dom,
                      gs_op op, unsigned transpose, const void *execdata,
                      const struct comm *comm, char *buf);
typedef void fin_fun(void *data);

typedef struct
{
    unsigned int buffer_size, mem_size;
    void *data;
    exec_fun *exec;
    fin_fun *fin;
} gs_remote;

typedef struct
{
    struct comm comm;
    const unsigned int *map_local[2]; /* 0=unflagged, 1=all */
    const unsigned int *flagged_primaries;
    gs_remote r;
    unsigned int handle_size;
} gs_data;

typedef enum {
    gs_auto,
    gs_pairwise,
    gs_crystal_router,
    gs_all_reduce
} gs_method;

extern "C" {
void nektar_gs(void *u, gs_dom dom, gs_op op, unsigned transpose, gs_data *gsh,
               buffer *buf);
gs_data *nektar_gs_setup(const long *id, unsigned int n,
                         const struct comm *comm, int unique, gs_method method,
                         int verbose);
void nektar_gs_free(gs_data *gsh);
void nektar_gs_unique(const long *id, unsigned int n, const struct comm *comm);
}

/**
 * @brief Initialise Gather-Scatter map.
 *
 * On each process an array of IDs for each global degree of freedom is
 * supplied which corresponds to a unique numbering of universal degrees of
 * freedom. This is used to initialise the GSLib mapping between process-
 * boundary degrees of freedom on different processes.
 * @param   pId         Array of integers providing universal IDs for each
 *                      global DOF on the process.
 * @param   pComm       Communication object used for inter-process
 *                      communication.
 * @return GSLib data structure containing mapping information.
 */
static inline gs_data *Init(const Nektar::Array<OneD, long> pId,
                            const LibUtilities::CommSharedPtr &pComm,
                            bool verbose = true)
{
#ifdef NEKTAR_USE_MPI
    if (pComm->GetSize() == 1)
    {
        return 0;
    }
    LibUtilities::CommMpiSharedPtr vCommMpi =
        boost::dynamic_pointer_cast<LibUtilities::CommMpi>(pComm);
    ASSERTL1(vCommMpi, "Failed to cast MPI Comm object.");
    comm vComm;
    MPI_Comm_dup(vCommMpi->GetComm(), &vComm.c);
    vComm.id        = vCommMpi->GetRank();
    vComm.np        = vCommMpi->GetSize();
    gs_data *result = nektar_gs_setup(pId.get(), pId.num_elements(), &vComm, 0,
                                      gs_auto, (int)verbose);
    MPI_Comm_free(&vComm.c);
    return result;
#else
    return 0;
#endif
}

/**
 * @brief Updates pId to negate all-but-one references to each universal ID.
 *
 * The array of universal IDs corresponding to the process-local DOF are
 * updated such that the ID of only one instance of each universal ID
 * remains positive. This allows the consistent formulation of universally
 * -distributed dot products, for which the contributions of each DOF must
 * be included only once.
 */
static inline void Unique(const Nektar::Array<OneD, long> pId,
                          const LibUtilities::CommSharedPtr &pComm)
{
#ifdef NEKTAR_USE_MPI
    if (pComm->GetSize() == 1)
    {
        return;
    }
    LibUtilities::CommMpiSharedPtr vCommMpi =
        boost::dynamic_pointer_cast<LibUtilities::CommMpi>(pComm);
    ASSERTL1(vCommMpi, "Failed to cast MPI Comm object.");
    comm vComm;
    vComm.c  = vCommMpi->GetComm();
    vComm.id = vCommMpi->GetRank();
    vComm.np = vCommMpi->GetSize();
    nektar_gs_unique(pId.get(), pId.num_elements(), &vComm);
#endif
}

/**
 * @brief Deallocates the GSLib mapping data.
 */
static inline void Finalise(gs_data *pGsh)
{
#ifdef NEKTAR_USE_MPI
    int finalized;
    MPI_Finalized(&finalized);
    if (pGsh && !finalized)
    {
        nektar_gs_free(pGsh);
    }
#endif
}

/**
 * @brief Performs a gather-scatter operation of the provided values.
 *
 * The
 */
static inline void Gather(
    Nektar::Array<OneD, NekDouble> pU, gs_op pOp, gs_data *pGsh,
    Nektar::Array<OneD, NekDouble> pBuffer = NullNekDouble1DArray)
{
#ifdef NEKTAR_USE_MPI
    if (!pGsh)
    {
        return;
    }
    if (pBuffer.num_elements() == 0)
    {
        nektar_gs(pU.get(), gs_double, pOp, false, pGsh, 0);
    }
    else
    {
        array buf;
        buf.ptr = &pBuffer[0];
        buf.n   = pBuffer.num_elements();
        nektar_gs(pU.get(), gs_double, pOp, false, pGsh, &buf);
    }
#endif
}
}

#endif