include/gmx_parallel_3dfft.h

   1 /* -*- mode: c; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4; c-file-style: "stroustrup"; -*-
   2  *
   3  *
   4  * Gromacs                               Copyright (c) 1991-2005
   5  * David van der Spoel, Erik Lindahl, University of Groningen.
   6  *
   7  * This program is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU General Public License
   9  * as published by the Free Software Foundation; either version 2
  10  * of the License, or (at your option) any later version.
  11  *
  12  * To help us fund GROMACS development, we humbly ask that you cite
  13  * the research papers on the package. Check out http://www.gromacs.org
  14  *
  15  * And Hey:
  16  * Gnomes, ROck Monsters And Chili Sauce
  17  */
  18
  19 #ifndef _gmx_parallel_3dfft_h_
  20 #define _gmx_parallel_3dfft_h_
  21
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26
  27 #include "types/simple.h"
  28 #include "gmxcomplex.h"
  29 #include "gmx_fft.h"
  30
  31 /* We NEED MPI Comm here. But if no MPI is used then void pointer for MPI_Comm is enough*/
  32 /* This saves on many ifdefs */
  33 #ifdef GMX_LIB_MPI
  34 #include <mpi.h>
  35 #elif defined GMX_THREADS
  36 #include "tmpi.h"
  37 #else
  38 #ifndef FFT5D_H_
  39 typedef void* MPI_Comm;
  40 #endif
  41 #endif
  42
  43 typedef struct gmx_parallel_3dfft *
  44 gmx_parallel_3dfft_t;
  45
  46
  47
  48 /*! \brief Initialize parallel MPI-based 3D-FFT.
  49  *
  50  *  This routine performs real-to-complex and complex-to-real parallel 3D FFTs,
  51  *  but not complex-to-complex.
  52  *
  53  *  The routine is optimized for small-to-medium size FFTs used for PME and
  54  *  PPPM algorithms, and do allocate extra workspace whenever it might improve
  55  *  performance.
  56  *
  57  *  \param pfft_setup     Pointer to parallel 3dfft setup structure, previously
  58  *                        allocated or with automatic storage.
  59  *  \param ngridx         Global number of grid cells in the x direction. Must be
  60  *                        divisible by the number of nodes.
  61  *  \param ngridy         Global number of grid cells in the y direction. Must be
  62  *                        divisible by the number of nodes.
  63  *  \param ngridz         Global number of grid cells in the z direction.
  64  *  \param node2slab      Node id to slab index array, can be NULL.
  65  *  \param slab2grid_x    Slab index to grid_x array (nnodes+1), can be NULL.
  66  *  \param comm           MPI communicator, must have been initialized.
  67  *  \param bReproducible  Try to avoid FFT timing optimizations and other stuff
  68  *                        that could make results differ for two runs with
  69  *                        identical input (reproducibility for debugging).
  70  *
  71  *  \return 0 or a standard error code.
  72  */
  73 int
  74 gmx_parallel_3dfft_init   (gmx_parallel_3dfft_t *    pfft_setup,
  75                            ivec                      ndata,
  76                                                    real **                   real_data,
  77                                                    t_complex **              complex_data,
  78                            MPI_Comm                  comm[2],
  79                            int *                     slab2index_major,
  80                            int *                     slab2index_minor,
  81                            bool                      bReproducible);
  82
  83
  84
  85
  86
  87 /*! \brief Get direct space grid index limits
  88  */
  89 int
  90 gmx_parallel_3dfft_real_limits(gmx_parallel_3dfft_t      pfft_setup,
  91                                                            ivec                      local_ndata,
  92                                                            ivec                      local_offset,
  93                                                            ivec                      local_size);
  94
  95
  96 /*! \brief Get reciprocal space grid index limits
  97  */
  98 int
  99 gmx_parallel_3dfft_complex_limits(gmx_parallel_3dfft_t      pfft_setup,
 100                                   ivec                      complex_order,
 101                                                                   ivec                      local_ndata,
 102                                                                   ivec                      local_offset,
 103                                                                   ivec                      local_size);
 104
 105
 106 int
 107 gmx_parallel_3dfft_execute(gmx_parallel_3dfft_t    pfft_setup,
 108                                                    enum gmx_fft_direction  dir,
 109                                                    void *                  in_data,
 110                                                    void *                  out_data);
 111
 112
 113 /*! \brief Release all data in parallel fft setup
 114  *
 115  *  All temporary storage and FFT plans are released. The structure itself
 116  *  is not released, but the contents is invalid after this call.
 117  *
 118  *  \param pfft_setup Parallel 3dfft setup.
 119  *
 120  *  \return 0 or a standard error code.
 121  */
 122 int
 123 gmx_parallel_3dfft_destroy(gmx_parallel_3dfft_t    pfft_setup);
 124
 125
 126
 127
 128 #endif /* _gmx_parallel_3dfft_h_ */
 129