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
  23 #include "types/simple.h"
  24 #include "types/commrec.h"
  25 #include "gmxcomplex.h"
  26 #include "gmx_fft.h"
  27
  28 typedef struct gmx_parallel_3dfft *
  29 gmx_parallel_3dfft_t;
  30
  31
  32
  33 /*! \brief Initialize parallel MPI-based 3D-FFT.
  34  *
  35  *  This routine performs real-to-complex and complex-to-real parallel 3D FFTs,
  36  *  but not complex-to-complex.
  37  *
  38  *  The routine is optimized for small-to-medium size FFTs used for PME and
  39  *  PPPM algorithms, and do allocate extra workspace whenever it might improve
  40  *  performance.
  41  *
  42  *  \param pfft_setup     Pointer to parallel 3dfft setup structure, previously
  43  *                        allocated or with automatic storage.
  44  *  \param ngridx         Global number of grid cells in the x direction. Must be
  45  *                        divisible by the number of nodes.
  46  *  \param ngridy         Global number of grid cells in the y direction. Must be
  47  *                        divisible by the number of nodes.
  48  *  \param ngridz         Global number of grid cells in the z direction.
  49  *  \param node2slab      Node id to slab index array, can be NULL.
  50  *  \param slab2grid_x    Slab index to grid_x array (nnodes+1), can be NULL.
  51  *  \param comm           MPI communicator, must have been initialized.
  52  *  \param bReproducible  Try to avoid FFT timing optimizations and other stuff
  53  *                        that could make results differ for two runs with
  54  *                        identical input (reproducibility for debugging).
  55  *  \param nthreads       Run in parallel using n threads
  56  *
  57  *  \return 0 or a standard error code.
  58  */
  59 int
  60 gmx_parallel_3dfft_init   (gmx_parallel_3dfft_t *    pfft_setup,
  61                            ivec                      ndata,
  62                            real **                   real_data,
  63                            t_complex **              complex_data,
  64                            MPI_Comm                  comm[2],
  65                            int *                     slab2index_major,
  66                            int *                     slab2index_minor,
  67                            gmx_bool                  bReproducible,
  68                            int                       nthreads);
  69
  70
  71
  72
  73
  74 /*! \brief Get direct space grid index limits
  75  */
  76 int
  77 gmx_parallel_3dfft_real_limits(gmx_parallel_3dfft_t      pfft_setup,
  78                                ivec                      local_ndata,
  79                                ivec                      local_offset,
  80                                ivec                      local_size);
  81
  82
  83 /*! \brief Get reciprocal space grid index limits
  84  */
  85 int
  86 gmx_parallel_3dfft_complex_limits(gmx_parallel_3dfft_t      pfft_setup,
  87                                   ivec                      complex_order,
  88                                   ivec                      local_ndata,
  89                                   ivec                      local_offset,
  90                                   ivec                      local_size);
  91
  92
  93 int
  94 gmx_parallel_3dfft_execute(gmx_parallel_3dfft_t    pfft_setup,
  95                            enum gmx_fft_direction  dir,
  96                            void *                  in_data,
  97                            void *                  out_data,
  98                            int                     thread,
  99                            gmx_wallcycle_t         wcycle);
 100
 101
 102 /*! \brief Release all data in parallel fft setup
 103  *
 104  *  All temporary storage and FFT plans are released. The structure itself
 105  *  is not released, but the contents is invalid after this call.
 106  *
 107  *  \param pfft_setup Parallel 3dfft setup.
 108  *  \param in_data    Input data.
 109  *  \param out_data   Output data.
 110  *  \param thread     Thread index of the calling thread, i.e. index to the part
 111  *                    of the data operated on last by the calling thread. This
 112  *                    is needed to start the FFT without an OpenMP barrier.
 113  *  \param wcycle     Wall cycle counters.
 114  *
 115  *  \return 0 or a standard error code.
 116  */
 117 int
 118 gmx_parallel_3dfft_destroy(gmx_parallel_3dfft_t    pfft_setup);
 119
 120
 121
 122
 123 #endif /* _gmx_parallel_3dfft_h_ */
 124