src/gromacs/fft/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_FFT_PARALLEL_3DFFT_H
  20 #define GMX_FFT_PARALLEL_3DFFT_H
  21
  22 #include "../legacyheaders/types/nrnb.h"
  23 #include "../legacyheaders/types/simple.h"
  24 #include "../legacyheaders/gmxcomplex.h"
  25 #include "../utility/gmxmpi.h"
  26 #include "fft.h"
  27
  28 #ifdef __cplusplus
  29 extern "C" {
  30 #endif
  31
  32 typedef struct gmx_parallel_3dfft *
  33     gmx_parallel_3dfft_t;
  34
  35
  36
  37 /*! \brief Initialize parallel MPI-based 3D-FFT.
  38  *
  39  *  This routine performs real-to-complex and complex-to-real parallel 3D FFTs,
  40  *  but not complex-to-complex.
  41  *
  42  *  The routine is optimized for small-to-medium size FFTs used for PME and
  43  *  PPPM algorithms, and do allocate extra workspace whenever it might improve
  44  *  performance.
  45  *
  46  *  \param pfft_setup     Pointer to parallel 3dfft setup structure, previously
  47  *                        allocated or with automatic storage.
  48  *  \param ndata          Number of grid cells in each direction
  49  *  \param real_data      Real data. Input for forward and output for backward.
  50  *  \param complex_data   Complex data.
  51  *  \param comm           MPI communicator for both parallelization axis.
  52  *                        Needs to be either initialized or MPI_NULL for
  53  *                        no parallelization in that axis.
  54  *  \param slab2index_major Not used
  55  *  \param slab2index_minor Not used
  56  *  \param bReproducible  Try to avoid FFT timing optimizations and other stuff
  57  *                        that could make results differ for two runs with
  58  *                        identical input (reproducibility for debugging).
  59  *  \param nthreads       Run in parallel using n threads
  60  *
  61  *  \return 0 or a standard error code.
  62  */
  63 int
  64     gmx_parallel_3dfft_init   (gmx_parallel_3dfft_t *    pfft_setup,
  65                                ivec                      ndata,
  66                                real **real_data,
  67                                t_complex **complex_data,
  68                                MPI_Comm                  comm[2],
  69                                int *                     slab2index_major,
  70                                int *                     slab2index_minor,
  71                                gmx_bool                  bReproducible,
  72                                int                       nthreads);
  73
  74
  75
  76
  77
  78 /*! \brief Get direct space grid index limits
  79  */
  80 int
  81 gmx_parallel_3dfft_real_limits(gmx_parallel_3dfft_t      pfft_setup,
  82                                ivec                      local_ndata,
  83                                ivec                      local_offset,
  84                                ivec                      local_size);
  85
  86
  87 /*! \brief Get reciprocal space grid index limits
  88  */
  89 int
  90 gmx_parallel_3dfft_complex_limits(gmx_parallel_3dfft_t      pfft_setup,
  91                                   ivec                      complex_order,
  92                                   ivec                      local_ndata,
  93                                   ivec                      local_offset,
  94                                   ivec                      local_size);
  95
  96
  97 int
  98 gmx_parallel_3dfft_execute(gmx_parallel_3dfft_t    pfft_setup,
  99                            enum gmx_fft_direction  dir,
 100                            void *                  in_data,
 101                            void *                  out_data,
 102                            int                     thread,
 103                            gmx_wallcycle_t         wcycle);
 104
 105
 106 /*! \brief Release all data in parallel fft setup
 107  *
 108  *  All temporary storage and FFT plans are released. The structure itself
 109  *  is not released, but the contents is invalid after this call.
 110  *
 111  *  \param pfft_setup Parallel 3dfft setup.
 112  *  \param in_data    Input data.
 113  *  \param out_data   Output data.
 114  *  \param thread     Thread index of the calling thread, i.e. index to the part
 115  *                    of the data operated on last by the calling thread. This
 116  *                    is needed to start the FFT without an OpenMP barrier.
 117  *  \param wcycle     Wall cycle counters.
 118  *
 119  *  \return 0 or a standard error code.
 120  */
 121 int
 122 gmx_parallel_3dfft_destroy(gmx_parallel_3dfft_t    pfft_setup);
 123
 124 #ifdef __cplusplus
 125 }
 126 #endif
 127
 128 #endif