1 /* -*- mode: c; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4; c-file-style: "stroustrup"; -*-
3 * This source code is part of
7 * GROningen MAchine for Chemical Simulations
9 * Written by David van der Spoel, Erik Lindahl, Berk Hess, and others.
10 * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
11 * Copyright (c) 2001-2012, The GROMACS development team,
12 * check out http://www.gromacs.org for more information.
14 * This program is free software; you can redistribute it and/or
15 * modify it under the terms of the GNU General Public License
16 * as published by the Free Software Foundation; either version 2
17 * of the License, or (at your option) any later version.
19 * If you want to redistribute modifications, please consider that
20 * scientific software is very special. Version control is crucial -
21 * bugs must be traceable. We will be happy to consider code for
22 * inclusion in the official distribution, but derived work must not
23 * be called official GROMACS. Details are found in the README & COPYING
24 * files - if they are missing, get the official version at www.gromacs.org.
26 * To help us fund GROMACS development, we humbly ask that you cite
27 * the papers on the package - you can find them in the top README file.
29 * For more info, check our website at http://www.gromacs.org
32 * Groningen Machine for Chemical Simulation
48 /*#define GMX_FFT_FFTW3*/
52 #include "gromacs/legacyheaders/gmxcomplex.h"
53 #include "gromacs/utility/gmxmpi.h"
54 #include "gromacs/fft/fft.h"
60 /*currently only special optimization for FFTE*/
66 #define FFTW(x) fftwf_ ## x
68 #define FFTW(x) fftw_ ## x
73 double fft, local, mpi1, mpi2;
75 typedef struct fft5d_time_t *fft5d_time;
77 #include "gromacs/timing/wallcycle.h"
78 typedef gmx_wallcycle_t fft5d_time;
81 typedef enum fft5d_flags_t {
84 FFT5D_REALCOMPLEX = 4,
93 t_complex *lout, *lout2, *lout3;
94 gmx_fft_t* p1d[3]; /*1D plans*/
96 FFTW(plan) p2d; /*2D plan: used for 1D decomposition if FFT supports transposed output*/
97 FFTW(plan) p3d; /*3D plan: used for 0D decomposition if FFT supports transposed output*/
102 int N[3], M[3], K[3]; /*local length in transposed coordinate system (if not divisisable max)*/
103 int pN[3], pM[3], pK[3]; /*local length - not max but length for this processor*/
104 int oM[3], oK[3]; /*offset for current processor*/
105 int *iNin[3], *oNin[3], *iNout[3], *oNout[3]; /*size for each processor (if divisisable=max) for out(=split)
106 and in (=join) and offsets in transposed coordinate system*/
107 int C[3], rC[3]; /*global length (of the one global axes) */
108 /* C!=rC for real<->complex. then C=rC/2 but with potential padding*/
109 int P[2]; /*size of processor grid*/
112 /* int realcomplex;*/
114 /*int N0,N1,M0,M1,K0,K1;*/
121 typedef struct fft5d_plan_t *fft5d_plan;
123 void fft5d_execute(fft5d_plan plan, int thread, fft5d_time times);
124 fft5d_plan fft5d_plan_3d(int N, int M, int K, MPI_Comm comm[2], int flags, t_complex**lin, t_complex**lin2, t_complex**lout2, t_complex**lout3, int nthreads);
125 void fft5d_local_size(fft5d_plan plan, int* N1, int* M0, int* K0, int* K1, int** coor);
126 void fft5d_destroy(fft5d_plan plan);
127 fft5d_plan fft5d_plan_3d_cart(int N, int M, int K, MPI_Comm comm, int P0, int flags, t_complex** lin, t_complex** lin2, t_complex** lout2, t_complex** lout3, int nthreads);
128 void fft5d_compare_data(const t_complex* lin, const t_complex* in, fft5d_plan plan, int bothLocal, int normarlize);