src/mdlib/fft5d.h

   1 #ifndef FFT5D_H_
   2 #define FFT5D_H_
   3
   4 #include <complex.h>
   5
   6 #ifdef HAVE_CONFIG_H
   7 #include <config.h>
   8 #endif
   9
  10 #ifdef NOGMX
  11 /*#define GMX_MPI*/
  12 /*#define GMX_FFT_FFTW3*/
  13 FILE* debug;
  14 #endif
  15
  16 #ifdef GMX_LIB_MPI
  17 #include <mpi.h>
  18 #elif defined GMX_THREADS
  19 #include "tmpi.h"
  20 #else
  21 #ifndef HAVE_MPICOMM
  22 #ifndef _gmx_parallel_3dfft_h_
  23 typedef void* MPI_Comm;
  24 #endif
  25 #define HAVE_MPICOMM
  26 #endif
  27 double MPI_Wtime();
  28 #endif
  29 #ifdef GMX_THREADS
  30 #include "tmpi.h"
  31 #endif
  32
  33 #include "gmx_fft.h"
  34
  35 #ifdef GMX_FFT_FFTW3
  36 #include <fftw3.h>
  37 #ifdef FFT5D_MPI_TRANSPOSE
  38 #include <fftw3-mpi.h>
  39 #endif
  40 #endif
  41 /* TODO: optional wrapper
  42 #ifdef GMX_FFT_MKL
  43 #include <fftw/fftw3.h>
  44 #ifdef FFT5D_MPI_TRANSPOSE
  45 #include <fftw/fftw3-mpi.h>
  46 #endif
  47 #endif
  48 */
  49
  50 #ifndef NOGMX
  51 #ifndef GMX_DOUBLE  /*TODO how to not how have to do this GMX specific in here?
  52 can't be in gmx_parallel_3dfft.h because it has to be also be set when included from fft5d.c */
  53 #define FFT5D_SINGLE
  54 #endif
  55 #endif
  56
  57 typedef t_complex fft5d_type;
  58 #ifdef FFT5D_SINGLE
  59 #define FFTW(x) fftwf_##x
  60 typedef float fft5d_rtype;
  61 #define FFT5D_EPS __FLT_EPSILON__
  62 #define FFT5D_MPI_RTYPE MPI_FLOAT
  63 #else
  64 #define FFTW(x) fftw_##x
  65 typedef double fft5d_rtype;
  66 #define FFT5D_EPS __DBL_EPSILON__
  67 #define FFT5D_MPI_RTYPE MPI_DOUBLE
  68 #endif
  69
  70 struct fft5d_time_t {
  71         double fft,local,mpi1,mpi2;
  72 };
  73 typedef struct fft5d_time_t *fft5d_time;
  74
  75 typedef enum fft5d_flags_t {
  76         FFT5D_ORDER_YZ=1,
  77         FFT5D_BACKWARD=2,
  78         FFT5D_REALCOMPLEX=4,
  79         FFT5D_DEBUG=8,
  80         FFT5D_NOMEASURE=16,
  81         FFT5D_INPLACE=32,
  82         FFT5D_NOMALLOC=64
  83 } fft5d_flags;
  84
  85 struct fft5d_plan_t {
  86         fft5d_type *lin;
  87         fft5d_type *lout;
  88         gmx_fft_t p1d[3];   /*1D plans*/
  89 #ifdef GMX_FFT_FFTW3
  90         FFTW(plan) p2d;  /*2D plan: used for 1D decomposition if FFT supports transposed output*/
  91         FFTW(plan) p3d;  /*3D plan: used for 0D decomposition if FFT supports transposed output*/
  92 #ifdef FFT5D_MPI_TRANSPOSE
  93         FFTW(plan) mpip[2];
  94 #endif
  95 #endif
  96         MPI_Comm cart[2];
  97
  98     int N[3],M[3],K[3]; /*local length in transposed coordinate system (if not divisisable max)*/
  99     int pN[3],pM[3], pK[3]; /*local length - not max but length for this processor*/
 100     int oM[3],oK[3]; /*offset for current processor*/
 101     int *iNin[3],*oNin[3],*iNout[3],*oNout[3]; /*size for each processor (if divisisable=max) for out(=split)
 102                                                  and in (=join) and offsets in transposed coordinate system*/
 103     int C[3],rC[3]; /*global length (of the one global axes) */
 104     /* C!=rC for real<->complex. then C=rC/2 but with potential padding*/
 105     int P[2]; /*size of processor grid*/
 106 /*      int fftorder;*/
 107 /*      int direction;*/
 108 /*      int realcomplex;*/
 109         int flags;
 110     /*int N0,N1,M0,M1,K0,K1;*/
 111         int NG,MG,KG;
 112     /*int P[2];*/
 113         int coor[2];
 114 };
 115
 116 typedef struct fft5d_plan_t *fft5d_plan;
 117
 118 void fft5d_execute(fft5d_plan plan,fft5d_time times);
 119 fft5d_plan fft5d_plan_3d(int N, int M, int K, MPI_Comm comm[2], int flags, fft5d_type** lin, fft5d_type** lin2);
 120 void fft5d_local_size(fft5d_plan plan,int* N1,int* M0,int* K0,int* K1,int** coor);
 121 void fft5d_destroy(fft5d_plan plan);
 122 fft5d_plan fft5d_plan_3d_cart(int N, int M, int K, MPI_Comm comm, int P0, int flags, fft5d_type** lin, fft5d_type** lin2);
 123 void fft5d_compare_data(const fft5d_type* lin, const fft5d_type* in, fft5d_plan plan, int bothLocal, int normarlize);
 124 #endif /*FFTLIB_H_*/