fftwnd_mpi.c

FFTW, a collection of fast C routines to compute the Discrete Fourier Transform in one or more dime

/*
 * Copyright (c) 1997-1999, 2003 Massachusetts Institute of Technology
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
 *
 */

#include <stdlib.h>

#include <mpi.h>

#include "fftw_mpi.h"

/***************************** Plan Creation ****************************/

fftwnd_mpi_plan fftwnd_mpi_create_plan(MPI_Comm comm,
				       int rank, const int *n,
				       fftw_direction dir,
				       int flags)
{
    fftwnd_mpi_plan p;

    if (rank < 2)
	return 0;

    p = (fftwnd_mpi_plan) fftw_malloc(sizeof(fftwnd_mpi_plan_data));
    p->p_fft_x = 0;
    p->p_fft = 0;
    p->p_transpose = 0;
    p->p_transpose_inv = 0;
    p->work = 0;

    p->p_fft_x = fftw_create_plan(n[0], dir, flags | FFTW_IN_PLACE);

    p->p_fft = fftwnd_create_plan(rank - 1, n + 1, dir, flags | FFTW_IN_PLACE);
    if (!p->p_fft)
	fftwnd_mpi_destroy_plan(p);

    p->p_transpose = transpose_mpi_create_plan(n[0], n[1], comm);
    if (!p->p_transpose)
	fftwnd_mpi_destroy_plan(p);

    p->p_transpose_inv = transpose_mpi_create_plan(n[1], n[0], comm);
    if (!p->p_transpose_inv)
	fftwnd_mpi_destroy_plan(p);

    if (n[0] > p->p_fft->nwork)
	 p->work = (fftw_complex *) fftw_malloc(n[0] * sizeof(fftw_complex));

    return p;
}

fftwnd_mpi_plan fftw2d_mpi_create_plan(MPI_Comm comm,
				       int nx, int ny,
				       fftw_direction dir, int flags)
{
    int n[2];

    n[0] = nx;
    n[1] = ny;

    return fftwnd_mpi_create_plan(comm, 2, n, dir, flags);
}

fftwnd_mpi_plan fftw3d_mpi_create_plan(MPI_Comm comm,
				       int nx, int ny, int nz,
				       fftw_direction dir, int flags)
{
    int n[3];

    n[0] = nx;
    n[1] = ny;
    n[2] = nz;

    return fftwnd_mpi_create_plan(comm, 3, n, dir, flags);
}

/********************** Plan Destruction ************************/

void fftwnd_mpi_destroy_plan(fftwnd_mpi_plan p)
{
    if (p) {
	if (p->p_fft_x)
	    fftw_destroy_plan(p->p_fft_x);
	if (p->p_fft)
	    fftwnd_destroy_plan(p->p_fft);
	if (p->p_transpose)
	    transpose_mpi_destroy_plan(p->p_transpose);
	if (p->p_transpose_inv)
	    transpose_mpi_destroy_plan(p->p_transpose_inv);
	if (p->work)
	     fftw_free(p->work);
	fftw_free(p);
    }
}

void fftw_mpi_die(const char *error_string)
{
     int my_pe;

     MPI_Comm_rank(MPI_COMM_WORLD, &my_pe);

     fprintf(stderr, "fftw process %d: %s", my_pe, error_string);

     MPI_Abort(MPI_COMM_WORLD, EXIT_FAILURE);
}

/********************* Getting Local Size ***********************/

void fftwnd_mpi_local_sizes(fftwnd_mpi_plan p,
			    int *local_nx,
			    int *local_x_start,
			    int *local_ny_after_transpose,
			    int *local_y_start_after_transpose,
			    int *total_local_size)
{
    if (p) {
	transpose_mpi_get_local_size(p->p_transpose->nx,
				     p->p_transpose->my_pe,
				     p->p_transpose->n_pes,
				     local_nx,
				     local_x_start);
	transpose_mpi_get_local_size(p->p_transpose->ny,
				     p->p_transpose->my_pe,
				     p->p_transpose->n_pes,
				     local_ny_after_transpose,
				     local_y_start_after_transpose);
	*total_local_size =
	    transpose_mpi_get_local_storage_size(p->p_transpose->nx,
						 p->p_transpose->ny,
						 p->p_transpose->my_pe,
						 p->p_transpose->n_pes);

	*total_local_size *= p->p_fft->n_after[0];
    }
}

/******************** Computing the Transform *******************/

void fftwnd_mpi(fftwnd_mpi_plan p,
		int n_fields, fftw_complex *local_data, fftw_complex *work,
		fftwnd_mpi_output_order output_order)
{
     int el_size = (sizeof(fftw_complex) / sizeof(TRANSPOSE_EL_TYPE))
	           * n_fields * p->p_fft->n_after[0];
     
     if (n_fields <= 0)
	  return;

     /* First, transform dimensions after the first, which are
	local to this process: */

     {
	  int local_nx = p->p_transpose->local_nx;
	  int n_after_x = p->p_fft->n[0] * p->p_fft->n_after[0];

	  if (n_fields > 1) {
	       fftwnd_plan p_fft = p->p_fft;
	       int fft_iter;
	       for (fft_iter = 0; fft_iter < local_nx; ++fft_iter)
		    fftwnd(p_fft, n_fields,
			   local_data + (n_after_x * n_fields) * fft_iter,
			   n_fields, 1,
			   NULL, 0, 0);
	  }
	  else
	       fftwnd(p->p_fft, local_nx,
		      local_data, 1, n_after_x, NULL, 0, 0);
     }
     
     /* Second, transpose the first dimension with the second dimension
	to bring the x dimension local to this process: */

     transpose_mpi(p->p_transpose, el_size, 
		   (TRANSPOSE_EL_TYPE *) local_data,
		   (TRANSPOSE_EL_TYPE *) work);

     /* Third, transform the x dimension, which is now local and contiguous: */
     
     n_fields *= p->p_fft->n_after[0]; /* dimensions after y 
					  no longer need be considered
					  separately from n_fields */
     {
	  int local_ny = p->p_transpose->local_ny;
	  int nx = p->p_fft_x->n;
	  fftw_complex *work_1d = p->work ? p->work : p->p_fft->work;

	  if (n_fields > 1) {
	       fftw_plan p_fft_x = p->p_fft_x;
	       int fft_iter;
	       for (fft_iter = 0; fft_iter < local_ny; ++fft_iter)
		    fftw(p_fft_x, n_fields,
			 local_data + (nx * n_fields) * fft_iter, n_fields, 1,
			 work_1d, 1, 0);
	  }
	  else
	       fftw(p->p_fft_x, local_ny,
		    local_data, 1, nx, work_1d, 1, 0);
     }
     
     /* transpose back, if desired: */
     if (output_order == FFTW_NORMAL_ORDER)
	  transpose_mpi(p->p_transpose_inv, el_size,
			(TRANSPOSE_EL_TYPE *) local_data,
			(TRANSPOSE_EL_TYPE *) work);
}