int main(int argc, char **argv)
{
int np[2];
ptrdiff_t n[3];
ptrdiff_t alloc_local;
ptrdiff_t local_ni[3], local_i_start[3];
ptrdiff_t local_no[3], local_o_start[3];
long double err;
pfftl_complex *in, *out;
pfftl_plan plan_forw=NULL, plan_back=NULL;
MPI_Comm comm_cart_2d;
/* Set size of FFT and process mesh */
n[0] = 29;
n[1] = 27;
n[2] = 31;
np[0] = 2;
np[1] = 2;
/* Initialize MPI and PFFT */
MPI_Init(&argc, &argv);
pfftl_init();
/* Create two-dimensional process grid of size np[0] x np[1], if possible */
if( pfftl_create_procmesh_2d(MPI_COMM_WORLD, np[0], np[1], &comm_cart_2d) ) {
pfftl_fprintf(MPI_COMM_WORLD, stderr, "Error: This test file only works with %d processes.\n", np[0]*np[1]);
MPI_Finalize();
return 1;
}
/* Get parameters of data distribution */
alloc_local = pfftl_local_size_dft_3d(n, comm_cart_2d, PFFT_TRANSPOSED_NONE,
local_ni, local_i_start, local_no, local_o_start);
/* Allocate memory */
in = pfftl_alloc_complex(alloc_local);
out = pfftl_alloc_complex(alloc_local);
/* Plan parallel forward FFT */
plan_forw = pfftl_plan_dft_3d(
n, in, out, comm_cart_2d, PFFT_FORWARD, PFFT_TRANSPOSED_NONE| PFFT_MEASURE| PFFT_DESTROY_INPUT);
/* Plan parallel backward FFT */
plan_back = pfftl_plan_dft_3d(
n, out, in, comm_cart_2d, PFFT_BACKWARD, PFFT_TRANSPOSED_NONE| PFFT_MEASURE| PFFT_DESTROY_INPUT);
/* Initialize input with random numbers */
pfftl_init_input_complex_3d(n, local_ni, local_i_start,
in);
/* execute parallel forward FFT */
pfftl_execute(plan_forw);
/* execute parallel backward FFT */
pfftl_execute(plan_back);
/* Scale data */
ptrdiff_t l;
for(l=0; l < local_ni[0] * local_ni[1] * local_ni[2]; l++)
in[l] /= (n[0]*n[1]*n[2]);
/* Print error of back transformed data */
err = pfftl_check_output_complex_3d(n, local_ni, local_i_start, in, comm_cart_2d);
pfftl_printf(comm_cart_2d, "Error after one forward and backward trafo of size n=(%td, %td, %td):\n", n[0], n[1], n[2]);
pfftl_printf(comm_cart_2d, "maxerror = %6.2Le;\n", err);
/* free mem and finalize */
pfftl_destroy_plan(plan_forw);
pfftl_destroy_plan(plan_back);
MPI_Comm_free(&comm_cart_2d);
pfftl_free(in);
pfftl_free(out);
MPI_Finalize();
return 0;
}
static void pnfft_perform_guru(
const ptrdiff_t *N, const ptrdiff_t *n, ptrdiff_t local_M,
int m, const long double *x_max, unsigned window_flag,
const int *np, MPI_Comm comm
)
{
int myrank;
ptrdiff_t local_N[3], local_N_start[3];
long double lower_border[3], upper_border[3];
long double local_sum = 0;
double time, time_max;
MPI_Comm comm_cart_3d;
pnfftl_complex *f_hat, *f, *f1;
long double *x, f_hat_sum;
pnfftl_plan pnfft;
/* create three-dimensional process grid of size np[0] x np[1] x np[2], if possible */
if( pnfftl_create_procmesh(3, comm, np, &comm_cart_3d) ){
pfftl_fprintf(comm, stderr, "Error: Procmesh of size %d x %d x %d does not fit to number of allocated processes.\n", np[0], np[1], np[2]);
pfftl_fprintf(comm, stderr, " Please allocate %d processes (mpiexec -np %d ...) or change the procmesh (with -pnfft_np * * *).\n", np[0]*np[1]*np[2], np[0]*np[1]*np[2]);
MPI_Finalize();
exit(1);
}
MPI_Comm_rank(comm_cart_3d, &myrank);
/* get parameters of data distribution */
pnfftl_local_size_guru(3, N, n, x_max, m, comm_cart_3d, PNFFT_TRANSPOSED_NONE,
local_N, local_N_start, lower_border, upper_border);
/* plan parallel NFFT */
pnfft = pnfftl_init_guru(3, N, n, x_max, local_M, m,
PNFFT_MALLOC_X| PNFFT_MALLOC_F_HAT| PNFFT_MALLOC_F| window_flag, PFFT_ESTIMATE,
comm_cart_3d);
/* get data pointers */
f_hat = pnfftl_get_f_hat(pnfft);
f = pnfftl_get_f(pnfft);
x = pnfftl_get_x(pnfft);
/* initialize Fourier coefficients */
pnfftl_init_f_hat_3d(N, local_N, local_N_start, PNFFT_TRANSPOSED_NONE,
f_hat);
/* initialize nonequispaced nodes */
srand(myrank);
init_random_x(lower_border, upper_border, x_max, local_M,
x);
/* execute parallel NFFT */
time = -MPI_Wtime();
pnfftl_trafo(pnfft);
time += MPI_Wtime();
/* print timing */
MPI_Reduce(&time, &time_max, 1, MPI_DOUBLE, MPI_MAX, 0, comm);
pfftl_printf(comm, "pnfftl_trafo needs %6.2e s\n", time_max);
/* calculate norm of Fourier coefficients for calculation of relative error */
for(ptrdiff_t k=0; k<local_N[0]*local_N[1]*local_N[2]; k++)
local_sum += cabsl(f_hat[k]);
MPI_Allreduce(&local_sum, &f_hat_sum, 1, MPI_LONG_DOUBLE, MPI_SUM, comm_cart_3d);
/* store results of NFFT */
f1 = pnfftl_alloc_complex(local_M);
for(ptrdiff_t j=0; j<local_M; j++) f1[j] = f[j];
/* execute parallel NDFT */
time = -MPI_Wtime();
pnfftl_direct_trafo(pnfft);
time += MPI_Wtime();
/* print timing */
MPI_Reduce(&time, &time_max, 1, MPI_DOUBLE, MPI_MAX, 0, comm);
pfftl_printf(comm, "pnfftl_direct_trafo needs %6.2e s\n", time_max);
/* calculate error of PNFFT */
compare_f(f1, f, local_M, f_hat_sum, "* Results in", MPI_COMM_WORLD);
/* free mem and finalize */
pnfftl_free(f1);
pnfftl_finalize(pnfft, PNFFT_FREE_X | PNFFT_FREE_F | PNFFT_FREE_F_HAT);
MPI_Comm_free(&comm_cart_3d);
}
