/**
* reconstruct makes a 2d-adjoint-nfft
*/
static void reconstruct(char* filename, int N, int M, int weight)
{
int j; /* some variables */
double weights; /* store one weight temporary */
double real,imag; /* to read the real and imag part of a complex number */
nfft_plan my_plan; /* plan for the two dimensional nfft */
FILE* fin; /* input file */
FILE* fweight; /* input file for the weights */
FILE *fout_real; /* output file */
FILE *fout_imag; /* output file */
int my_N[2],my_n[2];
int flags = PRE_PHI_HUT| PRE_PSI |MALLOC_X| MALLOC_F_HAT|
MALLOC_F| FFTW_INIT| FFT_OUT_OF_PLACE|
FFTW_MEASURE| FFTW_DESTROY_INPUT;
/* initialise nfft */
my_N[0]=N; my_n[0]=ceil(N*1.2);
my_N[1]=N; my_n[1]=ceil(N*1.2);
nfft_init_guru(&my_plan, 2, my_N, M, my_n, 6,flags,
FFTW_MEASURE| FFTW_DESTROY_INPUT);
fin=fopen(filename,"r");
fweight=fopen("weights.dat","r");
for(j=0;j<my_plan.M_total;j++)
{
fscanf(fweight,"%le ",&weights);
fscanf(fin,"%le %le %le %le",&my_plan.x[2*j+0],&my_plan.x[2*j+1],&real,&imag);
my_plan.f[j] = real + _Complex_I*imag;
if (weight)
my_plan.f[j] = my_plan.f[j] * weights;
}
fclose(fweight);
/* precompute psi */
if(my_plan.nfft_flags & PRE_PSI)
nfft_precompute_psi(&my_plan);
/* precompute full psi */
if(my_plan.nfft_flags & PRE_FULL_PSI)
nfft_precompute_full_psi(&my_plan);
/* compute the adjoint nfft */
nfft_adjoint(&my_plan);
fout_real=fopen("output_real.dat","w");
fout_imag=fopen("output_imag.dat","w");
for (j=0;j<N*N;j++) {
fprintf(fout_real,"%le ",creal(my_plan.f_hat[j]));
fprintf(fout_imag,"%le ",cimag(my_plan.f_hat[j]));
}
fclose(fin);
fclose(fout_real);
fclose(fout_imag);
nfft_finalize(&my_plan);
}
/**
* reconstruct makes an 2d-adjoint-nfft for every slice
*/
static void reconstruct(char* filename,int N,int M,int Z, int weight ,fftw_complex *mem)
{
int j,k,z; /* some variables */
double weights; /* store one weight temporary */
double tmp; /* tmp to read the obsolent z from the input file */
double real,imag; /* to read the real and imag part of a complex number */
nfft_plan my_plan; /* plan for the two dimensional nfft */
int my_N[2],my_n[2]; /* to init the nfft */
FILE* fin; /* input file */
FILE* fweight; /* input file for the weights */
/* initialise my_plan */
my_N[0]=N; my_n[0]=ceil(N*1.2);
my_N[1]=N; my_n[1]=ceil(N*1.2);
nfft_init_guru(&my_plan, 2, my_N, M/Z, my_n, 6, PRE_PHI_HUT| PRE_PSI|
MALLOC_X| MALLOC_F_HAT| MALLOC_F|
FFTW_INIT| FFT_OUT_OF_PLACE,
FFTW_MEASURE| FFTW_DESTROY_INPUT);
/* precompute lin psi if set */
if(my_plan.nfft_flags & PRE_LIN_PSI)
nfft_precompute_lin_psi(&my_plan);
fin=fopen(filename,"r");
for(z=0;z<Z;z++) {
fweight=fopen("weights.dat","r");
for(j=0;j<my_plan.M_total;j++)
{
fscanf(fweight,"%le ",&weights);
fscanf(fin,"%le %le %le %le %le",
&my_plan.x[2*j+0],&my_plan.x[2*j+1],&tmp,&real,&imag);
my_plan.f[j] = real + _Complex_I*imag;
if(weight)
my_plan.f[j] = my_plan.f[j] * weights;
}
fclose(fweight);
/* precompute psi if set just one time because the knots equal each slice */
if(z==0 && my_plan.nfft_flags & PRE_PSI)
nfft_precompute_psi(&my_plan);
/* precompute full psi if set just one time because the knots equal each slice */
if(z==0 && my_plan.nfft_flags & PRE_FULL_PSI)
nfft_precompute_full_psi(&my_plan);
/* compute the adjoint nfft */
nfft_adjoint(&my_plan);
for(k=0;k<my_plan.N_total;k++) {
/* write every slice in the memory.
here we make an fftshift direct */
mem[(Z*N*N/2+z*N*N+ k)%(Z*N*N)] = my_plan.f_hat[k];
}
}
fclose(fin);
nfft_finalize(&my_plan);
}
static void simple_test_nfft_1d(void)
{
nfft_plan p;
double t;
int N=14;
int M=19;
ticks t0, t1;
/** init an one dimensional plan */
nfft_init_1d(&p,N,M);
/** init pseudo random nodes */
nfft_vrand_shifted_unit_double(p.x,p.M_total);
/** precompute psi, the entries of the matrix B */
if(p.nfft_flags & PRE_ONE_PSI)
nfft_precompute_one_psi(&p);
/** init pseudo random Fourier coefficients and show them */
nfft_vrand_unit_complex(p.f_hat,p.N_total);
nfft_vpr_complex(p.f_hat,p.N_total,"given Fourier coefficients, vector f_hat");
/** direct trafo and show the result */
t0 = getticks();
nfft_trafo_direct(&p);
t1 = getticks();
t = nfft_elapsed_seconds(t1,t0);
nfft_vpr_complex(p.f,p.M_total,"ndft, vector f");
printf(" took %e seconds.\n",t);
/** approx. trafo and show the result */
nfft_trafo(&p);
nfft_vpr_complex(p.f,p.M_total,"nfft, vector f");
/** approx. adjoint and show the result */
nfft_adjoint_direct(&p);
nfft_vpr_complex(p.f_hat,p.N_total,"adjoint ndft, vector f_hat");
/** approx. adjoint and show the result */
nfft_adjoint(&p);
nfft_vpr_complex(p.f_hat,p.N_total,"adjoint nfft, vector f_hat");
/** finalise the one dimensional plan */
nfft_finalize(&p);
}
void // space to frequency
mad_cmat_nfft (const cnum_t x[], const num_t x_node[], cnum_t r[], ssz_t m, ssz_t n, ssz_t nr)
{
assert( x && r );
int precomp = 0;
if (m != p_n1 || n != p_n2 || nr != p_m) {
nfft_finalize(&p);
nfft_init_2d (&p, m, n, nr);
p_n1 = m, p_n2 = n, p_m = nr, precomp = 1;
}
if (x_node || precomp) {
for (ssz_t i=0; i < m*n; i++) // adjoint transform needs -x_node
p.x[i] = x_node[i] == -0.5 ? 0.4999999999999999 : -x_node[i];
if(p.flags & PRE_ONE_PSI) nfft_precompute_one_psi(&p);
}
mad_cvec_copy(x, p.f, m*n);
const char *error_str = nfft_check(&p);
if (error_str) error("%s", error_str);
nfft_adjoint(&p); // nfft_adjoint_direct(&p);
// mad_cvec_copy(p.f_hat, r, nr);
mad_cvec_copy(p.f_hat+nr/2, r, nr/2); // for compatibility with FFTW ?? (TBC)
mad_cvec_copy(p.f_hat, r+nr/2, nr/2);
}
/**
* Executes the fast Gauss transform.
*
* \arg ths The pointer to a fgt plan
*
* \author Stefan Kunis
*/
void fgt_trafo(fgt_plan *ths)
{
int l;
if(ths->flags & FGT_NDFT)
{
nfft_adjoint_direct(ths->nplan1);
for(l=0; l<ths->n; l++)
ths->nplan1->f_hat[l] *= ths->b[l];
nfft_trafo_direct(ths->nplan2);
}
else
{
nfft_adjoint(ths->nplan1);
for(l=0; l<ths->n; l++)
ths->nplan1->f_hat[l] *= ths->b[l];
nfft_trafo(ths->nplan2);
}
}
/** fast NFFT-based summation */
void fastsum_trafo(fastsum_plan *ths)
{
int j,k,t;
ticks t0, t1;
ths->MEASURE_TIME_t[4] = 0.0;
ths->MEASURE_TIME_t[5] = 0.0;
ths->MEASURE_TIME_t[6] = 0.0;
ths->MEASURE_TIME_t[7] = 0.0;
#ifdef MEASURE_TIME
t0 = getticks();
#endif
/** first step of algorithm */
nfft_adjoint(&(ths->mv1));
#ifdef MEASURE_TIME
t1 = getticks();
ths->MEASURE_TIME_t[4] += nfft_elapsed_seconds(t1,t0);
#endif
#ifdef MEASURE_TIME
t0 = getticks();
#endif
/** second step of algorithm */
#pragma omp parallel for default(shared) private(k)
for (k=0; k<ths->mv2.N_total; k++)
ths->mv2.f_hat[k] = ths->b[k] * ths->mv1.f_hat[k];
#ifdef MEASURE_TIME
t1 = getticks();
ths->MEASURE_TIME_t[5] += nfft_elapsed_seconds(t1,t0);
#endif
#ifdef MEASURE_TIME
t0 = getticks();
#endif
/** third step of algorithm */
nfft_trafo(&(ths->mv2));
#ifdef MEASURE_TIME
t1 = getticks();
ths->MEASURE_TIME_t[6] += nfft_elapsed_seconds(t1,t0);
#endif
#ifdef MEASURE_TIME
t0 = getticks();
#endif
/** add near field */
#pragma omp parallel for default(shared) private(j,k,t)
for (j=0; j<ths->M_total; j++)
{
double ymin[ths->d], ymax[ths->d]; /** limits for d-dimensional near field box */
if (ths->flags & NEARFIELD_BOXES)
{
ths->f[j] = ths->mv2.f[j] + SearchBox(ths->y + ths->d*j, ths);
}
else
{
for (t=0; t<ths->d; t++)
{
ymin[t] = ths->y[ths->d*j+t] - ths->eps_I;
ymax[t] = ths->y[ths->d*j+t] + ths->eps_I;
}
ths->f[j] = ths->mv2.f[j] + SearchTree(ths->d,0, ths->x, ths->alpha, ymin, ymax, ths->N_total, ths->k, ths->kernel_param, ths->Ad, ths->Add, ths->p, ths->flags);
}
/* ths->f[j] = ths->mv2.f[j]; */
/* ths->f[j] = SearchTree(ths->d,0, ths->x, ths->alpha, ymin, ymax, ths->N_total, ths->k, ths->kernel_param, ths->Ad, ths->Add, ths->p, ths->flags); */
}
#ifdef MEASURE_TIME
t1 = getticks();
ths->MEASURE_TIME_t[7] += nfft_elapsed_seconds(t1,t0);
#endif
}
void FC_FUNC(oct_nfft_adjoint,OCT_NFFT_ADJOINT)
(nfft_plan *ths)
{
nfft_adjoint (ths);
}
static void simple_test_nfft_2d(void)
{
int K,N[2],n[2],M;
double t;
ticks t0, t1;
nfft_plan p;
N[0]=32; n[0]=64;
N[1]=14; n[1]=32;
M=N[0]*N[1];
K=16;
t0 = getticks();
/** init a two dimensional plan */
nfft_init_guru(&p, 2, N, M, n, 7,
PRE_PHI_HUT| PRE_FULL_PSI| MALLOC_F_HAT| MALLOC_X| MALLOC_F |
FFTW_INIT| FFT_OUT_OF_PLACE,
FFTW_ESTIMATE| FFTW_DESTROY_INPUT);
/** init pseudo random nodes */
nfft_vrand_shifted_unit_double(p.x,p.d*p.M_total);
/** precompute psi, the entries of the matrix B */
if(p.nfft_flags & PRE_ONE_PSI)
nfft_precompute_one_psi(&p);
/** init pseudo random Fourier coefficients and show them */
nfft_vrand_unit_complex(p.f_hat,p.N_total);
t1 = getticks();
t = nfft_elapsed_seconds(t1,t0);
nfft_vpr_complex(p.f_hat,K,
"given Fourier coefficients, vector f_hat (first few entries)");
printf(" ... initialisation took %e seconds.\n",t);
/** direct trafo and show the result */
t0 = getticks();
nfft_trafo_direct(&p);
t1 = getticks();
t = nfft_elapsed_seconds(t1,t0);
nfft_vpr_complex(p.f,K,"ndft, vector f (first few entries)");
printf(" took %e seconds.\n",t);
/** approx. trafo and show the result */
t0 = getticks();
nfft_trafo(&p);
t1 = getticks();
t = nfft_elapsed_seconds(t1,t0);
nfft_vpr_complex(p.f,K,"nfft, vector f (first few entries)");
printf(" took %e seconds.\n",t);
/** direct adjoint and show the result */
t0 = getticks();
nfft_adjoint_direct(&p);
t1 = getticks();
t = nfft_elapsed_seconds(t1,t0);
nfft_vpr_complex(p.f_hat,K,"adjoint ndft, vector f_hat (first few entries)");
printf(" took %e seconds.\n",t);
/** approx. adjoint and show the result */
t0 = getticks();
nfft_adjoint(&p);
t1 = getticks();
t = nfft_elapsed_seconds(t1,t0);
nfft_vpr_complex(p.f_hat,K,"adjoint nfft, vector f_hat (first few entries)");
printf(" took %e seconds.\n",t);
/** finalise the two dimensional plan */
nfft_finalize(&p);
}
void bench_openmp(FILE *infile, int m, int psi_flag)
{
nfft_plan p;
int *N;
int *n;
int M, d, trafo_adjoint;
int t, j;
double re,im;
ticks t0, t1;
double tt_total, tt_preonepsi;
fscanf(infile, "%d %d", &d, &trafo_adjoint);
N = malloc(d*sizeof(int));
n = malloc(d*sizeof(int));
for (t=0; t<d; t++)
fscanf(infile, "%d", N+t);
for (t=0; t<d; t++)
fscanf(infile, "%d", n+t);
fscanf(infile, "%d", &M);
#ifdef _OPENMP
fftw_import_wisdom_from_filename("nfft_benchomp_detail_threads.plan");
#else
fftw_import_wisdom_from_filename("nfft_benchomp_detail_single.plan");
#endif
/** init an d-dimensional plan */
nfft_init_guru(&p, d, N, M, n, m,
PRE_PHI_HUT| psi_flag | MALLOC_X | MALLOC_F_HAT| MALLOC_F| FFTW_INIT | FFT_OUT_OF_PLACE,
FFTW_MEASURE| FFTW_DESTROY_INPUT);
#ifdef _OPENMP
fftw_export_wisdom_to_filename("nfft_benchomp_detail_threads.plan");
#else
fftw_export_wisdom_to_filename("nfft_benchomp_detail_single.plan");
#endif
for (j=0; j < p.M_total; j++)
{
for (t=0; t < p.d; t++)
fscanf(infile, "%lg", p.x+p.d*j+t);
}
if (trafo_adjoint==0)
{
for (j=0; j < p.N_total; j++)
{
fscanf(infile, "%lg %lg", &re, &im);
p.f_hat[j] = re + _Complex_I * im;
}
}
else
{
for (j=0; j < p.M_total; j++)
{
fscanf(infile, "%lg %lg", &re, &im);
p.f[j] = re + _Complex_I * im;
}
}
t0 = getticks();
/** precompute psi, the entries of the matrix B */
if(p.nfft_flags & PRE_ONE_PSI)
nfft_precompute_one_psi(&p);
t1 = getticks();
tt_preonepsi = nfft_elapsed_seconds(t1,t0);
if (trafo_adjoint==0)
nfft_trafo(&p);
else
nfft_adjoint(&p);
t1 = getticks();
tt_total = nfft_elapsed_seconds(t1,t0);
#ifndef MEASURE_TIME
p.MEASURE_TIME_t[0] = 0.0;
p.MEASURE_TIME_t[2] = 0.0;
#endif
#ifndef MEASURE_TIME_FFTW
p.MEASURE_TIME_t[1] = 0.0;
#endif
printf("%.6e %.6e %6e %.6e %.6e %.6e\n", tt_preonepsi, p.MEASURE_TIME_t[0], p.MEASURE_TIME_t[1], p.MEASURE_TIME_t[2], tt_total-tt_preonepsi-p.MEASURE_TIME_t[0]-p.MEASURE_TIME_t[1]-p.MEASURE_TIME_t[2], tt_total);
// printf("%.6e\n", tt);
free(N);
free(n);
/** finalise the one dimensional plan */
nfft_finalize(&p);
}
