void sararfftnd_one_complex_to_real(
sararfftnd_plan p, sarafft_complex *data
) {
#ifdef USE_GPUS
// TODO: GPU implementation
#else // #ifndef USE_GPUS
rfftwnd_one_complex_to_real( p, data, 0 );
#endif
}
void Wavelet::invFFT1DInPlace()
{
// use the operator version of the fourier transform
if(!isReal_) {
int flag;
rfftwnd_plan plan;
flag = FFTW_ESTIMATE | FFTW_IN_PLACE;
plan= rfftwnd_create_plan(1,&nzp_,FFTW_COMPLEX_TO_REAL,flag);
rfftwnd_one_complex_to_real(plan,cAmp_,rAmp_);
fftwnd_destroy_plan(plan);
isReal_=true;
double scale= static_cast<double>(1.0/static_cast<double>(nzp_));
for(int i=0; i < nzp_; i++)
rAmp_[i] = static_cast<fftw_real>(rAmp_[i]*scale);
}
}
int
gmx_fft_3d_real(gmx_fft_t fft,
enum gmx_fft_direction dir,
void * in_data,
void * out_data)
{
int inplace = (in_data == out_data);
int isforward = (dir == GMX_FFT_REAL_TO_COMPLEX);
int sz;
if((fft->ndim != 3) ||
((dir != GMX_FFT_REAL_TO_COMPLEX) && (dir != GMX_FFT_COMPLEX_TO_REAL)))
{
gmx_fatal(FARGS,"FFT plan mismatch - bad plan or direction.");
return EINVAL;
}
if(inplace == 0)
{
/* Copy data to avoid overwriting input, and redirect input ptr to work array */
sz = fft->nx*fft->ny*(fft->nz/2 + 1)*2;
memcpy(fft->work,in_data,sz*sizeof(real));
in_data = fft->work;
}
if(isforward)
{
rfftwnd_one_real_to_complex(fft->multi[inplace][isforward],(fftw_real *)in_data,(fftw_complex *)out_data);
}
else
{
rfftwnd_one_complex_to_real(fft->multi[inplace][isforward],(fftw_complex *)in_data,(fftw_real *)out_data);
}
return 0;
}
/*
* Class: jfftw_real_nd_Plan
* Method: transform
* Signature: ([D)[D
*/
JNIEXPORT jdoubleArray JNICALL Java_jfftw_real_nd_Plan_transform___3D( JNIEnv* env, jobject obj, jdoubleArray in )
{
jdouble *cin, *cout;
jdoubleArray out;
int i;
jclass clazz = (*env)->GetObjectClass( env, obj );
jfieldID id = (*env)->GetFieldID( env, clazz, "plan", "[B" );
jbyteArray arr = (jbyteArray)(*env)->GetObjectField( env, obj, id );
unsigned char* carr = (*env)->GetByteArrayElements( env, arr, 0 );
rfftwnd_plan plan = *(rfftwnd_plan*)carr;
int length = 1;
int clength = 2;
for( i = 0; i < plan->rank; ++i ) length *= plan->plans[i]->n;
for( i = 0; i < plan->rank; ++i ) clength *= plan->n[i];
if( (plan->dir==FFTW_REAL_TO_COMPLEX?length:clength) != (*env)->GetArrayLength( env, in ) )
{
(*env)->ThrowNew( env, (*env)->FindClass( env, "java/lang/IndexOutOfBoundsException" ), "the Plan was created for a different length" );
(*env)->ReleaseByteArrayElements( env, arr, carr, 0 );
return NULL;
}
cin = (*env)->GetDoubleArrayElements( env, in, 0 );
if( plan->is_in_place )
{
out = in;
cout = NULL;
}
else
{
out = (*env)->NewDoubleArray( env, (plan->dir==FFTW_REAL_TO_COMPLEX?clength:length) );
cout = (*env)->GetDoubleArrayElements( env, out, 0 );
}
if( plan->rank > 0 && ! plan->plans[0]->flags & FFTW_THREADSAFE )
{
// synchronization
(*env)->MonitorEnter( env, obj );
}
if( plan->dir == FFTW_REAL_TO_COMPLEX )
{
rfftwnd_one_real_to_complex( plan, cin, (fftw_complex*)cout );
}
else
{
rfftwnd_one_complex_to_real( plan, (fftw_complex*)cin, cout );
}
if( plan->rank > 0 && ! plan->plans[0]->flags & FFTW_THREADSAFE )
{
// synchronization
(*env)->MonitorExit( env, obj );
}
(*env)->ReleaseByteArrayElements( env, arr, carr, 0 );
(*env)->ReleaseDoubleArrayElements( env, in, cin, 0 );
if( plan->is_in_place )
{
(*env)->ReleaseDoubleArrayElements( env, out, cout, 0 );
}
return out;
}
void F77_FUNC_(rfftwnd_f77_one_complex_to_real,RFFTWND_F77_ONE_COMPLEX_TO_REAL)
(fftwnd_plan *p, fftw_complex *in, fftw_real *out)
{
rfftwnd_one_complex_to_real(*p,in,out);
}
main (int argc, char *argv[])
{
struct em_file inputdata1;
struct em_file inputdata2;
struct em_file inputdata3;
struct em_file inputdata4;
struct em_file outputdata;
fftw_real *Vol_tmpl_sort, *Volume, *e3, *PointCorr, *sqconv;
fftw_complex *C3, *PointVolume, *PointSq;
rfftwnd_plan p3, pi3, r3, ri3;
fftw_real scale;
struct tm *zeit;
struct tm start;
char name[200];
int Rx_max, Ry_max, Rz_max;
int Rx_min, Ry_min, Rz_min;
int Vx_min, Vy_min, Vz_min;
int Vx_max, Vy_max, Vz_max;
float Phi, Psi, Theta, winkel_lauf;
float *Rot_tmpl, *Vol_tmpl;
int i, j, k, tmpx, tmpy, tmpz,lauf_pe, ksub;
int ijk;
int lauf, n;
float max, eps;
time_t lt;
float Ctmp, Ctmpim, Dtmp, Dtmpim;
int dim_fft;
int sub[3],range[3],range_sub[3],subc[3],offset[3],dimarray[3];
int FullVolume_dims[3];
int nr[3];
int area[3];
/* MPI Variablen */
int winkel_max, winkel_min;
int winkel_max_pe, winkel_min_pe;
int winkel_step_pe;
int Phi_max, Psi_max, Theta_max;
int Phi_min, Psi_min, Theta_min;
int Phi_step, Psi_step, Theta_step;
int Theta_winkel_start, Psi_winkel_start, Phi_winkel_start;
int Theta_winkel_nr, Psi_winkel_nr, Phi_winkel_nr;
int Theta_winkel_end, Psi_winkel_end, Phi_winkel_end;
int Theta_steps, Psi_steps, Phi_steps;
float Theta_winkel_rest_nr, Psi_winkel_rest_nr, Phi_winkel_rest_nr;
int in_max;
float rms_wedge, tempccf;
float *Ergebnis, *conv;
float cycles;
int cycle;
/* MPI Variablen Ende*/
if (argc < 15)
{
printf ("\n\n");
printf (" 'OSCAR' is an Optimized SCanning AlgoRithm for \n");
printf (" local correlation.\n");
printf (" All files in EM-V-int4 format !!!\n\n");
printf (" Input: Volume to be searched, Template mask for local \n ");
printf (" correlation, pointspread function and angular search \n");
printf (" range. \n");
printf (" Output: locally normalized X-Correlation Function Out.ccf.norm, \n");
printf (" non-normalized X-Correlation Function Out.ccf, and Out.ang \n");
printf (" with the corresponding angles.\n\n");
printf (" usage: oscar Volume Template Out ...\n");
printf (" ... Phi_min Phi_max Phi_step Psi_min Psi_max Psi_step The_min The_max The_step\n");
printf (" ... Poinspread-function mask-file dim_of_fft\n\n");
printf (" with Message Passing Interface (MPI)\n");
printf (" the total number of angles must be modulo\n");
printf (" of used processors!\n\n");
printf (" Linux: 1.'lamboot' to start MPI\n");
printf (" 2.'mpirun -np 2 oscar Volume Templ Out 30 180 30 30 180 30 30 180 30 Poinspread-function mask-file 256'\n\n");
printf (" In this version asymmetric masks can be used ! \n");
printf (" last revision , 11.11.03, Friedrich Foerster");
printf (" \n\n");
exit (1);
}
MPI_Init (&argc, &argv);
MPI_Comm_size (MPI_COMM_WORLD, &mysize);
MPI_Comm_rank (MPI_COMM_WORLD, &myrank);
/* Dimensionen auslesen */
// Dimension of fft
dim_fft = atoi (argv[15]);
nr[0]=1;
nr[1]=1;
nr[2]=1;
area[0]=dim_fft;
area[1]=dim_fft;
area[2]=dim_fft;
read_em_header(argv[1], &inputdata1); /* Searchvolume */
read_em (argv[2], &inputdata2); /* Template */
FullVolume_dims[0]=inputdata1.dims[0];
FullVolume_dims[1]=inputdata1.dims[1];
FullVolume_dims[2]=inputdata1.dims[2];
Rx_min = 1;
Ry_min = 1;
Rz_min = 1;
Rx_max = (inputdata2.dims[0]);
Ry_max = (inputdata2.dims[1]);
Rz_max = (inputdata2.dims[2]);
Vx_min = 1;
Vy_min = 1;
Vz_min = 1;
Vx_max = dim_fft;
Vy_max = dim_fft;
Vz_max = dim_fft;
p3 = rfftw3d_create_plan (Vx_max, Vy_max, Vz_max, FFTW_REAL_TO_COMPLEX,
FFTW_MEASURE | FFTW_IN_PLACE); /*FFTW_ESTIMATE FFTW_MEASURE */
pi3 = rfftw3d_create_plan (Vx_max, Vy_max, Vz_max, FFTW_COMPLEX_TO_REAL,
FFTW_MEASURE | FFTW_IN_PLACE);
r3 = rfftw3d_create_plan (Rx_max, Rx_max, Rx_max, FFTW_REAL_TO_COMPLEX,
FFTW_MEASURE | FFTW_IN_PLACE); /*FFTW_ESTIMATE FFTW_MEASURE */
ri3 = rfftw3d_create_plan (Rx_max, Rx_max, Rx_max, FFTW_COMPLEX_TO_REAL,
FFTW_MEASURE | FFTW_IN_PLACE);
if (myrank == 0)
{
printf("Plans for FFTW created \n");fflush(stdout);
}
Volume = (fftw_real *) calloc (Vx_max * Vx_max * 2 * (Vx_max / 2 + 1),sizeof (fftw_real) );
Rot_tmpl = (float *) malloc (sizeof (float) * Rx_max * Ry_max * Rz_max);
Vol_tmpl = (float *) malloc (sizeof (float) * Vx_max * Vy_max * Vz_max);
conv = (float *) malloc (sizeof (float) * Vx_max * Vy_max * Vz_max);
sqconv = (fftw_real *) calloc(Vz_max * Vy_max * 2 * (Vx_max / 2 + 1), sizeof (fftw_real));
if (!
(inputdata1.floatdata =
(float *) malloc (sizeof (float) * Vx_max * Vy_max * Vz_max)))
{
printf ("Memory allocation failure in inputdata1.floatdata!!!");
fflush (stdout);
exit (1);
}
if (!
(outputdata.floatdata =
(float *) malloc (sizeof (float) * Vx_max * Vy_max * Vz_max)))
{
printf ("Memory allocation failure in outputdata.floatdata!!!");
fflush (stdout);
exit (1);
}
if (!
(Vol_tmpl_sort = (fftw_real *) calloc (Vz_max*Vy_max*2*(Vx_max / 2 + 1), sizeof (fftw_real) )))
{
printf ("Memory allocation failure in Volume_tmpl_sort!!!");
printf ("Nx = %i, Ny = %i, Nz = %i, bytes = %i \n",2 *(Vx_max / 2 + 1),Vy_max, Vz_max, sizeof (fftw_real));
fflush (stdout);
exit (1);
}
Ergebnis = (float *) calloc (Vz_max * Vy_max * Vx_max, sizeof (float));
/* Winkelraum */
Phi_min = atof (argv[4]);
Phi_max = atof (argv[5]);
Phi_step = atof (argv[6]);
Psi_min = atof (argv[7]);
Psi_max = atof (argv[8]);
Psi_step = atof (argv[9]);
Theta_min = atof (argv[10]);
Theta_max = atof (argv[11]);
Theta_step = atof (argv[12]);
/* Pointspread Function*/
read_em (argv[13], &inputdata3);
/* mask function */
read_em (argv[14], &inputdata4);
Phi_steps = (Phi_max - Phi_min) / Phi_step + 1;
Psi_steps = (Psi_max - Psi_min) / Psi_step + 1;
Theta_steps = (Theta_max - Theta_min) / Theta_step + 1;
winkel_max = Phi_steps * Psi_steps * Theta_steps;
winkel_min = 0;
range[0]=dim_fft-1;
range[1]=dim_fft-1;
range[2]=dim_fft-1;
range_sub[0]=range[0]-Rx_max;
range_sub[1]=range[1]-Rx_max;
range_sub[2]=range[2]-Rx_max;
sub[0]=1;
sub[1]=1;
sub[2]=1;
cycles=(int)(FullVolume_dims[2]/(dim_fft-Rx_max)+0.5);
cycles=(int)(FullVolume_dims[1]/(dim_fft-Rx_max)+0.5)*cycles;
cycles=(int)(FullVolume_dims[0]/(dim_fft-Rx_max)+0.5)*cycles;
cycle=0;
if (myrank == 0)
{
printf ("\n oscar starts to run ... ");tack (&start);fflush (stdout);
/* prepare Output */
strcpy (name, argv[3]);
strcat (name, ".ccf");
printf ("\nCreate outputfile: %s ... \n", name);fflush(stdout);
create_em (name, FullVolume_dims);
strcpy (name, argv[3]);
strcat (name, ".ang");
printf ("Create outputfile: %s ... \n", name);fflush(stdout);
create_em (name, FullVolume_dims);
strcpy (name, argv[3]);
strcat (name, ".ccf.norm");
printf ("Create outputfile: %s ... \n", name);fflush(stdout);
create_em (name, FullVolume_dims);
}
for (sub[2]=1; sub[2] < FullVolume_dims[2]-Rz_max;sub[2]=sub[2]+dim_fft-Rz_max)
{
if (myrank == 0)
{
tack (&start);
printf ("%f%%..", (float) (cycle / cycles * 100));
fflush (stdout);
}
for (sub[1]=1; sub[1] < FullVolume_dims[1]-Ry_max;sub[1]=sub[1]+dim_fft-Ry_max)
{
for (sub[0]=1; sub[0] < FullVolume_dims[0]-Rx_max;sub[0]=sub[0]+dim_fft-Rx_max)
{
cycle=cycle+1;
subc[0]=sub[0];
subc[1]=sub[1];
subc[2]=sub[2];
if (sub[2] + range[2] > FullVolume_dims[2]) subc[2]=FullVolume_dims[2]-range[2]; /* we are at the corner ?!*/
if (sub[1] + range[1] > FullVolume_dims[1]) subc[1]=FullVolume_dims[1]-range[1]; /* we are at the corner ?!*/
if (sub[0] + range[0] > FullVolume_dims[0]) subc[0]=FullVolume_dims[0]-range[0]; /* we are at the corner ?!*/
read_em_subregion (argv[1], &inputdata1,subc,range);
read_em_subregion (argv[1], &outputdata,subc,range);
/* Umsortieren der Daten */
lauf = 0;
for (k = 0; k < Vz_max; k++)
{
for (j = 0; j < Vy_max; j++)
{
for (i = 0; i < Vx_max; i++)
{
/* square - needed for normalization */
sqconv[i + 2 * (Vx_max / 2 + 1) * (j + Vy_max * k)] = inputdata1.floatdata[lauf]*inputdata1.floatdata[lauf];
Volume[i + 2 * (Vx_max / 2 + 1) * (j + Vy_max * k)] = inputdata1.floatdata[lauf];
inputdata1.floatdata[lauf] = -1.0; /* kleine Zahl wg Max-Op , hier kommen die CCFs rein*/
outputdata.floatdata[lauf] = -1.0; /* hier kommen die Winkel rein*/
lauf++;
}
}
}
rfftwnd_one_real_to_complex (p3, &Volume[0], NULL); /* einmalige fft von Suchvolumen */
rfftwnd_one_real_to_complex (p3, &sqconv[0], NULL); /* FFT of square*/
winkel_step_pe = (int) winkel_max / mysize;
winkel_min_pe = myrank * winkel_step_pe;
winkel_max_pe = winkel_min_pe + winkel_step_pe;
Theta_winkel_nr = (int) winkel_min_pe / (Psi_steps * Phi_steps);
Theta_winkel_rest_nr = winkel_min_pe - Theta_winkel_nr * (Psi_steps * Phi_steps);
Psi_winkel_nr = (int) Theta_winkel_rest_nr / (Phi_steps);
Psi_winkel_rest_nr = Theta_winkel_rest_nr - Psi_winkel_nr * (Phi_steps);
Phi_winkel_nr = (int) Psi_winkel_rest_nr;
Theta = Theta_winkel_nr * Theta_step + Theta_min;
Phi = Phi_winkel_nr * Phi_step + Phi_min - Phi_step;
Psi = Psi_winkel_nr * Psi_step + Psi_min;
eps = 0.001;
n = 0;
//Friedrich -> Zaehlung der voxels
n = countvoxel(inputdata4.dims[0], inputdata4.floatdata, eps);
eps = 0.001;
for (winkel_lauf = winkel_min_pe; winkel_lauf < winkel_max_pe;winkel_lauf++)
{
if (Phi < Phi_max)
Phi = Phi + Phi_step;
else
{
Phi = Phi_min;
Psi = Psi + Psi_step;
}
if (Psi > Psi_max)
{
Psi = Psi_min;
Theta = Theta + Theta_step;
}
tom_rotate3d (&Rot_tmpl[0], &inputdata2.floatdata[0], Phi, Psi, Theta, Rx_max, Ry_max, Rz_max);
/*calculate Ref variance */
rms_wedge = energizer (Rx_min, Rx_max, n, &Rot_tmpl[0], &inputdata3.floatdata[0], &inputdata4.floatdata[0], r3, ri3);
pastes (&Rot_tmpl[0], &Vol_tmpl[0], 1, 1, 1, Rx_max, Ry_max, Rz_max, Vx_max);
scale = 1.0 / ((double)Vx_max * (double)Vy_max * (double)Vz_max * ((double) rms_wedge) );
//printf("hippo1: scale = %.10f \n",scale);
sort4fftw(&Vol_tmpl_sort[0],&Vol_tmpl[0],Vx_max, Vy_max, Vz_max);
rfftwnd_one_real_to_complex (p3, &Vol_tmpl_sort[0], NULL);
PointVolume = (fftw_complex *) & Volume[0];
C3 = (fftw_complex *) & Vol_tmpl_sort[0];
/* Correlation */
correl(&PointVolume[0], &C3[0], Vx_max, Vy_max, Vz_max, scale);
/* back to real space */
rfftwnd_one_complex_to_real (pi3, &C3[0], NULL);
PointCorr = (fftw_real *) & C3[0];
/* Umsortieren der Daten */
sortback4fftw( &PointCorr[0], &Ergebnis[0], Vx_max, Vy_max, Vz_max);
// crossen
cross(&Ergebnis[0], Vx_max);
/* 3rd: divide */
lauf = 0;
for (k = 0 ; k < Vz_max ; k++)
{
for (j = 0; j < Vy_max; j++)
{
for (i = 0; i < Vx_max; i++)
{
if (inputdata1.floatdata[lauf] < Ergebnis[lauf] )
{
inputdata1.floatdata[lauf] = Ergebnis[lauf];
outputdata.floatdata[lauf] = (int) winkel_lauf;
}
lauf++;
}
}
}
} /* Ende winkel_lauf */
//FF
MPI_Barrier (MPI_COMM_WORLD);
/* Ergebnisse einsammeln (myrank 0)*/
if (myrank == 0)
{
for (lauf_pe = 1; lauf_pe < mysize; lauf_pe++)
{
MPI_Recv (&Ergebnis[0], Vx_max * Vy_max * Vz_max, MPI_FLOAT, lauf_pe,
99, MPI_COMM_WORLD, &status);
MPI_Recv (&conv[0], Vx_max * Vy_max * Vz_max, MPI_FLOAT,
lauf_pe, 98, MPI_COMM_WORLD, &status);
/* use conv as temporary memory for angles */
for (lauf = 0; lauf < Vx_max * Vy_max * Vz_max; lauf++)
{
if (inputdata1.floatdata[lauf] < Ergebnis[lauf])
{
inputdata1.floatdata[lauf] = Ergebnis[lauf];
outputdata.floatdata[lauf] = conv[lauf];
}
}
}
/*Ergebnisse eingesammelt */
}
// myrank > 0: Ergebnisse senden
else
{
MPI_Send (inputdata1.floatdata, Vx_max * Vy_max * Vz_max, MPI_FLOAT, 0,
99, MPI_COMM_WORLD);
MPI_Send (outputdata.floatdata, Vx_max * Vy_max * Vz_max, MPI_FLOAT, 0,
98, MPI_COMM_WORLD);
}
MPI_Barrier (MPI_COMM_WORLD);
// nicht normalisiertes Volumen und Winkel rausschreiben
subc[0]=subc[0]+Rx_max/2;
subc[1]=subc[1]+Rx_max/2;
subc[2]=subc[2]+Rx_max/2;
if (myrank==0)
{
offset[0]=Rx_max/2;
offset[1]=Rx_max/2;
offset[2]=Rx_max/2;
dimarray[0]=dim_fft;
dimarray[1]=dim_fft;
dimarray[2]=dim_fft;
strcpy (name, argv[3]);
strcat (name, ".ccf");
write_em_subsubregion (name, &inputdata1,subc,range_sub,offset,dimarray);
strcpy (name, argv[3]);
strcat (name, ".ang");
write_em_subsubregion (name, &outputdata,subc,range_sub,offset,dimarray);
/* ------------------- normalization - here only PE 0 ---------- */
pastes (&inputdata4.floatdata[0], &Vol_tmpl[0], 1, 1, 1, Rx_max, Ry_max, Rz_max, Vx_max); /* paste mask into zero volume*/
/* 1st local mean */
sort4fftw(&Vol_tmpl_sort[0], &Vol_tmpl[0], Vx_max, Vy_max, Vz_max);
rfftwnd_one_real_to_complex (p3, &Vol_tmpl_sort[0], NULL);
C3 = (fftw_complex *) & Vol_tmpl_sort[0];
/* Convolution of volume and mask */
scale = 1.0 / ((double)Vx_max * (double)Vy_max * (double)Vz_max );
convolve( &PointVolume[0], &C3[0], Vx_max, Vy_max, Vz_max, scale);
rfftwnd_one_complex_to_real (pi3, &C3[0], NULL);
PointCorr = (fftw_real *) & C3[0];
/* Umsortieren der Daten */
sortback4fftw( &PointCorr[0], &conv[0], Vx_max, Vy_max, Vz_max);
/* 2nd : convolution of square and resorting*/
pastes (&inputdata4.floatdata[0], &Vol_tmpl[0], 1, 1, 1, Rx_max, Ry_max, Rz_max, Vx_max); /* paste mask into zero volume*/
sort4fftw( &Vol_tmpl_sort[0], &Vol_tmpl[0], Vx_max, Vy_max, Vz_max);
rfftwnd_one_real_to_complex (p3, &Vol_tmpl_sort[0], NULL);
C3 = (fftw_complex *) & Vol_tmpl_sort[0];
PointSq = (fftw_complex *) & sqconv[0];// set pointer to FFT of square
convolve( &PointSq[0], &C3[0], Vx_max, Vy_max, Vz_max, scale);
rfftwnd_one_complex_to_real (pi3, &C3[0], NULL);
PointCorr = (fftw_real *) &C3[0];
//FF
lauf = 0;
for (k = 0; k < Vz_max; k++)
{
for (j = 0; j < Vy_max; j++)
{
for (i = 0; i < Vx_max; i++)
{
conv[lauf] = sqrt(PointCorr[i + 2 * (Vx_max / 2 + 1) * (j + Vy_max * k)] - conv[lauf]*conv[lauf]/((float) n) ) ;/*local variance*/
lauf++;
}
}
}
cross(&conv[0], Vx_max);
/* perform division */
for (lauf = 0; k < Vz_max*Vy_max*Vz_max; lauf++)
{
if (conv[lauf] > eps)
{
inputdata1[lauf].floatdata = inputdata1[lauf].floatdata/conv[lauf];
}
else
{
inputdata1[lauf].floatdata = 0;
}
}
strcpy (name, argv[3]);
strcat (name, ".ccf.norm");
write_em_subsubregion (name, &inputdata1,subc,range_sub,offset,dimarray);
}
MPI_Barrier (MPI_COMM_WORLD);
}
} /* these are the new brackets from the subregion_read , SN */
}
free(Ergebnis);
free(inputdata1.floatdata);
free(inputdata2.floatdata);
free(inputdata3.floatdata);
free(inputdata4.floatdata);
rfftwnd_destroy_plan(p3);
rfftwnd_destroy_plan(pi3);
rfftwnd_destroy_plan(r3);
rfftwnd_destroy_plan(ri3);
free(Volume);
free(sqconv);
free(conv);
free(Rot_tmpl);
free(Vol_tmpl_sort);
free(outputdata.floatdata);
if (myrank==0)
{
printf ("oscar finished. ");
tack (&start); fflush(stdout);
}
MPI_Finalize();
/* end main */
}
void testnd_in_place(int rank, int *n, fftwnd_plan validated_plan,
int alternate_api, int specific)
{
int istride, ostride, howmany;
int N, dim, i, j, k;
int nc, nhc, nr;
fftw_real *in1, *out3;
fftw_complex *in2, *out1, *out2;
fftwnd_plan p, ip;
int flags = measure_flag | wisdom_flag | FFTW_IN_PLACE;
if (coinflip())
flags |= FFTW_THREADSAFE;
N = nc = nr = nhc = 1;
for (dim = 0; dim < rank; ++dim)
N *= n[dim];
if (rank > 0) {
nr = n[rank - 1];
nc = N / nr;
nhc = nr / 2 + 1;
}
in1 = (fftw_real *) fftw_malloc(2 * nhc * nc * MAX_STRIDE * sizeof(fftw_real));
out3 = in1;
out1 = (fftw_complex *) in1;
in2 = (fftw_complex *) fftw_malloc(N * sizeof(fftw_complex));
out2 = (fftw_complex *) fftw_malloc(N * sizeof(fftw_complex));
if (alternate_api && specific && (rank == 2 || rank == 3)) {
if (rank == 2) {
p = rfftw2d_create_plan_specific(n[0], n[1],
FFTW_REAL_TO_COMPLEX, flags,
in1, MAX_STRIDE, 0, 0);
ip = rfftw2d_create_plan_specific(n[0], n[1],
FFTW_COMPLEX_TO_REAL, flags,
in1, MAX_STRIDE, 0, 0);
} else {
p = rfftw3d_create_plan_specific(n[0], n[1], n[2],
FFTW_REAL_TO_COMPLEX, flags,
in1, MAX_STRIDE, 0, 0);
ip = rfftw3d_create_plan_specific(n[0], n[1], n[2],
FFTW_COMPLEX_TO_REAL, flags,
in1, MAX_STRIDE, 0, 0);
}
} else if (specific) {
p = rfftwnd_create_plan_specific(rank, n, FFTW_REAL_TO_COMPLEX,
flags,
in1, MAX_STRIDE, in1, MAX_STRIDE);
ip = rfftwnd_create_plan_specific(rank, n, FFTW_COMPLEX_TO_REAL,
flags,
in1, MAX_STRIDE, in1, MAX_STRIDE);
} else if (alternate_api && (rank == 2 || rank == 3)) {
if (rank == 2) {
p = rfftw2d_create_plan(n[0], n[1], FFTW_REAL_TO_COMPLEX,
flags);
ip = rfftw2d_create_plan(n[0], n[1], FFTW_COMPLEX_TO_REAL,
flags);
} else {
p = rfftw3d_create_plan(n[0], n[1], n[2], FFTW_REAL_TO_COMPLEX,
flags);
ip = rfftw3d_create_plan(n[0], n[1], n[2], FFTW_COMPLEX_TO_REAL,
flags);
}
} else {
p = rfftwnd_create_plan(rank, n, FFTW_REAL_TO_COMPLEX, flags);
ip = rfftwnd_create_plan(rank, n, FFTW_COMPLEX_TO_REAL, flags);
}
CHECK(p != NULL && ip != NULL, "can't create plan");
for (i = 0; i < nc * nhc * 2 * MAX_STRIDE; ++i)
out3[i] = 0;
for (istride = 1; istride <= MAX_STRIDE; ++istride) {
/* generate random inputs */
for (i = 0; i < nc; ++i)
for (j = 0; j < nr; ++j) {
c_re(in2[i * nr + j]) = DRAND();
c_im(in2[i * nr + j]) = 0.0;
for (k = 0; k < istride; ++k)
in1[(i * nhc * 2 + j) * istride + k]
= c_re(in2[i * nr + j]);
}
fftwnd(validated_plan, 1, in2, 1, 1, out2, 1, 1);
howmany = ostride = istride;
WHEN_VERBOSE(2, printf("\n testing in-place stride %d...",
istride));
if (howmany != 1 || istride != 1 || ostride != 1 || coinflip())
rfftwnd_real_to_complex(p, howmany, in1, istride, 1,
out1, ostride, 1);
else
rfftwnd_one_real_to_complex(p, in1, NULL);
for (i = 0; i < nc; ++i)
for (k = 0; k < howmany; ++k)
CHECK(compute_error_complex(out1 + i * nhc * ostride + k,
ostride,
out2 + i * nr, 1,
nhc) < TOLERANCE,
"in-place (r2c): wrong answer");
if (howmany != 1 || istride != 1 || ostride != 1 || coinflip())
rfftwnd_complex_to_real(ip, howmany, out1, ostride, 1,
out3, istride, 1);
else
rfftwnd_one_complex_to_real(ip, out1, NULL);
for (i = 0; i < nc * nhc * 2 * istride; ++i)
out3[i] *= 1.0 / N;
for (i = 0; i < nc; ++i)
for (k = 0; k < howmany; ++k)
CHECK(compute_error(out3 + i * nhc * 2 * istride + k,
istride,
(fftw_real *) (in2 + i * nr), 2,
nr) < TOLERANCE,
"in-place (c2r): wrong answer (check 2)");
}
rfftwnd_destroy_plan(p);
rfftwnd_destroy_plan(ip);
fftw_free(out2);
fftw_free(in2);
fftw_free(in1);
}
void testnd_out_of_place(int rank, int *n, fftwnd_plan validated_plan)
{
int istride, ostride;
int N, dim, i, j, k;
int nc, nhc, nr;
fftw_real *in1, *out3;
fftw_complex *in2, *out1, *out2;
fftwnd_plan p, ip;
int flags = measure_flag | wisdom_flag;
if (coinflip())
flags |= FFTW_THREADSAFE;
N = nc = nr = nhc = 1;
for (dim = 0; dim < rank; ++dim)
N *= n[dim];
if (rank > 0) {
nr = n[rank - 1];
nc = N / nr;
nhc = nr / 2 + 1;
}
in1 = (fftw_real *) fftw_malloc(N * MAX_STRIDE * sizeof(fftw_real));
out3 = (fftw_real *) fftw_malloc(N * MAX_STRIDE * sizeof(fftw_real));
out1 = (fftw_complex *) fftw_malloc(nhc * nc * MAX_STRIDE
* sizeof(fftw_complex));
in2 = (fftw_complex *) fftw_malloc(N * sizeof(fftw_complex));
out2 = (fftw_complex *) fftw_malloc(N * sizeof(fftw_complex));
p = rfftwnd_create_plan(rank, n, FFTW_REAL_TO_COMPLEX, flags);
ip = rfftwnd_create_plan(rank, n, FFTW_COMPLEX_TO_REAL, flags);
CHECK(p != NULL && ip != NULL, "can't create plan");
for (istride = 1; istride <= MAX_STRIDE; ++istride) {
/* generate random inputs */
for (i = 0; i < nc; ++i)
for (j = 0; j < nr; ++j) {
c_re(in2[i * nr + j]) = DRAND();
c_im(in2[i * nr + j]) = 0.0;
for (k = 0; k < istride; ++k)
in1[(i * nr + j) * istride + k]
= c_re(in2[i * nr + j]);
}
for (i = 0; i < N * istride; ++i)
out3[i] = 0.0;
fftwnd(validated_plan, 1, in2, 1, 1, out2, 1, 1);
for (ostride = 1; ostride <= MAX_STRIDE; ++ostride) {
int howmany = (istride < ostride) ? istride : ostride;
WHEN_VERBOSE(2, printf("\n testing stride %d/%d...",
istride, ostride));
if (howmany != 1 || istride != 1 || ostride != 1 || coinflip())
rfftwnd_real_to_complex(p, howmany, in1, istride, 1,
out1, ostride, 1);
else
rfftwnd_one_real_to_complex(p, in1, out1);
for (i = 0; i < nc; ++i)
for (k = 0; k < howmany; ++k)
CHECK(compute_error_complex(out1 + i * nhc * ostride + k,
ostride,
out2 + i * nr, 1,
nhc) < TOLERANCE,
"out-of-place (r2c): wrong answer");
if (howmany != 1 || istride != 1 || ostride != 1 || coinflip())
rfftwnd_complex_to_real(ip, howmany, out1, ostride, 1,
out3, istride, 1);
else
rfftwnd_one_complex_to_real(ip, out1, out3);
for (i = 0; i < N * istride; ++i)
out3[i] *= 1.0 / N;
if (istride == howmany)
CHECK(compute_error(out3, 1, in1, 1, N * istride)
< TOLERANCE, "out-of-place (c2r): wrong answer");
for (i = 0; i < nc; ++i)
for (k = 0; k < howmany; ++k)
CHECK(compute_error(out3 + i * nr * istride + k,
istride,
(fftw_real *) (in2 + i * nr), 2,
nr) < TOLERANCE,
"out-of-place (c2r): wrong answer (check 2)");
}
}
rfftwnd_destroy_plan(p);
rfftwnd_destroy_plan(ip);
fftw_free(out3);
fftw_free(out2);
fftw_free(in2);
fftw_free(out1);
fftw_free(in1);
}
void Grid::ifft(bool) {
rfftwnd_one_complex_to_real(ifft_plan, data, NULL);
}
/* Use fftw2.
*/
static int
invfft1( IMAGE *dummy, IMAGE *in, IMAGE *out )
{
IMAGE *cmplx = im_open_local( dummy, "invfft1-1", "t" );
IMAGE *real = im_open_local( out, "invfft1-2", "t" );
const int half_width = in->Xsize / 2 + 1;
/* Transform to halfcomplex here.
*/
double *half_complex = IM_ARRAY( dummy,
in->Ysize * half_width * 2, double );
rfftwnd_plan plan;
int x, y;
double *q, *p;
if( !cmplx || !real || !half_complex || im_pincheck( in ) ||
im_poutcheck( out ) )
return( -1 );
if( in->Coding != IM_CODING_NONE || in->Bands != 1 ) {
im_error( "im_invfft", _( "one band uncoded only" ) );
return( -1 );
}
/* Make dp complex image for input.
*/
if( im_clip2fmt( in, cmplx, IM_BANDFMT_DPCOMPLEX ) )
return( -1 );
/* Make mem buffer real image for output.
*/
if( im_cp_desc( real, in ) )
return( -1 );
real->BandFmt = IM_BANDFMT_DOUBLE;
if( im_setupout( real ) )
return( -1 );
/* Build half-complex image.
*/
q = half_complex;
for( y = 0; y < cmplx->Ysize; y++ ) {
p = ((double *) cmplx->data) + y * in->Xsize * 2;
for( x = 0; x < half_width; x++ ) {
q[0] = p[0];
q[1] = p[1];
p += 2;
q += 2;
}
}
/* Make the plan for the transform. Yes, they really do use nx for
* height and ny for width.
*/
if( !(plan = rfftw2d_create_plan( in->Ysize, in->Xsize,
FFTW_BACKWARD, FFTW_MEASURE | FFTW_USE_WISDOM )) ) {
im_error( "im_invfft", _( "unable to create transform plan" ) );
return( -1 );
}
rfftwnd_one_complex_to_real( plan,
(fftw_complex *) half_complex, (fftw_real *) real->data );
rfftwnd_destroy_plan( plan );
/* Copy to out.
*/
if( im_copy( real, out ) )
return( -1 );
return( 0 );
}
void
RockPhysicsInversion4D::DivideAndSmoothTable(int tableInd,std::vector<std::vector<double> > priorDistribution, std::vector<fftw_complex*> smoothingFilter)
{
for (int j=0; j<nf_[0]; j++){
meanRockPrediction_(tableInd,j)->setAccessMode(FFTGrid::RANDOMACCESS);
}
int cnfp=nfp_/2+1;
int rnfp=2*cnfp;
fftw_real* rTemp = static_cast<fftw_real*>(fftw_malloc(sizeof(float)*rnfp));
fftw_complex* cTemp = reinterpret_cast<fftw_complex*>(rTemp);
double minDivisor = 1e-3;
LogKit::LogFormatted(LogKit::Low,"\n Smoothing direction 1 of 4\n");
float monitorSize = std::max(1.0f, static_cast<float>(nf_[0]*nf_[1]*nf_[2]*nf_[3])*0.02f);
float nextMonitor = monitorSize;
std::cout
<< "\n 0% 20% 40% 60% 80% 100%"
<< "\n | | | | | | | | | | | "
<< "\n ^";
// divide and smooth direction1
for(int i1=0;i1<nf_[1];i1++)
for(int i2=0;i2<nf_[2];i2++)
for(int i3=0;i3<nf_[3];i3++)
{
for(int i0=0;i0<nf_[0];i0++)
{
double divisor=std::max(minDivisor,priorDistribution_[0][i0]);
rTemp[i0]=float(GetGridValue(tableInd,i0,i1,i2,i3)/ divisor);
}
for(int i0=nf_[0];i0<nfp_;i0++)
rTemp[i0]=0.0f;
rfftwnd_one_real_to_complex(fftplan1_,rTemp,cTemp);
for(int i0=0;i0<cnfp;i0++)
{
cTemp[i0].re=cTemp[i0].re*smoothingFilter[0][i0].re;
cTemp[i0].im=cTemp[i0].im*smoothingFilter[0][i0].re;
}
rfftwnd_one_complex_to_real(fftplan2_,cTemp,rTemp);
for(int i0=0;i0<nf_[0];i0++)
{
SetGridValue(tableInd,i0,i1,i2,i3, rTemp[i0]);
if ( i1*nf_[2]*nf_[3]*nf_[0] +i2*nf_[3]*nf_[0]+ i3*nf_[0] + i0 + 1 >= static_cast<int>(nextMonitor)) {
nextMonitor += monitorSize;
std::cout << "^";
}
}
}
LogKit::LogFormatted(LogKit::Low,"\n\n Smoothing direction 2 of 4\n");
monitorSize = std::max(1.0f, static_cast<float>(nf_[0]*nf_[1]*nf_[2]*nf_[3])*0.02f);
nextMonitor = monitorSize;
std::cout
<< "\n 0% 20% 40% 60% 80% 100%"
<< "\n | | | | | | | | | | | "
<< "\n ^";
// divide and smoothdirection2
for(int i0=0;i0<nf_[0];i0++)
for(int i2=0;i2<nf_[2];i2++)
for(int i3=0;i3<nf_[3];i3++)
{
for(int i1=0;i1<nf_[1];i1++)
{
double divisor=std::max(minDivisor,priorDistribution_[1][i1]);
rTemp[i1]=float(GetGridValue(tableInd,i0,i1,i2,i3)/divisor);
}
for(int i1=nf_[1];i1<nfp_;i1++)
rTemp[i1]=0.0f;
rfftwnd_one_real_to_complex(fftplan1_,rTemp,cTemp);
for(int i1=0;i1<cnfp;i1++)
{
cTemp[i1].re=cTemp[i1].re*smoothingFilter[1][i1].re;
cTemp[i1].im=cTemp[i1].im*smoothingFilter[1][i1].re;
}
rfftwnd_one_complex_to_real(fftplan2_,cTemp,rTemp);
for(int i1=0;i1<nf_[1];i1++)
{
SetGridValue(tableInd,i0,i1,i2,i3, rTemp[i1]);
if ( i0*nf_[2]*nf_[3]*nf_[1] +i2*nf_[3]*nf_[1]+ i3*nf_[1] + i1 + 1 >= static_cast<int>(nextMonitor)) {
nextMonitor += monitorSize;
std::cout << "^";
}
}
}
LogKit::LogFormatted(LogKit::Low,"\n\n Smoothing direction 3 of 4\n");
monitorSize = std::max(1.0f, static_cast<float>(nf_[0]*nf_[1]*nf_[2]*nf_[3])*0.02f);
nextMonitor = monitorSize;
std::cout
<< "\n 0% 20% 40% 60% 80% 100%"
<< "\n | | | | | | | | | | | "
<< "\n ^";
// divide and smoothdirection 3
for(int i0=0;i0<nf_[0];i0++)
for(int i1=0;i1<nf_[1];i1++)
for(int i3=0;i3<nf_[3];i3++)
{
for(int i2=0;i2<nf_[2];i2++)
{
double divisor=std::max(minDivisor,priorDistribution_[2][i2]);
rTemp[i2]=float(GetGridValue(tableInd,i0,i1,i2,i3)/divisor);
}
for(int i2=nf_[2];i2<nfp_;i2++)
rTemp[i2]=0.0f;
rfftwnd_one_real_to_complex(fftplan1_,rTemp,cTemp);
for(int i2=0;i2<cnfp;i2++)
{
cTemp[i2].re=cTemp[i2].re*smoothingFilter[2][i2].re;
cTemp[i2].im=cTemp[i2].im*smoothingFilter[2][i2].re;
}
rfftwnd_one_complex_to_real(fftplan2_,cTemp,rTemp);
for(int i2=0;i2<nf_[2];i2++)
{
SetGridValue(tableInd,i0,i1,i2,i3, rTemp[i2]);
if ( i0*nf_[1]*nf_[3]*nf_[2] +i1*nf_[3]*nf_[2]+ i3*nf_[2] + i2 + 1 >= static_cast<int>(nextMonitor)) {
nextMonitor += monitorSize;
std::cout << "^";
}
}
}
LogKit::LogFormatted(LogKit::Low,"\n Smoothing last direction \n");
monitorSize = std::max(1.0f, static_cast<float>(nf_[0]*nf_[1]*nf_[2]*nf_[3])*0.02f);
nextMonitor = monitorSize;
std::cout
<< "\n 0% 20% 40% 60% 80% 100%"
<< "\n | | | | | | | | | | | "
<< "\n ^";
// divide and smoothdirection 4
for(int i0=0;i0<nf_[0];i0++)
for(int i1=0;i1<nf_[1];i1++)
for(int i2=0;i2<nf_[2];i2++)
{
for(int i3=0;i3<nf_[3];i3++)
{
double divisor=std::max(minDivisor,priorDistribution_[3][i3]);
rTemp[i3]=float(GetGridValue(tableInd,i0,i1,i2,i3)/divisor);
}
for(int i3=nf_[3];i3<nfp_;i3++)
rTemp[i3]=0.0f;
rfftwnd_one_real_to_complex(fftplan1_,rTemp,cTemp);
for(int i3=0;i3<cnfp;i3++)
{
cTemp[i3].re=cTemp[i3].re*smoothingFilter[3][i3].re;
cTemp[i3].im=cTemp[i3].im*smoothingFilter[3][i3].re;
}
rfftwnd_one_complex_to_real(fftplan2_,cTemp,rTemp);
for(int i3=0;i3<nf_[3];i3++)
{
SetGridValue(tableInd,i0,i1,i2,i3, rTemp[i3]);
if ( i0*nf_[1]*nf_[2]*nf_[3] +i1*nf_[2]*nf_[3]+ i2*nf_[3] + i3 + 1 >= static_cast<int>(nextMonitor)) {
nextMonitor += monitorSize;
std::cout << "^";
}
}
}
for (int j=0; j<nf_[0]; j++){
meanRockPrediction_(tableInd,j)->endAccess();
}
fftw_free(rTemp);
}
