/*
* Allocate FFT buffers for a given grid.
* This must be called before FFT can be carried out
* (called automatically, users don't need to call this)
*
* grid = 3D grid for which the FFT allocation is to be done (rgrid3d *).
*
* Notes:
* - This uses c2r transformation since it is much faster than r2r.
*
* No return value.
*
*/
EXPORT void rgrid3d_fftw_periodic_alloc(rgrid3d *grid) {
long s;
double *plan_temp;
s = 2 * grid->nx * grid->ny * (grid->nz/2 + 1);
if(!(plan_temp = fftw_malloc(sizeof(double) * s))) {
fprintf(stderr, "libgrid: Out of memory in rgrid3d_fft().\n");
return;
}
memcpy(plan_temp, grid->value, sizeof(double) * s);
if(s > temp_len) {
if(temp) free(temp);
if(!(temp = fftw_malloc(sizeof(double) * s))) {
fprintf(stderr, "libgrid: Out of memory in rgrid3d_fft().\n");
return;
}
temp_len = s;
temp2 = (double complex *) temp;
}
fftw_plan_with_nthreads(grid_threads());
grid->plan =
fftw_plan_dft_r2c_3d(grid->nx, grid->ny, grid->nz, grid->value, temp2, GRID_FFTW_PLAN | FFTW_DESTROY_INPUT);
fftw_plan_with_nthreads(grid_threads());
grid->iplan =
fftw_plan_dft_c2r_3d(grid->nx, grid->ny, grid->nz, grid->cint->value, temp, GRID_FFTW_PLAN | FFTW_DESTROY_INPUT);
grid->fft_norm = 1.0 / (grid->nx * grid->ny * grid->nz);
memcpy(grid->value, plan_temp, sizeof(double) * s);
free(plan_temp);
}
void init_gfft() {
int n_size2D_ZXY[] = {NX,NY};
#ifdef _OPENMP
fftw_plan_with_nthreads( nthreads );
#endif
r2cfft_mpi_t = fftw_mpi_plan_dft_r2c_3d( NY, NX, NZ, wr1, w1, MPI_COMM_WORLD, FFT_PLANNING | FFTW_MPI_TRANSPOSED_OUT);
if (r2cfft_mpi_t == NULL) ERROR_HANDLER( ERROR_CRITICAL, "FFTW R2C_T plan creation failed");
r2cfft_mpi = fftw_mpi_plan_dft_r2c_3d( NX, NY, NZ, wr1, w1, MPI_COMM_WORLD, FFT_PLANNING);
if (r2cfft_mpi == NULL) ERROR_HANDLER( ERROR_CRITICAL, "FFTW R2C plan creation failed");
c2rfft_mpi_t = fftw_mpi_plan_dft_c2r_3d( NY, NX, NZ, w1, wr1, MPI_COMM_WORLD, FFT_PLANNING | FFTW_MPI_TRANSPOSED_IN);
if (c2rfft_mpi_t == NULL) ERROR_HANDLER( ERROR_CRITICAL, "FFTW C2R_T plan creation failed");
c2rfft_mpi = fftw_mpi_plan_dft_c2r_3d( NX, NY, NZ, w1, wr1, MPI_COMM_WORLD, FFT_PLANNING);
if (c2rfft_mpi == NULL) ERROR_HANDLER( ERROR_CRITICAL, "FFTW C2R plan creation failed");
r2cfft_2Dslice = fftw_plan_dft_r2c_2d(NX,NY,wrh3,wh3,FFT_PLANNING);
if (r2cfft_2Dslice == NULL) ERROR_HANDLER( ERROR_CRITICAL, "FFTW r2c slice plan creation failed");
// init transpose routines (These are used by remap routines)
init_transpose();
fft_timer=0.0;
return;
}
DeconvolutionTool::DeconvolutionTool(QObject* parent):QObject(parent) {
// Init MultiThreading
threadsCount = QThread::idealThreadCount() > 0 ? QThread::idealThreadCount() : 2;
qDebug("Init Multi-Threading with threads count: %d", threadsCount);
fftw_plan_with_nthreads(threadsCount);
tvIterationsCount = 500;
previewMethod = 0;
inputImageMatrix = NULL;
outputImageMatrix = NULL;
kernelMatrix = NULL;
laplacianMatrix = NULL;
outLaplacianMatrix = NULL;
inputImageFFT = NULL;
outputImageFFT = NULL;
kernelFFT = NULL;
kernelTempFFT = NULL;
laplacianMatrixFFT = NULL;
realForwardPlan = NULL;
realForwardKernelPlan = NULL;
realBackwardPlan = NULL;
forwardLaplacianPlan = NULL;
backwardLaplacianPlan = NULL;
}
void dct(int N, double *in, double *out){
// compute variables
int ii;
fftw_plan my_plan;
fftw_init_threads();
// define plan
fftw_plan_with_nthreads(omp_get_max_threads());
my_plan = fftw_plan_r2r_1d(N, in, out, FFTW_REDFT00, FFTW_ESTIMATE);
//execute plan
fftw_execute(my_plan);
// scale output
for(ii=0; ii < N; ii++){
if(ii == 0 || ii == N-1){
out[ii] = out[ii]/(double)(N-1)/2.0;
}
else{
out[ii] = out[ii]/(double)(N-1);
}
}
// destroy plan
fftw_destroy_plan(my_plan);
fftw_cleanup_threads();
}
/**
* Initializes the library.
* @param nthreads The number of OpenMP threads to use for execution of local FFT.
* @return 0 if successful
*/
int accfft_init(int nthreads){
int threads_ok=1;
if (threads_ok) threads_ok = fftw_init_threads();
if (threads_ok) fftw_plan_with_nthreads(nthreads);
return (!threads_ok);
}
///Initialize FFTW
void FFTInit(int threads)
{
#ifdef FFTW_WITH_THREADS
fftw_init_threads();
fftw_plan_with_nthreads(threads);
#endif
}
convolution_plan::convolution_plan(int width, int height, int kw, int mode, int threadMaxCount) {
switch (mode) {
case 0:
this->width = width;
this->height = height;
break;
case 1:
this->width = width + kw - 1;
this->height = height + kw - 1;
break;
default:
throw std::invalid_argument("Warning: 2d convolution plan: Invalid mode");
}
if (threadMaxCount > 1) {
fftw_init_threads(); // This MUST come before all other fftw calls
fftw_plan_with_nthreads(threadMaxCount);
}
this->depth = 1;
this->dim = 2;
this->kw = kw;
this->threadMaxCount = threadMaxCount;
fftw_complex* benchmarkArray1 = (fftw_complex*) fftw_malloc(sizeof(fftw_complex) * this->width * this->height);
fftw_complex* benchmarkArray2 = (fftw_complex*) fftw_malloc(sizeof(fftw_complex) * this->width * this->height);
this->forwardPlan = fftw_plan_dft_2d(this->height, this->width, benchmarkArray1, benchmarkArray2, FFTW_FORWARD, FFTW_MEASURE);
this->backwardPlan = fftw_plan_dft_2d(this->height, this->width, benchmarkArray1, benchmarkArray2, FFTW_BACKWARD, FFTW_MEASURE);
fftw_free(benchmarkArray1);
fftw_free(benchmarkArray2);
this->staticKernel = NULL;
}
EXPORT void cgrid2d_fftw_alloc(cgrid2d *grid) {
double complex *temp;
long size = sizeof(double complex) * grid->nx * grid->ny;
temp = fftw_malloc(size);
memcpy(temp, grid->value, size);
fftw_plan_with_nthreads(grid_threads());
grid->plan = fftw_plan_dft_2d(grid->nx, grid->ny,
grid->value, grid->value, FFTW_FORWARD, GRID_FFTW_PLAN);
fftw_plan_with_nthreads(grid_threads());
grid->iplan = fftw_plan_dft_2d(grid->nx, grid->ny,
grid->value, grid->value, FFTW_BACKWARD, GRID_FFTW_PLAN);
memcpy(grid->value, temp, size);
fftw_free(temp);
}
void plan_fftw(
Search_settings *sett,
Command_line_opts *opts,
FFTW_plans *plans,
FFTW_arrays *fftw_arr,
Aux_arrays *aux_arr) {
char hostname[512], wfilename[512];
FILE *wisdom;
/* Imports a "wisdom file" containing information
* (previous tests) about how to optimally compute Fourier
* transforms on a given machine. If wisdom file is not present,
* it will be created after the test (measure) runs
* of the fft_plans are performed below
* (see http://www.fftw.org/fftw3_doc/Wisdom.html)
*/
fftw_init_threads();
gethostname(hostname, 512);
sprintf (wfilename, "wisdom-%s.dat", hostname);
if((wisdom = fopen (wfilename, "r")) != NULL) {
fftw_import_wisdom_from_file(wisdom);
fclose (wisdom);
}
sett->Ninterp = sett->interpftpad*sett->nfft;
// array length (xa, xb) is max{fftpad*nfft, Ninterp}
fftw_arr->arr_len = (sett->fftpad*sett->nfft > sett->Ninterp
? sett->fftpad*sett->nfft : sett->Ninterp);
// fftw_arr->xa = fftw_malloc(2*fftw_arr->arr_len*sizeof(fftw_complex));
//fftw_arr->xb = fftw_arr->xa + fftw_arr->arr_len;
fftw_arr->xa = fftw_malloc(fftw_arr->arr_len*sizeof(fftw_complex));
fftw_arr->xb = fftw_malloc(fftw_arr->arr_len*sizeof(fftw_complex));
sett->nfftf = sett->fftpad*sett->nfft;
// Change FFTW_MEASURE to FFTW_PATIENT for more optimized plan
// (takes more time to generate the wisdom file)
plans->plan = fftw_plan_dft_1d(sett->nfftf, fftw_arr->xa, fftw_arr->xa, FFTW_FORWARD, FFTW_MEASURE);
fftw_plan_with_nthreads(omp_get_max_threads());
plans->pl_int = fftw_plan_dft_1d(sett->nfft, fftw_arr->xa, fftw_arr->xa, FFTW_FORWARD, FFTW_MEASURE);
plans->pl_inv = fftw_plan_dft_1d(sett->Ninterp, fftw_arr->xa, fftw_arr->xa, FFTW_BACKWARD, FFTW_MEASURE);
// Generates a wisdom FFT file if there is none
if((wisdom = fopen(wfilename, "r")) == NULL) {
wisdom = fopen(wfilename, "w");
fftw_export_wisdom_to_file(wisdom);
}
fclose (wisdom);
} // end of FFT plans
void set_num_threads(int nr)
{
num_threads = nr;
omp_set_num_threads(nr);
int ret = fftw_init_threads();
if (ret == 0) {cout << "error" << endl; exit(1);}
fftw_plan_with_nthreads(nr);
}
void init_output_vtk() {
double complex *w2d;
const int n_size1D[1] = {NY};
#ifdef WITH_SHEAR
w2d = (double complex *) fftw_malloc( sizeof(double complex) * (NY/2+1) * NZ );
if (w2d == NULL) ERROR_HANDLER( ERROR_CRITICAL, "No memory for w2d allocation");
// FFT plans (we use dummy arrays since we use the "guru" interface of fft3 in the code)
// The in place/ out of place will be set automatically at this stage
// The Following Fourier transforms takes an array of size ( NX+1, NY+1, NZ+1) but consider only the "included" array
// of size ( NX+1, NY, NZ+1) and transforms it in y, in an array of size (NX+1, NY/2+1, NZ+1). The j=NY plane is therefore
// not modified by these calls
#ifdef _OPENMP
fftw_plan_with_nthreads( 1 );
#endif
#ifdef WITH_2D
fft_1d_forward = fftw_plan_many_dft_r2c(1, n_size1D, 1,
wr1, NULL, 1, 1,
w2d, NULL, 1, 1,
FFT_PLANNING || FFTW_UNALIGNED);
#else
fft_1d_forward = fftw_plan_many_dft_r2c(1, n_size1D, NZ,
wr1, NULL, NZ+2, 1,
w2d, NULL, NZ, 1,
FFT_PLANNING || FFTW_UNALIGNED);
#endif
if (fft_1d_forward == NULL) ERROR_HANDLER( ERROR_CRITICAL, "FFTW 1D forward plan creation failed");
#ifdef WITH_2D
fft_1d_backward = fftw_plan_many_dft_c2r(1, n_size1D, 1,
w2d, NULL, 1, 1,
wr1, NULL, 1, 1,
FFT_PLANNING || FFTW_UNALIGNED);
#else
fft_1d_backward = fftw_plan_many_dft_c2r(1, n_size1D, NZ,
w2d, NULL, NZ, 1,
wr1, NULL, NZ+2, 1,
FFT_PLANNING || FFTW_UNALIGNED);
#endif
if (fft_1d_backward == NULL) ERROR_HANDLER( ERROR_CRITICAL, "FFTW 1D backward plan creation failed");
fftw_free(w2d);
#endif
}
FFT::FFT(size_t threads)
: forward_plans(),
backward_plans()
{
if (!fftw_init_threads()) {
std::cerr << "Unable to init threads in fftw\n";
throw 1;
}
fftw_plan_with_nthreads(threads);
}
void cSystem::startNthreadsFFTW(void)
{
require( fftw_init_threads() != 0, "void cSystem::startNthreadsFFTW(void)");
require(fftwf_init_threads() != 0, "void cSystem::startNthreadsFFTW(void)");
fftw_plan_with_nthreads(getNumProcessors());
fftwf_plan_with_nthreads(getNumProcessors());
std::cout << "FFTW multithreading is turned on: " << getNumProcessors() << " threads\n\n";
}
JNIEXPORT void JNICALL Java_br_usp_ime_dspbenchmarking_algorithms_fftw_FFTW_removeThreadsJNI(JNIEnv *pEnv, jobject pObj) {
if (!threads_initialized) {
char buff[150];
sprintf(buff, "Threads weren't initialized");
(*pEnv)->ThrowNew(pEnv, (*pEnv)->FindClass(pEnv, "java/lang/Exception"), buff);
} else {
fftw_plan_with_nthreads(1);
threads_enabled = 0;
__android_log_print(ANDROID_LOG_INFO, LOG_TAG, "Threads disabled");
}
}
static void plan_with_threads(int nthreads = -1,
double timelimit=FFTW_NO_TIMELIMIT){
int av_procs = std::thread::hardware_concurrency();
if(nthreads<1 || nthreads>av_procs)
nthreads = av_procs;
fftw_plan_with_nthreads(nthreads);
fftw_set_timelimit(timelimit);
}
JNIEXPORT void JNICALL Java_br_usp_ime_dspbenchmarking_algorithms_fftw_FFTW_initThreadsJNI(JNIEnv *pEnv, jobject pObj, jint num_of_threads) {
if (!threads_initialized && !fftw_init_threads()) {
char buff[150];
sprintf(buff, "Failed to initialize thread");
(*pEnv)->ThrowNew(pEnv, (*pEnv)->FindClass(pEnv, "java/lang/Exception"), buff);
} else {
fftw_plan_with_nthreads(num_of_threads);
threads_enabled = 1;
threads_initialized = 1;
__android_log_print(ANDROID_LOG_INFO, LOG_TAG, "Threads enabled");
}
}
/* The use of REDFT00 is prefered over REDFT01 because
* REDFT00 starts at x=0 and REDFT01 at x=0.5*step. This makes
* REDFT00 a better choice to reproduce the
* coarse-graining and Lennard-Jones kernels.
* With REDFT00 their integral (or their k=0 FT) is closer
* to 1 and -b respectively.
*
* Note the normalization depends on the plan used:
* REDFT00 -> 1/(2*nx-1)
* REDFT01 -> 1/(2*nx)
*
*/
EXPORT void rgrid3d_fftw_nonperiodic_alloc(rgrid3d *grid) {
long s = grid-> nx * grid->ny * grid-> nz ;
double *temp ;
temp = malloc( sizeof(double)*s ) ;
memcpy(temp, grid->value, sizeof(double)*s ) ;
fftw_plan_with_nthreads(grid_threads());
grid->plan = fftw_plan_r2r_3d(grid->nx, grid->ny, grid->nz,
grid->value, grid->value,
FFTW_REDFT10 , FFTW_REDFT10 , FFTW_REDFT10 ,
GRID_FFTW_PLAN | FFTW_DESTROY_INPUT);
fftw_plan_with_nthreads(grid_threads());
grid->iplan = fftw_plan_r2r_3d(grid->nx, grid->ny, grid->nz,
grid->value, grid->value,
FFTW_REDFT01 , FFTW_REDFT01 , FFTW_REDFT01 ,
GRID_FFTW_PLAN | FFTW_DESTROY_INPUT);
grid->fft_norm = 1.0 / (2 * (grid->nx ) * 2 * (grid->ny ) * 2 * (grid->nz ));
memcpy(grid->value, temp, sizeof(double)*s );
free(temp);
}
void fftInitThreading() {
#ifdef _OPENMP
#ifdef INTEL_MKL_VERSION
// NOTE: Using Intel MKL (and threading particularly)
// could require a lot of additional setup
fftw3_mkl.number_of_user_threads = omp_get_max_threads();
#else
fftw_init_threads();
fftw_plan_with_nthreads(omp_get_max_threads());
#endif
#endif
}
void fft_init()
{
fftw_init_threads();
fftw_plan_with_nthreads(6);
int i;
for(i=0;i<2;i++){
fft_in[i] = (fftw_complex*) fftw_malloc(sizeof(fftw_complex) * width[i]*height[i]);
fft_out[i] = (fftw_complex*) fftw_malloc(sizeof(fftw_complex) * width[i]*height[i]);
fft_plan[i]=fftw_plan_dft_2d(height[i],width[i],fft_in[i],fft_out[i],
FFTW_FORWARD, FFTW_ESTIMATE);
}
}
void FFTHandler::init(long arg_n){
//#ifndef DEBUG
fftw_init_threads();
fftw_plan_with_nthreads(omp_get_max_threads());
//#endif
n = arg_n;
leased=0;
//fprintf(stderr, "Initializing fft plan of size [%ld]\n", n);
memoryPool.resize(1);
memoryPool[0] = (double*)fftw_malloc(sizeof(double)*2*(n/2+1));
fftForwardPlan = fftw_plan_dft_r2c_1d(n, memoryPool[0], (fftw_complex*)memoryPool[0],FFTW_MEASURE);
fftReversePlan = fftw_plan_dft_c2r_1d(n, (fftw_complex*)memoryPool[0], memoryPool[0],FFTW_MEASURE);
}
int cfft2_init(int pad1 /* padding on the first axis */,
int nx, int ny /* input data size */,
int *nx2, int *ny2 /* padded data size */)
/*< initialize >*/
{
#ifdef SF_HAS_FFTW
#ifdef _OPENMP
fftw_init_threads();
sf_warning("Using threaded FFTW3! \n");
fftw_plan_with_nthreads(omp_get_max_threads());
#endif
#endif
#ifndef SF_HAS_FFTW
int i2;
#endif
nk = n1 = kiss_fft_next_fast_size(nx*pad1);
#ifndef SF_HAS_FFTW
cfg1 = kiss_fft_alloc(n1,0,NULL,NULL);
icfg1 = kiss_fft_alloc(n1,1,NULL,NULL);
#endif
n2 = kiss_fft_next_fast_size(ny);
cc = sf_complexalloc2(n1,n2);
dd = sf_complexalloc2(nk,n2);
#ifndef SF_HAS_FFTW
cfg2 = kiss_fft_alloc(n2,0,NULL,NULL);
icfg2 = kiss_fft_alloc(n2,1,NULL,NULL);
tmp = (kiss_fft_cpx **) sf_alloc(n2,sizeof(*tmp));
tmp[0] = (kiss_fft_cpx *) sf_alloc(nk*n2,sizeof(kiss_fft_cpx));
for (i2=0; i2 < n2; i2++) {
tmp[i2] = tmp[0]+i2*nk;
}
trace2 = sf_complexalloc(n2);
ctrace2 = (kiss_fft_cpx *) trace2;
#endif
*nx2 = n1;
*ny2 = n2;
wt = 1.0/(n1*n2);
return (nk*n2);
}
/****** fft_init ************************************************************
PROTO void fft_init(void)
PURPOSE Initialize the FFT routines
INPUT -.
OUTPUT -.
NOTES Global preferences are used for multhreading.
AUTHOR E. Bertin (IAP)
VERSION 29/11/2006
***/
void fft_init(void)
{
if (!firsttimeflag)
{
#ifdef USE_THREADS
if (!fftw_init_threads())
error(EXIT_FAILURE, "*Error*: thread initialization failed in ", "FFTW");
fftw_plan_with_nthreads(prefs.nthreads);
QPTHREAD_MUTEX_INIT(&fftmutex, NULL);
#endif
firsttimeflag = 1;
}
return;
}
Space_trans::Space_trans(const Config & configSettings):Data(configSettings)
{
fftw_iodim dims[DIM], howmany_dims[1];
int rank = DIM;
int howmany_rank = 1;
if(DIM == 1){
dims[0].n = m[0];
dims[0].is = n3;
dims[0].os = n3;
}
if(DIM == 2){
dims[0].n = m[0];
dims[0].is = n3;
dims[0].os = n3;
dims[1].n = m[1];
dims[1].is = n3*m[0];
dims[1].os = n3*m[0];
}
howmany_dims[0].n = n3;
howmany_dims[0].is = 1;
howmany_dims[0].os = 1;
fftw_init_threads();
fftw_plan_with_nthreads(NUM_THREADS);
pf = fftw_plan_guru_dft( rank, dims,
howmany_rank, howmany_dims,
realdata, realdata,
FFTW_FORWARD, FFTW_ESTIMATE );
pi = fftw_plan_guru_dft( rank, dims,
howmany_rank, howmany_dims,
realdata, realdata,
FFTW_BACKWARD, FFTW_ESTIMATE );
fftw_plan_with_nthreads(1);
}
void FourierTransformer::setThreadsNumber(int tNumber)
{
if (tNumber != 1)
{
threadsSetOn = true;
nthreads = tNumber;
pthread_mutex_lock(&fftw_plan_mutex);
if (fftw_init_threads() == 0)
{
REPORT_ERROR("FFTW cannot init threads (setThreadsNumber)");
}
fftw_plan_with_nthreads(nthreads);
pthread_mutex_unlock(&fftw_plan_mutex);
}
}
void FFTWInterface::create_fftw_plan(int &has_plan, fftw_plan &plan,
const int fftw_direction, const int fftw_flag)
{
if(has_plan) return;
assert(f);
fftw_plan_with_nthreads(omp_get_max_threads());
double *f_copy = 0;
if(n && !m) {
// 1d
f_copy = new double [2*n];
assert(f_copy);
memcpy(f_copy, f, 2*n*sizeof(double));
} else if(n && m) {
// 2d case
f_copy = new double [2*m*n];
assert(f_copy);
memcpy(f_copy, f, 2*m*n*sizeof(double));
}
assert(f_copy);
if(n && !m) {
plan = fftw_plan_dft_1d(n,
reinterpret_cast <fftw_complex *> (f),
reinterpret_cast <fftw_complex *> (f),
fftw_direction, fftw_flag);
memcpy(f, f_copy, 2*n*sizeof(double));
} else if(m && n) {
int row = column_major ? n : m;
int col = column_major ? m : n;
plan = fftw_plan_dft_2d(row, col,
reinterpret_cast <fftw_complex *> (f),
reinterpret_cast <fftw_complex *> (f),
fftw_direction, fftw_flag);
memcpy(f, f_copy, 2*m*n*sizeof(double));
}
if(f_copy) { delete [] f_copy; f_copy = 0; }
has_plan = 1;
}
// have a look for parallel ffts
int
parallel_ffts( void )
{
// initialise parallel fft ?
#ifdef OMP_FFTW
if( fftw_init_threads() == 0 ) {
return GLU_FAILURE ;
} else {
// in here I set the number of fftw threads to be the same
// as the usual number of parallel threads ..
fftw_plan_with_nthreads( Latt.Nthreads ) ;
fprintf( stdout , "[PAR] FFTW using %d thread(s) \n" , Latt.Nthreads ) ;
}
#endif
return GLU_SUCCESS ;
}
void ini() {
LogLevel = 2;
fftw_init_threads(); //<-- Desini aufrufen
fftw_plan_with_nthreads(1);
iniTools(5517+Parameter_SD_Selector*12345);
physicalScale = NaN;
physicalScaleError = NaN;
GoldstoneRenormFactor = 1.0;
GoldstoneRenormFactorError = 0.0;
GoldstoneRenormFactorDetermined = false;
xmlOutput_Tag = new char*[10000];
xmlOutput_Description = new char*[10000];
xmlOutput_Value = new double[10000];
xmlOutput_ValueError = new double[10000];
xmlOutput_Count = 0;
}
// have a look for parallel ffts
static int
parallel_ffts( void )
{
// initialise parallel fft ?
#ifdef OMP_FFTW
if( fftw_init_threads( ) == 0 ) {
return FAILURE ;
} else {
// in here I set the number of fftw threads to be the same
// as the usual number of parallel threads ..
#pragma omp parallel
{ nthreads = omp_get_num_threads( ) ; } // set nthreads
}
fftw_plan_with_nthreads( nthreads ) ;
printf("[PAR] FFTW using %d thread(s) \n" , nthreads ) ;
#endif
return SUCCESS ;
}
/*!
this is a straight forward main function, that does the following
- load lua config file
- get configuration table from config
- contruct a preferences class from them
- start simulation according to the preferences
- dump the result as precified in preferences
*/
int main(int argc, char * argv[argc])
{
fftw_init_threads();
fftw_plan_with_nthreads(4);
if (argc != 3) {
fprintf(stderr, "usage: %s config.lua result.dat\n", argv[0]);
return -1;
}
lua_State * L = luaL_newstate();
luaL_openlibs(L);
preferences_t * prefs = preferences_new();
if (luaL_dofile(L, argv[1])) {
fprintf(stderr, "could not load '%s' : %s\n", argv[1], lua_tostring(L, 1));
} else {
// get config table
lua_getfield(L, LUA_GLOBALSINDEX, "config");
if (lua_isnil(L, -1)) {
fprintf(stderr, "table config undefined\n");
} else {
// ref config table, so we can access it in preferences_read()
prefs->config = luaL_ref(L, LUA_REGISTRYINDEX);
if (!preferences_read(L, prefs)) {
if (!start_simulation(prefs)) {
FILE * fp = fopen(argv[2], "wb"); assert(fp);
dump_results(prefs, fp);
fclose(fp);
}
}
}
}
preferences_free(prefs);
lua_close(L);
fftw_cleanup();
fftw_cleanup_threads();
pthread_exit(NULL);
}
void init_gfft() {
double complex *wi1;
double *wir1;
DEBUG_START_FUNC;
wi1 = (double complex *) fftw_malloc( sizeof(double complex) * NTOTAL_COMPLEX);
if (wi1 == NULL) ERROR_HANDLER( ERROR_CRITICAL, "No memory for wi1 allocation");
wir1 = (double *) wi1;
#ifdef _OPENMP
fftw_plan_with_nthreads( nthreads );
#endif
#ifdef WITH_2D
r2cfft = fftw_plan_dft_r2c_2d( NX, NY, wr1, w1, FFT_PLANNING);
if (r2cfft == NULL) ERROR_HANDLER( ERROR_CRITICAL, "FFTW R2C plan creation failed");
c2rfft = fftw_plan_dft_c2r_2d( NX, NY, w1, wr1, FFT_PLANNING);
if (c2rfft == NULL) ERROR_HANDLER( ERROR_CRITICAL, "FFTW C2R plan creation failed");
#else
r2cfft = fftw_plan_dft_r2c_3d( NX, NY, NZ, wr1, w1, FFT_PLANNING);
if (r2cfft == NULL) ERROR_HANDLER( ERROR_CRITICAL, "FFTW R2C plan creation failed");
c2rfft = fftw_plan_dft_c2r_3d( NX, NY, NZ, w1, wr1, FFT_PLANNING);
if (c2rfft == NULL) ERROR_HANDLER( ERROR_CRITICAL, "FFTW C2R plan creation failed");
r2cfft_2Dslice = fftw_plan_dft_r2c_2d(NX,NY,wrh3,wh3,FFT_PLANNING);
if (r2cfft_2Dslice == NULL) ERROR_HANDLER( ERROR_CRITICAL, "FFTW r2c slice plan creation failed");
#endif
fftw_free(wi1);
fft_timer=0.0;
DEBUG_END_FUNC;
return;
}