LV2_Handle PitchShifter::instantiate(const LV2_Descriptor* descriptor, double samplerate, const char* bundle_path, const LV2_Feature* const* features)
{
PitchShifter *plugin = new PitchShifter();
plugin->nBuffers = 20;
plugin->Qcolumn = plugin->nBuffers;
plugin->hopa = TAMANHO_DO_BUFFER;
plugin->N = plugin->nBuffers*plugin->hopa;
plugin->cont = 0;
plugin->s = 0;
plugin->g = 1;
plugin->Hops = (int*)calloc(plugin->Qcolumn,sizeof(int)); memset(plugin->Hops, plugin->hopa, plugin->Qcolumn );
plugin->frames = (double*)calloc(plugin->N,sizeof(double));
plugin->ysaida = (double*)calloc(2*(plugin->N + 2*(plugin->Qcolumn-1)*plugin->hopa),sizeof(double));
plugin->yshift = (double*)calloc(plugin->hopa,sizeof(double));
plugin->b = (double**)calloc(plugin->hopa,sizeof(double*));
plugin->frames2 = fftwf_alloc_real(plugin->N);
plugin->q = fftwf_alloc_real(plugin->N);
plugin->fXa = fftwf_alloc_complex(plugin->N/2 + 1);
plugin->fXs = fftwf_alloc_complex(plugin->N/2 + 1);
plugin->Xa.zeros(plugin->N/2 + 1);
plugin->Xs.zeros(plugin->N/2 + 1);
plugin->XaPrevious.zeros(plugin->N/2 + 1);
plugin->Xa_arg.zeros(plugin->N/2 + 1);
plugin->XaPrevious_arg.zeros(plugin->N/2 + 1);
plugin->Phi.zeros(plugin->N/2 + 1);
plugin->PhiPrevious.zeros(plugin->N/2 + 1);
plugin->d_phi.zeros(plugin->N/2 + 1);
plugin->d_phi_prime.zeros(plugin->N/2 + 1);
plugin->d_phi_wrapped.zeros(plugin->N/2 + 1);
plugin->omega_true_sobre_fs.zeros(plugin->N/2 + 1);
plugin->AUX.zeros(plugin->N/2 + 1);
plugin->Xa_abs.zeros(plugin->N/2 + 1);
plugin->w.zeros(plugin->N); hann(plugin->N,&plugin->w);
plugin->I.zeros(plugin->N/2 + 1); plugin->I = linspace(0, plugin->N/2,plugin->N/2 + 1);
for (int i=1 ; i<= (plugin->nBuffers); i++)
{
plugin->b[i-1] = &plugin->frames[(i-1)*plugin->hopa];
}
plugin->p = fftwf_plan_dft_r2c_1d(plugin->N, plugin->frames2, plugin->fXa, FFTW_ESTIMATE);
plugin->p2 = fftwf_plan_dft_c2r_1d(plugin->N, plugin->fXs, plugin->q, FFTW_ESTIMATE);
return (LV2_Handle)plugin;
}
fftwf_complex*
FFTW::alloc_complex<float>(size_t n) {
assert(n > 0);
fftwf_complex* p = fftwf_alloc_complex(n);
if (p == nullptr) THROW_EXCEPTION(FFTWException, "Error in fftw_alloc_complex (float, n = " << n << ").");
return p;
}
PSSinthesis::PSSinthesis(PSAnalysis *obj, const char* wisdomFile) //Construtor
{
Qcolumn = obj->Qcolumn;
hopa = obj->hopa;
N = obj->N;
omega_true_sobre_fs = &obj->omega_true_sobre_fs;
Xa_abs = &obj->Xa_abs;
w = &obj->w;
first = true;
hops = new int[Qcolumn]; fill_n(hops,Qcolumn,hopa);
ysaida = new double[2*N + 4*(Qcolumn-1)*hopa]; fill_n(ysaida,2*N + 4*(Qcolumn-1)*hopa,0);
yshift = new double[hopa]; fill_n(yshift,hopa,0);
q = fftwf_alloc_real(N);
fXs = fftwf_alloc_complex(N/2 + 1);
Xs.zeros(N/2 + 1);
Phi.zeros(N/2 + 1);
PhiPrevious.zeros(N/2 + 1);
if (fftwf_import_wisdom_from_filename(wisdomFile) != 0)
{
p2 = fftwf_plan_dft_c2r_1d(N, fXs, q, FFTW_WISDOM_ONLY);
}
else
{
p2 = NULL;
printf("PSSinthesis: failed to import wisdom file '%s'\n", wisdomFile);
}
}
ProcessSamples::ProcessSamples(uint32_t numSamples,
uint32_t sampleRate,
uint32_t enob,
float threshold,
gr::fft::window::win_type windowType,
Mode mode,
uint32_t threadCount,
std::string fileNameBase,
double useBandWidth,
double dcIgnoreWidth,
uint32_t preTrigger,
uint32_t postTrigger)
: m_sampleCount(numSamples),
m_sampleRate(sampleRate),
m_enob(enob),
m_fileCounter(0),
m_threshold(threshold),
m_fftWindow(windowType, numSamples),
m_correctDCOffset(false),
m_useWindow(uint32_t(useBandWidth * numSamples / 2.0)),
// This is only for hackRF.
m_dcIgnoreWindow(4),
// m_dcIgnoreWindow(uint32_t(dcIgnoreWidth * numSamples / 2.0)),
m_fft(numSamples),
m_mode(mode),
m_sampleQueue(nullptr),
m_fileNameBase(fileNameBase),
m_preTrigger(preTrigger),
m_postTrigger(postTrigger),
m_writing(false),
m_endSequenceId(0),
m_threadCount(threadCount)
{
assert(mode > Illegal && mode <= FrequencyDomain);
assert(threadCount <= MAX_THREADS);
for (uint32_t threadId = 0; threadId < threadCount; threadId++) {
this->m_inputSamples[threadId] =
reinterpret_cast<fftwf_complex *>(fftwf_alloc_complex(numSamples));
this->m_fftOutputBuffer[threadId] =
reinterpret_cast<fftwf_complex *>(fftwf_alloc_complex(numSamples));
this->m_threads[threadId] = nullptr;
}
}
int main(){
int nthreads = 4;
omp_set_num_threads(nthreads);
#pragma omp parallel
fprintf(stderr,"nthreads %d \n", omp_get_num_threads());
int n3 = 128;
int n2 = 128;
int n1 = 128;
// float ***array = sf_floatalloc3(n1,n2,n3);
float *array = fftwf_alloc_real(n3*n2*n1);
fftwf_complex* cout = fftwf_alloc_complex(n3*n2*n1);
int err = fftwf_init_threads();
if (err == 0) {
fprintf(stderr,"something went wrong with fftw\n");
}
fprintf(stderr,"Got here\n");
double start,end;
start = omp_get_wtime()*omp_get_wtick();
fftwf_plan_with_nthreads(nthreads);
fftwf_plan plan = fftwf_plan_dft_r2c_3d(
n1,n2,n3,
array,cout,
FFTW_MEASURE);
end = omp_get_wtime()*omp_get_wtick();
fprintf(stderr,"elapsed time: %f %f %f\n",end,start,end-start);
for(int i = 0; i < n3*n2*n1; ++i)
array[i] = rand()/RAND_MAX;
//float start = clock()/CLOCKS_PER_SEC;
start = omp_get_wtime();
for(int i=0; i < 1001; ++i)
fftwf_execute(plan);
//float end = clock()/CLOCKS_PER_SEC;
end = omp_get_wtime();
fprintf(stderr,"elapsed time: %f time/calc %f\n",
end-start,(end-start)/100.0);
fftwf_cleanup_threads();
fftwf_cleanup();
fftwf_destroy_plan(plan);
fftwf_free(cout);
fftwf_free(array);
//free(**array); free(*array); free(array);
return 0;
}
int main(int argc, char **argv) {
fftwf_plan plan;
fftwf_complex *data;
ptrdiff_t alloc_local, local_n0, local_0_start, i, j;
if (argc != 2) { printf("usage: ./fft_mpi MATRIX_SIZE\n"); exit(1); }
const ptrdiff_t N0 = atoi(argv[1]);
const ptrdiff_t N1 = N0;
int id;
double startTime, totalTime;
totalTime = 0;
MPI_Init(&argc, &argv);
MPI_Comm_rank(MPI_COMM_WORLD, &id);
fftwf_mpi_init();
/* get local data size and allocate */
alloc_local = fftwf_mpi_local_size_2d(N0, N1, MPI_COMM_WORLD, &local_n0, &local_0_start);
data = fftwf_alloc_complex(alloc_local);//(fftwf_complex *) fftwf_malloc(sizeof(fftw_complex) * alloc_local);
/* create plan for in-place forward DFT */
plan = fftwf_mpi_plan_dft_2d(N0, N1, data, data, MPI_COMM_WORLD, FFTW_FORWARD, FFTW_ESTIMATE);
/* initialize data to some function my_function(x,y) */
for (i = 0; i < local_n0; ++i) for (j = 0; j < N1; ++j){
data[i*N1 + j][0] = local_0_start;;//my_function(local_0_start + i, j);
data[i*N1 + j][1]=i;
}
/* compute transforms, in-place, as many times as desired */
MPI_Barrier(MPI_COMM_WORLD);
if (id == 0) {
startTime = getTime();
}
fftwf_execute(plan);
MPI_Barrier(MPI_COMM_WORLD);
if (id == 0) {
totalTime += getTime() - startTime;
}
fftwf_destroy_plan(plan);
fftwf_mpi_cleanup();
if (id == 0) {
printf("%.5f\n", totalTime);
}
MPI_Finalize();
return 0;
}
PitchDetection::PitchDetection(uint32_t n_samples, int nBuffers, double SampleRate, const char* wisdomFile) //Constructor
{
hopa = n_samples;
N = nBuffers*n_samples;
fs = SampleRate;
frames = fftwf_alloc_real(2*N); memset(frames, 0, 2*N );
b = new float*[hopa];
for (int i=0 ; i< nBuffers; i++)
{
b[i] = &frames[i*hopa];
}
q = fftwf_alloc_real(2*N);
fXa = fftwf_alloc_complex(N + 1);
fXs = fftwf_alloc_complex(N + 1);
Xa.zeros(N + 1);
Xs.zeros(N + 1);
R.zeros(N);
NORM.zeros(N);
F.zeros(N);
AUTO.zeros(N);
if (fftwf_import_wisdom_from_filename(wisdomFile) != 0)
{
p = fftwf_plan_dft_r2c_1d(2*N, frames, fXa, FFTW_WISDOM_ONLY);
p2 = fftwf_plan_dft_c2r_1d(2*N, fXs, q, FFTW_WISDOM_ONLY);
}
else
{
p = fftwf_plan_dft_r2c_1d(2*N, frames, fXa, FFTW_ESTIMATE);
p2 = fftwf_plan_dft_c2r_1d(2*N, fXs, q, FFTW_ESTIMATE);
printf("PitchDetection: failed to import wisdom file '%s', using estimate instead\n", wisdomFile);
}
}
PSAnalysis::PSAnalysis(uint32_t n_samples, int nBuffers, const char* wisdomFile) //Construtor
{
Qcolumn = nBuffers;
hopa = n_samples;
N = nBuffers*n_samples;
frames = new double[N]; fill_n(frames,N,0);
b = new double*[hopa];
for (int i=0 ; i< nBuffers; i++)
b[i] = &frames[i*hopa];
frames2 = fftwf_alloc_real(N);
fXa = fftwf_alloc_complex(N/2 + 1);
Xa.zeros(N/2 + 1);
XaPrevious.zeros(N/2 + 1);
Xa_arg.zeros(N/2 + 1);
XaPrevious_arg.zeros(N/2 + 1);
d_phi.zeros(N/2 + 1);
d_phi_prime.zeros(N/2 + 1);
d_phi_wrapped.zeros(N/2 + 1);
omega_true_sobre_fs.zeros(N/2 + 1);
AUX.zeros(N/2 + 1);
Xa_abs.zeros(N/2 + 1);
w.zeros(N); hann(N,&w);
I.zeros(N/2 + 1); I = linspace(0, N/2, N/2 + 1);
if (fftwf_import_wisdom_from_filename(wisdomFile) != 0)
{
p = fftwf_plan_dft_r2c_1d(N, frames2, fXa, FFTW_WISDOM_ONLY);
}
else
{
p = NULL;
printf("PSAnalysis: failed to import wisdom file '%s'\n", wisdomFile);
}
}
float _Complex*
malloc_vector_c(size_t n)
{
return fftwf_alloc_complex(n);
}
void PitchShifter::run(LV2_Handle instance, uint32_t n_samples)
{
PitchShifter *plugin;
plugin = (PitchShifter *) instance;
float media = 0;
for (uint32_t i=1; i<n_samples; i++)
{
media = media + abs(plugin->in[i-1]);
}
if (media == 0)
{
for (uint32_t i=1; i<n_samples; i++)
{
plugin->out_1[i-1] = 0;
}
}
else
{
int hops;
double g_before = plugin->g;
plugin->g = pow(10, (float)(*(plugin->gain))/20.0);
plugin->s = (double)(*(plugin->step));
hops = round(plugin->hopa*(pow(2,(plugin->s/12))));
for (int k=1; k<= plugin->Qcolumn-1; k++)
{
plugin->Hops[k-1] = plugin->Hops[k];
}
plugin->Hops[plugin->Qcolumn-1] = hops;
if ( ((plugin->hopa) != (int)n_samples) )
{
plugin->hopa = n_samples;
plugin->N = plugin->nBuffers*plugin->hopa;
plugin->frames = (double*)realloc(plugin->frames,plugin->N*sizeof(double)); memset(plugin->frames, 0, plugin->N );
plugin->ysaida = (double*)realloc(plugin->ysaida,2*(plugin->N + 2*(plugin->Qcolumn-1)*plugin->hopa)*sizeof(double)); memset(plugin->ysaida, 0, 2*(plugin->N + 2*(plugin->Qcolumn-1)*plugin->hopa) );
plugin->yshift = (double*)realloc(plugin->yshift,plugin->hopa*sizeof(double)); memset(plugin->yshift, 0, plugin->hopa );
plugin->b = (double**)realloc(plugin->b,plugin->hopa*sizeof(double*));
fftwf_free(plugin->frames2); plugin->frames2 = fftwf_alloc_real(plugin->N);
fftwf_free(plugin->q); plugin->q = fftwf_alloc_real(plugin->N);
fftwf_free(plugin->fXa); plugin->fXa = fftwf_alloc_complex(plugin->N/2 + 1);
fftwf_free(plugin->fXs); plugin->fXs = fftwf_alloc_complex(plugin->N/2 + 1);
plugin->Xa.zeros(plugin->N/2 + 1);
plugin->Xs.zeros(plugin->N/2 + 1);
plugin->XaPrevious.zeros(plugin->N/2 + 1);
plugin->Xa_arg.zeros(plugin->N/2 + 1);
plugin->XaPrevious_arg.zeros(plugin->N/2 + 1);
plugin->Phi.zeros(plugin->N/2 + 1);
plugin->PhiPrevious.zeros(plugin->N/2 + 1);
plugin->d_phi.zeros(plugin->N/2 + 1);
plugin->d_phi_prime.zeros(plugin->N/2 + 1);
plugin->d_phi_wrapped.zeros(plugin->N/2 + 1);
plugin->omega_true_sobre_fs.zeros(plugin->N/2 + 1);
plugin->AUX.zeros(plugin->N/2 + 1);
plugin->Xa_abs.zeros(plugin->N/2 + 1);
plugin->w.zeros(plugin->N); hann(plugin->N,&plugin->w);
plugin->I.zeros(plugin->N/2 + 1); plugin->I = linspace(0, plugin->N/2,plugin->N/2 + 1);
for (int i=1 ; i<= plugin->nBuffers; i++)
{
plugin->b[i-1] = &plugin->frames[(i-1)*plugin->hopa];
}
fftwf_destroy_plan(plugin->p); plugin->p = fftwf_plan_dft_r2c_1d(plugin->N, plugin->frames2, plugin->fXa, FFTW_ESTIMATE);
fftwf_destroy_plan(plugin->p2); plugin->p2 = fftwf_plan_dft_c2r_1d(plugin->N, plugin->fXs, plugin->q, FFTW_ESTIMATE);
return;
}
for (int i=1; i<=plugin->hopa; i++)
{
for (int j=1; j<=(plugin->nBuffers-1); j++)
{
plugin->b[j-1][i-1] = plugin->b[j][i-1];
}
plugin->b[plugin->nBuffers-1][i-1] = plugin->in[i-1];
}
if ( plugin->cont < plugin->nBuffers-1)
{
plugin->cont = plugin->cont + 1;
}
else
{
shift(plugin->N, plugin->hopa, plugin->Hops, plugin->frames, plugin->frames2, &plugin->w, &plugin->XaPrevious, &plugin->Xa_arg, &plugin->Xa_abs, &plugin->XaPrevious_arg, &plugin->PhiPrevious, plugin->yshift, &plugin->Xa, &plugin->Xs, plugin->q, &plugin->Phi, plugin->ysaida, plugin->ysaida2, plugin->Qcolumn, &plugin->d_phi, &plugin->d_phi_prime, &plugin->d_phi_wrapped, &plugin->omega_true_sobre_fs, &plugin->I, &plugin->AUX, plugin->p, plugin->p2, plugin->fXa, plugin->fXs);
for (int i=1; i<=plugin->hopa; i++)
{
plugin->out_1[i-1] = (g_before + ((plugin->g - g_before)/(plugin->hopa - 1))*(i-1) )*(float)plugin->yshift[i-1];
}
}
}
}
// Setup variables for 2LPT initial condition
void lpt_init(const int nc, const void* mem, const size_t size)
{
// nc: number of mesh per dimension
ptrdiff_t local_nx, local_x_start;
ptrdiff_t total_size=
fftwf_mpi_local_size_3d(nc, nc, nc/2+1, MPI_COMM_WORLD,
&local_nx, &local_x_start);
Local_nx= local_nx;
Local_x_start= local_x_start;
//
// Allocate memory
//
if(mem == 0) {
// allocate memory here
size_t bytes= sizeof(fftwf_complex)*total_size;
int allocation_failed= 0;
// 1&2 displacement
for(int axes=0; axes < 3; axes++) {
cdisp[axes]= fftwf_alloc_complex(total_size);
disp[axes] = (float*) cdisp[axes];
cdisp2[axes]= fftwf_alloc_complex(total_size);
disp2[axes] = (float*) cdisp2[axes];
bytes += 2*sizeof(fftwf_complex)*total_size;
allocation_failed = allocation_failed ||
(cdisp[axes] == 0) || (cdisp2[axes] == 0);
}
// 2LPT
for(int i=0; i<6; i++) {
cdigrad[i] = (fftwf_complex *) fftwf_alloc_complex(total_size);
digrad[i] = (float*) cdigrad[i];
bytes += sizeof(fftwf_complex)*total_size;
allocation_failed = allocation_failed || (digrad[i] == 0);
}
if(allocation_failed)
msg_abort(2003, "Error: Failed to allocate memory for 2LPT."
"Tried to allocate %d Mbytes\n", (int)(bytes/(1024*1024)));
msg_printf(info, "%d Mbytes allocated for LPT\n", (int)(bytes/(1024*1024)));
}
else {
size_t bytes= 0;
fftwf_complex* p= (fftwf_complex*) mem;
for(int axes=0; axes<3; axes++) {
cdisp[axes]= p;
disp[axes]= (float*) p;
bytes += sizeof(fftwf_complex)*total_size*2;
p += total_size;
}
for(int i=0; i<6; i++) {
cdigrad[i]= p;
digrad[i]= (float*) p;
bytes += sizeof(fftwf_complex)*total_size;
p += total_size;
}
assert(bytes <= size);
}
//
// FFTW3 plans
//
for(int i=0; i<6; ++i)
Inverse_plan[i]=
fftwf_mpi_plan_dft_c2r_3d(nc, nc, nc, cdigrad[i], digrad[i],
MPI_COMM_WORLD, FFTW_ESTIMATE);
Forward_plan=
fftwf_mpi_plan_dft_r2c_3d(nc, nc, nc, digrad[3], cdigrad[3],
MPI_COMM_WORLD, FFTW_ESTIMATE);
for(int i=0; i<3; ++i) {
Disp_plan[i]=
fftwf_mpi_plan_dft_c2r_3d(nc, nc, nc, cdisp[i], disp[i],
MPI_COMM_WORLD, FFTW_ESTIMATE);
Disp2_plan[i]=
fftwf_mpi_plan_dft_c2r_3d(nc, nc, nc, cdisp2[i], disp2[i],
MPI_COMM_WORLD, FFTW_ESTIMATE);
}
// FFTW_MPI_TRANSPOSED_IN/FFTW_MPI_TRANSPOSED_OUT would be faster
// FFTW_MEASURE is probably better for multiple realization
// misc data
Nmesh= nc;
Nsample= nc;
seedtable = malloc(Nmesh * Nmesh * sizeof(unsigned int)); assert(seedtable);
}
int shrinkWrap
(
float * const & rIntensity,
const std::vector<unsigned> & rSize,
unsigned rnCycles,
float rTargetError,
float rHioBeta,
float rIntensityCutOffAutoCorel,
float rIntensityCutOff,
float rSigma0,
float rSigmaChange,
unsigned rnHioCycles
)
{
if ( rSize.size() != 2 ) return 1;
const unsigned & Ny = rSize[1];
const unsigned & Nx = rSize[0];
/* Evaluate input parameters and fill with default values if necessary */
if ( rIntensity == NULL ) return 1;
if ( rTargetError <= 0 ) rTargetError = 1e-5;
if ( rnHioCycles == 0 ) rnHioCycles = 20;
if ( rHioBeta <= 0 ) rHioBeta = 0.9;
if ( rIntensityCutOffAutoCorel <= 0 ) rIntensityCutOffAutoCorel = 0.04;
if ( rIntensityCutOff <= 0 ) rIntensityCutOff = 0.2;
if ( rSigma0 <= 0 ) rSigma0 = 3.0;
if ( rSigmaChange <= 0 ) rSigmaChange = 0.01;
float sigma = rSigma0;
/* calculate this (length of array) often needed value */
unsigned nElements = 1;
for ( unsigned i = 0; i < rSize.size(); ++i )
{
assert( rSize[i] > 0 );
nElements *= rSize[i];
}
/* allocate needed memory so that HIO doesn't need to allocate and
* deallocate on each call */
fftwf_complex * const curData = fftwf_alloc_complex( nElements );
fftwf_complex * const gPrevious = fftwf_alloc_complex( nElements );
auto const isMasked = new float[nElements];
/* create fft plans G' to g' and g to G */
auto toRealSpace = fftwf_plan_dft( rSize.size(),
(int*) &rSize[0], curData, curData, FFTW_BACKWARD, FFTW_ESTIMATE );
auto toFreqSpace = fftwf_plan_dft( rSize.size(),
(int*) &rSize[0], gPrevious, curData, FFTW_FORWARD, FFTW_ESTIMATE );
/* create first guess for mask from autocorrelation (fourier transform
* of the intensity @see
* https://en.wikipedia.org/wiki/Wiener%E2%80%93Khinchin_theorem */
#pragma omp parallel for
for ( unsigned i = 0; i < nElements; ++i )
{
curData[i][0] = rIntensity[i]; /* Re */
curData[i][1] = 0;
}
fftwf_execute( toRealSpace );
complexNormElementwise( isMasked, curData, nElements );
/* fftShift is not necessary, but I introduced this, because for the
* example it shifted the result to a better looking position ... */
//fftShift( isMasked, Nx,Ny );
libs::gaussianBlur( isMasked, Nx, Ny, sigma );
#if DEBUG_SHRINKWRAPP_CPP == 1
std::ofstream file;
std::string fname = std::string("shrinkWrap-init-mask-blurred");
file.open( ( fname + std::string(".dat") ).c_str() );
for ( unsigned ix = 0; ix < rSize[0]; ++ix )
{
for ( unsigned iy = 0; iy < rSize[1]; ++iy )
file << std::setw(10) << isMasked[ iy*rSize[0] + ix ] << " ";
file << "\n";
}
file.close();
std::cout << "Written out " << fname << ".png\n";
#endif
/* apply threshold to make binary mask */
{
const auto absMax = vectorMax( isMasked, nElements );
const float threshold = rIntensityCutOffAutoCorel * absMax;
#pragma omp parallel for
for ( unsigned i = 0; i < nElements; ++i )
isMasked[i] = isMasked[i] < threshold ? 1 : 0;
}
#if DEBUG_SHRINKWRAPP_CPP == 1
fname = std::string("shrinkWrap-init-mask");
file.open( ( fname + std::string(".dat") ).c_str() );
for ( unsigned ix = 0; ix < rSize[0]; ++ix )
{
for ( unsigned iy = 0; iy < rSize[1]; ++iy )
file << std::setw(10) << isMasked[ iy*rSize[0] + ix ] << " ";
file << "\n";
}
file.close();
std::cout << "Written out " << fname << ".png\n";
#endif
/* copy original image into fftw_complex array and add random phase */
#pragma omp parallel for
for ( unsigned i = 0; i < nElements; ++i )
{
curData[i][0] = rIntensity[i]; /* Re */
curData[i][1] = 0;
}
/* in the first step the last value for g is to be approximated
* by g'. The last value for g, called g_k is needed, because
* g_{k+1} = g_k - hioBeta * g' ! This is inside the loop
* because the fft is needed */
#pragma omp parallel for
for ( unsigned i = 0; i < nElements; ++i )
{
gPrevious[i][0] = curData[i][0];
gPrevious[i][1] = curData[i][1];
}
/* repeatedly call HIO algorithm and change mask */
for ( unsigned iCycleShrinkWrap = 0; iCycleShrinkWrap < rnCycles; ++iCycleShrinkWrap )
{
/************************** Update Mask ***************************/
std::cout << "Update Mask with sigma=" << sigma << "\n";
/* blur |g'| (normally g' should be real!, so |.| not necessary) */
complexNormElementwise( isMasked, curData, nElements );
libs::gaussianBlur( isMasked, Nx, Ny, sigma );
const auto absMax = vectorMax( isMasked, nElements );
/* apply threshold to make binary mask */
const float threshold = rIntensityCutOff * absMax;
#pragma omp parallel for
for ( unsigned i = 0; i < nElements; ++i )
isMasked[i] = isMasked[i] < threshold ? 1 : 0;
/* update the blurring sigma */
sigma = fmax( 1.5, ( 1 - rSigmaChange ) * sigma );
for ( unsigned iHioCycle = 0; iHioCycle < rnHioCycles; ++iHioCycle )
{
/* apply domain constraints to g' to get g */
#pragma omp parallel for
for ( unsigned i = 0; i < nElements; ++i )
{
if ( isMasked[i] == 1 or /* g' */ curData[i][0] < 0 )
{
gPrevious[i][0] -= rHioBeta * curData[i][0];
gPrevious[i][1] -= rHioBeta * curData[i][1];
}
else
{
gPrevious[i][0] = curData[i][0];
gPrevious[i][1] = curData[i][1];
}
}
/* Transform new guess g for f back into frequency space G' */
fftwf_execute( toFreqSpace );
/* Replace absolute of G' with measured absolute |F| */
applyComplexModulus( curData, curData, rIntensity, nElements );
fftwf_execute( toRealSpace );
} // HIO loop
/* check if we are done */
const float currentError = imresh::libs::calculateHioError( curData /*g'*/, isMasked, nElements );
std::cout << "[Error " << currentError << "/" << rTargetError << "] "
<< "[Cycle " << iCycleShrinkWrap << "/" << rnCycles-1 << "]"
<< "\n";
if ( rTargetError > 0 && currentError < rTargetError )
break;
if ( iCycleShrinkWrap >= rnCycles )
break;
} // shrink wrap loop
for ( unsigned i = 0; i < nElements; ++i )
rIntensity[i] = curData[i][0];
/* free buffers and plans */
fftwf_destroy_plan( toFreqSpace );
fftwf_destroy_plan( toRealSpace );
fftwf_free( curData );
fftwf_free( gPrevious);
delete[] isMasked;
return 0;
}
void conv_tilde_init_fft_plans(t_conv_tilde* x)
{
post("Allocating FFT arrays..");
/* Allocate complex arrays for the fftw plans. */
x->input_complex = fftwf_alloc_complex(sizeof(fftwf_complex) * x->fft_size);
x->ir_complex = fftwf_alloc_complex(sizeof(fftwf_complex) * x->fft_size);
x->out_complex = fftwf_alloc_complex(sizeof(fftwf_complex) * x->fft_size);
x->outTemp = (float*) fftwf_malloc(sizeof(float) * x->fft_size);
post("Allocating plans. This might take a while...");
/* Create plans for in-place or out-of-place transform according to the
INPLACE macro. */
#ifdef INPLACE
x->fftplan_in = fftwf_plan_dft_1d(x->fft_size,
x->input_complex, x->input_complex, FFTW_FORWARD, FFTW_MEASURE);
x->fftplan_ir = fftwf_plan_dft_1d(x->fft_size,
x->ir_complex, x->ir_complex, FFTW_FORWARD, FFTW_MEASURE);
#else
/* Need additional real arrays for out-of-place transform. */
x->input_to_fft = (float*) fftwf_malloc(sizeof(float) * x->fft_size);
x->ir_to_fft = (float*) fftwf_malloc(sizeof(float) * x->fft_size);
x->fftplan_in = fftwf_plan_dft_r2c_1d(x->fft_size,
x->input_to_fft, x->input_complex, FFTW_MEASURE);
x->fftplan_ir = fftwf_plan_dft_r2c_1d(x->fft_size,
x->ir_to_fft, x->ir_complex, FFTW_MEASURE);
#endif
x->fftplan_inverse = fftwf_plan_dft_c2r_1d(x->fft_size,
x->out_complex, x->outTemp, FFTW_MEASURE);
/* With threads. */
#ifdef THREADS
post("Allocating arrays for thread 2...");
x->input_complex_2 = fftwf_alloc_complex(sizeof(fftwf_complex) * x->fft_size);
x->ir_complex_2 = fftwf_alloc_complex(sizeof(fftwf_complex) * x->fft_size);
x->out_complex_2 = fftwf_alloc_complex(sizeof(fftwf_complex) * x->fft_size);
x->outTemp_2 = (float*) fftwf_malloc(sizeof(float) * x->fft_size);
post("Allocating plans for thread 2. This might take a while...");
#ifdef INPLACE
x->fftplan_in_2 = fftwf_plan_dft_1d(x->fft_size,
x->input_complex_2, x->input_complex_2, FFTW_FORWARD, FFTW_MEASURE);
x->fftplan_ir_2 = fftwf_plan_dft_1d(x->fft_size,
x->ir_complex_2, x->ir_complex_2, FFTW_FORWARD, FFTW_MEASURE);
#else
/* Need additional real arrays for out-of-place transform. */
x->input_to_fft_2 = (float*) fftwf_malloc(sizeof(float) * x->fft_size);
x->ir_to_fft_2 = (float*) fftwf_malloc(sizeof(float) * x->fft_size);
x->fftplan_in_2 = fftwf_plan_dft_r2c_1d(x->fft_size,
x->input_to_fft_2, x->input_complex_2, FFTW_MEASURE);
x->fftplan_ir_2 = fftwf_plan_dft_r2c_1d(x->fft_size,
x->ir_to_fft_2, x->ir_complex_2, FFTW_MEASURE);
#endif
x->fftplan_inverse_2 = fftwf_plan_dft_c2r_1d(x->fft_size,
x->out_complex_2, x->outTemp_2, FFTW_MEASURE);
#endif
post("Done!");
}
/* Constructor */
void *conv_tilde_new(t_floatarg channels)
{
int i, n;
t_conv_tilde *x = (t_conv_tilde *)pd_new(conv_tilde_class);
x->f = 0;
x->framesize = DEFAULTFRAMESIZE; /* update this later if needed */
x->fft_size = 2 * x->framesize;
x->out_gain = 1.0 / x->fft_size; /* Needs the decimal point! */
/* Clip the channel between 1 and default maximum channels */
x->channels = CLIP((int)channels, 1, DEFAULT_AUDIO_CHANNELS);
/* Threads are not worth with only one channel. */
//if ( x->channels <= 1 )
// #undef THREADS
x->all_channels = x->channels * 3;
/* Assign an inlet pair (audio + IR) for each channel.
One inlet is in place by default. */
for (i = 1; i < x->channels * 2; i++)
inlet_new(&x->x_obj, &x->x_obj.ob_pd, &s_signal, &s_signal);
/* Assign outlets for each output channel. */
for (i = 0; i < x->channels; i++)
outlet_new(&x->x_obj, &s_signal);
/* Allocate memory for the DSP vector */
x->x_myvec = (t_int **)t_getbytes(sizeof(t_int *) * (x->all_channels + 3));
if (!x->x_myvec)
{
error("conv~: out of memory");
return NULL;
}
/* Allocate memory for the parameters vector. */
//x->parameters = (t_float*) malloc(sizeof(t_float) * NUMBER_OF_PARAMETERS);
/* Allocate memory for channel variables. 1 variable per channel. */
x->input_rw_ptr = (int*) malloc(sizeof(int) * x->channels);
x->ir_end_ptr = (int*) malloc(sizeof(int) * x->channels);
x->ir_frames = (int*) malloc(sizeof(int) * x->channels);
x->frames_stored = (int*) malloc(sizeof(int) * x->channels);
x->irlength = (int*) malloc(sizeof(t_float) * x->channels);
x->ircount = (int*) malloc(sizeof(t_float) * x->channels);
x->ircount_prev = (int*) malloc(sizeof(int) * x->channels);
x->bypass = (int*) malloc(sizeof(int) * x->channels);
/* Allocate multi-dimensional storage arrays and the overlap-save array. */
x->stored_input = (fftwf_complex**) malloc( sizeof(fftwf_complex*) * x->channels );
x->stored_ir = (fftwf_complex**) malloc( sizeof(fftwf_complex*) * x->channels );
x->overlap_save = (float**) malloc( sizeof(float*) * x->channels );
for ( i = 0; i < x->channels; i++)
{
x->stored_input[i] = fftwf_alloc_complex(sizeof(fftwf_complex) * 2 * STORAGELENGTH);
x->stored_ir[i] = fftwf_alloc_complex(sizeof(fftwf_complex) * 2 * STORAGELENGTH);
x->overlap_save[i] = (float*) fftwf_malloc( sizeof(float) * x->framesize );
}
/* Init pointers and variables. */
for ( i = 0; i < x->channels; i++ )
{
x->ir_end_ptr[i] = 0;
x->input_rw_ptr[i] = 0;
x->ircount[i] = 0;
x->ircount_prev[i] = -1;
x->irlength[i] = 0;
x->ir_frames[i] = 0;
x->frames_stored[i] = 0;
x->bypass[i] = 0;
}
#ifdef THREADS
/* round up the half way point. */
x->channels_halved = ( x->channels + 1 ) / 2;
#endif
/* Initialize FFTW arrays and plans. */
conv_tilde_init_fft_plans(x);
return (void *)x;
}