void FftClFft::
computeC2R( Tfr::ChunkData::Ptr input, DataStorage<float>::Ptr output )
{
unsigned denseWidth = input->size().width;
unsigned redundantWidth = output->size().width;
EXCEPTION_ASSERT( denseWidth == redundantWidth/2+1 );
Tfr::ChunkData::Ptr redundantInput( new Tfr::ChunkData( redundantWidth, input->size().height ));
{
Tfr::ChunkElement* in = CpuMemoryStorage::ReadOnly<1>( input ).ptr();
Tfr::ChunkElement* out = CpuMemoryStorage::WriteAll<1>( redundantInput ).ptr();
unsigned x;
for (x=0; x<denseWidth; ++x)
out[x] = in[x];
for (; x<redundantWidth; ++x)
out[x] = conj(in[redundantWidth - x]);
}
Tfr::ChunkData::Ptr complexoutput( new Tfr::ChunkData( output->size()));
computeWithClFft(redundantInput, complexoutput, FftDirection_Inverse);
::stftDiscardImag( complexoutput, output );
}
void FftClFft::
computeR2C( DataStorage<float>::Ptr input, Tfr::ChunkData::Ptr output )
{
unsigned denseWidth = output->size().width;
unsigned redundantWidth = input->size().width;
EXCEPTION_ASSERT( denseWidth == redundantWidth/2+1 );
// interleave input to complex data
Tfr::ChunkData::Ptr complexinput( new Tfr::ChunkData( input->size()));
::stftToComplex( input, complexinput );
// make room for full output
Tfr::ChunkData::Ptr redundantOutput( new Tfr::ChunkData( redundantWidth ));
// compute
computeWithClFft(complexinput, redundantOutput, FftDirection_Forward);
// discard redundant output
{
Tfr::ChunkElement* in = CpuMemoryStorage::ReadOnly<1>( redundantOutput ).ptr();
Tfr::ChunkElement* out = CpuMemoryStorage::WriteAll<1>( output ).ptr();
unsigned x;
for (x=0; x<denseWidth; ++x)
out[x] = in[x];
}
}
void FftClFft::
compute(DataStorage<float>::Ptr input, Tfr::ChunkData::Ptr output, DataStorageSize n )
{
unsigned denseWidth = n.width/2+1;
EXCEPTION_ASSERT( output->numberOfElements()/denseWidth == n.height );
EXCEPTION_ASSERT( input->numberOfElements()/n.width == n.height );
// interleave input to complex data
Tfr::ChunkData::Ptr complexinput( new Tfr::ChunkData( input->size()));
::stftToComplex( input, complexinput );
// make room for full output
Tfr::ChunkData::Ptr redundantOutput( new Tfr::ChunkData( n.width*n.height ));
// compute
computeWithClFft(complexinput, redundantOutput, n, FftDirection_Forward);
// discard redundant output
Tfr::ChunkElement* in = CpuMemoryStorage::ReadOnly<1>( redundantOutput ).ptr();
Tfr::ChunkElement* out = CpuMemoryStorage::WriteAll<1>( output ).ptr();
#pragma omp parallel for
for (int i=0; i < (int)n.height; ++i)
{
unsigned x;
for (x=0; x<denseWidth; ++x)
out[i*denseWidth + x] = in[i*n.width + x];
}
}
std::string DataStorageString::
printDataStorageStats(DataStorage<float>::ptr data)
{
std::stringstream ss;
Statistics<float> s(data, false, true);
ss << "size = " << data->size () << ", min = " << *s.getMin () << ", max = " << *s.getMax ()
<< ", mean = " << s.getMean () << ", std = " << s.getStd ();
return ss.str();
}
std::string DataStorageString::
printDataStorage(DataStorage<float>::ptr data)
{
EXCEPTION_ASSERT(data);
std::stringstream ss;
DataStorageSize sz = data->size ();
ss << sz;
float *p = sz.width ? data->getCpuMemory () : 0;
if (sz.depth>1) {
for (int z=0; z<sz.depth; ++z) {
for (int y=0; y<sz.height; ++y) {
ss << std::endl << "[z:" << z << ", y:" << y << "] = { ";
float *q = p + y*sz.width + z*sz.height*sz.width;
if (sz.width)
ss << q[0];
for (int x=1; x<sz.width; ++x)
ss << ", " << q[x];
ss << " }";
}
}
} else if (sz.height>1) {
for (int y=0; y<sz.height; ++y) {
ss << std::endl << "[y:" << y << "] = { ";
float *q = p + y*sz.width;
if (sz.width)
ss << q[0];
for (int x=1; x<sz.width; ++x)
ss << ", " << q[x];
ss << " }";
}
} else {
ss << " = { ";
if (sz.width)
ss << p[0];
for (int x=1; x<sz.width; ++x)
ss << ", " << p[x];
ss << " }";
}
return ss.str ();
}
void FftClFft::
inverse(Tfr::ChunkData::Ptr input, DataStorage<float>::Ptr output, DataStorageSize n )
{
unsigned denseWidth = n.width/2+1;
unsigned redundantWidth = n.width;
unsigned batchcount1 = output->numberOfElements()/redundantWidth,
batchcount2 = input->numberOfElements()/denseWidth;
EXCEPTION_ASSERT( batchcount1 == batchcount2 );
EXCEPTION_ASSERT( (denseWidth-1)*2 == redundantWidth );
EXCEPTION_ASSERT( redundantWidth*n.height == output->numberOfElements() );
Tfr::ChunkData::Ptr redundantInput( new Tfr::ChunkData( n.height*redundantWidth ));
{
Tfr::ChunkElement* in = CpuMemoryStorage::ReadOnly<1>( input ).ptr();
Tfr::ChunkElement* out = CpuMemoryStorage::WriteAll<1>( redundantInput ).ptr();
#pragma omp parallel for
for (int i=0; i < (int)n.height; ++i)
{
unsigned x;
for (x=0; x<denseWidth; ++x)
out[i*redundantWidth + x] = in[i*denseWidth + x];
for (; x<redundantWidth; ++x)
out[i*redundantWidth + x] = conj(in[i*denseWidth + redundantWidth - x]);
}
}
Tfr::ChunkData::Ptr complexoutput( new Tfr::ChunkData( output->size()));
computeWithClFft(redundantInput, complexoutput, DataStorageSize( redundantWidth, n.height), FftDirection_Inverse);
::stftDiscardImag( complexoutput, output );
TIME_STFT ComputationSynchronize();
}
