void Pulsar::PolnProfileFit::solve () try
{
RealTimer clock;
clock.start();
equation->solve ();
clock.stop();
if (verbose)
{
float chisq = equation->get_solver()->get_chisq();
float nfree = equation->get_solver()->get_nfree();
cerr << "Pulsar::PolnProfileFit::solve solved in " << clock << "."
" chisq=" << chisq/nfree << endl;
}
}
catch (Error& error) {
throw error += "Pulsar::PolnProfileFit::solve";
}
void dsp::Archiver::unload (const PhaseSeries* _profiles) try
{
if (!single_archive && archive_class_name.size() == 0)
throw Error (InvalidState, "dsp::Archiver::unload",
"neither Archive nor class name specified");
if (!_profiles)
throw Error (InvalidState, "dsp::Archiver::unload",
"Profile data not provided");
if (verbose > 2)
cerr << "dsp::Archiver::unload profiles=" << _profiles << endl;
if (_profiles->get_nbin() == 0 || _profiles->get_ndat_folded() == 0)
{
if (verbose > 2)
cerr << "dsp::Archiver::unload ignoring empty sub-integration" << endl;
return;
}
on_host( _profiles, profiles );
uint64_t ndat_folded = profiles->get_ndat_folded();
uint64_t ndat_total = profiles->get_ndat_total();
double percent = double(ndat_folded)/double(ndat_total) * 100.0;
if (verbose > 2)
cerr << "dsp::Archiver::unload folded " << ndat_folded << " out of "
<< ndat_total << " total samples: " << percent << "%" << endl;
uint64_t ndat_expected = profiles->get_ndat_expected();
if (ndat_expected && ndat_expected < 0.9 * ndat_total)
{
/*
ndat_expected is the number of samples expected to be totalled in
the sub-integration. This number can possibly differ from ndat_total
due to different rounding in different thread (an untested assertion).
*/
if (verbose > 2)
cerr << "dsp::Archiver::unload ignoring incomplete sub-integration \n\t"
"expected=" << ndat_expected << " total=" << ndat_total << endl;
return;
}
if (profiles->get_integration_length() < minimum_integration_length)
{
cerr << "dsp::Archiver::unload ignoring "
<< profiles->get_integration_length() << " seconds of data" << endl;
return;
}
if (use_single_archive)
{
// Generate new archive if needed
if (!single_archive)
{
if (verbose > 2)
cerr << "dsp::Archiver::unload creating new single Archive" << endl;
single_archive = new_Archive();
}
// refer to the single archive to which all sub-integration will be written
archive = single_archive;
// add the profile data
add (archive, profiles);
// See if we've reached max subints per file, if not, we're done for now
if (subints_per_file == 0 || (archive->get_nsubint() < subints_per_file))
return;
// Max subint limit reached so we need to unload this file and
// start a new one next time.
finish();
single_archive = 0;
return;
}
if (!archive)
archive = new_Archive();
if (verbose > 2)
cerr << "dsp::Archiver::unload set Pulsar::Archive" << endl;
set (archive, profiles);
if (script.size()) try
{
if (verbose > 2)
cerr << "dsp::Archiver::unload post-processing" << endl;
if (!interpreter)
interpreter = standard_shell();
interpreter->set( archive );
interpreter->script( script );
}
catch (Error& error)
{
if (verbose)
cerr << "dsp::Archiver::unload post-processing "
<< archive->get_filename() << " failed:\n"
<< error.get_message() << endl;
return;
}
if (verbose > 2)
cerr << "dsp::Archiver::unload archive '"
<< archive->get_filename() << "'" << endl;
RealTimer timer;
if (dsp::Operation::record_time)
timer.start();
archive -> unload();
if (dsp::Operation::record_time)
{
timer.stop();
cerr << "dsp::Archiver::unload in " << timer << endl;
}
}
catch (Error& error)
{
throw error += "dsp::Archiver::unload";
}
void Speed::runTest ()
{
unsigned nfloat = nchan * nfft;
if (!real_to_complex)
nfloat *= 2;
unsigned size = sizeof(float) * nfloat;
if (!nloop)
{
nloop = (1024*1024*256) / size;
if (nloop > 2000)
nloop = 2000;
cerr << "Speed::runTest nloop=" << nloop << endl;
}
dsp::Filterbank::Engine* engine = 0;
dsp::Memory* memory = 0;
#if HAVE_CUFFT
cudaStream_t stream = 0;
// cudaStreamCreate( &stream );
engine = new CUDA::FilterbankEngine (stream);
memory = new CUDA::DeviceMemory;
#endif
if (!memory)
memory = new dsp::Memory;
if (!engine)
throw Error (InvalidState, "Speed::runTest",
"engine not set");
float* in = NULL;
if (do_fwd_fft)
{
in = (float*) memory->do_allocate (size);
memory->do_zero (in, size);
}
engine->scratch = (float*) memory->do_allocate (size + 4*sizeof(float));
dsp::TimeSeries ts;
ts.set_state( Signal::Analytic );
dsp::Filterbank temp;
temp.set_nchan (nchan);
temp.set_frequency_resolution (nfft);
temp.set_input (&ts);
engine->setup (&temp);
cerr << "entering loop" << endl;
double total_time = 0;
for (unsigned j=0; j<niter; j++)
{
RealTimer timer;
timer.start ();
for (unsigned i=0; i<nloop; i++)
engine->perform (in);
engine->finish ();
timer.stop ();
total_time += timer.get_elapsed();
}
double time_us = total_time * 1e6 / (nloop*niter);
// cerr << "time=" << time << endl;
if (in)
memory->do_free (in);
memory->do_free (engine->scratch);
double log2_nfft = log2(nfft);
double log2_nchan = log2(nchan);
double bwd = 2;
if (nchan == 1)
bwd = 1;
double mflops = 5.0 * nfft * nchan * (bwd*log2_nfft + log2_nchan) / time_us;
cerr << "nchan=" << nchan << " nfft=" << nfft << " time=" << time_us << "us"
" log2(nfft)=" << log2_nfft << " log2(nchan)=" << log2_nchan <<
" mflops=" << mflops << endl;
cout << nchan << " " << nfft << " " << time_us << " "
<< log2_nchan << " " << log2_nfft << " " << mflops << endl;
}
static PyObject *
solve_slae(PyObject *dummy, PyObject *args)
{
Perm vecPerm(nVectors);
bemcluster<vec3d>* clTreeVec =
new bemcluster<vec3d>(vectors, vecPerm.op_perm, 0, nVectors);
clTreeVec->createClusterTree(bmin, vecPerm.op_perm, vecPerm.po_perm);
const unsigned nClustersPan = clTreeVec->getncl();
std::cout << "done, " << nClustersPan << " clusters -- ";
std::cout << inMB(nClustersPan * sizeof(bemcluster<vec3d>)) << " MB." << std::endl;
bemblcluster<vec3d,vec3d>* blclTreeVec =
new bemblcluster<vec3d,vec3d>(clTreeVec, clTreeVec);
unsigned nblcksVec = 0;
blclTreeVec->subdivide(clTreeVec,clTreeVec, eta * eta, nblcksVec);
std::cout << "done, " << nblcksVec << " blocks -- ";
std::cout << inMB(blclTreeVec->size()) << " MB." << std::endl;
RealTimer timer;
kernel = &callback_kernel;
MATGENKERNEL MatGen(vectors, vecPerm.op_perm, kernel);
BEMMatrix<vec3d,vec3d> A(nVectors, blclTreeVec);
matgenGeH_omp(MatGen, nblcksVec, A.blclTree, eps_matgen, rankmax, A.blcks);
{
const double allmem = sizeH(A.blclTree, A.blcks);
io::displayInfo(allmem, nVectors, timer.current(), sizeof(dcomp));
}
// std::cout << "Agglomerating matrix ... " << std::flush;
// timer.restart();
// agglH(A.blclTree, A.blcks, eps_aggl, rankmax);
// std::cout << "done." << std::endl;
// {
// const double allmem = sizeH(A.blclTree, A.blcks);
// io::displayInfo(allmem, nVectors, timer.current(), sizeof(dcomp));
// }
dcomp *b = new dcomp[nVectors];
unsigned i = 0;
do {
b[i] = callback_vectorb(vecPerm.op_perm[i] + 1);
i++;
} while( i < nVectors );
dcomp *x = new dcomp[nVectors];
std::fill_n(x, nVectors, dcomp(0.,0.));
double acc = eps_gmres;
unsigned steps = steps_gmres;
if (GMRes(A, b, x, acc, 100, steps))
std::cout << "GMRes: iteration did not converge.";
else
std::cout << "GMRes converged to " << acc << " in "
<< steps << " steps.";
std::cout << " Solution took " << timer << "." << std::endl;
solution = new dcomp[nVectors];
i = 0;
do {
solution[i] = x[vecPerm.po_perm[i]];
i++;
} while( i < nVectors );
delete clTreeVec;
return Py_None;
}