std::vector<continuous_result> continuous::split_input(size_t num_threads) {
flexible_type current_min = m_transformer.min;
flexible_type current_max = m_transformer.max;
double current_scale_min = m_transformer.scale_min;
double current_scale_max = m_transformer.scale_max;
std::vector<continuous_result> thread_results(num_threads);
for (auto& thread_result : thread_results) {
thread_result.init(current_min, current_max, current_scale_min, current_scale_max);
}
return thread_results;
}
CRef<CSearchResultSet> CLocalRPSBlast::RunThreadedSearch(void)
{
s_ModifyVolumePaths(m_rps_databases);
if((kAutoThreadedSearch == m_num_of_threads) ||
(m_num_of_threads > m_rps_databases.size()))
{
//Default num of thread : a thread for each db
m_num_of_threads = m_rps_databases.size();
}
else if(m_num_of_threads < m_rps_databases.size())
{
// Combine databases, modified the size of rps_database
s_MapDbToThread(m_rps_databases, m_num_of_threads);
}
vector<CRef<CSearchResultSet> * > thread_results(m_num_of_threads, NULL);
vector <CRPSThread* > thread(m_num_of_threads, NULL);
vector<CRef<CSearchResultSet> > results;
for(unsigned int t=0; t < m_num_of_threads; t++)
{
// CThread destructor is protected, all threads destory themselves when terminated
thread[t] = (new CRPSThread(m_query_vector, m_rps_databases[t], m_opt_handle->SetOptions().Clone()));
thread[t]->Run();
}
for(unsigned int t=0; t < m_num_of_threads; t++)
{
thread[t]->Join(reinterpret_cast<void**> (&thread_results[t]));
}
for(unsigned int t=0; t < m_num_of_threads; t++)
{
results.push_back(*(thread_results[t]));
}
CRef<CBlastRPSInfo> rpsInfo = CSetupFactory::CreateRpsStructures(m_db_name,
CRef<CBlastOptions> (&(m_opt_handle->SetOptions())));
return s_CombineSearchSets(results, m_num_of_threads);
}
void vector_scan_impl(vector_base<NumericT> const & vec1,
vector_base<NumericT> & vec2,
bool is_inclusive)
{
NumericT const * data_vec1 = detail::extract_raw_pointer<NumericT>(vec1);
NumericT * data_vec2 = detail::extract_raw_pointer<NumericT>(vec2);
vcl_size_t start1 = viennacl::traits::start(vec1);
vcl_size_t inc1 = viennacl::traits::stride(vec1);
vcl_size_t size1 = viennacl::traits::size(vec1);
if (size1 < 1)
return;
vcl_size_t start2 = viennacl::traits::start(vec2);
vcl_size_t inc2 = viennacl::traits::stride(vec2);
#ifdef VIENNACL_WITH_OPENMP
if (size1 > VIENNACL_OPENMP_VECTOR_MIN_SIZE)
{
std::vector<NumericT> thread_results(omp_get_max_threads());
// inclusive scan each thread segment:
#pragma omp parallel
{
vcl_size_t work_per_thread = (size1 - 1) / thread_results.size() + 1;
vcl_size_t thread_start = work_per_thread * omp_get_thread_num();
vcl_size_t thread_stop = std::min<vcl_size_t>(thread_start + work_per_thread, size1);
NumericT thread_sum = 0;
for(vcl_size_t i = thread_start; i < thread_stop; i++)
thread_sum += data_vec1[i * inc1 + start1];
thread_results[omp_get_thread_num()] = thread_sum;
}
// exclusive-scan of thread results:
NumericT current_offset = 0;
for (vcl_size_t i=0; i<thread_results.size(); ++i)
{
NumericT tmp = thread_results[i];
thread_results[i] = current_offset;
current_offset += tmp;
}
// exclusive/inclusive scan of each segment with correct offset:
#pragma omp parallel
{
vcl_size_t work_per_thread = (size1 - 1) / thread_results.size() + 1;
vcl_size_t thread_start = work_per_thread * omp_get_thread_num();
vcl_size_t thread_stop = std::min<vcl_size_t>(thread_start + work_per_thread, size1);
NumericT thread_sum = thread_results[omp_get_thread_num()];
if (is_inclusive)
{
for(vcl_size_t i = thread_start; i < thread_stop; i++)
{
thread_sum += data_vec1[i * inc1 + start1];
data_vec2[i * inc2 + start2] = thread_sum;
}
}
else
{
for(vcl_size_t i = thread_start; i < thread_stop; i++)
{
NumericT tmp = data_vec1[i * inc1 + start1];
data_vec2[i * inc2 + start2] = thread_sum;
thread_sum += tmp;
}
}
}
} else
#endif
{
NumericT sum = 0;
if (is_inclusive)
{
for(vcl_size_t i = 0; i < size1; i++)
{
sum += data_vec1[i * inc1 + start1];
data_vec2[i * inc2 + start2] = sum;
}
}
else
{
for(vcl_size_t i = 0; i < size1; i++)
{
NumericT tmp = data_vec1[i * inc1 + start1];
data_vec2[i * inc2 + start2] = sum;
sum += tmp;
}
}
}
}