Csr/OpenMP/Morpheus_Convert_Impl.hpp

 
#ifndef MORPHEUS_CSR_OPENMP_CONVERT_IMPL_HPP
#define MORPHEUS_CSR_OPENMP_CONVERT_IMPL_HPP
 
#include <Morpheus_Macros.hpp>
#if defined(MORPHEUS_ENABLE_OPENMP)
 
#include <Morpheus_Exceptions.hpp>
#include <Morpheus_SpaceTraits.hpp>
#include <Morpheus_FormatTraits.hpp>
#include <Morpheus_FormatTags.hpp>
#include <Morpheus_Spaces.hpp>
 
namespace Morpheus {
namespace Impl {
 
template <typename ExecSpace, typename SourceType, typename DestinationType>
void convert(
    const SourceType& src, DestinationType& dst,
    typename std::enable_if<
        Morpheus::is_csr_matrix_format_container_v<SourceType> &&
        Morpheus::is_csr_matrix_format_container_v<DestinationType> &&
        Morpheus::has_custom_backend_v<ExecSpace> &&
        Morpheus::has_openmp_execution_space_v<ExecSpace> &&
        Morpheus::has_access_v<ExecSpace, SourceType, DestinationType>>::type* =
        nullptr) {
  using size_type = typename SourceType::size_type;
 
  dst.resize(src.nrows(), src.ncols(), src.nnnz());
 
// element-wise copy of indices and values
#pragma omp parallel for
  for (size_type n = 0; n < src.nrows() + 1; n++) {
    dst.row_offsets(n) = src.crow_offsets(n);
  }
 
#pragma omp parallel for
  for (size_type n = 0; n < src.nnnz(); n++) {
    dst.column_indices(n) = src.ccolumn_indices(n);
  }
 
#pragma omp parallel for
  for (size_type n = 0; n < src.nnnz(); n++) {
    dst.values(n) = src.cvalues(n);
  }
}
 
template <typename ExecSpace, typename SourceType, typename DestinationType>
void convert(
    const SourceType& src, DestinationType& dst,
    typename std::enable_if<
        Morpheus::is_csr_matrix_format_container_v<SourceType> &&
        Morpheus::is_coo_matrix_format_container_v<DestinationType> &&
        Morpheus::has_custom_backend_v<ExecSpace> &&
        Morpheus::has_openmp_execution_space_v<ExecSpace> &&
        Morpheus::has_access_v<ExecSpace, SourceType, DestinationType>>::type* =
        nullptr) {
  using size_type  = typename SourceType::size_type;
  using index_type = typename SourceType::index_type;
 
  dst.resize(src.nrows(), src.ncols(), src.nnnz());
 
// expand compressed indices
#pragma omp parallel for
  for (size_type i = 0; i < src.nrows(); i++) {
    for (index_type jj = src.crow_offsets(i); jj < src.crow_offsets(i + 1);
         jj++) {
      dst.row_indices(jj) = i;
    }
  }
 
#pragma omp parallel for
  for (size_type n = 0; n < src.nnnz(); n++) {
    dst.column_indices(n) = src.ccolumn_indices(n);
  }
 
#pragma omp parallel for
  for (size_type n = 0; n < src.nnnz(); n++) {
    dst.values(n) = src.cvalues(n);
  }
}
 
template <typename ExecSpace, typename SourceType, typename DestinationType>
void convert(
    const SourceType&, DestinationType&,
    typename std::enable_if<
        Morpheus::is_coo_matrix_format_container_v<SourceType> &&
        Morpheus::is_csr_matrix_format_container_v<DestinationType> &&
        Morpheus::has_custom_backend_v<ExecSpace> &&
        Morpheus::has_openmp_execution_space_v<ExecSpace> &&
        Morpheus::has_access_v<ExecSpace, SourceType, DestinationType>>::type* =
        nullptr) {
  throw Morpheus::NotImplementedException("convert<Kokkos::OpenMP>");
}
 
}  // namespace Impl
}  // namespace Morpheus
 
#endif  // MORPHEUS_ENABLE_OPENMP
#endif  // MORPHEUS_CSR_OPENMP_CONVERT_IMPL_HPP