cuda_precondition.h

// ---------------------------------------------------------------------
//
// Copyright (C) 2018 by the deal.II authors
//
// This file is part of the deal.II library.
//
// The deal.II library is free software; you can use it, redistribute
// it, and/or modify it under the terms of the GNU Lesser General
// Public License as published by the Free Software Foundation; either
// version 2.1 of the License, or (at your option) any later version.
// The full text of the license can be found in the file LICENSE.md at
// the top level directory of deal.II.
//
// ---------------------------------------------------------------------

#ifndef dealii_cuda_precondition_h
#define dealii_cuda_precondition_h

#include <deal.II/base/config.h>

#include <deal.II/base/cuda.h>
#include <deal.II/base/smartpointer.h>

#include <memory>

#ifdef DEAL_II_COMPILER_CUDA_AWARE

DEAL_II_NAMESPACE_OPEN

// forward-definition
namespace LinearAlgebra
{
  namespace CUDAWrappers
  {
    template <typename Number>
    class Vector;
  }
} // namespace LinearAlgebra

namespace CUDAWrappers
{
  // forward definition
  template <typename Number>
  class SparseMatrix;

  template <typename Number>
  class PreconditionIC
  {
  public:
    using size_type = int;

    struct AdditionalData
    {
      AdditionalData(bool use_level_analysis = true);

      bool use_level_analysis;
    };

    PreconditionIC(const Utilities::CUDA::Handle &handle);

    PreconditionIC(const PreconditionIC<Number> &) = delete;

    PreconditionIC &
    operator=(const PreconditionIC<Number> &) = delete;

    ~PreconditionIC();

    void
    initialize(const SparseMatrix<Number> &matrix,
               const AdditionalData &      additional_data = AdditionalData());

    void
    vmult(LinearAlgebra::CUDAWrappers::Vector<Number> &      dst,
          const LinearAlgebra::CUDAWrappers::Vector<Number> &src) const;

    void
    Tvmult(LinearAlgebra::CUDAWrappers::Vector<Number> &      dst,
           const LinearAlgebra::CUDAWrappers::Vector<Number> &src) const;

    size_type
    m() const;

    size_type
    n() const;

  private:
    cusparseHandle_t cusparse_handle;

    cusparseMatDescr_t descr_M;

    cusparseMatDescr_t descr_L;

    csric02Info_t info_M;

    csrsv2Info_t info_L;

    csrsv2Info_t info_Lt;

    SmartPointer<const SparseMatrix<Number>> matrix_pointer;

    std::unique_ptr<Number[], void (*)(Number *)> P_val_dev;

    const int *P_row_ptr_dev;

    const int *P_column_index_dev;

    std::unique_ptr<Number[], void (*)(Number *)> tmp_dev;

    std::unique_ptr<void, void (*)(void *)> buffer_dev;

    cusparseSolvePolicy_t policy_L;

    cusparseSolvePolicy_t policy_Lt;

    cusparseSolvePolicy_t policy_M;

    int n_rows;

    int n_nonzero_elements;
  };

  template <typename Number>
  class PreconditionILU
  {
  public:
    using size_type = int;

    struct AdditionalData
    {
      AdditionalData(bool use_level_analysis = true);

      bool use_level_analysis;
    };

    PreconditionILU(const Utilities::CUDA::Handle &handle);

    PreconditionILU(const PreconditionILU<Number> &) = delete;

    PreconditionILU &
    operator=(const PreconditionILU<Number> &) = delete;

    ~PreconditionILU();

    void
    initialize(const SparseMatrix<Number> &matrix,
               const AdditionalData &      additional_data = AdditionalData());

    void
    vmult(LinearAlgebra::CUDAWrappers::Vector<Number> &      dst,
          const LinearAlgebra::CUDAWrappers::Vector<Number> &src) const;

    void
    Tvmult(LinearAlgebra::CUDAWrappers::Vector<Number> &      dst,
           const LinearAlgebra::CUDAWrappers::Vector<Number> &src) const;

    size_type
    m() const;

    size_type
    n() const;

  private:
    cusparseHandle_t cusparse_handle;

    cusparseMatDescr_t descr_M;

    cusparseMatDescr_t descr_L;

    cusparseMatDescr_t descr_U;

    csrilu02Info_t info_M;

    csrsv2Info_t info_L;

    csrsv2Info_t info_U;

    SmartPointer<const SparseMatrix<Number>> matrix_pointer;

    std::unique_ptr<Number[], void (*)(Number *)> P_val_dev;

    const int *P_row_ptr_dev;

    const int *P_column_index_dev;

    std::unique_ptr<Number[], void (*)(Number *)> tmp_dev;

    std::unique_ptr<void, void (*)(void *)> buffer_dev;

    cusparseSolvePolicy_t policy_L;

    cusparseSolvePolicy_t policy_U;

    cusparseSolvePolicy_t policy_M;

    int n_rows;

    int n_nonzero_elements;
  };

  /*--------------------------- inline functions ----------------------------*/

#  ifndef DOXYGEN
  template <typename Number>
  inline typename PreconditionIC<Number>::size_type
  PreconditionIC<Number>::m() const
  {
    return n_rows;
  }



  template <typename Number>
  inline typename PreconditionIC<Number>::size_type
  PreconditionIC<Number>::n() const
  {
    return n_rows;
  }



  template <typename Number>
  inline typename PreconditionILU<Number>::size_type
  PreconditionILU<Number>::m() const
  {
    return n_rows;
  }



  template <typename Number>
  inline typename PreconditionILU<Number>::size_type
  PreconditionILU<Number>::n() const
  {
    return n_rows;
  }
#  endif // DOXYGEN

} // namespace CUDAWrappers

DEAL_II_NAMESPACE_CLOSE

#endif // DEAL_II_WITH_CUDA

#endif // dealii_cuda_precondition_h