GpuSeqILU0.hpp

Go to the documentation of this file.

/*
  Copyright 2022-2023 SINTEF AS
 
  This file is part of the Open Porous Media project (OPM).
 
  OPM is free software: you can redistribute it and/or modify
  it under the terms of the GNU General Public License as published by
  the Free Software Foundation, either version 3 of the License, or
  (at your option) any later version.
 
  OPM is distributed in the hope that it will be useful,
  but WITHOUT ANY WARRANTY; without even the implied warranty of
  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  GNU General Public License for more details.
 
  You should have received a copy of the GNU General Public License
  along with OPM.  If not, see <http://www.gnu.org/licenses/>.
*/
#ifndef OPM_GPUSEQILU0_HPP
#define OPM_GPUSEQILU0_HPP
 
#include <dune/istl/preconditioner.hh>
#include <opm/simulators/linalg/PreconditionerWithUpdate.hpp>
#include <opm/simulators/linalg/gpuistl/GpuSparseMatrixWrapper.hpp>
#include <opm/simulators/linalg/gpuistl/detail/CuMatrixDescription.hpp>
#include <opm/simulators/linalg/gpuistl/detail/CuSparseHandle.hpp>
#include <opm/simulators/linalg/gpuistl/detail/CuSparseResource.hpp>
#include <opm/simulators/linalg/gpuistl/detail/fix_zero_diagonal.hpp>
#include <opm/simulators/linalg/is_gpu_operator.hpp>
 
 
 
namespace Opm::gpuistl
{
template <class M, class X, class Y, int l = 1>
class GpuSeqILU0 : public Dune::PreconditionerWithUpdate<X, Y>
{
public:
    using matrix_type = typename std::remove_const<M>::type ;
    using domain_type = X;
    using range_type = Y;
    using field_type = typename X::field_type;
 
    GpuSeqILU0(const M& A, field_type w)
        requires (is_gpu_matrix_v<M>);
 
 
    GpuSeqILU0(const M& A, field_type w)
        requires (!is_gpu_matrix_v<M>);
 
 
    void pre(X& x, Y& b) override;
 
    void apply(X& v, const Y& d) override;
 
    void post(X& x) override;
 
    Dune::SolverCategory::Category category() const override;
 
    void update() override;
 
 
    static constexpr bool shouldCallPre()
    {
        return false;
    }
 
    static constexpr bool shouldCallPost()
    {
        return false;
    }
 
    bool hasPerfectUpdate() const override {
        return true;
    }
 
 
private:
    const M& m_underlyingMatrix;
    field_type m_w;
 
    GpuSparseMatrixWrapper<field_type, true> m_LU;
 
    GpuVector<field_type> m_temporaryStorage;
 
 
    detail::GpuSparseMatrixDescriptionPtr m_descriptionL;
    detail::GpuSparseMatrixDescriptionPtr m_descriptionU;
    detail::CuSparseResource<bsrsv2Info_t> m_infoL;
    detail::CuSparseResource<bsrsv2Info_t> m_infoU;
    detail::CuSparseResource<bsrilu02Info_t> m_infoM;
 
    std::unique_ptr<GpuVector<field_type>> m_buffer;
    detail::CuSparseHandle& m_cuSparseHandle;
 
    bool m_analysisDone = false;
 
    void analyzeMatrix();
    size_t findBufferSize();
 
    void createILU();
 
    void updateILUConfiguration();
};
 
 
// Case where matrix is a CPU matrix as input
template <class M, class X, class Y, int l>
GpuSeqILU0<M, X, Y, l>::GpuSeqILU0(const M& A, field_type w)
    requires (!is_gpu_matrix_v<M>)
    : m_underlyingMatrix(A)
    , m_w(w)
    , m_LU(GpuSparseMatrixWrapper<field_type, true>::fromMatrix(detail::makeMatrixWithNonzeroDiagonal(A)))
    , m_temporaryStorage(m_LU.N() * m_LU.blockSize())
    , m_descriptionL(detail::createLowerDiagonalDescription())
    , m_descriptionU(detail::createUpperDiagonalDescription())
    , m_cuSparseHandle(detail::CuSparseHandle::getInstance())
{
    // Some sanity check
    OPM_ERROR_IF(A.N() != m_LU.N(),
                 fmt::format("CuSparse matrix not same size as DUNE matrix. {} vs {}.", m_LU.N(), A.N()));
    OPM_ERROR_IF(
        A[0][0].N() != m_LU.blockSize(),
        fmt::format("CuSparse matrix not same blocksize as DUNE matrix. {} vs {}.", m_LU.blockSize(), A[0][0].N()));
    OPM_ERROR_IF(
        A.N() * A[0][0].N() != m_LU.dim(),
        fmt::format("CuSparse matrix not same dimension as DUNE matrix. {} vs {}.", m_LU.dim(), A.N() * A[0][0].N()));
    OPM_ERROR_IF(A.nonzeroes() != m_LU.nonzeroes(),
                 fmt::format("CuSparse matrix not same number of non zeroes as DUNE matrix. {} vs {}. ",
                             m_LU.nonzeroes(),
                             A.nonzeroes()));
 
 
    updateILUConfiguration();
}
 
// Case where matrix is a GPU matrix as input
template <class M, class X, class Y, int l>
GpuSeqILU0<M, X, Y, l>::GpuSeqILU0(const M& A, field_type w)
    requires (is_gpu_matrix_v<M>)
    : m_underlyingMatrix(A)
    , m_w(w)
    , m_LU(A.getRowIndices(), A.getColumnIndices(), A.blockSize())
    , m_temporaryStorage(m_LU.N() * m_LU.blockSize())
    , m_descriptionL(detail::createLowerDiagonalDescription())
    , m_descriptionU(detail::createUpperDiagonalDescription())
    , m_cuSparseHandle(detail::CuSparseHandle::getInstance())
{
    m_LU.updateNonzeroValues(A);
 
    // Some sanity check
    OPM_ERROR_IF(A.N() != m_LU.N(),
                 fmt::format("CuSparse matrix not same size as DUNE matrix. {} vs {}.", m_LU.N(), A.N()));
    OPM_ERROR_IF(A.nonzeroes() != m_LU.nonzeroes(),
                 fmt::format("CuSparse matrix not same number of non zeroes as DUNE matrix. {} vs {}. ",
                             m_LU.nonzeroes(),
                             A.nonzeroes()));
 
 
    updateILUConfiguration();
}
 
} // end namespace Opm::gpuistl
 
#endif