cusparseSolverBackend.hpp

/*
  Copyright 2019 Equinor ASA

  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_CUSPARSESOLVER_BACKEND_HEADER_INCLUDED
#define OPM_CUSPARSESOLVER_BACKEND_HEADER_INCLUDED


#include "cublas_v2.h"
#include "cusparse_v2.h"

#include <opm/simulators/linalg/gpubridge/GpuResult.hpp>
#include <opm/simulators/linalg/gpubridge/GpuSolver.hpp>
#include <opm/simulators/linalg/gpubridge/WellContributions.hpp>

namespace Opm::Accelerator {

template<class Scalar, unsigned int block_size>
class cusparseSolverBackend : public GpuSolver<Scalar,block_size>
{
    using Base = GpuSolver<Scalar,block_size>;

    using Base::N;
    using Base::Nb;
    using Base::nnz;
    using Base::nnzb;
    using Base::verbosity;
    using Base::deviceID;
    using Base::maxit;
    using Base::tolerance;
    using Base::initialized;

private:
    cublasHandle_t cublasHandle;
    cusparseHandle_t cusparseHandle;
    cudaStream_t stream;
    cusparseMatDescr_t descr_B, descr_M, descr_L, descr_U;
    bsrilu02Info_t info_M;
    bsrsv2Info_t info_L, info_U;
    // b: bsr matrix, m: preconditioner
    Scalar *d_bVals, *d_mVals;
    int *d_bCols, *d_mCols;
    int *d_bRows, *d_mRows;
    Scalar *d_x, *d_b, *d_r, *d_rw, *d_p;     // vectors, used during linear solve
    Scalar *d_pw, *d_s, *d_t, *d_v;
    void *d_buffer;
    Scalar *vals_contiguous;                  // only used if COPY_ROW_BY_ROW is true in cusparseSolverBackend.cpp

    bool analysis_done = false;

    bool useJacMatrix = false;
    int nnzbs_prec;             // number of nonzero blocks in the matrix for preconditioner
                                // could be jacMatrix or matrix

    double c_copy = 0.0; // cummulative timer measuring the total time it takes to transfer the data to the GPU

    void gpu_pbicgstab(WellContributions<Scalar>& wellContribs, GpuResult& res);

    void initialize(std::shared_ptr<BlockedMatrix<Scalar>> matrix,
                    std::shared_ptr<BlockedMatrix<Scalar>> jacMatrix);

    void finalize();

    void copy_system_to_gpu(std::shared_ptr<BlockedMatrix<Scalar>> matrix,
                            Scalar* b,
                            std::shared_ptr<BlockedMatrix<Scalar>> jacMatrix);

    void update_system_on_gpu(std::shared_ptr<BlockedMatrix<Scalar>> matrix,
                              Scalar* b,
                              std::shared_ptr<BlockedMatrix<Scalar>> jacMatrix);

    bool analyse_matrix();

    bool create_preconditioner();

    void solve_system(WellContributions<Scalar>& wellContribs, GpuResult &res);

public:
    cusparseSolverBackend(int linear_solver_verbosity, int maxit,
                          Scalar tolerance, unsigned int deviceID);

    ~cusparseSolverBackend();

    SolverStatus solve_system(std::shared_ptr<BlockedMatrix<Scalar>> matrix,
                              Scalar* b,
                              std::shared_ptr<BlockedMatrix<Scalar>> jacMatrix,
                              WellContributions<Scalar>& wellContribs,
                              GpuResult& res) override;

    void get_result(Scalar* x) override;

}; // end class cusparseSolverBackend

} // namespace Opm::Accelerator

#endif