rocsparseSolverBackend.hpp

Go to the documentation of this file.

/*
  Copyright 2023 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_ROCSPARSESOLVER_BACKEND_HEADER_INCLUDED
#define OPM_ROCSPARSESOLVER_BACKEND_HEADER_INCLUDED
 
#include <memory>
 
#include <opm/simulators/linalg/bda/BdaResult.hpp>
#include <opm/simulators/linalg/bda/BdaSolver.hpp>
#include <opm/simulators/linalg/bda/WellContributions.hpp>
 
#include <opm/simulators/linalg/bda/rocm/rocsparsePreconditioner.hpp>
 
#include <rocblas/rocblas.h>
#include <rocsparse/rocsparse.h>
 
#include <hip/hip_version.h>
 
namespace Opm::Accelerator {
 
template<class Scalar, unsigned int block_size>
class rocsparseSolverBackend : public BdaSolver<Scalar,block_size>
{
    using Base = BdaSolver<Scalar,block_size>;
 
    using Base::N;
    using Base::Nb;
    using Base::nnz;
    using Base::nnzb;
    using Base::verbosity;
    using Base::platformID;
    using Base::deviceID;
    using Base::maxit;
    using Base::tolerance;
    using Base::initialized;
 
private:
    double c_copy = 0.0; // cummulative timer measuring the total time it takes to transfer the data to the GPU
 
    bool useJacMatrix = false;
 
    bool analysis_done = false;
    std::shared_ptr<BlockedMatrix<Scalar>> mat{};                 // original matrix
 
    rocsparse_direction dir = rocsparse_direction_row;
    rocsparse_operation operation = rocsparse_operation_none;
    rocsparse_handle handle;
    rocblas_handle blas_handle;
    rocsparse_mat_descr descr_A;
#if HIP_VERSION >= 50400000
    rocsparse_mat_info spmv_info;
#endif
    hipStream_t stream;
 
    rocsparse_int *d_Arows, *d_Acols;
    Scalar *d_Avals;
    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;
    int  ver;
    char rev[64];
 
    std::unique_ptr<rocsparsePreconditioner<Scalar, block_size> > prec; // can perform blocked ILU0 and AMG on pressure component
    
    void gpu_pbicgstab(WellContributions<Scalar>& wellContribs, BdaResult& res);
 
    void initialize(std::shared_ptr<BlockedMatrix<Scalar>> matrix,
                    std::shared_ptr<BlockedMatrix<Scalar>> jacMatrix);
 
    void copy_system_to_gpu(Scalar* b);
 
    void update_system_on_gpu(Scalar* b);
 
    bool analyze_matrix();
 
    bool create_preconditioner();
 
    void solve_system(WellContributions<Scalar>& wellContribs, BdaResult& res);
 
public:
    rocsparseSolverBackend(int linear_solver_verbosity,
                           int maxit,
                           Scalar tolerance,
                           unsigned int platformID,
                           unsigned int deviceID,
                           std::string linsolver);
 
    rocsparseSolverBackend(int linear_solver_verbosity, int maxit, Scalar tolerance, bool opencl_ilu_reorder);
 
    ~rocsparseSolverBackend();
 
    SolverStatus solve_system(std::shared_ptr<BlockedMatrix<Scalar>> matrix,
                              Scalar* b,
                              std::shared_ptr<BlockedMatrix<Scalar>> jacMatrix,
                              WellContributions<Scalar>& wellContribs,
                              BdaResult& res) override;
 
    void get_result(Scalar* x) override;
 
}; // end class rocsparseSolverBackend
 
} // namespace Opm::Accelerator
 
#endif