openclSolverBackend.hpp

/*
  Copyright 2020 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_OPENCLSOLVER_BACKEND_HEADER_INCLUDED
#define OPM_OPENCLSOLVER_BACKEND_HEADER_INCLUDED

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

#include <opm/simulators/linalg/gpubridge/opencl/openclPreconditioner.hpp>

namespace Opm::Accelerator {

template<class Scalar, unsigned int block_size>
class openclSolverBackend : 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::platformID;
    using Base::deviceID;
    using Base::maxit;
    using Base::tolerance;
    using Base::initialized;

private:
    Scalar* h_b = nullptr;                // b vector, on host
    std::vector<Scalar> vals_contiguous;  // only used if COPY_ROW_BY_ROW is true in openclSolverBackend.cpp

    // OpenCL variables must be reusable, they are initialized in initialize()
    cl::Buffer d_Avals, d_Acols, d_Arows;        // matrix in BSR format on GPU
    cl::Buffer d_x, d_b, d_rb, d_r, d_rw, d_p;   // vectors, used during linear solve
    cl::Buffer d_pw, d_s, d_t, d_v;              // vectors, used during linear solve
    cl::Buffer d_tmp;                            // used as tmp GPU buffer for dot() and norm()

    std::vector<cl::Device> devices;

    bool useJacMatrix = false;

    std::unique_ptr<openclPreconditioner<Scalar,block_size>> prec;
                                                                  // can perform blocked ILU0 and AMG on pressure component
    bool is_root;                                                 // allow for nested solvers, the root solver is called by GpuBridge
    bool analysis_done = false;
    std::shared_ptr<BlockedMatrix<Scalar>> mat{};                 // original matrix
    std::shared_ptr<BlockedMatrix<Scalar>> jacMat{};              // matrix for preconditioner
    bool opencl_ilu_parallel;                                     // parallelize ILU operations (with level_scheduling)
    std::vector<cl::Event> events;
    cl_int err;

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

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

    void copy_system_to_gpu();

    void update_system(Scalar* vals, Scalar* b);

    void update_system_on_gpu();

    bool analyze_matrix();

    bool create_preconditioner();

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

public:
    std::shared_ptr<cl::Context> context{};
    std::shared_ptr<cl::CommandQueue> queue{};

    openclSolverBackend(int linear_solver_verbosity, int maxit, Scalar tolerance,
                        unsigned int platformID, unsigned int deviceID,
                        bool opencl_ilu_parallel, std::string linsolver);

    openclSolverBackend(int linear_solver_verbosity, int maxit,
                        Scalar tolerance, bool opencl_ilu_parallel);

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

    // SolverStatus solve_system(GpuResult &res);

    void get_result(Scalar* x) override;

    void setOpencl(std::shared_ptr<cl::Context>& context,
                   std::shared_ptr<cl::CommandQueue>& queue);
}; // end class openclSolverBackend

} // namespace Opm::Accelerator

#endif