gpujacobi.h

Go to the documentation of this file.

/*
 * Copyright (c) 2013-2015:  G-CSC, Goethe University Frankfurt
 * Author: Martin Rupp
 * 
 * This file is part of UG4.
 * 
 * UG4 is free software: you can redistribute it and/or modify it under the
 * terms of the GNU Lesser General Public License version 3 (as published by the
 * Free Software Foundation) with the following additional attribution
 * requirements (according to LGPL/GPL v3 §7):
 * 
 * (1) The following notice must be displayed in the Appropriate Legal Notices
 * of covered and combined works: "Based on UG4 (www.ug4.org/license)".
 * 
 * (2) The following notice must be displayed at a prominent place in the
 * terminal output of covered works: "Based on UG4 (www.ug4.org/license)".
 * 
 * (3) The following bibliography is recommended for citation and must be
 * preserved in all covered files:
 * "Reiter, S., Vogel, A., Heppner, I., Rupp, M., and Wittum, G. A massively
 *   parallel geometric multigrid solver on hierarchically distributed grids.
 *   Computing and visualization in science 16, 4 (2013), 151-164"
 * "Vogel, A., Reiter, S., Rupp, M., Nägel, A., and Wittum, G. UG4 -- a novel
 *   flexible software system for simulating pde based models on high performance
 *   computers. Computing and visualization in science 16, 4 (2013), 165-179"
 * 
 * This program 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 Lesser General Public License for more details.
 */
 
#ifdef CUDA_AVAILABLE
#ifdef UG_GPU
#ifndef __H__UG__LIB_DISC__OPERATOR__LINEAR_OPERATOR__GPUJACOBI__
#define __H__UG__LIB_DISC__OPERATOR__LINEAR_OPERATOR__GPUJACOBI__
 
#include "lib_algebra/operator/interface/preconditioner.h"
#include "lib_algebra/gpu_algebra/gpusparsematrix.h"
#include "lib_algebra/gpu_algebra/gpuvector.h"
 
#ifdef UG_PARALLEL
  #include "lib_algebra/parallelization/parallelization.h"
#endif
 
#include "lib_algebra/small_algebra/blocks.h"
#include "lib_algebra/gpu_algebra/gpuvector.h"
#include "lib_algebra/cpu_algebra_types.h"
 
 
extern "C" bool
CUDA_JacobiApply(const double *diagInv, double *corr, const double *defect, const int N);
 
namespace ug{
 
 
class GPUJacobi : public IPreconditioner<GPUAlgebra>
{
  public:
    typedef GPUAlgebra TAlgebra;
    typedef TAlgebra algebra_type;
 
    typedef TAlgebra::vector_type vector_type;
 
    typedef TAlgebra::matrix_type matrix_type;
 
    typedef IPreconditioner<TAlgebra>::matrix_operator_type matrix_operator_type;
 
    typedef IPreconditioner<TAlgebra> base_type;
 
  protected:
    using base_type::set_debug;
    using base_type::debug_writer;
    using base_type::write_debug;
    using base_type::damping;
 
  public:
    GPUJacobi() {this->set_damp(1.0);};
 
    GPUJacobi(number damp) {this->set_damp(damp);};
 
    GPUJacobi( const GPUJacobi &parent )
      : base_type(parent)
    {
    }
 
    virtual SmartPtr<ILinearIterator<vector_type> > clone()
    {
      return make_sp(new GPUJacobi(*this));
    }
 
 
    virtual ~GPUJacobi()
    {};
 
    virtual bool supports_parallel() const { return false; }
  protected:
    virtual const char* name() const {return "GPUJacobi";}
 
    virtual bool preprocess(SmartPtr<MatrixOperator<matrix_type, vector_type> > pOp)
    {
      PROFILE_BEGIN_GROUP(GPUJacobi_preprocess, "algebra GPUJacobi");
 
      matrix_type &mat = *pOp;
    //  create help vector to apply diagonal
      size_t size = mat.num_rows();
      if(size != mat.num_cols())
      {
        UG_LOG("Square Matrix needed for GPUJacobi Iteration.\n");
        return false;
      }
 
    //  get damping in constant case to damp at once
      number damp = 1.0;
      if(damping()->constant_damping())
        damp = damping()->damping();
 
      if(CheckDiagonalInvertible(mat)==false)
        return false;
//      UG_ASSERT(CheckDiagonalInvertible(mat), "GPUJacobi: A has noninvertible diagonal");
 
 
      m_diagInv.resize(mat.num_rows());
    //  invert diagonal and multiply by damping
      for(size_t i = 0; i < mat.num_rows(); ++i)
        m_diagInv[i] = damp/mat(i,i);
 
      CUDAHelper::get_instance();
    //  done
      return true;
    }
 
    virtual bool step(SmartPtr<MatrixOperator<matrix_type, vector_type> > pOp, vector_type& c, const vector_type& d)
    {
#ifdef UG_PARALLEL
    //  the computed correction is additive
      c.set_storage_type(PST_ADDITIVE);
 
    //  we make it consistent
      if(!c.change_storage_type(PST_CONSISTENT))
      {
        UG_LOG("ERROR in 'JacobiPreconditioner::apply': "
            "Cannot change parallel status of correction to consistent.\n");
        return false;
      }
#endif
 
      const double *devDiagInv = m_diagInv.get_dev_ptr();
      double *devC = c.get_dev_ptr();
      const double *devD = d.get_dev_ptr();
      int N = c.size();
//      UG_LOG("devDiagInv = " << devDiagInv<< " " << m_diagInv.size()  << " devC = " << devC << " " << c.size() << " devD = "
//          << devD << " " << d.size() << " N = " << N << "\n");
      CUDA_JacobiApply(devDiagInv, devC, devD, N);
 
 
      return true;
    }
 
    virtual bool postprocess() {return true;}
 
  protected:
    GPUVector<double> m_diagInv;
 
};
 
template<>
class Jacobi<GPUAlgebra> : public GPUJacobi
{
  using GPUJacobi::set_damp;
public:
    Jacobi() {this->set_damp(1.0);};
 
    Jacobi(number damp) {this->set_damp(damp);};
 
    void set_block(bool b)
    {
      // always scalar
    }
};
 
 
} // end namespace ug
 
#endif
#endif
#endif // USE_CUDA