Nous avons procédé ce jeudi matin 08 avril 2021 à une MAJ de sécurité urgente. Nous sommes passé de la version 13.9.3 à la version 13.9.5 les releases notes correspondantes sont ici:
https://about.gitlab.com/releases/2021/03/17/security-release-gitlab-13-9-4-released/
https://about.gitlab.com/releases/2021/03/31/security-release-gitlab-13-10-1-released/

FUnifTensorialKernel.hpp 10.6 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
// ===================================================================================
// Copyright ScalFmm 2011 INRIA, Olivier Coulaud, Bérenger Bramas, Matthias Messner
// olivier.coulaud@inria.fr, berenger.bramas@inria.fr
// This software is a computer program whose purpose is to compute the FMM.
//
// This software is governed by the CeCILL-C and LGPL licenses and
// abiding by the rules of distribution of free software.
//
// 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 General Public and CeCILL-C Licenses for more details.
// "http://www.cecill.info".
// "http://www.gnu.org/licenses".
// ===================================================================================
#ifndef FUNIFTENSORIALKERNEL_HPP
#define FUNIFTENSORIALKERNEL_HPP

#include "../../Utils/FGlobal.hpp"
#include "../../Utils/FTrace.hpp"
#include "../../Utils/FSmartPointer.hpp"

#include "./FAbstractUnifKernel.hpp"
24 25
#include "./FUnifM2LHandler.hpp"
#include "./FUnifTensorialM2LHandler.hpp" //PB: temporary version
26 27 28 29 30 31 32 33 34 35 36 37 38 39 40

class FTreeCoordinate;

/**
 * @author Pierre Blanchard (pierre.blanchard@inria.fr)
 * @class FUnifTensorialKernel
 * @brief
 * Please read the license
 *
 * This kernels implement the Lagrange interpolation based FMM operators. It
 * implements all interfaces (P2P,P2M,M2M,M2L,L2L,L2P) which are required by
 * the FFmmAlgorithm and FFmmAlgorithmThread.
 *
 * PB: 3 IMPORTANT remarks !!!
 *
41 42
 * 1) Handling tensorial kernels (DIM,NRHS,NLHS) and having multiple rhs 
 * (NVALS) are considered 2 distinct features and are currently combined.
43
 *
44 45 46 47 48 49
 * 2) When it comes to applying M2L it is NOT much faster to loop over 
 * NRHSxNLHS inside applyM2L (at least for the Lagrange case).
 * 2-bis) During precomputation the tensorial matrix kernels are evaluated 
 * blockwise, but this is not always possible. 
 * In fact, in the ChebyshevSym variant the matrix kernel needs to be 
 * evaluated compo-by-compo since we currently use a scalar ACA.
50
 *
51 52 53
 * 3) We currently use multiple 1D FFT instead of multidim FFT since embedding
 * is circulant. Multidim FFT could be used if embedding were block circulant.
 * TODO investigate possibility of block circulant embedding
54 55 56 57 58 59 60 61 62 63
 *
 * @tparam CellClass Type of cell
 * @tparam ContainerClass Type of container to store particles
 * @tparam MatrixKernelClass Type of matrix kernel function
 * @tparam ORDER Lagrange interpolation order
 */
template < class CellClass,	class ContainerClass,	class MatrixKernelClass, int ORDER, int NVALS = 1>
class FUnifTensorialKernel
  : public FAbstractUnifKernel< CellClass, ContainerClass, MatrixKernelClass, ORDER, NVALS>
{
64
  enum {nRhs = MatrixKernelClass::NRHS,
65 66 67
        nLhs = MatrixKernelClass::NLHS,
        nPot = MatrixKernelClass::NPOT,
        nPV = MatrixKernelClass::NPV};
68 69 70

protected://PB: for OptiDis

71
  // private types
72
  typedef FUnifTensorialM2LHandler<ORDER,MatrixKernelClass,MatrixKernelClass::Type> M2LHandlerClass;
73 74 75 76 77

  // using from
  typedef FAbstractUnifKernel< CellClass, ContainerClass, MatrixKernelClass, ORDER, NVALS>
  AbstractBaseClass;

78 79 80
  /// Needed for P2P and M2L operators
  const MatrixKernelClass *const MatrixKernel;

81
  /// Needed for M2L operator
82
  const M2LHandlerClass M2LHandler;
83 84 85 86 87 88 89 90

public:
  /**
   * The constructor initializes all constant attributes and it reads the
   * precomputed and compressed M2L operators from a binary file (an
   * runtime_error is thrown if the required file is not valid).
   */
  FUnifTensorialKernel(const int inTreeHeight,
91
                       const FReal inBoxWidth,
92
                       const FPoint& inBoxCenter,
93 94 95 96 97
                       const MatrixKernelClass *const inMatrixKernel,
                       const FReal inBoxWidthExtension)
    : FAbstractUnifKernel< CellClass, ContainerClass, MatrixKernelClass, ORDER, NVALS>(inTreeHeight,inBoxWidth,inBoxCenter,inBoxWidthExtension),
      MatrixKernel(inMatrixKernel),
      M2LHandler(MatrixKernel,
98
                 inTreeHeight,
99 100
                 inBoxWidth,
                 inBoxWidthExtension) 
101
  { }
102 103 104 105 106 107


  void P2M(CellClass* const LeafCell,
           const ContainerClass* const SourceParticles)
  {
    const FPoint LeafCellCenter(AbstractBaseClass::getLeafCellCenter(LeafCell->getCoordinate())); 
108 109
    const FReal ExtendedLeafCellWidth(AbstractBaseClass::BoxWidthLeaf 
                                      + AbstractBaseClass::BoxWidthExtension);
110

111
    for(int idxV = 0 ; idxV < NVALS ; ++idxV){
112 113

      // 1) apply Sy
114
      AbstractBaseClass::Interpolator->applyP2M(LeafCellCenter, ExtendedLeafCellWidth,
115 116
                                                LeafCell->getMultipole(idxV*nRhs), SourceParticles);

117 118 119 120 121
      for(int idxRhs = 0 ; idxRhs < nRhs ; ++idxRhs){
        // update multipole index
        int idxMul = idxV*nRhs + idxRhs;

        // 2) apply Discrete Fourier Transform
122 123
        M2LHandler.applyZeroPaddingAndDFT(LeafCell->getMultipole(idxMul), 
                                          LeafCell->getTransformedMultipole(idxMul));
124 125 126

      }
    }// NVALS
127 128 129 130 131
  }


  void M2M(CellClass* const FRestrict ParentCell,
           const CellClass*const FRestrict *const FRestrict ChildCells,
132
           const int TreeLevel)
133
  {
134 135 136 137 138 139 140 141 142
    for(int idxV = 0 ; idxV < NVALS ; ++idxV){
      for(int idxRhs = 0 ; idxRhs < nRhs ; ++idxRhs){
        // update multipole index
        int idxMul = idxV*nRhs + idxRhs;

        // 1) apply Sy
        FBlas::scal(AbstractBaseClass::nnodes, FReal(0.), ParentCell->getMultipole(idxMul));
        for (unsigned int ChildIndex=0; ChildIndex < 8; ++ChildIndex){
          if (ChildCells[ChildIndex]){
143 144 145 146
            AbstractBaseClass::Interpolator->applyM2M(ChildIndex, 
                                                      ChildCells[ChildIndex]->getMultipole(idxMul),
                                                      ParentCell->getMultipole(idxMul), 
                                                      TreeLevel/*Cell width extension specific*/);
147
          }
148
        }
149
        // 2) Apply Discete Fourier Transform
150 151
        M2LHandler.applyZeroPaddingAndDFT(ParentCell->getMultipole(idxMul), 
                                          ParentCell->getTransformedMultipole(idxMul));
152
      }
153
    }// NVALS
154 155 156 157 158 159 160 161
  }


  void M2L(CellClass* const FRestrict TargetCell,
           const CellClass* SourceCells[343],
           const int /*NumSourceCells*/,
           const int TreeLevel)
  {
162
    const FReal CellWidth(AbstractBaseClass::BoxWidth / FReal(FMath::pow(2, TreeLevel)));
163
    const FReal ExtendedCellWidth(CellWidth + AbstractBaseClass::BoxWidthExtension);
164
    const FReal scale(MatrixKernel->getScaleFactor(ExtendedCellWidth));
165 166 167

    for(int idxV = 0 ; idxV < NVALS ; ++idxV){
      for (int idxLhs=0; idxLhs < nLhs; ++idxLhs){
168

169 170 171 172 173
          // update local index
          const int idxLoc = idxV*nLhs + idxLhs;

          // load transformed local expansion
          FComplexe *const TransformedLocalExpansion = TargetCell->getTransformedLocal(idxLoc);
174

175 176
          // update idxRhs
          const int idxRhs = idxLhs % nPV; 
177

178
          // update multipole index
179 180
          const int idxMul = idxV*nRhs + idxRhs;

181
          // get index in matrix kernel
182
          const unsigned int d = MatrixKernel->getPosition(idxLhs);
183

184 185
          for (int idx=0; idx<343; ++idx){
            if (SourceCells[idx]){
186

187
              M2LHandler.applyFC(idx, TreeLevel, scale, d,
188
                                  SourceCells[idx]->getTransformedMultipole(idxMul),
189
                                  TransformedLocalExpansion);
190

191
            }
192
          }
193
      }// NLHS=NPOT*NPV
194
    }// NVALS
195 196 197 198 199
  }


  void L2L(const CellClass* const FRestrict ParentCell,
           CellClass* FRestrict *const FRestrict ChildCells,
200
           const int TreeLevel)
201
  {
202 203 204 205
    for(int idxV = 0 ; idxV < NVALS ; ++idxV){
      for(int idxLhs = 0 ; idxLhs < nLhs ; ++idxLhs){
        int idxLoc = idxV*nLhs + idxLhs;
        // 1) Apply Inverse Discete Fourier Transform
206 207
        M2LHandler.unapplyZeroPaddingAndDFT(ParentCell->getTransformedLocal(idxLoc),
                                            const_cast<CellClass*>(ParentCell)->getLocal(idxLoc));
208 209 210
        // 2) apply Sx
        for (unsigned int ChildIndex=0; ChildIndex < 8; ++ChildIndex){
          if (ChildCells[ChildIndex]){
211 212 213 214
            AbstractBaseClass::Interpolator->applyL2L(ChildIndex, 
                                                      ParentCell->getLocal(idxLoc), 
                                                      ChildCells[ChildIndex]->getLocal(idxLoc),
                                                      TreeLevel/*Cell width extension specific*/);
215
          }
216 217
        }
      }
218
    }// NVALS
219 220 221 222 223 224
  }

  void L2P(const CellClass* const LeafCell,
           ContainerClass* const TargetParticles)
  {
    const FPoint LeafCellCenter(AbstractBaseClass::getLeafCellCenter(LeafCell->getCoordinate()));
225 226
    const FReal ExtendedLeafCellWidth(AbstractBaseClass::BoxWidthLeaf 
                                      + AbstractBaseClass::BoxWidthExtension);
227

228 229 230 231
    for(int idxV = 0 ; idxV < NVALS ; ++idxV){
      for(int idxLhs = 0 ; idxLhs < nLhs ; ++idxLhs){
        int idxLoc = idxV*nLhs + idxLhs;
        // 1)  Apply Inverse Discete Fourier Transform
232 233
        M2LHandler.unapplyZeroPaddingAndDFT(LeafCell->getTransformedLocal(idxLoc), 
                                            const_cast<CellClass*>(LeafCell)->getLocal(idxLoc));
234

235 236 237
      }

      // 2.a) apply Sx
238
      AbstractBaseClass::Interpolator->applyL2P(LeafCellCenter, ExtendedLeafCellWidth,
239
                                                LeafCell->getLocal(idxV*nLhs), TargetParticles);
240

241
      // 2.b) apply Px (grad Sx)
242
      AbstractBaseClass::Interpolator->applyL2PGradient(LeafCellCenter, ExtendedLeafCellWidth,
243
                                                        LeafCell->getLocal(idxV*nLhs), TargetParticles);
244

245
    }// NVALS
246 247 248 249 250 251 252 253
  }

  void P2P(const FTreeCoordinate& /* LeafCellCoordinate */, // needed for periodic boundary conditions
           ContainerClass* const FRestrict TargetParticles,
           const ContainerClass* const FRestrict /*SourceParticles*/,
           ContainerClass* const NeighborSourceParticles[27],
           const int /* size */)
  {
254
    DirectInteractionComputer<MatrixKernelClass::Identifier, NVALS>::P2P(TargetParticles,NeighborSourceParticles,MatrixKernel);
255 256 257 258 259 260
  }


  void P2PRemote(const FTreeCoordinate& /*inPosition*/,
                 ContainerClass* const FRestrict inTargets, const ContainerClass* const FRestrict /*inSources*/,
                 ContainerClass* const inNeighbors[27], const int /*inSize*/){
261
    DirectInteractionComputer<MatrixKernelClass::Identifier, NVALS>::P2PRemote(inTargets,inNeighbors,27,MatrixKernel);
262 263 264 265 266
  }

};


267
#endif //FUNIFTENSORIALKERNEL_HPP
268 269

// [--END--]