utestChebyshevThread.cpp 12.1 KB
Newer Older
COULAUD Olivier's avatar
COULAUD Olivier committed
1
// ===================================================================================
2 3 4 5
// Copyright ScalFmm 2016 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.
6
//
7
// This software is governed by the CeCILL-C and LGPL licenses and
8
// abiding by the rules of distribution of free software.
9 10 11
// An extension to the license is given to allow static linking of scalfmm
// inside a proprietary application (no matter its license).
// See the main license file for more details.
12
//
COULAUD Olivier's avatar
COULAUD Olivier committed
13 14 15
// 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
16 17 18
// GNU General Public and CeCILL-C Licenses for more details.
// "http://www.cecill.info".
// "http://www.gnu.org/licenses".
COULAUD Olivier's avatar
COULAUD Olivier committed
19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53
// ===================================================================================

// ==== CMAKE =====
// @FUSE_BLAS
// ================
#include "ScalFmmConfig.h"
#include "Utils/FGlobal.hpp"

#include "Containers/FOctree.hpp"

#include "Files/FFmaGenericLoader.hpp"

#include "Core/FFmmAlgorithmThread.hpp"

#include "FUTester.hpp"

#include "Components/FSimpleLeaf.hpp"


#include "Kernels/Chebyshev/FChebCell.hpp"
#include "Kernels/Interpolation/FInterpMatrixKernel.hpp"
#include "Kernels/Chebyshev/FChebKernel.hpp"
#include "Kernels/Chebyshev/FChebSymKernel.hpp"

#include "Kernels/P2P/FP2PParticleContainerIndexed.hpp"
/*
  In this test we compare the Chebyschev fmm results and the direct results.
 */


/** the test class
 *
 */
class TestChebyshevDirect : public FUTester<TestChebyshevDirect> {

54 55 56 57 58 59 60 61
    ///////////////////////////////////////////////////////////
    // The tests!
    ///////////////////////////////////////////////////////////

    template <class FReal, class CellClass, class ContainerClass, class KernelClass, class MatrixKernelClass,
              class LeafClass, class OctreeClass, class FmmClass>
    void RunTest()	{
        //
COULAUD Olivier's avatar
COULAUD Olivier committed
62
#ifdef _OPENMP
63
        std::cout << "\n>> Using " << omp_get_max_threads() << " threads.\n" << std::endl;
COULAUD Olivier's avatar
COULAUD Olivier committed
64
#else
65 66
        std::cout << "\n>> OpenMP test !!!\n" << std::endl;
        exit(EXIT_FAILURE);
COULAUD Olivier's avatar
COULAUD Olivier committed
67
#endif
68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101
        //
        // Load particles
        //
        if(sizeof(FReal) == sizeof(float) ) {
            std::cerr << "No input data available for Float "<< std::endl;
            exit(EXIT_FAILURE);
        }
        const std::string parFile( (sizeof(FReal) == sizeof(float))?
                                       "Test/DirectFloatbfma":
                                       "UTest/DirectDouble.bfma");
        //
        std::string filename(SCALFMMDataPath+parFile);
        //
        FFmaGenericLoader<FReal> loader(filename);
        Print("Number of particles:");
        Print(loader.getNumberOfParticles());

        const int NbLevels        = 4;
        const int SizeSubLevels = 2;

        // Create Matrix Kernel
        const MatrixKernelClass MatrixKernel; // FUKernelTester is only designed to work with 1/R, i.e. matrix kernel ctor takes no argument.

        // Load particles
        FSize nbParticles = loader.getNumberOfParticles() ;
        FmaRWParticle<FReal, 8,8>* const particles = new FmaRWParticle<FReal, 8,8>[nbParticles];

        loader.fillParticle(particles,nbParticles);

        // Create octree
        OctreeClass tree(NbLevels, SizeSubLevels, loader.getBoxWidth(), loader.getCenterOfBox());
        //
        //   Insert particle in the tree
        //
102
        for(FSize idxPart = 0 ; idxPart < loader.getNumberOfParticles() ; ++idxPart){
103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118
            tree.insert(particles[idxPart].getPosition() , idxPart, particles[idxPart].getPhysicalValue() );
        }
        //
        /////////////////////////////////////////////////////////////////////////////////////////////////
        // Run FMM computation
        /////////////////////////////////////////////////////////////////////////////////////////////////
        Print("Fmm...");
        KernelClass kernels(NbLevels, loader.getBoxWidth(), loader.getCenterOfBox(),&MatrixKernel);
        FmmClass algo(&tree,&kernels);
        algo.execute();
        //0
        FReal energy= 0.0 , energyD = 0.0 ;
        /////////////////////////////////////////////////////////////////////////////////////////////////
        // Compute direct energy
        /////////////////////////////////////////////////////////////////////////////////////////////////

119
        for(FSize idx = 0 ; idx < loader.getNumberOfParticles()  ; ++idx){
120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135
            energyD +=  particles[idx].getPotential()*particles[idx].getPhysicalValue() ;
        }
        /////////////////////////////////////////////////////////////////////////////////////////////////
        // Compare
        /////////////////////////////////////////////////////////////////////////////////////////////////
        Print("Compute Diff...");
        FMath::FAccurater<FReal> potentialDiff;
        FMath::FAccurater<FReal> fx, fy, fz;
        { // Check that each particle has been summed with all other

            tree.forEachLeaf([&](LeafClass* leaf){
                const FReal*const potentials        = leaf->getTargets()->getPotentials();
                const FReal*const physicalValues = leaf->getTargets()->getPhysicalValues();
                const FReal*const forcesX            = leaf->getTargets()->getForcesX();
                const FReal*const forcesY            = leaf->getTargets()->getForcesY();
                const FReal*const forcesZ            = leaf->getTargets()->getForcesZ();
136 137
                const FSize nbParticlesInLeaf           = leaf->getTargets()->getNbParticles();
                const FVector<FSize>& indexes = leaf->getTargets()->getIndexes();
138

139 140
                for(FSize idxPart = 0 ; idxPart < nbParticlesInLeaf ; ++idxPart){
                    const FSize indexPartOrig = indexes[idxPart];
141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233
                    potentialDiff.add(particles[indexPartOrig].getPotential(),potentials[idxPart]);
                    fx.add(particles[indexPartOrig].getForces()[0],forcesX[idxPart]);
                    fy.add(particles[indexPartOrig].getForces()[1],forcesY[idxPart]);
                    fz.add(particles[indexPartOrig].getForces()[2],forcesZ[idxPart]);
                    energy   += potentials[idxPart]*physicalValues[idxPart];
                }
            });
        }

        delete[] particles;

        // Print for information

        Print("Potential diff is = ");
        printf("         Pot L2Norm     %e\n",potentialDiff.getL2Norm());
        printf("         Pot RL2Norm   %e\n",potentialDiff.getRelativeL2Norm());
        printf("         Pot RMSError   %e\n",potentialDiff.getRMSError());
        Print("Fx diff is = ");
        printf("         Fx L2Norm     %e\n",fx.getL2Norm());
        printf("         Fx RL2Norm   %e\n",fx.getRelativeL2Norm());
        printf("         Fx RMSError   %e\n",fx.getRMSError());
        Print("Fy diff is = ");
        printf("        Fy L2Norm     %e\n",fy.getL2Norm());
        printf("        Fy RL2Norm   %e\n",fy.getRelativeL2Norm());
        printf("        Fy RMSError   %e\n",fy.getRMSError());
        Print("Fz diff is = ");
        printf("        Fz L2Norm     %e\n",fz.getL2Norm());
        printf("        Fz RL2Norm   %e\n",fz.getRelativeL2Norm());
        printf("        Fz RMSError   %e\n",fz.getRMSError());
        FReal L2error = (fx.getRelativeL2Norm()*fx.getRelativeL2Norm() + fy.getRelativeL2Norm()*fy.getRelativeL2Norm()  + fz.getRelativeL2Norm() *fz.getRelativeL2Norm()  );
        printf(" Total L2 Force Error= %e\n",FMath::Sqrt(L2error)) ;
        printf("  Energy Error  =   %.12e\n",FMath::Abs(energy-energyD));
        printf("  Energy FMM    =   %.12e\n",FMath::Abs(energy));
        printf("  Energy DIRECT =   %.12e\n",FMath::Abs(energyD));

        // Assert
        const FReal MaximumDiffPotential = FReal(9e-3);
        const FReal MaximumDiffForces     = FReal(9e-2);

        Print("Test1 - Error Relative L2 norm Potential ");
        uassert(potentialDiff.getRelativeL2Norm() < MaximumDiffPotential);    //1
        Print("Test2 - Error RMS L2 norm Potential ");
        uassert(potentialDiff.getRMSError() < MaximumDiffPotential);  //2
        Print("Test3 - Error Relative L2 norm FX ");
        uassert(fx.getRelativeL2Norm()  < MaximumDiffForces);                       //3
        Print("Test4 - Error RMS L2 norm FX ");
        uassert(fx.getRMSError() < MaximumDiffForces);                      //4
        Print("Test5 - Error Relative L2 norm FY ");
        uassert(fy.getRelativeL2Norm()  < MaximumDiffForces);                       //5
        Print("Test6 - Error RMS L2 norm FY ");
        uassert(fy.getRMSError() < MaximumDiffForces);                      //6
        Print("Test7 - Error Relative L2 norm FZ ");
        uassert(fz.getRelativeL2Norm()  < MaximumDiffForces);                      //8
        Print("Test8 - Error RMS L2 norm FZ ");
        uassert(fz.getRMSError() < MaximumDiffForces);                                           //8
        Print("Test9 - Error Relative L2 norm F ");
        uassert(L2error              < MaximumDiffForces);                                            //9   Total Force
        Print("Test10 - Relative error Energy ");
        uassert(FMath::Abs(energy-energyD) /energyD< MaximumDiffPotential);                     //10  Total Energy


    }

    /** If memstas is running print the memory used */
    void PostTest() {
        if( FMemStats::controler.isUsed() ){
            std::cout << "Memory used at the end " << FMemStats::controler.getCurrentAllocated()
                      << " Bytes (" << FMemStats::controler.getCurrentAllocatedMB() << "MB)\n";
            std::cout << "Max memory used " << FMemStats::controler.getMaxAllocated()
                      << " Bytes (" << FMemStats::controler.getMaxAllocatedMB() << "MB)\n";
            std::cout << "Total memory used " << FMemStats::controler.getTotalAllocated()
                      << " Bytes (" << FMemStats::controler.getTotalAllocatedMB() << "MB)\n";
        }
    }


    ///////////////////////////////////////////////////////////
    // Set the tests!
    ///////////////////////////////////////////////////////////


    /** TestChebKernel */
    void TestChebKernel(){
        typedef double FReal;
        const unsigned int ORDER = 6;
        typedef FP2PParticleContainerIndexed<FReal> ContainerClass;
        typedef FSimpleLeaf<FReal, ContainerClass> LeafClass;
        typedef FInterpMatrixKernelR<FReal> MatrixKernelClass;
        typedef FChebCell<FReal,ORDER> CellClass;
        typedef FOctree<FReal, CellClass,ContainerClass,LeafClass> OctreeClass;
        typedef FChebKernel<FReal,CellClass,ContainerClass,MatrixKernelClass,ORDER> KernelClass;
        typedef FFmmAlgorithmThread<OctreeClass,CellClass,ContainerClass,KernelClass,LeafClass> FmmClass;
        // run test
234
        RunTest<FReal,CellClass,ContainerClass,KernelClass,MatrixKernelClass,LeafClass,OctreeClass,FmmClass>();
235 236 237 238 239 240 241 242 243 244 245 246 247 248
    }

    /** TestChebSymKernel */
    void TestChebSymKernel(){
        typedef double FReal;
        const unsigned int ORDER = 6;
        typedef FP2PParticleContainerIndexed<FReal> ContainerClass;
        typedef FSimpleLeaf<FReal, ContainerClass> LeafClass;
        typedef FInterpMatrixKernelR<FReal> MatrixKernelClass;
        typedef FChebCell<FReal,ORDER> CellClass;
        typedef FOctree<FReal, CellClass,ContainerClass,LeafClass> OctreeClass;
        typedef FChebSymKernel<FReal,CellClass,ContainerClass,MatrixKernelClass,ORDER> KernelClass;
        typedef FFmmAlgorithmThread<OctreeClass,CellClass,ContainerClass,KernelClass,LeafClass> FmmClass;
        // run test
249
        RunTest<FReal,CellClass,ContainerClass,KernelClass,MatrixKernelClass,LeafClass,OctreeClass,FmmClass>();
250 251 252 253 254 255 256 257 258 259 260 261 262
    }



    ///////////////////////////////////////////////////////////
    // Set the tests!
    ///////////////////////////////////////////////////////////

    /** set test */
    void SetTests(){
        AddTest(&TestChebyshevDirect::TestChebKernel,"Test Chebyshev Kernel with one big SVD");
        AddTest(&TestChebyshevDirect::TestChebSymKernel,"Test Chebyshev Kernel with 16 small SVDs and symmetries");
    }
COULAUD Olivier's avatar
COULAUD Olivier committed
263 264 265 266 267 268 269 270 271
};


// You must do this
TestClass(TestChebyshevDirect)