ParticleHandlerSpeedTest.cpp File Reference

#include <Boundaries/CubeInsertionBoundary.h>
#include <Mercury3D.h>
#include <Species/LinearViscoelasticSpecies.h>
#include <MercuryTime.h>

Functions
int	main ()

Function Documentation

main()

int main

(

)

{
    // create a particle handler
    Mercury3D dpm;
    auto species = dpm.speciesHandler.copyAndAddObject(LinearViscoelasticSpecies());
    SphericalParticle particle(species);
    CubeInsertionBoundary insertionBoundary;
    insertionBoundary.set(particle, 1e6, {0, 0, 0}, {10000, 10000, 10000}, {0, 0, 0}, {0, 0, 0});
    insertionBoundary.setInitialVolume(1e5 * constants::pi / 6.);
    insertionBoundary.checkBoundaryBeforeTimeStep(&dpm);
    size_t n = dpm.particleHandler.getSize();
    logger(INFO, "Number of particles: %", n);
    logger(INFO, "Size of particles: % bytes, % doubles", sizeof(particle), sizeof(particle) / sizeof(double));
    
    // create a data vector
    std::vector<std::array<double, 82>> dataVector(n);
    for (int i = 0; i < n; ++i)
    {
        dataVector[i][0] = dpm.particleHandler.getObject(i)->getPosition().X;
    }
    
    // create a small data vector
    std::vector<std::array<double, 8>> smallDataVector(n);
    for (int i = 0; i < n; ++i)
    {
        smallDataVector[i][0] = dpm.particleHandler.getObject(i)->getPosition().X;
    }
    
    // create a particle vector
    std::vector<SphericalParticle> particleVector(n);
    for (int i = 0; i < n; ++i) {
        particleVector[i].setPosition(dpm.particleHandler.getObject(i)->getPosition());
    }
 
    Time timer;
    double sum=0;
    int repetitions = 1e8;
 
    for (int i = 0; i < repetitions; ++i) {
        sum += dpm.particleHandler.getObject(rand() % n)->getPosition().X;
    }
    double refTime = timer.toctic();
    logger(INFO, "Time to access randomly data in the particle handler: % s (% pct)", refTime, 100);
 
    for (int i = 0; i < repetitions; ++i) {
        sum += dpm.particleHandler.getObject(i % n)->getPosition().X;
    }
    double time = timer.toctic();
    int relTime = 100. * time / refTime;
    logger(INFO, "Time to access ordered data in a particle handler: % s (% pct)", time, relTime);
 
    for (int i = 0; i < repetitions; ++i) {
        sum += particleVector[rand() % n].getPosition().X;
    }
    time = timer.toctic();
    relTime = 100. * time / refTime;
    logger(INFO, "Time to access randomly data in a particle vector: % s (% pct)", time, relTime);
 
    for (int i = 0; i < repetitions; ++i) {
        sum += dataVector[rand() % n][0];
    }
    time = timer.toctic();
    relTime = 100. * time / refTime;
    logger(INFO, "Time to access randomly data in a vector of arrays: % s (% pct)", time, relTime);
 
    for (int i = 0; i < repetitions; ++i) {
        sum += smallDataVector[rand() % n][0];
    }
    time = timer.toctic();
    relTime = 100. * time / refTime;
    logger(INFO, "Time to access randomly data in a small vector of small arrays: % s (% pct)", time, relTime);
 
    logger(INFO,"Computation: %",sum);
 
    logger(INFO,"Conclusions:\n"
                " - 50\% gain: Ordered data access is much quicker than random\n"
                " - 20\% gain: Vectors are quicker than handlers\n"
                " -  0\% gain: Class type (SphericalParticle vs array<double,83>) makes no difference\n"
                " - 20\% gain: A lean array (8 doubles) is read quicker than a fat array (83 doubles) ");
 
}

References InsertionBoundary::checkBoundaryBeforeTimeStep(), BaseHandler< T >::copyAndAddObject(), BaseHandler< T >::getObject(), BaseInteractable::getPosition(), BaseHandler< T >::getSize(), i, INFO, logger, n, DPMBase::particleHandler, constants::pi, CubeInsertionBoundary::set(), InsertionBoundary::setInitialVolume(), DPMBase::speciesHandler, and Vec3D::X.