FFmmAlgorithmThread.hpp 14.1 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 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 54 55 56 57 58 59 60 61 62 63 64 65 66 67 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352
#ifndef FFMMALGORITHMTHREAD_HPP
#define FFMMALGORITHMTHREAD_HPP
// /!\ Please, you must read the license at the bottom of this page

#include "../Utils/FAssertable.hpp"
#include "../Utils/FDebug.hpp"
#include "../Utils/FTrace.hpp"
#include "../Utils/FTic.hpp"
#include "../Utils/FGlobal.hpp"

#include "../Containers/FOctree.hpp"


#include <omp.h>

/**
* @author Berenger Bramas (berenger.bramas@inria.fr)
* @class FFmmAlgorithmThread
* @brief
* Please read the license
*
* This class is a threaded FMM algorithm
* It just iterates on a tree and call the kernels with good arguments.
* It used the inspector-executor model :
* iterates on the tree and builds an array to work in parallel on this array
*
* Of course this class does not deallocate pointer given in arguements.
*
* Threaded & based on the inspector-executor model
* schedule(runtime)
*/
template<template< class ParticleClass, class CellClass, int OctreeHeight> class KernelClass,
        class ParticleClass, class CellClass,
        template<class ParticleClass> class LeafClass,
        int OctreeHeight, int SubtreeHeight>
class FFmmAlgorithmThread : protected FAssertable{
    // To reduce the size of variable type based on foctree in this file
    typedef FOctree<ParticleClass, CellClass, LeafClass, OctreeHeight, SubtreeHeight> Octree;
    typedef typename FOctree<ParticleClass, CellClass,LeafClass, OctreeHeight, SubtreeHeight>::Iterator OctreeIterator;
    typedef KernelClass<ParticleClass, CellClass, OctreeHeight> Kernel;

    Octree* const tree;                  //< The octree to work on
    Kernel* kernels[FThreadNumbers];          //< The kernels

    FDEBUG(FTic counterTime);                //< In case of debug: to count the elapsed time
    FDEBUG(FTic computationCounter);     //< In case of debug: to  count computation time

    OctreeIterator* iterArray;

    static const int SizeShape = 3*3*3;
    int shapeLeaf[SizeShape];

public:
    /** The constructor need the octree and the kernels used for computation
      * @param inTree the octree to work on
      * @param inKernels the kernels to call
      * An assert is launched if one of the arguments is null
      */
    FFmmAlgorithmThread(Octree* const inTree, Kernel* const inKernels)
                      : tree(inTree) , iterArray(0) {

        assert(tree, "tree cannot be null", __LINE__, __FILE__);
        assert(kernels, "kernels cannot be null", __LINE__, __FILE__);

        for(int idxThread = 0 ; idxThread < FThreadNumbers ; ++idxThread){
            this->kernels[idxThread] = new KernelClass<ParticleClass, CellClass, OctreeHeight>(*inKernels);
        }

        FDEBUG(FDebug::Controller << "FFmmAlgorithmThread\n");
    }

    /** Default destructor */
    virtual ~FFmmAlgorithmThread(){
        for(int idxThread = 0 ; idxThread < FThreadNumbers ; ++idxThread){
            delete this->kernels[idxThread];
        }
    }

    /**
      * To execute the fmm algorithm
      * Call this function to run the complete algorithm
      */
    void execute(){
        FTRACE( FTrace::Controller.enterFunction(FTrace::FMM, __FUNCTION__ , __FILE__ , __LINE__) );

        for(int idxShape = 0 ; idxShape < SizeShape ; ++idxShape){
            this->shapeLeaf[idxShape] = 0;
        }
        const int LeafIndex = OctreeHeight - 1;

        // Count leaf
        int leafs = 0;
        OctreeIterator octreeIterator(tree);
        octreeIterator.gotoBottomLeft();
        do{
            ++leafs;
            const MortonIndex index = octreeIterator.getCurrentGlobalIndex();
            FTreeCoordinate coord;
            coord.setPositionFromMorton(index, LeafIndex);
            ++this->shapeLeaf[(coord.getX()%3)*9 + (coord.getY()%3)*3 + (coord.getZ()%3)];

        } while(octreeIterator.moveRight());
        iterArray = new OctreeIterator[leafs];
        assert(iterArray, "iterArray bad alloc", __LINE__, __FILE__);

        for(int idxThread = 0 ; idxThread < FThreadNumbers ; ++idxThread){
            this->kernels[idxThread]->init();
        }

        bottomPass();
        upwardPass();

        downardPass();

        directPass();

        delete [] iterArray;
        iterArray = 0;

        FTRACE( FTrace::Controller.leaveFunction(FTrace::FMM) );
    }

    /** P2M */
    void bottomPass(){
        FTRACE( FTrace::Controller.enterFunction(FTrace::FMM, __FUNCTION__ , __FILE__ , __LINE__) );
        FDEBUG( FDebug::Controller.write("\tStart Bottom Pass\n").write(FDebug::Flush) );
        FDEBUG( counterTime.tic() );

        OctreeIterator octreeIterator(tree);
        int leafs = 0;
        // Iterate on leafs
        octreeIterator.gotoBottomLeft();
        do{
            iterArray[leafs] = octreeIterator;
            ++leafs;
        } while(octreeIterator.moveRight());

        FDEBUG(computationCounter.tic());
        #pragma omp parallel num_threads(FThreadNumbers)
        {
            Kernel * const myThreadkernels = kernels[omp_get_thread_num()];
            #pragma omp for
            for(int idxLeafs = 0 ; idxLeafs < leafs ; ++idxLeafs){
                // We need the current cell that represent the leaf
                // and the list of particles
                myThreadkernels->P2M( iterArray[idxLeafs].getCurrentCell() , iterArray[idxLeafs].getCurrentListSrc());
            }
        }
        FDEBUG(computationCounter.tac());

        FDEBUG( counterTime.tac() );
        FDEBUG( FDebug::Controller << "\tFinished ("  << counterTime.elapsed() << "s)\n" );
        FDEBUG( FDebug::Controller << "\t\t Computation : " << computationCounter.elapsed() << " s\n" );
        FTRACE( FTrace::Controller.leaveFunction(FTrace::FMM) );
    }

    /** M2M */
    void upwardPass(){
        FTRACE( FTrace::Controller.enterFunction(FTrace::FMM, __FUNCTION__ , __FILE__ , __LINE__) );
        FDEBUG( FDebug::Controller.write("\tStart Upward Pass\n").write(FDebug::Flush); );
        FDEBUG( counterTime.tic() );
        FDEBUG( double totalComputation = 0 );

        // Start from leal level - 1
        OctreeIterator octreeIterator(tree);
        octreeIterator.gotoBottomLeft();
        octreeIterator.moveUp();
        OctreeIterator avoidGotoLeftIterator(octreeIterator);

        // for each levels
        for(int idxLevel = OctreeHeight - 2 ; idxLevel > 1 ; --idxLevel ){
            int leafs = 0;
            // for each cells
            do{
                iterArray[leafs] = octreeIterator;
                ++leafs;
            } while(octreeIterator.moveRight());
            avoidGotoLeftIterator.moveUp();
            octreeIterator = avoidGotoLeftIterator;// equal octreeIterator.moveUp(); octreeIterator.gotoLeft();

            FDEBUG(computationCounter.tic());
            #pragma omp parallel num_threads(FThreadNumbers)
            {
                Kernel * const myThreadkernels = kernels[omp_get_thread_num()];
                #pragma omp for
                for(int idxLeafs = 0 ; idxLeafs < leafs ; ++idxLeafs){
                    // We need the current cell and the child
                    // child is an array (of 8 child) that may be null
                    myThreadkernels->M2M( iterArray[idxLeafs].getCurrentCell() , iterArray[idxLeafs].getCurrentChild(), idxLevel);
                }
            }
            FDEBUG(computationCounter.tac());
            FDEBUG(totalComputation += computationCounter.elapsed());
        }

        FDEBUG( counterTime.tac() );
        FDEBUG( FDebug::Controller << "\tFinished ("  << counterTime.elapsed() << "s)\n" );
        FDEBUG( FDebug::Controller << "\t\t Computation : " << totalComputation << " s\n" );
        FTRACE( FTrace::Controller.leaveFunction(FTrace::FMM) );
    }

    /** M2L L2L */
    void downardPass(){
        FTRACE( FTrace::Controller.enterFunction(FTrace::FMM, __FUNCTION__ , __FILE__ , __LINE__) );
        FDEBUG( FDebug::Controller.write("\tStart Downward Pass (M2L)\n").write(FDebug::Flush); );
        FDEBUG( counterTime.tic() );
        FDEBUG( double totalComputation = 0 );

        { // first M2L
            OctreeIterator octreeIterator(tree);
            octreeIterator.moveDown();
            OctreeIterator avoidGotoLeftIterator(octreeIterator);

            // for each levels
            for(int idxLevel = 2 ; idxLevel < OctreeHeight ; ++idxLevel ){
                int leafs = 0;
                // for each cells
                do{
                    iterArray[leafs] = octreeIterator;
                    ++leafs;
                } while(octreeIterator.moveRight());
                avoidGotoLeftIterator.moveDown();
                octreeIterator = avoidGotoLeftIterator;

                FDEBUG(computationCounter.tic());
                #pragma omp parallel num_threads(FThreadNumbers)
                {
                    Kernel * const myThreadkernels = kernels[omp_get_thread_num()];
                    CellClass* neighbors[208];
                    #pragma omp for
                    for(int idxLeafs = 0 ; idxLeafs < leafs ; ++idxLeafs){
                        const int counter = tree->getDistantNeighbors(neighbors,  iterArray[idxLeafs].getCurrentGlobalIndex(),idxLevel);
                        if(counter) myThreadkernels->M2L(  iterArray[idxLeafs].getCurrentCell() , neighbors, counter, idxLevel);
                    }
                }
                FDEBUG(computationCounter.tac());
                FDEBUG(totalComputation += computationCounter.elapsed());
            }
        }
        FDEBUG( counterTime.tac() );
        FDEBUG( FDebug::Controller << "\tFinished ("  << counterTime.elapsed() << "s)\n" );
        FDEBUG( FDebug::Controller << "\t\t Computation : " << totalComputation << " s\n" );

        FDEBUG( FDebug::Controller.write("\tStart Downward Pass (L2L)\n").write(FDebug::Flush); );
        FDEBUG( counterTime.tic() );
        FDEBUG( totalComputation = 0 );
        { // second L2L
            OctreeIterator octreeIterator(tree);
            octreeIterator.moveDown();

            OctreeIterator avoidGotoLeftIterator(octreeIterator);

            const int heightMinusOne = OctreeHeight - 1;
            // for each levels exepted leaf level
            for(int idxLevel = 2 ; idxLevel < heightMinusOne ; ++idxLevel ){
                int leafs = 0;
                // for each cells
                do{
                    iterArray[leafs] = octreeIterator;
                    ++leafs;
                } while(octreeIterator.moveRight());
                avoidGotoLeftIterator.moveDown();
                octreeIterator = avoidGotoLeftIterator;

                FDEBUG(computationCounter.tic());
                #pragma omp parallel num_threads(FThreadNumbers)
                {
                    Kernel * const myThreadkernels = kernels[omp_get_thread_num()];
                    #pragma omp for
                    for(int idxLeafs = 0 ; idxLeafs < leafs ; ++idxLeafs){
                        myThreadkernels->L2L( iterArray[idxLeafs].getCurrentCell() , iterArray[idxLeafs].getCurrentChild(), idxLevel);
                    }
                }
                FDEBUG(computationCounter.tac());
                FDEBUG(totalComputation += computationCounter.elapsed());
            }
        }

        FDEBUG( counterTime.tac() );
        FDEBUG( FDebug::Controller << "\tFinished ("  << counterTime.elapsed() << "s)\n" );
        FDEBUG( FDebug::Controller << "\t\t Computation : " << totalComputation << " s\n" );
        FTRACE( FTrace::Controller.leaveFunction(FTrace::FMM) );
    }

    /** P2P */
    void directPass(){
        FTRACE( FTrace::Controller.enterFunction(FTrace::FMM, __FUNCTION__ , __FILE__ , __LINE__) );
        FDEBUG( FDebug::Controller.write("\tStart Direct Pass\n").write(FDebug::Flush); );
        FDEBUG( counterTime.tic() );

        OctreeIterator* shapeArray[SizeShape];
        int countShape[SizeShape];
        for(int idxShape = 0 ; idxShape < SizeShape ; ++idxShape){
            shapeArray[idxShape] = new OctreeIterator[this->shapeLeaf[idxShape]];
            countShape[idxShape] = 0;
        }

        const int LeafIndex = OctreeHeight - 1;
        //int leafs = 0;
        {
            OctreeIterator octreeIterator(tree);
            octreeIterator.gotoBottomLeft();
            // for each leafs
            do{
                //iterArray[leafs] = octreeIterator;
                //++leafs;
                const MortonIndex index = octreeIterator.getCurrentGlobalIndex();
                FTreeCoordinate coord;
                coord.setPositionFromMorton(index, LeafIndex);
                const int shapePosition = (coord.getX()%3)*9 + (coord.getY()%3)*3 + (coord.getZ()%3);
                shapeArray[shapePosition][countShape[shapePosition]] = octreeIterator;
                ++countShape[shapePosition];

            } while(octreeIterator.moveRight());
        }

        FDEBUG(computationCounter.tic());
        for(int idxShape = 0 ; idxShape < SizeShape ; ++idxShape){
            const int leafAtThisShape = this->shapeLeaf[idxShape];
            #pragma omp parallel num_threads(FThreadNumbers)
            {
                Kernel * const myThreadkernels = kernels[omp_get_thread_num()];
                // There is a maximum of 26 neighbors
                FList<ParticleClass*>* neighbors[26];

                #pragma omp for
                for(int idxLeafs = 0 ; idxLeafs < leafAtThisShape ; ++idxLeafs){
                    myThreadkernels->L2P(shapeArray[idxShape][idxLeafs].getCurrentCell(), shapeArray[idxShape][idxLeafs].getCurrentListTargets());
                    // need the current particles and neighbors particles
                    const int counter = tree->getLeafsNeighbors(neighbors, shapeArray[idxShape][idxLeafs].getCurrentGlobalIndex(),LeafIndex);
                    myThreadkernels->P2P( shapeArray[idxShape][idxLeafs].getCurrentListTargets(), shapeArray[idxShape][idxLeafs].getCurrentListSrc() , neighbors, counter);
                }
            }
        }
        FDEBUG(computationCounter.tac());

        for(int idxShape = 0 ; idxShape < SizeShape ; ++idxShape){
            delete [] shapeArray[idxShape];
        }

        FDEBUG( counterTime.tac() );
        FDEBUG( FDebug::Controller << "\tFinished ("  << counterTime.elapsed() << "s)\n" );
        FDEBUG( FDebug::Controller << "\t\t Computation : " << computationCounter.elapsed() << " s\n" );
        FTRACE( FTrace::Controller.leaveFunction(FTrace::FMM) );
    }

};


#endif //FFMMALGORITHMTHREAD_HPP

// [--LICENSE--]