Skip to content

GitLab

S
ScalFMM
Commit 48d763c1 authored by BRAMAS Berenger's avatar BRAMAS Berenger
Browse files
Options

The block algorithms are now able to work on specific level

parent 64e09b61
Hide whitespace changes
Inline Side-by-side
Showing with 60 additions and 32 deletions
Src/GroupTree/Core/FGroupSeqAlgorithm.hpp
View file @ 48d763c1
......@@ -17,7 +17,7 @@
#include <vector>
template <class OctreeClass, class CellContainerClass, class CellClass, class KernelClass, class ParticleGroupClass, class ParticleContainerClass>
class FGroupSeqAlgorithm {
class FGroupSeqAlgorithm : public FAbstractAlgorithm {
protected:
const int MaxThreads; //< The number of threads
OctreeClass*const tree; //< The Tree
......@@ -28,13 +28,19 @@ public:
FAssertLF(tree, "tree cannot be null");
FAssertLF(kernels, "kernels cannot be null");
FAbstractAlgorithm::setNbLevelsInTree(tree->getHeight());
FLOG(FLog::Controller << "FGroupSeqAlgorithm (Max Thread " << MaxThreads << ")\n");
}
~FGroupSeqAlgorithm(){
}
void execute(const unsigned operationsToProceed = FFmmNearAndFarFields){
protected:
/**
* Runs the complete algorithm.
*/
void executeCore(const unsigned operationsToProceed) override {
FLOG( FLog::Controller << "\tStart FGroupSeqAlgorithm\n" );
if(operationsToProceed & FFmmP2M) bottomPass();
......@@ -48,7 +54,6 @@ public:
if( (operationsToProceed & FFmmP2P) || (operationsToProceed & FFmmL2P) ) directPass();
}
protected:
void bottomPass(){
FLOG( FTic timer; );
typename OctreeClass::ParticleGroupIterator iterParticles = tree->leavesBegin();
......@@ -83,7 +88,7 @@ protected:
void upwardPass(){
FLOG( FTic timer; );
for(int idxLevel = tree->getHeight()-2 ; idxLevel >= 2 ; --idxLevel){
for(int idxLevel = FMath::Min(tree->getHeight() - 2, FAbstractAlgorithm::lowerWorkingLevel - 1) ; idxLevel >= FAbstractAlgorithm::upperWorkingLevel ; --idxLevel){
typename OctreeClass::CellGroupIterator iterCells = tree->cellsBegin(idxLevel);
const typename OctreeClass::CellGroupIterator endCells = tree->cellsEnd(idxLevel);
......@@ -133,7 +138,7 @@ protected:
void transferPass(){
FLOG( FTic timer; );
for(int idxLevel = tree->getHeight()-1 ; idxLevel >= 2 ; --idxLevel){
for(int idxLevel = FAbstractAlgorithm::lowerWorkingLevel-1 ; idxLevel >= FAbstractAlgorithm::upperWorkingLevel ; --idxLevel){
typename OctreeClass::CellGroupIterator iterCells = tree->cellsBegin(idxLevel);
const typename OctreeClass::CellGroupIterator endCells = tree->cellsEnd(idxLevel);
......@@ -238,7 +243,7 @@ protected:
void downardPass(){
FLOG( FTic timer; );
for(int idxLevel = 2 ; idxLevel <= tree->getHeight()-2 ; ++idxLevel){
for(int idxLevel = FAbstractAlgorithm::upperWorkingLevel ; idxLevel < FAbstractAlgorithm::lowerWorkingLevel - 1 ; ++idxLevel){
typename OctreeClass::CellGroupIterator iterCells = tree->cellsBegin(idxLevel);
const typename OctreeClass::CellGroupIterator endCells = tree->cellsEnd(idxLevel);
......
Src/GroupTree/Core/FGroupTaskAlgorithm.hpp
View file @ 48d763c1
......@@ -19,7 +19,7 @@
#include <omp.h>
template <class OctreeClass, class CellContainerClass, class CellClass, class KernelClass, class ParticleGroupClass, class ParticleContainerClass>
class FGroupTaskAlgorithm {
class FGroupTaskAlgorithm : public FAbstractAlgorithm {
protected:
template <class OtherBlockClass>
struct BlockInteractions{
......@@ -40,6 +40,8 @@ public:
FAssertLF(tree, "tree cannot be null");
FAssertLF(inKernels, "kernels cannot be null");
FAbstractAlgorithm::setNbLevelsInTree(tree->getHeight());
kernels = new KernelClass*[MaxThreads];
#pragma omp parallel for schedule(static) num_threads(MaxThreads)
for(int idxThread = 0 ; idxThread < MaxThreads ; ++idxThread){
......@@ -61,7 +63,11 @@ public:
delete[] kernels;
}
void execute(const unsigned operationsToProceed = FFmmNearAndFarFields){
protected:
/**
* Runs the complete algorithm.
*/
void executeCore(const unsigned operationsToProceed) override {
FLOG( FLog::Controller << "\tStart FGroupTaskAlgorithm\n" );
#pragma omp parallel num_threads(MaxThreads)
......@@ -101,7 +107,6 @@ public:
}
}
protected:
/**
* This function is creating the interactions vector between blocks.
* It fills externalInteractionsAllLevel and externalInteractionsLeafLevel.
......@@ -308,7 +313,7 @@ protected:
void upwardPass(){
FLOG( FTic timer; );
for(int idxLevel = tree->getHeight()-2 ; idxLevel >= 2 ; --idxLevel){
for(int idxLevel = FMath::Min(tree->getHeight() - 2, FAbstractAlgorithm::lowerWorkingLevel - 1) ; idxLevel >= FAbstractAlgorithm::upperWorkingLevel ; --idxLevel){
typename OctreeClass::CellGroupIterator iterCells = tree->cellsBegin(idxLevel);
const typename OctreeClass::CellGroupIterator endCells = tree->cellsEnd(idxLevel);
......@@ -388,7 +393,7 @@ protected:
void transferPass(){
FLOG( FTic timer; );
FLOG( FTic timerInBlock; FTic timerOutBlock; );
for(int idxLevel = tree->getHeight()-1 ; idxLevel >= 2 ; --idxLevel){
for(int idxLevel = FAbstractAlgorithm::lowerWorkingLevel-1 ; idxLevel >= FAbstractAlgorithm::upperWorkingLevel ; --idxLevel){
FLOG( timerInBlock.tic() );
{
typename OctreeClass::CellGroupIterator iterCells = tree->cellsBegin(idxLevel);
......@@ -499,7 +504,7 @@ protected:
void downardPass(){
FLOG( FTic timer; );
for(int idxLevel = 2 ; idxLevel <= tree->getHeight()-2 ; ++idxLevel){
for(int idxLevel = FAbstractAlgorithm::upperWorkingLevel ; idxLevel < FAbstractAlgorithm::lowerWorkingLevel - 1 ; ++idxLevel){
typename OctreeClass::CellGroupIterator iterCells = tree->cellsBegin(idxLevel);
const typename OctreeClass::CellGroupIterator endCells = tree->cellsEnd(idxLevel);
......
Src/GroupTree/Core/FGroupTaskDepAlgorithm.hpp
View file @ 48d763c1
......@@ -21,7 +21,7 @@
template <class OctreeClass, class CellContainerClass, class CellClass,
class SymboleCellClass, class PoleCellClass, class LocalCellClass, class KernelClass, class ParticleGroupClass, class ParticleContainerClass>
class FGroupTaskDepAlgorithm {
class FGroupTaskDepAlgorithm : public FAbstractAlgorithm {
protected:
template <class OtherBlockClass>
struct BlockInteractions{
......@@ -42,6 +42,8 @@ public:
FAssertLF(tree, "tree cannot be null");
FAssertLF(inKernels, "kernels cannot be null");
FAbstractAlgorithm::setNbLevelsInTree(tree->getHeight());
kernels = new KernelClass*[MaxThreads];
#pragma omp parallel for schedule(static) num_threads(MaxThreads)
for(int idxThread = 0 ; idxThread < MaxThreads ; ++idxThread){
......@@ -63,7 +65,11 @@ public:
delete[] kernels;
}
void execute(const unsigned operationsToProceed = FFmmNearAndFarFields){
protected:
/**
* Runs the complete algorithm.
*/
void executeCore(const unsigned operationsToProceed) override {
FLOG( FLog::Controller << "\tStart FGroupTaskDepAlgorithm\n" );
#pragma omp parallel num_threads(MaxThreads)
......@@ -96,7 +102,7 @@ public:
}
}
protected:
/**
* This function is creating the interactions vector between blocks.
* It fills externalInteractionsAllLevel and externalInteractionsLeafLevel.
......@@ -304,7 +310,7 @@ protected:
void upwardPass(){
FLOG( FTic timer; );
for(int idxLevel = tree->getHeight()-2 ; idxLevel >= 2 ; --idxLevel){
for(int idxLevel = FMath::Min(tree->getHeight() - 2, FAbstractAlgorithm::lowerWorkingLevel - 1) ; idxLevel >= FAbstractAlgorithm::upperWorkingLevel ; --idxLevel){
typename OctreeClass::CellGroupIterator iterCells = tree->cellsBegin(idxLevel);
const typename OctreeClass::CellGroupIterator endCells = tree->cellsEnd(idxLevel);
......@@ -388,7 +394,7 @@ protected:
void transferPass(){
FLOG( FTic timer; );
FLOG( FTic timerInBlock; FTic timerOutBlock; );
for(int idxLevel = tree->getHeight()-1 ; idxLevel >= 2 ; --idxLevel){
for(int idxLevel = FAbstractAlgorithm::lowerWorkingLevel-1 ; idxLevel >= FAbstractAlgorithm::upperWorkingLevel ; --idxLevel){
FLOG( timerInBlock.tic() );
{
typename OctreeClass::CellGroupIterator iterCells = tree->cellsBegin(idxLevel);
......@@ -502,7 +508,7 @@ protected:
void downardPass(){
FLOG( FTic timer; );
for(int idxLevel = 2 ; idxLevel <= tree->getHeight()-2 ; ++idxLevel){
for(int idxLevel = FAbstractAlgorithm::upperWorkingLevel ; idxLevel < FAbstractAlgorithm::lowerWorkingLevel - 1 ; ++idxLevel){
typename OctreeClass::CellGroupIterator iterCells = tree->cellsBegin(idxLevel);
const typename OctreeClass::CellGroupIterator endCells = tree->cellsEnd(idxLevel);
......
Src/GroupTree/Core/FGroupTaskStarpuAlgorithm.hpp
View file @ 48d763c1
......@@ -48,7 +48,7 @@ template <class OctreeClass, class CellContainerClass, class KernelClass, class
, class StarPUOpenClWrapperClass = FStarPUOpenClWrapper<KernelClass, FOpenCLDeviceWrapper<KernelClass>>
#endif
>
class FGroupTaskStarPUAlgorithm {
class FGroupTaskStarPUAlgorithm : public FAbstractAlgorithm {
protected:
typedef FGroupTaskStarPUAlgorithm<OctreeClass, CellContainerClass, KernelClass, ParticleGroupClass, StarPUCpuWrapperClass
#ifdef SCALFMM_ENABLE_CUDA_KERNEL
......@@ -134,6 +134,8 @@ public:
FAssertLF(tree, "tree cannot be null");
FAssertLF(inKernels, "kernels cannot be null");
FAbstractAlgorithm::setNbLevelsInTree(tree->getHeight());
struct starpu_conf conf;
FAssertLF(starpu_conf_init(&conf) == 0);
FStarPUFmmPriorities::Controller().init(&conf, tree->getHeight(), inKernels);
......@@ -235,7 +237,11 @@ public:
starpu_shutdown();
}
void execute(const unsigned operationsToProceed = FFmmNearAndFarFields){
protected:
/**
* Runs the complete algorithm.
*/
void executeCore(const unsigned operationsToProceed) override {
FLOG( FLog::Controller << "\tStart FGroupTaskStarPUAlgorithm\n" );
const bool directOnly = (tree->getHeight() <= 2);
......@@ -263,7 +269,7 @@ public:
starpu_pause();
}
protected:
void initCodelet(){
memset(&p2m_cl, 0, sizeof(p2m_cl));
#ifdef STARPU_USE_CPU
......@@ -738,7 +744,7 @@ protected:
void upwardPass(){
FLOG( FTic timer; );
for(int idxLevel = tree->getHeight()-2 ; idxLevel >= 2 ; --idxLevel){
for(int idxLevel = FMath::Min(tree->getHeight() - 2, FAbstractAlgorithm::lowerWorkingLevel - 1) ; idxLevel >= FAbstractAlgorithm::upperWorkingLevel ; --idxLevel){
int idxSubGroup = 0;
for(int idxGroup = 0 ; idxGroup < tree->getNbCellGroupAtLevel(idxLevel) ; ++idxGroup){
......@@ -799,7 +805,7 @@ protected:
void transferPass(){
FLOG( FTic timer; );
FLOG( FTic timerInBlock; FTic timerOutBlock; );
for(int idxLevel = tree->getHeight()-1 ; idxLevel >= 2 ; --idxLevel){
for(int idxLevel = FAbstractAlgorithm::lowerWorkingLevel-1 ; idxLevel >= FAbstractAlgorithm::upperWorkingLevel ; --idxLevel){
FLOG( timerInBlock.tic() );
for(int idxGroup = 0 ; idxGroup < tree->getNbCellGroupAtLevel(idxLevel) ; ++idxGroup){
starpu_insert_task(&m2l_cl_in,
......@@ -848,7 +854,7 @@ protected:
void downardPass(){
FLOG( FTic timer; );
for(int idxLevel = 2 ; idxLevel <= tree->getHeight()-2 ; ++idxLevel){
for(int idxLevel = FAbstractAlgorithm::upperWorkingLevel ; idxLevel < FAbstractAlgorithm::lowerWorkingLevel - 1 ; ++idxLevel){
int idxSubGroup = 0;
for(int idxGroup = 0 ; idxGroup < tree->getNbCellGroupAtLevel(idxLevel) ; ++idxGroup){
......
Src/GroupTree/Core/FGroupTaskStarpuMpiAlgorithm.hpp
View file @ 48d763c1
......@@ -54,7 +54,7 @@ template <class OctreeClass, class CellContainerClass, class KernelClass, class
, class StarPUOpenClWrapperClass = FStarPUOpenClWrapper<KernelClass, FOpenCLDeviceWrapper<KernelClass>>
#endif
>
class FGroupTaskStarPUMpiAlgorithm {
class FGroupTaskStarPUMpiAlgorithm : public FAbstractAlgorithm {
protected:
typedef FGroupTaskStarPUMpiAlgorithm<OctreeClass, CellContainerClass, KernelClass, ParticleGroupClass, StarPUCpuWrapperClass
#ifdef SCALFMM_ENABLE_CUDA_KERNEL
......@@ -151,6 +151,8 @@ public:
FAssertLF(tree, "tree cannot be null");
FAssertLF(inKernels, "kernels cannot be null");
FAbstractAlgorithm::setNbLevelsInTree(tree->getHeight());
struct starpu_conf conf;
FAssertLF(starpu_conf_init(&conf) == 0);
FStarPUFmmPriorities::Controller().init(&conf, tree->getHeight(), inKernels);
......@@ -254,7 +256,11 @@ public:
starpu_shutdown();
}
void execute(const unsigned operationsToProceed = FFmmNearAndFarFields){
protected:
/**
* Runs the complete algorithm.
*/
void executeCore(const unsigned operationsToProceed) override {
FLOG( FLog::Controller << "\tStart FGroupTaskStarPUMpiAlgorithm\n" );
const bool directOnly = (tree->getHeight() <= 2);
......@@ -291,7 +297,7 @@ public:
starpu_pause();
}
protected:
void initCodelet(){
memset(&p2m_cl, 0, sizeof(p2m_cl));
#ifdef STARPU_USE_CPU
......@@ -921,7 +927,7 @@ protected:
FAssertLF(tree->getHeight() == int(remoteCellGroups.size()));
const bool directOnly = (tree->getHeight() <= 2);
if(!directOnly){
for(int idxLevel = 0 ; idxLevel < tree->getHeight() ; ++idxLevel){
for(int idxLevel = FAbstractAlgorithm::upperWorkingLevel ; idxLevel < FAbstractAlgorithm::lowerWorkingLevel ; ++idxLevel){
for(int idxHandle = 0 ; idxHandle < int(remoteCellGroups[idxLevel].size()) ; ++idxHandle){
if(remoteCellGroups[idxLevel][idxHandle].ptrSymb){
FAssertLF(remoteCellGroups[idxLevel][idxHandle].ptrUp);
......@@ -1328,7 +1334,7 @@ protected:
void upwardPass(){
FLOG( FTic timer; );
for(int idxLevel = tree->getHeight()-2 ; idxLevel >= 2 ; --idxLevel){
for(int idxLevel = FMath::Min(tree->getHeight() - 2, FAbstractAlgorithm::lowerWorkingLevel - 1) ; idxLevel >= FAbstractAlgorithm::upperWorkingLevel ; --idxLevel){
int idxSubGroup = 0;
for(int idxGroup = 0 ; idxGroup < tree->getNbCellGroupAtLevel(idxLevel)
......@@ -1507,7 +1513,7 @@ protected:
void transferPassMpi(){
FLOG( FTic timer; );
for(int idxLevel = tree->getHeight()-1 ; idxLevel >= 2 ; --idxLevel){
for(int idxLevel = FAbstractAlgorithm::lowerWorkingLevel-1 ; idxLevel >= FAbstractAlgorithm::upperWorkingLevel ; --idxLevel){
for(int idxGroup = 0 ; idxGroup < tree->getNbCellGroupAtLevel(idxLevel) ; ++idxGroup){
for(int idxInteraction = 0; idxInteraction < int(externalInteractionsAllLevelMpi[idxLevel][idxGroup].size()) ; ++idxInteraction){
const int interactionid = externalInteractionsAllLevelMpi[idxLevel][idxGroup][idxInteraction].otherBlockId;
......@@ -1537,7 +1543,7 @@ protected:
void transferPass(){
FLOG( FTic timer; );
FLOG( FTic timerInBlock; FTic timerOutBlock; );
for(int idxLevel = tree->getHeight()-1 ; idxLevel >= 2 ; --idxLevel){
for(int idxLevel = FAbstractAlgorithm::lowerWorkingLevel-1 ; idxLevel >= FAbstractAlgorithm::upperWorkingLevel ; --idxLevel){
FLOG( timerInBlock.tic() );
for(int idxGroup = 0 ; idxGroup < tree->getNbCellGroupAtLevel(idxLevel) ; ++idxGroup){
starpu_insert_task(&m2l_cl_in,
......@@ -1586,7 +1592,7 @@ protected:
void downardPass(){
FLOG( FTic timer; );
for(int idxLevel = 2 ; idxLevel <= tree->getHeight()-2 ; ++idxLevel){
for(int idxLevel = FAbstractAlgorithm::upperWorkingLevel ; idxLevel < FAbstractAlgorithm::lowerWorkingLevel - 1 ; ++idxLevel){
/////////////////////////////////////////////////////////////
// Exchange for MPI
/////////////////////////////////////////////////////////////
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment