Skip to content

GitLab

S
ScalFMM
Commit 8067ae8b authored by COULAUD Olivier's avatar COULAUD Olivier
Browse files
Options

Remove some errors+warning with gcc 8

parent 98cc61e5
Hide whitespace changes
Inline Side-by-side
Showing with 271 additions and 224 deletions
Contribs/json.hpp
View file @ 8067ae8b
......@@ -1913,8 +1913,8 @@ class basic_json
@since version 1.0.0
*/
basic_json(const value_t value_type)
: m_type(value_type), m_value(value_type)
basic_json(const value_t value_type1)
: m_type(value_type1), m_value(value_type1)
{
assert_invariant();
}
......
README.md
View file @ 8067ae8b
......@@ -28,15 +28,14 @@ The following are optional:
- Custom BLAS, FFT implementations.
- [StarPU](http://starpu.gforge.inria.fr/) for the relevant FMM implementations.
### To get and Build ScalFMM
To obtain ScalFMM (develop branch) and its git submodules do
### Get and Build ScalFMM
To use last development states of ScalFMM, please clone the develop
branch. Note that ScalFMM contains a git submodule `morse_cmake`.
To get sources please use these commands:
``` bash
git clone --recursive git@gitlab.inria.fr:solverstack/ScalFMM.git -b develop
```
or
```bash
git clone git@gitlab.inria.fr:solverstack/ScalFMM.git
cd ScalFMM
......@@ -55,9 +54,9 @@ The build may be configured after the first CMake invocation using, for instance
```bash
# Still in the Build folder
ccmake .
ccmake ../
# Or
cmake-gui .
cmake-gui ../
```
The binaries are then compiled calling `make`. They can be found in `scalfmm/Build/Tests/{Release,Debug}/...`
......
Src/Containers/FBufferReader.hpp
View file @ 8067ae8b
......@@ -3,6 +3,7 @@
#define FBUFFERREADER_HPP
#include <memory>
#include <algorithm>
#include "FAbstractBuffer.hpp"
#include "FBufferWriter.hpp"
#include "Utils/FAssert.hpp"
......@@ -213,6 +214,7 @@ public :
template <class T>
void fillArray(T* const inArray, const FSize count){
FAssertLF(currentIndex + FSize(sizeof(T))*count <= arrayCapacity );
// std::copy(&(array[currentIndex]),&(array[currentIndex])+count, inArray);
memcpy(inArray, &array[currentIndex], sizeof(T)*count);
currentIndex += sizeof(T)*count;
}
......
Src/Core/FFmmAlgorithmThreadProc.hpp
View file @ 8067ae8b
......@@ -4,36 +4,39 @@
#define SCALFMM_DISTRIBUTED_ALGORITHM
#include <omp.h>
#include <algorithm>
#include <vector>
#include <memory>
//#include <sys/time.h>
//
#ifdef _OPENMP
#include <omp.h>
#endif
//
#include "Utils/FGlobal.hpp"
#include "Utils/FAssert.hpp"
#include "Utils/FLog.hpp"
#include "Utils/FTic.hpp"
#include "Utils/FAlgorithmTimers.hpp"
#include "Utils/FGlobal.hpp"
#include "../Containers/FBoolArray.hpp"
#include "../Containers/FOctree.hpp"
#include "../Containers/FLightOctree.hpp"
#include "Utils/FEnv.hpp"
#include "Utils/FMpi.hpp"
#include "../Containers/FBufferWriter.hpp"
#include "../Containers/FBufferReader.hpp"
#include "../Containers/FVector.hpp"
#include "Containers/FVector.hpp"
#include "Containers/FBoolArray.hpp"
#include "Containers/FOctree.hpp"
#include "Containers/FLightOctree.hpp"
#include "Utils/FMpi.hpp"
#include <sys/time.h>
#include "Containers/FBufferWriter.hpp"
#include "Containers/FBufferReader.hpp"
#include "FCoreCommon.hpp"
#include "FP2PExclusion.hpp"
#include <memory>
#include <vector>
#include "Utils/FAlgorithmTimers.hpp"
/**
* @author Berenger Bramas (berenger.bramas@inria.fr)
......@@ -66,21 +69,21 @@ private:
OctreeClass* const tree; ///< The octree to work on
KernelClass** kernels; ///< The kernels
const FMpi::FComm comm; ///< MPI comm
FMpi::FComm fcomCompute;
const int OctreeHeight; ///< Tree height
/// Used to store pointers to cells/leafs to work with
typename OctreeClass::Iterator* iterArray;
/// Used to store pointers to cells/leafs to send/rcv
typename OctreeClass::Iterator* iterArrayComm;
typename OctreeClass::Iterator* iterArray; ///< Used to store pointers to cells/leafs to work with
typename OctreeClass::Iterator* iterArrayComm; ///< Used to store pointers to cells/leafs to send/rcv
const FMpi::FComm comm; ///< MPI communicator
FMpi::FComm fcomCompute;
int numberOfLeafs; ///< To store the size at the previous level
const int MaxThreads; ///< Max number of thread allowed by openmp
const int nbProcessOrig; ///< Process count
const int idProcessOrig; ///< Current process id
int nbProcess; ///< Process count
int idProcess; ///< Current process id
const int OctreeHeight; ///< Tree height
int nbProcess; ///< Process count
int idProcess; ///< Current process id
const int nbProcessOrig; ///< Process count
const int idProcessOrig; ///< Current process id
const int userChunkSize;
const int leafLevelSeparationCriteria;
......@@ -106,8 +109,6 @@ private:
const Interval& getWorkingInterval( int level, int proc) const {
return workingIntervalsPerLevel[OctreeHeight * proc + level];
}
/// Does \a procIdx have work at given \a idxLevel
/** i.e. does it hold cells and is responsible of them ? */
bool procHasWorkAtLevel(const int idxLevel , const int idxProc) const {
......@@ -125,21 +126,23 @@ private:
}
public:
/// Get an interval from a process id and tree level
// Interval& getWorkingInterval( int level, int proc){
// return workingIntervalsPerLevel[OctreeHeight * proc + level];
// }
// /// Get an interval from a process id and tree level
// const Interval& getWorkingInterval( int level, int proc) const {
// return workingIntervalsPerLevel[OctreeHeight * proc + level];
// }
/// Get current process interval at given \a level
///
/// \brief getWorkingInterval
/// \param level level in th tree
/// \return the interval from current process at tree level
///
Interval& getWorkingInterval( int level){
return getWorkingInterval(level, idProcess);
}
/// Build and dill vector of the MortonIndex Distribution at Leaf level
/// p = mpi process id then
/// [mortonLeafDistribution[2*p], mortonLeafDistribution[2*p+1] is the morton index shared by process p
/// Get the Morton index Distribution at the leaf level
///
/// Fill the vector mortonDistribution
///
/// p = mpi process id then
/// Processor p owns indexes between [mortonLeafDistribution[2*p], mortonLeafDistribution[2*p]+1]
///
/// parameter[out] mortonLeafDistribution
///
void getMortonLeafDistribution(std::vector<MortonIndex> & mortonLeafDistribution) final {
mortonLeafDistribution.resize(2*nbProcess) ;
auto level = OctreeHeight - 1;
......@@ -148,14 +151,48 @@ public:
mortonLeafDistribution[2*p] = inter.leftIndex;
mortonLeafDistribution[2*p+1] = inter.rightIndex;
}
}
// /// Build and dill vector of the MortonIndex Distribution at Leaf level
// /// p = mpi process id then
// /// [mortonLeafDistribution[2*p], mortonLeafDistribution[2*p+1] is the morton index shared by process p
// void getMortonLeafDistribution(std::vector<MortonIndex> & mortonLeafDistribution) final {
// mortonLeafDistribution.resize(2*nbProcess) ;
// auto level = OctreeHeight - 1;
// for (int p=0 ; p< nbProcess ; ++p ){
// auto inter = this->getWorkingInterval(level, p );
// mortonLeafDistribution[2*p] = inter.leftIndex;
// mortonLeafDistribution[2*p+1] = inter.rightIndex;
// }
// }
///
/// \brief setKernel
/// \param inKernels a pointer on the computational kernel used in the algorithm
///
/// @todo move it in private section
void setKernel(KernelClass*const inKernels){
this->kernels = new KernelClass*[MaxThreads]{};
#pragma omp parallel num_threads(MaxThreads)
{
#pragma omp critical (InitFFmmAlgorithmThreadProcPeriodic)
{
this->kernels[omp_get_thread_num()] = new KernelClass(*inKernels);
}
}
}
///
/// \brief getPtrOnMortonIndexAtLeaf
/// \return
///
const MortonIndex * getPtrOnMortonIndexAtLeaf() {
return &workingIntervalsPerLevel[0].leftIndex ;
}
/// Does the current process has some work at this level ?
bool hasWorkAtLevel( int level){
///
/// \brief hasWorkAtLevel - Does the current process have some work at this level ?
/// \param level
/// \return true if the current process have some work at this level
bool hasWorkAtLevel( int level){
return idProcess == 0 || (getWorkingInterval(level, idProcess - 1).rightIndex) < (getWorkingInterval(level, idProcess).rightIndex);
}
......@@ -171,17 +208,17 @@ public:
const int inLeafLevelSeperationCriteria = 1) :
tree(inTree),
kernels(nullptr),
comm(inComm),
fcomCompute(inComm),
OctreeHeight(tree->getHeight()),
iterArray(nullptr),
iterArrayComm(nullptr),
comm(inComm),
fcomCompute(inComm),
numberOfLeafs(0),
MaxThreads(FEnv::GetValue("SCALFMM_ALGO_NUM_THREADS",omp_get_max_threads())),
nbProcessOrig(inComm.processCount()),
idProcessOrig(inComm.processId()),
nbProcess(0),
idProcess(0),
OctreeHeight(tree->getHeight()),
nbProcessOrig(inComm.processCount()),
idProcessOrig(inComm.processId()),
userChunkSize(inUserChunkSize),
leafLevelSeparationCriteria(inLeafLevelSeperationCriteria),
intervals(new Interval[inComm.processCount()]),
......@@ -189,6 +226,7 @@ public:
FAssertLF(tree, "tree cannot be null");
FAssertLF(leafLevelSeparationCriteria < 3, "Separation criteria should be < 3");
// this->setKernel(inKernels) ;
this->kernels = new KernelClass*[MaxThreads];
#pragma omp parallel num_threads(MaxThreads)
{
......@@ -1703,7 +1741,7 @@ protected:
}
}
// Wait the come to finish (and the previous computation also)
// Wait the communications to finish (and the previous computation also)
#pragma omp barrier
......
Src/Core/FFmmAlgorithmThreadProcPeriodic.hpp
View file @ 8067ae8b
......@@ -8,14 +8,19 @@
#include <array>
#include <vector>
#include <memory>
//
#ifdef _OPENMP
#include <omp.h>
#endif
//
#include "Utils/FGlobal.hpp"
#include "Utils/FAssert.hpp"
#include "Utils/FLog.hpp"
#include "Utils/FTic.hpp"
#include "Utils/FGlobal.hpp"
#include "Utils/FMemUtils.hpp"
#include "Utils/FEnv.hpp"
#include "Utils/FMpi.hpp"
#include "Containers/FVector.hpp"
#include "Containers/FBoolArray.hpp"
......@@ -25,13 +30,6 @@
#include "Containers/FBufferWriter.hpp"
#include "Containers/FBufferReader.hpp"
#include "Utils/FEnv.hpp"
#include "Utils/FMpi.hpp"
#ifdef _OPENMP
#include <omp.h>
#endif
#include "FCoreCommon.hpp"
#include "FP2PExclusion.hpp"
......@@ -61,9 +59,9 @@
template<class FReal, class OctreeClass, class CellClass, class ContainerClass, class KernelClass, class LeafClass, class P2PExclusionClass = FP2PMiddleExclusion>
class FFmmAlgorithmThreadProcPeriodic : public FAbstractAlgorithm, public FAlgorithmTimers {
using multipole_t = typename CellClass::multipole_t;
using multipole_t = typename CellClass::multipole_t;
using local_expansion_t = typename CellClass::local_expansion_t;
using symbolic_data_t = CellClass;
using symbolic_data_t = CellClass;
OctreeClass* const tree; ///< The octree to work on
......@@ -97,33 +95,10 @@ class FFmmAlgorithmThreadProcPeriodic : public FAbstractAlgorithm, public FAlgor
/// All processes intervals
Interval*const workingIntervalsPerLevel;
public:
/// Get the Morton index Distribution at the leaf level
///
/// Fill the vector mortonDistribution
///
/// p = mpi process id then
/// Processor p owns indexes between [mortonLeafDistribution[2*p], mortonLeafDistribution[2*p]+1]
///
/// parameter[out] mortonLeafDistribution
///
void getMortonLeafDistribution(std::vector<MortonIndex> & mortonLeafDistribution) final {
mortonLeafDistribution.resize(2*nbProcess) ;
auto level = OctreeHeight - 1;
for (int p=0 ; p< nbProcess ; ++p ){
auto inter = this->getWorkingInterval(level, p );
mortonLeafDistribution[2*p] = inter.leftIndex;
mortonLeafDistribution[2*p+1] = inter.rightIndex;
}
}
private:
/// Get an interval from a process id and tree level
Interval& getWorkingInterval( int level, int proc){
return workingIntervalsPerLevel[OctreeHeight * proc + level];
}
/// Get an interval from a process id and tree level
const Interval& getWorkingInterval( int level, int proc) const {
return workingIntervalsPerLevel[OctreeHeight * proc + level];
......@@ -145,9 +120,41 @@ private:
}
public:
///
/// \brief getWorkingInterval
/// \param level level in th tree
/// \return the interval from current process at tree level
///
Interval& getWorkingInterval( int level){
return getWorkingInterval(level, idProcess);
}
/// Get the Morton index Distribution at the leaf level
///
/// Fill the vector mortonDistribution
///
/// p = mpi process id then
/// Processor p owns indexes between [mortonLeafDistribution[2*p], mortonLeafDistribution[2*p]+1]
///
/// parameter[out] mortonLeafDistribution
///
void getMortonLeafDistribution(std::vector<MortonIndex> & mortonLeafDistribution) final {
mortonLeafDistribution.resize(2*nbProcess) ;
auto level = OctreeHeight - 1;
for (int p=0 ; p< nbProcess ; ++p ){
auto inter = this->getWorkingInterval(level, p );
mortonLeafDistribution[2*p] = inter.leftIndex;
mortonLeafDistribution[2*p+1] = inter.rightIndex;
}
}
///
/// \brief setKernel
/// \param inKernels a pointer on the computational kernel used in the algorithm
///
/// @todo move it in private section --> warning we need to extend the box
///
void setKernel(KernelClass*const inKernels){
this->kernels = new KernelClass*[MaxThreads];
this->kernels = new KernelClass*[MaxThreads]{};
#pragma omp parallel num_threads(MaxThreads)
{
#pragma omp critical (InitFFmmAlgorithmThreadProcPeriodic)
......@@ -156,20 +163,18 @@ public:
}
}
}
// /// Build and dill vector of the MortonIndex Distribution at Leaf level
// /// p = mpi process id then
// /// [mortonLeafDistribution[2*p], mortonLeafDistribution[2*p+1] is the morton index shared by process p
// virtual void getAndBuildMortonIndexAtLeaf(std::vector<MortonIndex> & mortonLeafDistribution) {
// mortonLeafDistribution.resize(2*nbProcess) ;
// } ;
Interval& getWorkingInterval(const int level){
return getWorkingInterval(level, idProcess);
///
/// \brief getPtrOnMortonIndexAtLeaf
/// \return
///
const MortonIndex * getPtrOnMortonIndexAtLeaf() {
return &workingIntervalsPerLevel[0].leftIndex ;
}
/// Does the current process have some work at this level ?
///
/// \brief hasWorkAtLevel - Does the current process have some work at this level ?
/// \param level
/// \return true if the current process have some work at this level
///
bool hasWorkAtLevel(int level){
return idProcess == 0 || (getWorkingInterval(level, idProcess - 1).rightIndex) < (getWorkingInterval(level, idProcess).rightIndex);
}
......@@ -222,120 +227,6 @@ public:
delete [] workingIntervalsPerLevel;
}
long long int theoricalRepetition() const {
if( nbLevelsAboveRoot == -1 ){
// we know it is 3 (-1;+1)
return 3;
}
// Else we find the repetition in one dir and double it
return 6 * (1<<(nbLevelsAboveRoot));
}
//
//
//
void repetitionsIntervals(FTreeCoordinate*const min, FTreeCoordinate*const max) const {
if( nbLevelsAboveRoot == -1 ){
// We know it is (-1;1)
min->setPosition(-1,-1,-1);
max->setPosition(1,1,1);
}
else{
const int halfRepeated = int(theoricalRepetition()/2);
min->setPosition(-halfRepeated,-halfRepeated,-halfRepeated);
// if we repeat the box 8 times, we go from [-4 to 3]
max->setPosition(halfRepeated-1,halfRepeated-1,halfRepeated-1);
}
}
FReal extendedBoxWidth() const {
if( nbLevelsAboveRoot == -1 ){
return tree->getBoxWidth()*2;
}
else{
return tree->getBoxWidth() * FReal(4<<(nbLevelsAboveRoot));
}
}
FReal extendedBoxWidthBoundary() const {
if( nbLevelsAboveRoot == -1 ){
return tree->getBoxWidth()*4;
}
else{
return tree->getBoxWidth() * FReal(8<<(nbLevelsAboveRoot));
}
}
/** This function has to be used to init the kernel with correct args
* It returns the box center seen from a kernel point of view from the periodicity the user ask for
* this is computed using the originalBoxWidth and originalBoxCenter given in parameter
*
* @param originalBoxCenter the real system center
* @param originalBoxWidth the real system size
*
* @return the center the kernel should use
*/
FPoint<FReal> extendedBoxCenter() const {
if( nbLevelsAboveRoot == -1 ){
const FReal originalBoxWidth = tree->getBoxWidth();
const FPoint<FReal> originalBoxCenter = tree->getBoxCenter();
const FReal originalBoxWidthDiv2 = originalBoxWidth/2.0;
return FPoint<FReal>( originalBoxCenter.getX() + originalBoxWidthDiv2,
originalBoxCenter.getY() + originalBoxWidthDiv2,
originalBoxCenter.getZ() + originalBoxWidthDiv2);
} else {
const FReal originalBoxWidth = tree->getBoxWidth();
const FReal originalBoxWidthDiv2 = originalBoxWidth/2.0;
const FPoint<FReal> originalBoxCenter = tree->getBoxCenter();
const FReal offset = extendedBoxWidth()/FReal(2.0);
return FPoint<FReal>( originalBoxCenter.getX() - originalBoxWidthDiv2 + offset,
originalBoxCenter.getY() - originalBoxWidthDiv2 + offset,
originalBoxCenter.getZ() - originalBoxWidthDiv2 + offset);
}
}
FPoint<FReal> extendedBoxCenterBoundary() const {
if( nbLevelsAboveRoot == -1 ){
const FReal originalBoxWidth = tree->getBoxWidth();
const FPoint<FReal> originalBoxCenter = tree->getBoxCenter();
const FReal originalBoxWidthDiv2 = originalBoxWidth/2.0;
return FPoint<FReal>( originalBoxCenter.getX() + originalBoxWidthDiv2,
originalBoxCenter.getY() + originalBoxWidthDiv2,
originalBoxCenter.getZ() + originalBoxWidthDiv2);
} else {
const FReal originalBoxWidth = tree->getBoxWidth();
const FReal originalBoxWidthDiv2 = originalBoxWidth/2.0;
const FPoint<FReal> originalBoxCenter = tree->getBoxCenter();
return FPoint<FReal>( originalBoxCenter.getX() + originalBoxWidthDiv2,
originalBoxCenter.getY() + originalBoxWidthDiv2,
originalBoxCenter.getZ() + originalBoxWidthDiv2);
}
}
/** This function has to be used to init the kernel with correct args
* it returns the tree height seen from a kernel point of view from the periodicity the user ask for
* this is computed using the originalTreeHeight given in parameter
*
* @param originalTreeHeight the real tree heigh
*
* @return the height the kernel should use
*/
int extendedTreeHeight() const {
// The real height
return OctreeHeight + offsetRealTree;
}
int extendedTreeHeightBoundary() const {
// The real height
return OctreeHeight + offsetRealTree + 1;
}
protected:
/**
* To execute the fmm algorithm
......@@ -2142,6 +2033,120 @@ protected:
}
public:
long long int theoricalRepetition() const {
if( nbLevelsAboveRoot == -1 ){
// we know it is 3 (-1;+1)
return 3;
}
// Else we find the repetition in one dir and double it
return 6 * (1<<(nbLevelsAboveRoot));
}
//
//
//
void repetitionsIntervals(FTreeCoordinate*const min, FTreeCoordinate*const max) const {
if( nbLevelsAboveRoot == -1 ){
// We know it is (-1;1)
min->setPosition(-1,-1,-1);
max->setPosition(1,1,1);
}
else{
const int halfRepeated = int(theoricalRepetition()/2);
min->setPosition(-halfRepeated,-halfRepeated,-halfRepeated);
// if we repeat the box 8 times, we go from [-4 to 3]
max->setPosition(halfRepeated-1,halfRepeated-1,halfRepeated-1);
}
}
FReal extendedBoxWidth() const {
if( nbLevelsAboveRoot == -1 ){
return tree->getBoxWidth()*2;
}
else{
return tree->getBoxWidth() * FReal(4<<(nbLevelsAboveRoot));
}
}
FReal extendedBoxWidthBoundary() const {
if( nbLevelsAboveRoot == -1 ){
return tree->getBoxWidth()*4;
}
else{
return tree->getBoxWidth() * FReal(8<<(nbLevelsAboveRoot));
}
}
/** This function has to be used to init the kernel with correct args
* It returns the box center seen from a kernel point of view from the periodicity the user ask for
* this is computed using the originalBoxWidth and originalBoxCenter given in parameter
*
* @param originalBoxCenter the real system center
* @param originalBoxWidth the real system size
*
* @return the center the kernel should use
*/
FPoint<FReal> extendedBoxCenter() const {
if( nbLevelsAboveRoot == -1 ){
const FReal originalBoxWidth = tree->getBoxWidth();
const FPoint<FReal> originalBoxCenter = tree->getBoxCenter();
const FReal originalBoxWidthDiv2 = originalBoxWidth/2.0;
return FPoint<FReal>( originalBoxCenter.getX() + originalBoxWidthDiv2,
originalBoxCenter.getY() + originalBoxWidthDiv2,
originalBoxCenter.getZ() + originalBoxWidthDiv2);
} else {
const FReal originalBoxWidth = tree->getBoxWidth();
const FReal originalBoxWidthDiv2 = originalBoxWidth/2.0;
const FPoint<FReal> originalBoxCenter = tree->getBoxCenter();
const FReal offset = extendedBoxWidth()/FReal(2.0);
return FPoint<FReal>( originalBoxCenter.getX() - originalBoxWidthDiv2 + offset,
originalBoxCenter.getY() - originalBoxWidthDiv2 + offset,
originalBoxCenter.getZ() - originalBoxWidthDiv2 + offset);
}
}
FPoint<FReal> extendedBoxCenterBoundary() const {
if( nbLevelsAboveRoot == -1 ){