Skip to content
GitLab
Commit 914b3dc1 authored by PIACIBELLO Cyrille's avatar PIACIBELLO Cyrille
Browse files

Api woks with threads now

parent 5c486c56
Hide whitespace changes
Inline Side-by-side
Addons/CKernelApi/Src/FScalfmmApiInit.cpp
View file @ 914b3dc1
......@@ -24,16 +24,16 @@ extern "C" scalfmm_handle scalfmm_init(/*int TreeHeight,double BoxWidth,double*
handle->engine = new FInterEngine<ChebCell,ChebKernel>(/*TreeHeight,BoxWidth,BoxCenter, */KernelType);
break;
// case 2:
// //TODO typedefs
// typedef FP2PParticleContainerIndexed<> ContainerClass;
// typedef FUnifCell<7> UnifCell;
// case 2:
// //TODO typedefs
// typedef FP2PParticleContainerIndexed<> ContainerClass;
// typedef FUnifCell<7> UnifCell;
// typedef FInterpMatrixKernelR MatrixKernelClass;
// typedef FUnifKernel<UnifCell,ContainerClass,MatrixKernelClass,7> UnifKernel;
// typedef FInterpMatrixKernelR MatrixKernelClass;
// typedef FUnifKernel<UnifCell,ContainerClass,MatrixKernelClass,7> UnifKernel;
// handle->engine = new FInterEngine<UnifCell,UnifKernel>(/*TreeHeight,BoxWidth,BoxCenter, */KernelType);
// break;
// handle->engine = new FInterEngine<UnifCell,UnifKernel>(/*TreeHeight,BoxWidth,BoxCenter, */KernelType);
// break;
default:
std::cout<< "Kernel type unsupported" << std::endl;
......@@ -67,14 +67,12 @@ extern "C" {
typedef struct FChebCell_struct{
//Store what's needed
FChebCell<7> * cell;
long long int index;
}ChebCellStruct;
//How to create/destroy cells
extern "C" ChebCellStruct * ChebCellStruct_create(long long int inIndex,int * position){
ChebCellStruct * newCell = new ChebCellStruct();
newCell->index = inIndex;
newCell->cell = new FChebCell<7>();
newCell->cell->setMortonIndex(inIndex);
newCell->cell->setCoordinate(position[0],position[1],position[2]);
......@@ -92,7 +90,7 @@ extern "C" void ChebCellStruct_free(ChebCellStruct * inCell){
typedef struct FChebKernel_struct{
//Be ready full duplication go there !!!
FChebSymKernel<FChebCell<7>,FP2PParticleContainerIndexed<>,FInterpMatrixKernelR,7> * kernel;
FChebSymKernel<FChebCell<7>,FP2PParticleContainerIndexed<>,FInterpMatrixKernelR,7> ** kernel;
FInterpMatrixKernelR * matrix;
} ChebKernelStruct;
......@@ -100,21 +98,30 @@ typedef struct FChebKernel_struct{
extern "C" ChebKernelStruct * ChebKernelStruct_create(int inTreeHeight,
double inBoxWidth,
double* inBoxCenter){
//typedef to lighten the writing
typedef FChebSymKernel<FChebCell<7>, FP2PParticleContainerIndexed<>, FInterpMatrixKernelR,7> KernelClass;
ChebKernelStruct * newKernel = new ChebKernelStruct();
newKernel->matrix= new FInterpMatrixKernelR();
newKernel->kernel =
new FChebSymKernel<FChebCell<7>,FP2PParticleContainerIndexed<>,
FInterpMatrixKernelR,7>(inTreeHeight,
inBoxWidth,
FPoint(inBoxCenter[0],inBoxCenter[1],inBoxCenter[2]),
newKernel->matrix);
int nb_threads = omp_get_max_threads();
newKernel->kernel = new KernelClass*[nb_threads];
newKernel->kernel[0]=new KernelClass(inTreeHeight,
inBoxWidth,
FPoint(inBoxCenter[0], inBoxCenter[1], inBoxCenter[2]),
newKernel->matrix);
for(int idThreads=1 ; idThreads<nb_threads ; ++idThreads){
newKernel->kernel[idThreads] = new KernelClass(*newKernel->kernel[0]);
}
return newKernel;
}
extern "C" void ChebKernelStruct_free(void *inKernel){
delete reinterpret_cast<ChebKernelStruct *>(inKernel)->matrix;
delete reinterpret_cast<ChebKernelStruct *>(inKernel)->kernel;
int nb_threads = omp_get_max_threads();
for(int idThreads=0 ; idThreads<nb_threads ; ++idThreads){
delete reinterpret_cast<ChebKernelStruct *>(inKernel)->kernel[idThreads];
}
delete [] reinterpret_cast<ChebKernelStruct *>(inKernel)->kernel;
delete reinterpret_cast<ChebKernelStruct *>(inKernel);
}
......@@ -136,11 +143,14 @@ extern "C" void ChebKernel_P2M(void * leafCell, int nbParticles, const int *part
ChebCellStruct * realCellStruct = reinterpret_cast<ChebCellStruct *>(leafCell);
FChebCell<7> * realCell = realCellStruct->cell;
//Identify thread number
int id_thread = omp_get_thread_num();
//get the real chebyshev struct
UserData * userDataKernel = reinterpret_cast<UserData *>(inKernel);
ChebKernelStruct * realKernel = userDataKernel->kernelStruct;
realKernel->kernel->P2M(realCell, tempContainer);
realKernel->kernel[id_thread]->P2M(realCell, tempContainer);
delete tempContainer;
}
......@@ -152,11 +162,14 @@ extern "C" void ChebKernel_M2M(int level, void* parentCell, int childPosition,
FChebCell<7>* parentChebCell = parentCellStruct->cell;
FChebCell<7>* childChebCell = childCellStruct->cell;
//Identify thread number
int id_thread = omp_get_thread_num();
//Get the kernel
ChebKernelStruct * inKernelStruct = reinterpret_cast<UserData*>(inKernel)->kernelStruct;
inKernelStruct->kernel->getPtrToInterpolator()->applyM2M(childPosition,
childChebCell->getMultipole(0),
parentChebCell->getMultipole(0));
inKernelStruct->kernel[id_thread]->getPtrToInterpolator()->applyM2M(childPosition,
childChebCell->getMultipole(0),
parentChebCell->getMultipole(0));
}
extern "C" void ChebKernel_M2L(int level, void* targetCell, void* sourceCell[343], void* inKernel){
......@@ -175,24 +188,31 @@ extern "C" void ChebKernel_M2L(int level, void* targetCell, void* sourceCell[34
arrayOfChebCell[i] = nullptr;
}
}
//Identify thread number
int id_thread = omp_get_thread_num();
//Get the kernel
ChebKernelStruct * inKernelStruct = reinterpret_cast<UserData*>(inKernel)->kernelStruct;
inKernelStruct->kernel->M2L(targetChebCell,arrayOfChebCell,0,level);
inKernelStruct->kernel[id_thread]->M2L(targetChebCell,arrayOfChebCell,0,level);
}
extern "C" void ChebKernel_L2L(int level, void* parentCell, int childPosition, void* childCell, void* inKernel){
//Get our structures
//Get our structures
ChebCellStruct * parentCellStruct = reinterpret_cast<ChebCellStruct *>(parentCell);
ChebCellStruct * childCellStruct = reinterpret_cast<ChebCellStruct *>(childCell);
//get real cheb cell
FChebCell<7>* parentChebCell = parentCellStruct->cell;
FChebCell<7>* childChebCell = childCellStruct->cell;
//Identify thread number
int id_thread = omp_get_thread_num();
//Get the kernel
ChebKernelStruct * inKernelStruct = reinterpret_cast<UserData*>(inKernel)->kernelStruct;
inKernelStruct->kernel->getPtrToInterpolator()->applyL2L(childPosition,
parentChebCell->getLocal(0),
childChebCell->getLocal(0));
inKernelStruct->kernel[id_thread]->getPtrToInterpolator()->applyL2L(childPosition,
parentChebCell->getLocal(0),
childChebCell->getLocal(0));
}
extern "C" void ChebKernel_L2P(void* leafCell, int nbParticles, const int* particleIndexes, void* inKernel){
......@@ -202,8 +222,8 @@ extern "C" void ChebKernel_L2P(void* leafCell, int nbParticles, const int* parti
FPoint pos;
for(int i=0 ; i<nbParticles ; ++i){
pos = FPoint(reinterpret_cast<UserData *>(inKernel)->insertedPositions[particleIndexes[i]*3 ],
reinterpret_cast<UserData *>(inKernel)->insertedPositions[particleIndexes[i]*3+1],
reinterpret_cast<UserData *>(inKernel)->insertedPositions[particleIndexes[i]*3+2]);
reinterpret_cast<UserData *>(inKernel)->insertedPositions[particleIndexes[i]*3+1],
reinterpret_cast<UserData *>(inKernel)->insertedPositions[particleIndexes[i]*3+2]);
double Phi = reinterpret_cast<UserData *>(inKernel)->myPhyValues[particleIndexes[i]];
tempContainer->push(pos,particleIndexes[i],Phi);
}
......@@ -212,13 +232,16 @@ extern "C" void ChebKernel_L2P(void* leafCell, int nbParticles, const int* parti
//get real cheb cell
FChebCell<7>* leafChebCell = leafCellStruct->cell;
//Identify thread number
int id_thread = omp_get_thread_num();
//Get the kernel
ChebKernelStruct * inKernelStruct = reinterpret_cast<UserData*>(inKernel)->kernelStruct;
inKernelStruct->kernel->L2P(leafChebCell,tempContainer);
inKernelStruct->kernel[id_thread]->L2P(leafChebCell,tempContainer);
//Then retrieve the results
double * forcesToFill = reinterpret_cast<UserData *>(inKernel)->forcesComputed;
double * forcesToFill = reinterpret_cast<UserData *>(inKernel)->forcesComputed[id_thread];
const FVector<int>& indexes = tempContainer->getIndexes();
for(int idxPart = 0 ; idxPart<nbParticles ; ++idxPart){
forcesToFill[indexes[idxPart]*3+0] += tempContainer->getForcesX()[idxPart];
......@@ -233,6 +256,7 @@ extern "C" void ChebKernel_L2P(void* leafCell, int nbParticles, const int* parti
void ChebKernel_P2P(int nbParticles, const int* particleIndexes,
const int * sourceParticleIndexes[27],int sourceNbPart[27],void* inKernel){
//Create temporary FSimpleLeaf for target
FP2PParticleContainerIndexed<>* tempContTarget = new FP2PParticleContainerIndexed<>();
tempContTarget->reserve(nbParticles);
......@@ -269,16 +293,19 @@ void ChebKernel_P2P(int nbParticles, const int* particleIndexes,
}
//Everything is fine, now, call Chebyshev P2P
//Identify thread number
int id_thread = omp_get_thread_num();
//Get the kernel
ChebKernelStruct * inKernelStruct = reinterpret_cast<UserData*>(inKernel)->kernelStruct;
//Empty tree coordinate
int coord[3] = {0,0,0};
inKernelStruct->kernel->P2P(FTreeCoordinate(coord),tempContTarget,nullptr,tempContSources,0);
inKernelStruct->kernel[id_thread]->P2P(FTreeCoordinate(coord),tempContTarget,nullptr,tempContSources,0);
//get back forces
double * forcesToFill = reinterpret_cast<UserData *>(inKernel)->forcesComputed;
double * forcesToFill = reinterpret_cast<UserData *>(inKernel)->forcesComputed[id_thread];
const FVector<int>& indexes = tempContTarget->getIndexes();
for(int idxPart = 0 ; idxPart<nbParticles ; ++idxPart){
forcesToFill[indexes[idxPart]*3+0] += tempContTarget->getForcesX()[idxPart];
......
Addons/CKernelApi/Tests/testChebInterface.c
View file @ 914b3dc1
......@@ -2,6 +2,7 @@
#include <stdlib.h>
#include <string.h>
//For timing monitoring
#include <time.h>
#include <sys/time.h>
......@@ -106,17 +107,22 @@ void print_difference_elapsed(Timer* inTimer1,Timer*inTimer2){
* @param number of particle (no default value)
*/
int main(int argc, char ** av){
omp_set_num_threads(1);
//omp_set_num_threads(1);
printf("Start\n");
if(argc<2){
printf("Use : %s nb_part\n exiting\n",av[0]);
printf("Use : %s nb_part (optionnal : TreeHeight) \nexiting\n",av[0]);
exit(0);
}
int nbPart= atoi(av[1]);
int treeHeight = 5 ;
if(argc>2){
int treeHeight = atoi(av[2]);
}
double* particleXYZ = malloc(sizeof(double)*3*nbPart);
double* physicalValues = malloc(sizeof(double)*nbPart);
int treeHeight = 5;
double boxWidth = 1.0;
double boxCenter[3];
boxCenter[0] = boxCenter[1] = boxCenter[2] = 0.0;
......@@ -193,12 +199,29 @@ int main(int argc, char ** av){
//Set my datas
UserData userDatas;
userDatas.kernelStruct = ChebKernelStruct_create(treeHeight,boxWidth,boxCenter); // Set my kernel inside userDatas
int nb_threads = omp_get_max_threads();
int idThreads= 0;
//Create as many kernels as there are threads in order to void concurrent write
userDatas.kernelStruct = ChebKernelStruct_create(treeHeight,boxWidth,boxCenter);
/* malloc(sizeof(void *)*nb_threads); */
/* for(idThreads=0 ; idThreads<nb_threads ; ++idThreads){ */
/* userDatas.kernelStruct[idThreads] = ChebKernelStruct_create(treeHeight,boxWidth,boxCenter); // Set my kernel inside userDatas */
/* } */
//Only read, so no split needed
userDatas.insertedPositions = particleXYZ; // Set the position
userDatas.myPhyValues = physicalValues; // Set the physical values
double * forcesToStore = malloc(sizeof(double)*nbPart*3); // Create array to store results
memset(forcesToStore,0,sizeof(double)*nbPart*3); // memset array
userDatas.forcesComputed = forcesToStore; // Set the forces array
//Create as many array of forces as there are threads in order to void concurrent write
double ** forcesToStore = malloc(sizeof(double*)*nb_threads);
//For each array, initialisation
for(idThreads=0 ; idThreads<nb_threads ; ++idThreads){
forcesToStore[idThreads] = malloc(sizeof(double)*nbPart*3); //allocate memory
memset(forcesToStore[idThreads],0,sizeof(double)*nbPart*3); //set to zero (IMPORTANT, as operators usually "+=" on forces)
}
userDatas.forcesComputed = forcesToStore;
//Give ScalFMM the datas before calling fmm (this will set as well the kernel)
scalfmm_user_kernel_config(handle,kernel,&userDatas);
......@@ -211,6 +234,17 @@ int main(int argc, char ** av){
scalfmm_execute_fmm(handle/*, kernel, &my_data*/);
tac(&interface_timer);
//Reduction on forces array
{
int idxPart;
for(idThreads=1 ; idThreads<nb_threads ; ++idThreads){
for(idxPart=0 ; idxPart<nbPart*3 ; ++idxPart){
//Everything is stored in first array
forcesToStore[0][idxPart] += forcesToStore[idThreads][idxPart];
}
}
}
printf("User defined Chebyshev done\n");
print_elapsed(&interface_timer);
......@@ -228,34 +262,49 @@ int main(int argc, char ** av){
double * forcesRef = malloc(sizeof(double)*3*nbPart);
memset(forcesRef,0,sizeof(double)*3*nbPart);
scalfmm_get_forces_xyz(handle_ref,nbPart,forcesRef);
{
{//Comparison part
int idxPart;
int nbPartOkay = 0;
for(idxPart=0 ; idxPart<nbPart ; ++idxPart ){
double diffX,diffY,diffZ;
diffX = userDatas.forcesComputed[idxPart*3+0]-forcesRef[idxPart*3+0];
diffY = userDatas.forcesComputed[idxPart*3+1]-forcesRef[idxPart*3+1];
diffZ = userDatas.forcesComputed[idxPart*3+2]-forcesRef[idxPart*3+2];
/* printf("id : %d : %e, %e, %e\n",idxPart,diffX,diffY,diffZ); */
if(diffX < 0.00000001 || diffY < 0.00000001 || diffZ < 0.00000001){
diffX = forcesToStore[0][idxPart*3+0]-forcesRef[idxPart*3+0];
diffY = forcesToStore[0][idxPart*3+1]-forcesRef[idxPart*3+1];
diffZ = forcesToStore[0][idxPart*3+2]-forcesRef[idxPart*3+2];
if(diffX < 0.00000001 && diffY < 0.00000001 && diffZ < 0.00000001){
nbPartOkay++;
}
else{
//printf("id : %d : %e, %e, %e\n",idxPart,diffX,diffY,diffZ);
}
//That part is to verify with our usual exec' if everything is alright
if(idxPart == 0 || idxPart == nbPart/2 || idxPart == nbPart-1){
printf("id : %d : %e, %e, %e\n",idxPart,userDatas.forcesComputed[idxPart*3+0],
userDatas.forcesComputed[idxPart*3+1],
userDatas.forcesComputed[idxPart*3+2]);
printf("User one's id : %d : %e, %e, %e\n",idxPart,
forcesToStore[0][idxPart*3+0],
forcesToStore[0][idxPart*3+1],
forcesToStore[0][idxPart*3+2]);
printf("Chebyshev one's id : %d : %e, %e, %e\n",idxPart,
forcesRef[idxPart*3+0],
forcesRef[idxPart*3+1],
forcesRef[idxPart*3+2]);
}
}
printf("End of simulation \n \t Percentage of good parts : %d/%d (%f %%) \n",
nbPartOkay,nbPart,(((double) nbPartOkay)/(double)nbPart)*100);
}
printf("Free the kernels\n");
printf("Free the Handle ...\n");
printf("Free the Handles ...\n");
scalfmm_dealloc_handle(handle,cheb_free_cell);
scalfmm_dealloc_handle(handle_ref,NULL);
free(particleXYZ);
free(physicalValues);
free(forcesRef);
//free the thread' specific datas
for(idThreads=0 ; idThreads<nb_threads ; ++idThreads){
free(forcesToStore[idThreads]);
}
ChebKernelStruct_free(userDatas.kernelStruct);
free(userDatas.forcesComputed);
}
Src/Kernels/Chebyshev/FChebInterface.h
View file @ 914b3dc1
......@@ -60,7 +60,7 @@ typedef struct myUserDatas{
ChebKernelStruct * kernelStruct;
double * insertedPositions;
double * myPhyValues;
double * forcesComputed;
double ** forcesComputed;
}UserData;
......
Supports Markdown
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