Skip to content

GitLab

Nous avons procédé ce jeudi matin 08 avril 2021 à une MAJ de sécurité urgente. Nous sommes passé de la version 13.9.3 à la version 13.9.5 les releases notes correspondantes sont ici:
https://about.gitlab.com/releases/2021/03/17/security-release-gitlab-13-9-4-released/
https://about.gitlab.com/releases/2021/03/31/security-release-gitlab-13-10-1-released/

Commit 6514e229 authored by Berenger Bramas's avatar Berenger Bramas
Browse files
Options

update mpi tree builder test to make it simple

parent 14ea2817
Hide whitespace changes
Inline Side-by-side
Showing with 85 additions and 204 deletions
UTests/utestMpiTreeBuilder.cpp
View file @ 6514e229
......@@ -53,28 +53,17 @@ class TestMpiTreeBuilder : public FUTesterMpi< class TestMpiTreeBuilder> {
template <class FReal>
struct TestParticle{
MortonIndex index;
MortonIndex mindex;
FSize indexInFile;
FPoint<FReal> position;
FReal physicalValue;
const FPoint<FReal>& getPosition()const{
return position;
}
TestParticle& operator=(const TestParticle& other){
index=other.index;
indexInFile=other.indexInFile;
position=other.position;
physicalValue=other.physicalValue;
return *this;
}
bool operator<(const TestParticle& rhs)const{
if(rhs.index < this->index){return false;}
else{
if(rhs.index > this->index){return true;}
else{
return this->indexInFile < rhs.indexInFile;
}
}
return rhs.mindex < this->mindex || (rhs.mindex == this->mindex && this->indexInFile < rhs.indexInFile);
}
};
......@@ -95,232 +84,124 @@ class TestMpiTreeBuilder : public FUTesterMpi< class TestMpiTreeBuilder> {
std::string filename(SCALFMMDataPath+parFile);
//std::string filename("../Data/unitCubeXYZQ100.bfma");
int TreeHeight =3;
int TreeHeight = 3;
//First part is Sequential
//Since there is no MPI Loader with ascii datas, the file need to be in binary format.
FFmaGenericLoader<FReal> loaderSeq(filename,true);
if(!loaderSeq.isOpen()) throw std::runtime_error("Particle file couldn't be opened!") ;
FMpiFmaGenericLoader<FReal> loader(filename,app.global());
if(!loader.isOpen()) throw std::runtime_error("Particle file couldn't be opened!") ;
//Get the needed informations
FReal boxWidth = loaderSeq.getBoxWidth();
FReal boxWidthAtLeafLevel = boxWidth/FReal(1 << (TreeHeight - 1));
const FReal boxWidth = loader.getBoxWidth();
const FReal boxWidthAtLeafLevel = boxWidth/FReal(1 << (TreeHeight - 1));
FPoint<FReal> centerOfBox = loaderSeq.getCenterOfBox();
FPoint<FReal> boxCorner = centerOfBox - boxWidth/2;
FTreeCoordinate host;
//Copy from the file to the array that will be sorted
FSize nbOfParticles = loaderSeq.getNumberOfParticles();
printf("nbOfParticles : %lld \n",nbOfParticles);
struct TestParticle<FReal>* arrayOfParticles = new TestParticle<FReal>[nbOfParticles];
memset(arrayOfParticles,0,sizeof(struct TestParticle<FReal>)*nbOfParticles);
for(FSize idxParts=0 ; idxParts<nbOfParticles ; ++idxParts){
const FPoint<FReal> centerOfBox = loader.getCenterOfBox();
const FPoint<FReal> boxCorner = centerOfBox - boxWidth/2;
//Now, we sort again the particles with MPI QuickSort
const FSize idxStart = loader.getStart();
FAssertLF(idxStart + loader.getMyNumberOfParticles() <= loader.getNumberOfParticles());
FMpiTreeBuilder<FReal,TestParticle<FReal>>::IndexedParticle * arrayToBeSorted = new FMpiTreeBuilder<FReal,TestParticle<FReal>>::IndexedParticle[loader.getMyNumberOfParticles()];
//Copy the TestParticles into an array of indexedParticle
for(FSize i=0 ; i<loader.getMyNumberOfParticles() ; ++i){
//Fill automatically position AND physicalValue attributes
loaderSeq.fillParticle(&(arrayOfParticles[idxParts].position),&(arrayOfParticles[idxParts].physicalValue));
loader.fillParticle(&(arrayToBeSorted[i].particle.position),&(arrayToBeSorted[i].particle.physicalValue));
//We store the index in the file
arrayOfParticles[idxParts].indexInFile = idxParts;
arrayToBeSorted[i].particle.indexInFile = i + idxStart;
//Build temporary TreeCoordinate
host.setX( FCoordinateComputer::GetTreeCoordinate<FReal>( arrayOfParticles[idxParts].getPosition().getX() - boxCorner.getX(), boxWidth, boxWidthAtLeafLevel, TreeHeight ));
host.setY( FCoordinateComputer::GetTreeCoordinate<FReal>( arrayOfParticles[idxParts].getPosition().getY() - boxCorner.getY(), boxWidth, boxWidthAtLeafLevel, TreeHeight ));
host.setZ( FCoordinateComputer::GetTreeCoordinate<FReal>( arrayOfParticles[idxParts].getPosition().getZ() - boxCorner.getZ(), boxWidth, boxWidthAtLeafLevel, TreeHeight ));
FTreeCoordinate host;
host.setX( FCoordinateComputer::GetTreeCoordinate<FReal>( arrayToBeSorted[i].particle.getPosition().getX() - boxCorner.getX(), boxWidth, boxWidthAtLeafLevel, TreeHeight ));
host.setY( FCoordinateComputer::GetTreeCoordinate<FReal>( arrayToBeSorted[i].particle.getPosition().getY() - boxCorner.getY(), boxWidth, boxWidthAtLeafLevel, TreeHeight ));
host.setZ( FCoordinateComputer::GetTreeCoordinate<FReal>( arrayToBeSorted[i].particle.getPosition().getZ() - boxCorner.getZ(), boxWidth, boxWidthAtLeafLevel, TreeHeight ));
//Set Morton index from Tree Coordinate
arrayOfParticles[idxParts].index = host.getMortonIndex();
}
//Save the original array
struct TestParticle<FReal> * originalArray = new TestParticle<FReal>[nbOfParticles];
memcpy(originalArray,arrayOfParticles,sizeof(struct TestParticle<FReal>)*nbOfParticles);
//Sort the array
std::sort(arrayOfParticles,arrayOfParticles+nbOfParticles);
// for(int k=0 ; k< nbOfParticles ; ++k){
// printf("arrayOfParticles[].index %lld \n",arrayOfParticles[k].index);
// }
//Start of the parallel part :
MortonIndex ref = -1;
int numMort = 0;
if(app.global().processId()==0){
for(FSize i=0 ; i<loaderSeq.getNumberOfParticles() ; ++i){
if (arrayOfParticles[i].index !=ref){
numMort++;
ref = arrayOfParticles[i].index;
}
}
printf("Total leaf : %d Last : %lld \n",numMort,ref);
}
FSize outputSize;
//Refer to ChebyshevInterpolationAlgorithmProc to know how to FMpiFmaLoader + index
FMpiFmaGenericLoader<FReal> loader(filename,app.global());
if(!loader.isOpen()) throw std::runtime_error("Particle file couldn't be opened!") ;
//Now, we sort again the particles with MPI QuickSort
FSize idxStart = loader.getStart();
FMpiTreeBuilder<FReal,TestParticle<FReal>>::IndexedParticle * arrayToBeSorted = new FMpiTreeBuilder<FReal,TestParticle<FReal>>::IndexedParticle[loader.getMyNumberOfParticles()];
//Copy the TestParticles into an array of indexedParticle
for(FSize i=0 ; i<loader.getMyNumberOfParticles() ; ++i){
arrayToBeSorted[i].particle = originalArray[i+idxStart];
arrayToBeSorted[i].index = arrayToBeSorted[i].particle.index;
arrayToBeSorted[i].particle.mindex = host.getMortonIndex();
arrayToBeSorted[i].index = arrayToBeSorted[i].particle.mindex;
}
FMpiTreeBuilder<FReal,TestParticle<FReal>>::IndexedParticle* outputArray = nullptr;
FSize outputSize;
FQuickSortMpi<FMpiTreeBuilder<FReal,TestParticle<FReal>>::IndexedParticle,MortonIndex,FSize>::QsMpi(arrayToBeSorted,loader.getMyNumberOfParticles(),&outputArray,&outputSize,app.global());
//FBitonicSort<FMpiTreeBuilder<FReal,TestParticle<FReal>>::IndexedParticle,MortonIndex, FSize>::Sort(arrayToBeSorted,loader.getMyNumberOfParticles(),app.global());
//Sum the outputSize of every body for knowing where to start inside the sorted array
FSize starter = 0;
//We use a prefix sum
MPI_Exscan(&outputSize,&starter,1,MPI_LONG_LONG_INT,MPI_SUM,app.global().getComm());
//We sort the output array relatvely to line number in origin file
FSize inc = 0;
FMpiTreeBuilder<FReal,TestParticle<FReal>>::IndexedParticle * saveForSort = outputArray;
int nbOfPartsInLeaf = 0;
while(inc < outputSize){
while(outputArray[inc].index == saveForSort->index && inc < outputSize){
inc++;
nbOfPartsInLeaf++;
}
std::sort(saveForSort,saveForSort+nbOfPartsInLeaf,
[&](FMpiTreeBuilder<FReal,TestParticle<FReal>>::IndexedParticle a,FMpiTreeBuilder<FReal,TestParticle<FReal>>::IndexedParticle b) -> bool {
return (a.particle.indexInFile)<(b.particle.indexInFile);
});
nbOfPartsInLeaf = 0;
saveForSort = &outputArray[inc];
}
bool resultQsMpi = true; //passed..
//Test
//Gather size of output
int * nbPartPerProcess = new int[app.global().processCount()];
nbPartPerProcess[app.global().processId()] = int(outputSize);
MPI_Gather(&nbPartPerProcess[app.global().processId()],1,FMpi::GetType(*nbPartPerProcess),nbPartPerProcess,1,FMpi::GetType(*nbPartPerProcess),0,app.global().getComm());
FSize * toSend = new FSize[outputSize];
int * displ = nullptr;
int * recvParts = nullptr;
FSize * myPart = nullptr;
//Prepare the indexInFile to send
for(int idPart=0 ; idPart<outputSize ; ++idPart){
toSend[idPart] = FSize(outputArray[idPart].particle.indexInFile);
}
if(app.global().processId() == 0){
//There, we build the array of displacement
displ = new int[app.global().processCount()];
displ[0] = 0;
for(int idProc = 1 ; idProc < app.global().processCount() ; ++idProc){
displ[idProc] = int( nbPartPerProcess[idProc-1] + displ[idProc-1]);
}
//Buffer to recv into
recvParts = new int[loader.getNumberOfParticles()];
FAssertLF(outputSize < std::numeric_limits<int>::max());
MPI_Gatherv(toSend,int(outputSize),FMpi::GetType(*toSend),recvParts,nbPartPerProcess,displ,FMpi::GetType(*toSend),0,app.global().getComm());
//Buffer to put result into
myPart = new FSize[loader.getNumberOfParticles()];
memset(myPart,0,sizeof(FSize)*loader.getNumberOfParticles());
for(FSize idP = 0 ; idP < loader.getNumberOfParticles() ; ++idP){
myPart[recvParts[idP]] += 1;
}
//Check if everything is set to 1
for(FSize idP = 0 ; idP < loader.getNumberOfParticles() ; ++idP){
if(myPart[idP] != 1){
std::cout << "Part number "<< idP << " in file is lost or duplicated : "<< myPart[idP]<< std::endl;
resultQsMpi = false;
}
}
FSize allparts;
MPI_Reduce(&outputSize, &allparts, 1, FMpi::GetType(outputSize), MPI_SUM, 0, app.global().getComm());
FAssertLF(allparts == loader.getNumberOfParticles());
}
else{
FAssertLF(outputSize < std::numeric_limits<int>::max());
MPI_Gatherv(toSend,int(outputSize),FMpi::GetType(*toSend),recvParts,nbPartPerProcess,displ,FMpi::GetType(*toSend),0,app.global().getComm());
MPI_Reduce(&outputSize, nullptr, 1, FMpi::GetType(outputSize), MPI_SUM, 0, app.global().getComm());
}
int* allPartsCount = new int[loader.getNumberOfParticles()];
memset(allPartsCount, 0, sizeof(int)*loader.getNumberOfParticles());
Print("Test 1 : is QsMpi really sorting the array");
uassert(resultQsMpi);
if(app.global().processId() == 0){
delete [] myPart;
delete [] recvParts;
delete [] displ;
for(int idxPart = 0 ; idxPart < outputSize ; ++idxPart){
FAssertLF(allPartsCount[outputArray[idxPart].particle.indexInFile] == 0);
allPartsCount[outputArray[idxPart].particle.indexInFile] += 1;
}
delete [] toSend;
if(app.global().processId() == 0){
int* allPartsCountReduced = new int[loader.getNumberOfParticles()];
MPI_Reduce(allPartsCount, allPartsCountReduced, int(loader.getNumberOfParticles()), FMpi::GetType(*allPartsCount), MPI_SUM, 0, app.global().getComm());
//Test MergeLeaves
bool resultMergeLeaves= true;
for(int idxPart = 0 ; idxPart < loader.getNumberOfParticles() ; ++idxPart){
FAssertLF(allPartsCountReduced[idxPart] == 1);
}
//inputs needed
TestParticle<FReal> * leavesArray = nullptr;
FSize * leavesIndices = nullptr;
FSize leaveSize = 0;
delete[] allPartsCountReduced;
}
else{
MPI_Reduce(allPartsCount, nullptr, int(loader.getNumberOfParticles()), FMpi::GetType(*allPartsCount), MPI_SUM, 0, app.global().getComm());
}
FMpiTreeBuilder<FReal,TestParticle<FReal>>::MergeSplitedLeaves(app.global(),&outputArray,&outputSize,&leavesIndices,&leavesArray,&leaveSize);
//Compare again the results with the output of std::qsort
MortonIndex leftright[2] = {-1, -1};
//we need to know how many parts still remains
FSize CounterStart = 0;
//We use a prefix sum
MPI_Exscan(&outputSize,&CounterStart,1,FMpi::GetType(outputSize),MPI_SUM,app.global().getComm());
if(outputSize){
leftright[0] = outputArray[0].particle.mindex;
leftright[1] = outputArray[0].particle.mindex;
//Test if no problems
for(FSize k=0 ; k<outputSize ; ++k){
if(leavesArray[k].indexInFile != arrayOfParticles[k+CounterStart].indexInFile){
printf("MergeLeaves :: Proc %d, finalParticles : %lld,%lld, sortedArray %lld,%lld \n",
app.global().processId(),
leavesArray[k].index,leavesArray[k].indexInFile,
arrayOfParticles[k+CounterStart].index,arrayOfParticles[k+CounterStart].indexInFile);
resultMergeLeaves = false;
for(int idxPart = 1 ; idxPart < outputSize ; ++idxPart){
leftright[0] = std::min(leftright[0], outputArray[idxPart].particle.mindex);
leftright[1] = std::max(leftright[1], outputArray[idxPart].particle.mindex);
}
}
Print("Test 2 : Output of merging leaves is tested");
uassert(resultMergeLeaves);
//Test the Equalize and Fill tree
FLeafBalance balancer;
FVector<TestParticle<FReal>> finalParticles;
bool resultEqualize = true;
FMpiTreeBuilder<FReal,TestParticle<FReal>>::EqualizeAndFillContainer(app.global(),&finalParticles,leavesIndices,leavesArray,leaveSize,outputSize,&balancer);
//Ok now count the Particles at the end of the Equalize
FSize finalNbPart = finalParticles.getSize();
FSize finalStart = 0;
MPI_Exscan(&finalNbPart,&finalStart,1,FMpi::GetType(finalNbPart),MPI_SUM,app.global().getComm());
for (int k=0; k<finalNbPart ; k++){
if(finalParticles[k].indexInFile != arrayOfParticles[k+finalStart].indexInFile){
printf("Equalize :: Proc %d, k=[%d+%lld] finalParticles : %lld,%lld, sortedArray %lld,%lld \n",
app.global().processId(),k,finalStart,
finalParticles[k].index,finalParticles[k].indexInFile,
arrayOfParticles[k+finalStart].index,arrayOfParticles[k+finalStart].indexInFile);
resultEqualize = false;
if(app.global().processId() == 0){
MortonIndex* allintervals = new MortonIndex[app.global().processCount()*2];
MPI_Gather(leftright, 2, FMpi::GetType(*leftright),
allintervals, 2, FMpi::GetType(*allintervals), 0, app.global().getComm());
MortonIndex currentLimit = -1;
for(int idxProc = 0 ; idxProc < app.global().processCount() ; ++idxProc){
FAssertLF(allintervals[idxProc*2] != -1 || allintervals[idxProc*2+1] == -1);
if(allintervals[idxProc*2] != -1){
FAssertLF(allintervals[idxProc*2] <= allintervals[idxProc*2+1]);
if(idxProc && allintervals[idxProc*2-1] != -1){
FAssertLF(allintervals[idxProc*2-1] < allintervals[idxProc*2]);
}
if(idxProc != app.global().processCount()-1 && allintervals[idxProc*2+1] != -1){
FAssertLF(allintervals[idxProc*2] < allintervals[idxProc*2+1]);
}
FAssertLF(currentLimit < allintervals[idxProc*2]);
currentLimit = allintervals[idxProc*2+1];
}
}
}
Print("Test 3 : Output of Equalize Tree is tested");
MPI_Barrier(MPI_COMM_WORLD);
uassert(resultEqualize);
delete[] allintervals;
}
else{
MPI_Gather(leftright, 2, FMpi::GetType(*leftright),
nullptr, 2, FMpi::GetType(*leftright), 0, app.global().getComm());
}
delete [] originalArray;
delete [] arrayOfParticles;
delete [] arrayToBeSorted;
delete [] outputArray;
delete[] allPartsCount;
delete[] outputArray;
}
/** If memstas is running print the memory used */
void PostTest() {
if( FMemStats::controler.isUsed() ){
......
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