void DMCOMP::benchMark()
{
//set the collection mode for the estimator
Estimators->setCollectionMode(branchEngine->SwapMode);
IndexType PopIndex = Estimators->addColumn("Population");
IndexType EtrialIndex = Estimators->addColumn("Etrial");
//Estimators->reportHeader(AppendRun);
//Estimators->reset();
IndexType block = 0;
RealType Eest = branchEngine->E_T;
//resetRun();
for(int ip=0; ip<NumThreads; ip++)
{
char fname[16];
sprintf(fname,"test.%i",ip);
ofstream fout(fname);
}
for(int istep=0; istep<nSteps; istep++)
{
FairDivideLow(W.getActiveWalkers(),NumThreads,wPerNode);
#pragma omp parallel
{
int ip = omp_get_thread_num();
Movers[ip]->benchMark(W.begin()+wPerNode[ip],W.begin()+wPerNode[ip+1],ip);
}
}
}
bool RNDMCOMP::run()
{
resetUpdateEngines();
//estimator does not need to collect data
Estimators->setCollectionMode(true);
Estimators->start(nBlocks);
for(int ip=0; ip<NumThreads; ip++)
Movers[ip]->startRun(nBlocks,false);
Timer myclock;
IndexType block = 0;
IndexType updatePeriod=(QMCDriverMode[QMC_UPDATE_MODE])?Period4CheckProperties:(nBlocks+1)*nSteps;
do // block
{
Estimators->startBlock(nSteps);
for(int ip=0; ip<NumThreads; ip++)
Movers[ip]->startBlock(nSteps);
IndexType step = 0;
for(IndexType step=0; step< nSteps; ++step, CurrentStep+=BranchInterval)
{
#pragma omp parallel for
for(int ip=0; ip<NumThreads; ++ip)
{
int now=CurrentStep;
MCWalkerConfiguration::iterator
wit(W.begin()+wPerNode[ip]), wit_end(W.begin()+wPerNode[ip+1]);
for(int interval = 0; interval<BranchInterval-1; ++interval,++now)
Movers[ip]->advanceWalkers(wit,wit_end,false);
wClones[ip]->resetCollectables();
Movers[ip]->advanceWalkers(wit,wit_end,false);
Movers[ip]->setReleasedNodeMultiplicity(wit,wit_end);
if(QMCDriverMode[QMC_UPDATE_MODE] && now%updatePeriod == 0)
Movers[ip]->updateWalkers(wit, wit_end);
}//#pragma omp parallel
for(int ip=1; ip<NumThreads; ++ip)
{
for(int j=0; j<W.Collectables.size(); ++j)
W.Collectables[j]+=wClones[ip]->Collectables[j];
}
branchEngine->branch(CurrentStep,W, branchClones);
FairDivideLow(W.getActiveWalkers(),NumThreads,wPerNode);
}
// #pragma omp parallel for
// for(int ip=0; ip<NumThreads; ++ip)
// {
// MCWalkerConfiguration::iterator wit(W.begin()+wPerNode[ip]), wit_end(W.begin()+wPerNode[ip+1]);
// Movers[ip]->resetbuffers(wit, wit_end);
// }
Estimators->stopBlock(acceptRatio());
block++;
recordBlock(block);
}
while(block<nBlocks && myclock.elapsed()<MaxCPUSecs);
//for(int ip=0; ip<NumThreads; ip++) Movers[ip]->stopRun();
for(int ip=0; ip<NumThreads; ip++)
*(RandomNumberControl::Children[ip])=*(Rng[ip]);
Estimators->stop();
return finalize(block);
}
bool DMCOMP::run() {
bool variablePop = (Reconfiguration == "no");
resetUpdateEngines();
//set the collection mode for the estimator
Estimators->setCollectionMode(branchEngine->SwapMode);
Estimators->reportHeader(AppendRun);
Estimators->reset();
IndexType block = 0;
do // block
{
Estimators->startBlock();
for(int ip=0; ip<NumThreads; ip++) Movers[ip]->startBlock();
IndexType step = 0;
do //step
{
#pragma omp parallel
{
int ip = omp_get_thread_num();
IndexType interval = 0;
do // interval
{
Movers[ip]->advanceWalkers(W.begin()+wPerNode[ip],W.begin()+wPerNode[ip+1]);
++interval;
} while(interval<BranchInterval);
}//#pragma omp parallel
Movers[0]->setMultiplicity(W.begin(),W.end());
Estimators->accumulate(W);
branchEngine->branch(CurrentStep,W, branchClones);
if(variablePop) FairDivideLow(W.getActiveWalkers(),NumThreads,wPerNode);
++step;
CurrentStep+=BranchInterval;
} while(step<nSteps);
Estimators->stopBlock(acceptRatio());
block++;
recordBlock(block);
if(QMCDriverMode[QMC_UPDATE_MODE] && CurrentStep%100 == 0)
{
#pragma omp parallel
{
int ip = omp_get_thread_num();
Movers[ip]->updateWalkers(W.begin()+wPerNode[ip], W.begin()+wPerNode[ip+1]);
}
}
} while(block<nBlocks);
return finalize(block);
}
void DMCOMP::resetUpdateEngines()
{
bool fixW = (Reconfiguration == "yes");
makeClones(W,Psi,H);
if(Movers.empty())
{
branchEngine->initWalkerController(Tau,fixW);
//if(QMCDriverMode[QMC_UPDATE_MODE]) W.clearAuxDataSet();
Movers.resize(NumThreads,0);
branchClones.resize(NumThreads,0);
estimatorClones.resize(NumThreads,0);
Rng.resize(NumThreads,0);
FairDivideLow(W.getActiveWalkers(),NumThreads,wPerNode);
app_log() << " Initial partition of walkers ";
std::copy(wPerNode.begin(),wPerNode.end(),ostream_iterator<int>(app_log()," "));
app_log() << endl;
#pragma omp parallel
{
int ip = omp_get_thread_num();
if(ip)
hClones[ip]->add2WalkerProperty(*wClones[ip]);
estimatorClones[ip]= new ScalarEstimatorManager(*Estimators,*hClones[ip]);
Rng[ip]=new RandomGenerator_t();
Rng[ip]->init(ip,NumThreads,-1);
hClones[ip]->setRandomGenerator(Rng[ip]);
branchClones[ip] = new BranchEngineType(*branchEngine);
if(QMCDriverMode[QMC_UPDATE_MODE])
{
if(NonLocalMove == "yes")
{
DMCNonLocalUpdatePbyP* nlocMover=
new DMCNonLocalUpdatePbyP(*wClones[ip],*psiClones[ip],*hClones[ip],*Rng[ip]);
nlocMover->put(qmcNode);
Movers[ip]=nlocMover;
}
else
{
Movers[ip]= new DMCUpdatePbyPWithRejection(*wClones[ip],*psiClones[ip],*hClones[ip],*Rng[ip]);
}
Movers[ip]->resetRun(branchClones[ip],estimatorClones[ip]);
//Movers[ip]->initWalkersForPbyP(W.begin()+wPerNode[ip],W.begin()+wPerNode[ip+1]);
}
else
{
if(NonLocalMove == "yes")
{
app_log() << " Non-local update is used." << endl;
DMCNonLocalUpdate* nlocMover= new DMCNonLocalUpdate(W,Psi,H,Random);
nlocMover->put(qmcNode);
Movers[ip]=nlocMover;
}
else
{
if(KillNodeCrossing)
{
Movers[ip] = new DMCUpdateAllWithKill(W,Psi,H,Random);
}
else
{
Movers[ip] = new DMCUpdateAllWithRejection(W,Psi,H,Random);
}
}
Movers[ip]->resetRun(branchClones[ip],estimatorClones[ip]);
//Movers[ip]->initWalkers(W.begin()+wPerNode[ip],W.begin()+wPerNode[ip+1]);
}
}
}
if(fixW)
{
if(QMCDriverMode[QMC_UPDATE_MODE])
app_log() << " DMCOMP PbyP Update with reconfigurations" << endl;
else
app_log() << " DMCOMP walker Update with reconfigurations" << endl;
for(int ip=0; ip<Movers.size(); ip++)
Movers[ip]->MaxAge=0;
if(BranchInterval<0)
{
BranchInterval=nSteps;
}
}
else
{
if(QMCDriverMode[QMC_UPDATE_MODE])
{
app_log() << " DMCOMP PbyP Update with a fluctuating population" << endl;
for(int ip=0; ip<Movers.size(); ip++)
Movers[ip]->MaxAge=1;
}
else
{
app_log() << " DMCOMP walker Update with a fluctuating population" << endl;
for(int ip=0; ip<Movers.size(); ip++)
Movers[ip]->MaxAge=3;
}
if(BranchInterval<0)
BranchInterval=1;
}
app_log() << " BranchInterval = " << BranchInterval << endl;
app_log() << " Steps per block = " << nSteps << endl;
app_log() << " Number of blocks = " << nBlocks << endl;
}
bool DMCOMP::run()
{
bool variablePop = (Reconfiguration == "no");
resetUpdateEngines();
Estimators->start(nBlocks);
#pragma omp parallel
{
int ip = omp_get_thread_num();
int np = omp_get_num_threads();
vector<int> wpart(np+1);
FairDivideLow(W.getActiveWalkers(),np,wpart);
int firstW=wpart[ip];
int lastW=wpart[ip+1];
//char fname[16];
//sprintf(fname,"debug.%d",ip);
//ofstream fout(fname);
if(QMCDriverMode[QMC_UPDATE_MODE])
Movers[ip]->initWalkersForPbyP(W.begin()+firstW,W.begin()+lastW);
else
Movers[ip]->initWalkers(W.begin()+firstW,W.begin()+lastW);
Movers[ip]->startRun(nBlocks,false);
IndexType block = 0;
do // block
{
Movers[ip]->startBlock(nSteps);
IndexType step = 0;
do //step
{
IndexType interval = 0;
IndexType bi=BranchInterval;
do // interval
{
++interval;
Movers[ip]->advanceWalkers(W.begin()+firstW,W.begin()+lastW);
}
while(interval<bi);
Movers[ip]->setMultiplicity(W.begin()+firstW,W.begin()+lastW);
#pragma omp barrier
#pragma omp master
{
//branchEngine->branch(CurrentStep,W, branchClones);
branchEngine->branch(CurrentStep,W);
if(storeConfigs && (CurrentStep%storeConfigs == 0))
{
branchEngine->storeConfigsForForwardWalking(W);
W.resetWalkerParents();
}
}
#pragma omp barrier
branchClones[ip]->E_T=branchEngine->E_T;
if(variablePop)
{
FairDivideLow(W.getActiveWalkers(),np,wpart);
firstW=wpart[ip];
lastW=wpart[ip+1];
}
++step;
//fout << step << " " << firstW << " " << lastW << " " << W.getActiveWalkers() << " " << branchClones[ip]->E_T << endl;
CurrentStep+=BranchInterval;
}
while(step<nSteps);
block++;
#pragma omp master
{
Estimators->stopBlock(Movers[0]->acceptRatio());
recordBlock(block);
}
if(QMCDriverMode[QMC_UPDATE_MODE] && CurrentStep%100 == 0)
{
Movers[ip]->updateWalkers(W.begin()+firstW, W.begin()+lastW);
}
}
while(block<nBlocks);
Movers[ip]->stopRun();
}
Estimators->stop();
return finalize(nBlocks);
}
void RNDMCOMP::resetUpdateEngines()
{
ReportEngine PRE("RNDMCOMP","resetUpdateEngines");
Timer init_timer;
makeClones(W,Psi,H);
if(Movers.empty())
{
//load walkers
W.loadEnsemble();
for(int ip=1;ip<NumThreads;++ip) wClones[ip]->loadEnsemble(W);
if (useAlternate=="yes")
branchEngine->initWalkerController(W,false,false);
//branchEngine->initWalkerController(W,Tau,false,false);
else
branchEngine->initWalkerController(W,false,true);
//branchEngine->initWalkerController(W,Tau,false,true);
branchEngine->setRN(true);
//if(QMCDriverMode[QMC_UPDATE_MODE]) W.clearAuxDataSet();
Movers.resize(NumThreads,0);
branchClones.resize(NumThreads,0);
Rng.resize(NumThreads,0);
estimatorClones.resize(NumThreads,0);
FairDivideLow(W.getActiveWalkers(),NumThreads,wPerNode);
bool klw=(KillWalker=="yes");
{//log file
ostringstream o;
o << " Initial partition of walkers on a node: ";
std::copy(wPerNode.begin(),wPerNode.end(),ostream_iterator<int>(o," "));
o << "\n";
o << "Killing Walkers at nodes " << (useAlternate!="yes") <<endl;
o << "Running the released node driver."<<endl;
app_log() << o.str();
}
#pragma omp parallel for
for(int ip=0; ip<NumThreads; ++ip)
{
estimatorClones[ip]= new EstimatorManager(*Estimators);
estimatorClones[ip]->setCollectionMode(false);
Rng[ip]=new RandomGenerator_t(*RandomNumberControl::Children[ip]);
hClones[ip]->setRandomGenerator(Rng[ip]);
branchClones[ip] = new BranchEngineType(*branchEngine);
// Movers[ip] = new RNDMCUpdatePbyPWithRejectionFast(*wClones[ip],*psiClones[ip],*hClones[ip],*Rng[ip]);
if (useAlternate=="yes")
Movers[ip] = new RNDMCUpdatePbyPAlternate(*wClones[ip],*psiClones[ip],*hClones[ip],*Rng[ip]);
else
Movers[ip] = new RNDMCUpdatePbyPCeperley(*wClones[ip],*psiClones[ip],*hClones[ip],*Rng[ip]);
Movers[ip]->put(qmcNode);
Movers[ip]->resetRun(branchClones[ip],estimatorClones[ip]);
MCWalkerConfiguration::iterator wit(W.begin()+wPerNode[ip]), wit_end(W.begin()+wPerNode[ip+1]);
Movers[ip]->initWalkersForPbyP(wit, wit_end);
}
}
// branchEngine->checkParameters(W);
if(BranchInterval<0) BranchInterval=1;
{
ostringstream o;
if (useAlternate=="yes") o << " Using Alternate Mover"<<endl;
else o << " Using Ceperley Mover"<<endl;
o << " BranchInterval = " << BranchInterval << "\n";
o << " Steps per block = " << nSteps << "\n";
o << " Number of blocks = " << nBlocks << "\n";
app_log() << endl << o.str() << endl;
}
app_log() << " RNDMC Engine Initialization = " << init_timer.elapsed() << " secs " << endl;
}
bool DMCOMPOPT::run()
{
bool variablePop = (Reconfiguration == "no");
resetUpdateEngines();
//estimator does not need to collect data
Estimators->setCollectionMode(true);
Estimators->start(nBlocks);
for(int ip=0; ip<NumThreads; ip++)
Movers[ip]->startRun(nBlocks,false);
Timer myclock;
IndexType block = 0;
IndexType updatePeriod=(QMCDriverMode[QMC_UPDATE_MODE])?Period4CheckProperties:(nBlocks+1)*nSteps;
int nsampls(0);
int g_nsampls(0);
do // block
{
Estimators->startBlock(nSteps);
for(int ip=0; ip<NumThreads; ip++)
Movers[ip]->startBlock(nSteps);
IndexType step = 0;
for(IndexType step=0; step< nSteps; ++step, CurrentStep+=BranchInterval)
{
#pragma omp parallel
{
int ip=omp_get_thread_num();
int now=CurrentStep;
MCWalkerConfiguration::iterator wit(W.begin()+wPerNode[ip]), wit_first(W.begin()+wPerNode[ip]), wit2(W.begin()+wPerNode[ip]), wit_end(W.begin()+wPerNode[ip+1]);
for(int interval = 0; interval<BranchInterval-1; ++interval,++now)
{
Movers[ip]->advanceWalkers(wit,wit_end,false);
while(wit2!=wit_end)
{
(**wit2).addPropertyHistoryPoint(Eindx,(**wit2).getPropertyBase()[LOCALENERGY]);
wit2++;
}
wit2=wit_first;
if (Period4WalkerDump&&((now+1)%myPeriod4WalkerDump==0))
{
wClones[ip]->saveEnsemble(wit2,wit_end);
#pragma omp master
nsampls+=W.getActiveWalkers();
}
}
wClones[ip]->resetCollectables();
Movers[ip]->advanceWalkers(wit,wit_end,false);
wit2=wit_first;
while(wit2!=wit_end)
{
(**wit2).addPropertyHistoryPoint(Eindx,(**wit2).getPropertyBase()[LOCALENERGY]);
wit2++;
}
wit2=wit_first;
if (myPeriod4WalkerDump&&((now+1)%myPeriod4WalkerDump==0))
{
wClones[ip]->saveEnsemble(wit2,wit_end);
#pragma omp master
nsampls+=W.getActiveWalkers();
}
Movers[ip]->setMultiplicity(wit,wit_end);
if(QMCDriverMode[QMC_UPDATE_MODE] && now%updatePeriod == 0)
Movers[ip]->updateWalkers(wit, wit_end);
}//#pragma omp parallel
}
// branchEngine->debugFWconfig();
#pragma omp parallel for
for (int ip=0; ip<NumThreads; ++ip)
{
MCWalkerConfiguration::iterator wit(W.begin()+wPerNode[ip]), wit_end(W.begin()+wPerNode[ip+1]);
while (wit!=wit_end)
{
Walker_t& thisWalker(**wit);
Walker_t::Buffer_t& w_buffer(thisWalker.DataSet);
wClones[ip]->loadWalker(thisWalker,true);
psiClones[ip]->copyFromBuffer(*wClones[ip],w_buffer);
vector<RealType> Dsaved(NumOptimizables,0);
vector<RealType> HDsaved(NumOptimizables,0);
psiClones[ip]->evaluateDerivatives(*wClones[ip],dummyOptVars[ip],Dsaved,HDsaved,true);//SH like deriv style
#pragma omp critical
fillComponentMatrices(Dsaved,HDsaved,thisWalker);
wit++;
}
}
branchEngine->branch(CurrentStep,W, branchClones);
if(variablePop)
FairDivideLow(W.getActiveWalkers(),NumThreads,wPerNode);
Estimators->stopBlock(acceptRatio());
block++;
recordBlock(block);
g_nsampls=nsampls;
myComm->allreduce(g_nsampls);
}
while(((block<nBlocks) || (g_nsampls<nTargetSamples)) && myclock.elapsed()<MaxCPUSecs);
//for(int ip=0; ip<NumThreads; ip++) Movers[ip]->stopRun();
for(int ip=0; ip<NumThreads; ip++)
*(RandomNumberControl::Children[ip])=*(Rng[ip]);
// adding weight index
MCWalkerConfiguration::iterator wit(W.begin()), wit_end(W.end());
while(wit!=wit_end)
{
(**wit).deletePropertyHistory();
wit++;
}
Estimators->stop();
return finalize(block);
}
void DMCOMPOPT::resetUpdateEngines()
{
ReportEngine PRE("DMCOMPOPT","resetUpdateEngines");
resetComponents(qmcNode);
bool fixW = (Reconfiguration == "yes");
Timer init_timer;
// HACK HACK HACK
// This is so ugly it's probably a crime. It must be fixed.
if(firsttime)
{
// for(int ip=1; ip<NumThreads; ++ip)
// {
// delete wClones[ip];
// delete psiClones[ip];
// delete hClones[ip];
// }
// wClones.resize(0);
wClones.clear();
psiClones.clear();
hClones.clear();
firsttime=false;
}
for (int ip=0; ip<NumThreads; ++ip)
{
opt_variables_type dummy;
psiClones[ip]->checkInVariables(dummy);
dummy.resetIndex();
psiClones[ip]->checkOutVariables(dummy);
dummyOptVars.push_back(dummy);
}
NumOptimizables=dummyOptVars[0].size();
resizeForOpt(NumOptimizables);
makeClones(W,Psi,H);
if(Movers.empty())
{
// //load walkers
Eindx = W.addPropertyHistory(wlen);
W.loadEnsemble(wClones);
for(int ip=1; ip<NumThreads; ++ip)
wClones[ip]->addPropertyHistory(wlen);
// m_param.put(qmcNode);
// put(qmcNode);
// //app_log()<<"DMCOMPOPT::resetComponents"<<endl;
// Estimators->reset();
// branchEngine->resetRun(qmcNode);
// branchEngine->checkParameters(W);
branchEngine->initWalkerController(W,fixW,false);
//if(QMCDriverMode[QMC_UPDATE_MODE]) W.clearAuxDataSet();
Movers.resize(NumThreads,0);
branchClones.resize(NumThreads,0);
Rng.resize(NumThreads,0);
estimatorClones.resize(NumThreads,0);
FairDivideLow(W.getActiveWalkers(),NumThreads,wPerNode);
{
//log file
ostringstream o;
o << " Initial partition of walkers on a node: ";
std::copy(wPerNode.begin(),wPerNode.end(),ostream_iterator<int>(o," "));
o << "\n";
if(QMCDriverMode[QMC_UPDATE_MODE])
o << " Updates by particle-by-particle moves";
else
o << " Updates by walker moves";
if(UseFastGrad == "yes")
o << " using fast gradient version ";
else
o << " using full-ratio version ";
if(KillNodeCrossing)
o << "\n Walkers are killed when a node crossing is detected";
else
o << "\n DMC moves are rejected when a node crossing is detected";
app_log() << o.str() << endl;
}
#pragma omp parallel for
for(int ip=0; ip<NumThreads; ++ip)
{
estimatorClones[ip]= new EstimatorManager(*Estimators);
estimatorClones[ip]->setCollectionMode(false);
Rng[ip]=new RandomGenerator_t(*RandomNumberControl::Children[ip]);
hClones[ip]->setRandomGenerator(Rng[ip]);
branchClones[ip] = new BranchEngineType(*branchEngine);
if(QMCDriverMode[QMC_UPDATE_MODE])
{
if(UseFastGrad == "yes")
Movers[ip] = new DMCUpdatePbyPWithRejectionFast(*wClones[ip],*psiClones[ip],*hClones[ip],*Rng[ip]);
else
Movers[ip] = new DMCUpdatePbyPWithRejection(*wClones[ip],*psiClones[ip],*hClones[ip],*Rng[ip]);
// Movers[ip]->put(qmcNode);
// Movers[ip]->resetRun(branchClones[ip],estimatorClones[ip]);
// Movers[ip]->initWalkersForPbyP(W.begin()+wPerNode[ip],W.begin()+wPerNode[ip+1]);
}
else
{
if(KillNodeCrossing)
Movers[ip] = new DMCUpdateAllWithKill(*wClones[ip],*psiClones[ip],*hClones[ip],*Rng[ip]);
else
Movers[ip] = new DMCUpdateAllWithRejection(*wClones[ip],*psiClones[ip],*hClones[ip],*Rng[ip]);
// Movers[ip]->put(qmcNode);
// Movers[ip]->resetRun(branchClones[ip],estimatorClones[ip]);
// Movers[ip]->initWalkers(W.begin()+wPerNode[ip],W.begin()+wPerNode[ip+1]);
}
}
// #pragma omp parallel for
// for (int ip=0; ip<NumThreads; ++ip)
// {
// MCWalkerConfiguration::iterator wit(W.begin()+wPerNode[ip]), wit_end(W.begin()+wPerNode[ip+1]);
// while (wit!=wit_end)
// {
// Walker_t& thisWalker(**wit);
// Walker_t::Buffer_t& w_buffer(thisWalker.DataSet);
// wClones[ip]->loadWalker(thisWalker,true);
// psiClones[ip]->copyFromBuffer(*wClones[ip],w_buffer);
// psiClones[ip]->updateBuffer(*wClones[ip],w_buffer,true);
// wit++;
// }
// }
}
#pragma omp parallel for
for(int ip=0; ip<NumThreads; ++ip)
{
if(QMCDriverMode[QMC_UPDATE_MODE])
{
Movers[ip]->put(qmcNode);
Movers[ip]->resetRun(branchClones[ip],estimatorClones[ip]);
Movers[ip]->initWalkersForPbyP(W.begin()+wPerNode[ip],W.begin()+wPerNode[ip+1]);
}
else
{
Movers[ip]->put(qmcNode);
Movers[ip]->resetRun(branchClones[ip],estimatorClones[ip]);
Movers[ip]->initWalkers(W.begin()+wPerNode[ip],W.begin()+wPerNode[ip+1]);
}
}
// adding weight index
MCWalkerConfiguration::iterator wit(W.begin()), wit_end(W.end());
(**wit).addPropertyHistory(wlen);
wit++;
while(wit!=wit_end)
{
(**wit).addPropertyHistory(wlen);
wit++;
}
std::vector<IndexType> samples_th(omp_get_max_threads(),0);
myPeriod4WalkerDump=(Period4WalkerDump>0)?Period4WalkerDump:(nBlocks+1)*nSteps;
samples_this_node = nTargetSamples/myComm->size();
if (nTargetSamples%myComm->size() > myComm->rank())
samples_this_node+=1;
int samples_each_thread = samples_this_node/omp_get_max_threads();
for (int ip=0; ip<omp_get_max_threads(); ++ip)
samples_th[ip]=samples_each_thread;
if(samples_this_node%omp_get_max_threads())
for (int ip=0; ip < samples_this_node%omp_get_max_threads(); ++ip)
samples_th[ip] +=1;
int ndumps = std::max(samples_this_node/W.getActiveWalkers() + 1,2);
myPeriod4WalkerDump = nBlocks*nSteps/ndumps;
app_log() << " Samples are dumped every " << myPeriod4WalkerDump << " steps " << endl;
app_log() << " Total Sample Size =" << nTargetSamples << endl;
app_log() << " Nodes Sample Size =" << samples_this_node << endl;
for (int ip=0; ip<NumThreads; ++ip)
app_log() << " Sample size for thread " <<ip<<" = " << samples_th[ip] << endl;
#pragma omp critical
for(int ip=0; ip<NumThreads; ++ip)
{
wClones[ip]->clearEnsemble();
wClones[ip]->setNumSamples(samples_th[ip]);
}
t= Movers[0]->getTau();
clearComponentMatrices();
branchEngine->checkParameters(W);
int mxage=mover_MaxAge;
if(fixW)
{
if(BranchInterval<0)
BranchInterval=nSteps;
mxage=(mover_MaxAge<0)?0:mover_MaxAge;
for(int ip=0; ip<Movers.size(); ++ip)
Movers[ip]->MaxAge=mxage;
}
else
{
if(BranchInterval<0)
BranchInterval=1;
int miage=(QMCDriverMode[QMC_UPDATE_MODE])?1:5;
mxage=(mover_MaxAge<0)?miage:mover_MaxAge;
for(int ip=0; ip<Movers.size(); ++ip)
Movers[ip]->MaxAge=mxage;
}
{
ostringstream o;
if(fixW)
o << " Fixed population using reconfiguration method\n";
else
o << " Fluctuating population\n";
o << " Persisent walkers are killed after " << mxage << " MC sweeps\n";
o << " BranchInterval = " << BranchInterval << "\n";
o << " Steps per block = " << nSteps << "\n";
o << " Number of blocks = " << nBlocks << "\n";
app_log() << o.str() << endl;
}
app_log() << " DMC Engine Initialization = " << init_timer.elapsed() << " secs " << endl;
}
void DMCOMP::resetUpdateEngines()
{
ReportEngine PRE("DMCOMP","resetUpdateEngines");
bool fixW = (Reconfiguration == "yes");
Timer init_timer;
makeClones(W,Psi,H);
if(Movers.empty())
{
//load walkers
W.loadEnsemble();
for(int ip=1;ip<NumThreads;++ip) wClones[ip]->loadEnsemble(W);
branchEngine->initWalkerController(W,fixW,false);
//if(QMCDriverMode[QMC_UPDATE_MODE]) W.clearAuxDataSet();
Movers.resize(NumThreads,0);
branchClones.resize(NumThreads,0);
Rng.resize(NumThreads,0);
estimatorClones.resize(NumThreads,0);
FairDivideLow(W.getActiveWalkers(),NumThreads,wPerNode);
{//log file
ostringstream o;
o << " Initial partition of walkers on a node: ";
std::copy(wPerNode.begin(),wPerNode.end(),ostream_iterator<int>(o," "));
o << "\n";
if(QMCDriverMode[QMC_UPDATE_MODE])
o << " Updates by particle-by-particle moves";
else
o << " Updates by walker moves";
if(UseFastGrad == "yes")
o << " using fast gradient version ";
else
o << " using full-ratio version ";
if(KillNodeCrossing)
o << "\n Walkers are killed when a node crossing is detected";
else
o << "\n DMC moves are rejected when a node crossing is detected";
app_log() << o.str() << endl;
}
#pragma omp parallel for
for(int ip=0; ip<NumThreads; ++ip)
{
estimatorClones[ip]= new EstimatorManager(*Estimators);
estimatorClones[ip]->setCollectionMode(false);
Rng[ip]=new RandomGenerator_t(*RandomNumberControl::Children[ip]);
hClones[ip]->setRandomGenerator(Rng[ip]);
branchClones[ip] = new BranchEngineType(*branchEngine);
if(QMCDriverMode[QMC_UPDATE_MODE])
{
if(UseFastGrad == "yes")
Movers[ip] = new DMCUpdatePbyPWithRejectionFast(*wClones[ip],*psiClones[ip],*hClones[ip],*Rng[ip]);
else
Movers[ip] = new DMCUpdatePbyPWithRejection(*wClones[ip],*psiClones[ip],*hClones[ip],*Rng[ip]);
Movers[ip]->put(qmcNode);
Movers[ip]->resetRun(branchClones[ip],estimatorClones[ip]);
Movers[ip]->initWalkersForPbyP(W.begin()+wPerNode[ip],W.begin()+wPerNode[ip+1]);
}
else
{
if(KillNodeCrossing)
Movers[ip] = new DMCUpdateAllWithKill(*wClones[ip],*psiClones[ip],*hClones[ip],*Rng[ip]);
else
Movers[ip] = new DMCUpdateAllWithRejection(*wClones[ip],*psiClones[ip],*hClones[ip],*Rng[ip]);
Movers[ip]->put(qmcNode);
Movers[ip]->resetRun(branchClones[ip],estimatorClones[ip]);
Movers[ip]->initWalkers(W.begin()+wPerNode[ip],W.begin()+wPerNode[ip+1]);
}
}
}
branchEngine->checkParameters(W);
int mxage=mover_MaxAge;
if(fixW)
{
if(BranchInterval<0) BranchInterval=nSteps;
mxage=(mover_MaxAge<0)?0:mover_MaxAge;
for(int ip=0; ip<Movers.size(); ++ip) Movers[ip]->MaxAge=mxage;
}
else
{
if(BranchInterval<0) BranchInterval=1;
int miage=(QMCDriverMode[QMC_UPDATE_MODE])?1:5;
mxage=(mover_MaxAge<0)?miage:mover_MaxAge;
for(int ip=0; ip<Movers.size(); ++ip) Movers[ip]->MaxAge=mxage;
}
{
ostringstream o;
if(fixW)
o << " Fixed population using reconfiguration method\n";
else
o << " Fluctuating population\n";
o << " Persisent walkers are killed after " << mxage << " MC sweeps\n";
o << " BranchInterval = " << BranchInterval << "\n";
o << " Steps per block = " << nSteps << "\n";
o << " Number of blocks = " << nBlocks << "\n";
app_log() << o.str() << endl;
}
app_log() << " DMC Engine Initialization = " << init_timer.elapsed() << " secs " << endl;
}
bool DMCOMP::run() {
bool variablePop = (Reconfiguration == "no");
resetUpdateEngines();
//estimator does not need to collect data
Estimators->setCollectionMode(true);
Estimators->start(nBlocks);
for(int ip=0; ip<NumThreads; ip++) Movers[ip]->startRun(nBlocks,false);
Timer myclock;
IndexType block = 0;
IndexType updatePeriod=(QMCDriverMode[QMC_UPDATE_MODE])?Period4CheckProperties:(nBlocks+1)*nSteps;
do // block
{
Estimators->startBlock(nSteps);
for(int ip=0; ip<NumThreads; ip++) Movers[ip]->startBlock(nSteps);
IndexType step = 0;
for(IndexType step=0; step< nSteps; ++step, CurrentStep+=BranchInterval)
{
if(storeConfigs && (CurrentStep%storeConfigs == 0)) {
ForwardWalkingHistory.storeConfigsForForwardWalking(W);
W.resetWalkerParents();
}
#pragma omp parallel for
for(int ip=0; ip<NumThreads; ++ip)
{
int now=CurrentStep;
MCWalkerConfiguration::iterator
wit(W.begin()+wPerNode[ip]), wit_end(W.begin()+wPerNode[ip+1]);
for(int interval = 0;interval<BranchInterval-1; ++interval,++now)
Movers[ip]->advanceWalkers(wit,wit_end,false);
wClones[ip]->resetCollectables();
Movers[ip]->advanceWalkers(wit,wit_end,false);
Movers[ip]->setMultiplicity(wit,wit_end);
if(QMCDriverMode[QMC_UPDATE_MODE] && now%updatePeriod == 0) Movers[ip]->updateWalkers(wit, wit_end);
}//#pragma omp parallel
Estimators->accumulateCollectables(wClones,1.0);
branchEngine->branch(CurrentStep,W, branchClones);
// if(storeConfigs && (CurrentStep%storeConfigs == 0)) {
// ForwardWalkingHistory.storeConfigsForForwardWalking(W);
// W.resetWalkerParents();
// }
if(variablePop) FairDivideLow(W.getActiveWalkers(),NumThreads,wPerNode);
}
// branchEngine->debugFWconfig();
Estimators->stopBlock(acceptRatio());
block++;
recordBlock(block);
} while(block<nBlocks && myclock.elapsed()<MaxCPUSecs);
//for(int ip=0; ip<NumThreads; ip++) Movers[ip]->stopRun();
for(int ip=0; ip<NumThreads; ip++)
*(RandomNumberControl::Children[ip])=*(Rng[ip]);
Estimators->stop();
return finalize(block);
}
void VMCSingleOMP::resetRun()
{
makeClones(W,Psi,H);
//determine dump period for walkers
int samples_tot=W.getActiveWalkers()*nBlocks*nSteps*myComm->size();
myPeriod4WalkerDump=(nTargetSamples>0)?samples_tot/nTargetSamples:Period4WalkerDump;
//fall back to the default
if(myPeriod4WalkerDump==0) myPeriod4WalkerDump=Period4WalkerDump;
if(QMCDriverMode[QMC_WARMUP]) myPeriod4WalkerDump=nBlocks*nSteps;
int samples_th=nTargetSamples/myComm->size()/NumThreads;
for(int ip=0; ip<NumThreads;++ip)
{
wClones[ip]->clearEnsemble();
wClones[ip]->setNumSamples(samples_th);
}
app_log() << " Samples are dumped at every " << myPeriod4WalkerDump << " step " << endl;
app_log() << " Total Sample Size =" << nTargetSamples
<< "\n Sample size per node per thread = " << samples_th << endl;
app_log() << " Warmup Steps " << myWarmupSteps << endl;
if(Movers.empty())
{
Movers.resize(NumThreads,0);
branchClones.resize(NumThreads,0);
estimatorClones.resize(NumThreads,0);
Rng.resize(NumThreads,0);
int nwtot=(W.getActiveWalkers()/NumThreads)*NumThreads;
FairDivideLow(nwtot,NumThreads,wPerNode);
app_log() << " Initial partition of walkers ";
std::copy(wPerNode.begin(),wPerNode.end(),ostream_iterator<int>(app_log()," "));
app_log() << endl;
//#pragma omp parallel for
for(int ip=0; ip<NumThreads; ++ip)
{
estimatorClones[ip]= new EstimatorManager(*Estimators);//,*hClones[ip]);
estimatorClones[ip]->resetTargetParticleSet(*wClones[ip]);
estimatorClones[ip]->setCollectionMode(false);
Rng[ip]=new RandomGenerator_t(*(RandomNumberControl::Children[ip]));
hClones[ip]->setRandomGenerator(Rng[ip]);
branchClones[ip] = new BranchEngineType(*branchEngine);
if(QMCDriverMode[QMC_UPDATE_MODE])
{
if(UseDrift == "yes")
Movers[ip]=new VMCUpdatePbyPWithDrift(*wClones[ip],*psiClones[ip],*hClones[ip],*Rng[ip]);
else
Movers[ip]=new VMCUpdatePbyP(*wClones[ip],*psiClones[ip],*hClones[ip],*Rng[ip]);
Movers[ip]->resetRun(branchClones[ip],estimatorClones[ip]);
}
else
{
if(UseDrift == "yes")
Movers[ip]=new VMCUpdateAllWithDrift(*wClones[ip],*psiClones[ip],*hClones[ip],*Rng[ip]);
else
Movers[ip]=new VMCUpdateAll(*wClones[ip],*psiClones[ip],*hClones[ip],*Rng[ip]);
Movers[ip]->resetRun(branchClones[ip],estimatorClones[ip]);
}
}
}
//#pragma omp parallel for
for(int ip=0; ip<NumThreads; ++ip)
{
if(QMCDriverMode[QMC_UPDATE_MODE])
Movers[ip]->initWalkersForPbyP(W.begin()+wPerNode[ip],W.begin()+wPerNode[ip+1]);
else
Movers[ip]->initWalkers(W.begin()+wPerNode[ip],W.begin()+wPerNode[ip+1]);
for(int prestep=0; prestep<myWarmupSteps; ++prestep)
Movers[ip]->advanceWalkers(W.begin()+wPerNode[ip],W.begin()+wPerNode[ip+1],true);
if(myWarmupSteps && QMCDriverMode[QMC_UPDATE_MODE])
Movers[ip]->updateWalkers(W.begin()+wPerNode[ip],W.begin()+wPerNode[ip+1]);
}
myWarmupSteps=0;
//Used to debug and benchmark opnemp
//#pragma omp parallel for
// for(int ip=0; ip<NumThreads; ip++)
// {
// Movers[ip]->benchMark(W.begin()+wPerNode[ip],W.begin()+wPerNode[ip+1],ip);
// }
}
void RNDMCOMP::resetUpdateEngines()
{
ReportEngine PRE("RNDMCOMP","resetUpdateEngines");
Timer init_timer;
makeClones(W,Psi,H);
makeClones(W,Guide);
if(Movers.empty())
{
W.loadEnsemble(wClones);
branchEngine->initWalkerController(W,false,true);
branchEngine->setRN(true);
//if(QMCDriverMode[QMC_UPDATE_MODE]) W.clearAuxDataSet();
Movers.resize(NumThreads,0);
branchClones.resize(NumThreads,0);
Rng.resize(NumThreads,0);
estimatorClones.resize(NumThreads,0);
FairDivideLow(W.getActiveWalkers(),NumThreads,wPerNode);
bool klw=(KillWalker=="yes");
{
//log file
ostringstream o;
o << " Initial partition of walkers on a node: ";
std::copy(wPerNode.begin(),wPerNode.end(),ostream_iterator<int>(o," "));
o << "\n";
o << "Killing Walkers at nodes " << (useAlternate!="yes") <<endl;
o << "Running the released node driver."<<endl;
app_log() << o.str();
}
#pragma omp parallel for
for(int ip=0; ip<NumThreads; ++ip)
{
estimatorClones[ip]= new EstimatorManager(*Estimators);
estimatorClones[ip]->setCollectionMode(false);
Rng[ip]=new RandomGenerator_t(*RandomNumberControl::Children[ip]);
hClones[ip]->setRandomGenerator(Rng[ip]);
branchClones[ip] = new BranchEngineType(*branchEngine);
// Movers[ip] = new RNDMCUpdatePbyPWithRejectionFast(*wClones[ip],*psiClones[ip],*hClones[ip],*Rng[ip]);
Movers[ip] = new RNDMCUpdatePbyPFast(*wClones[ip],*wgClones[ip],*psiClones[ip],*guideClones[ip],*hClones[ip],*Rng[ip]);
Movers[ip]->put(qmcNode);
Movers[ip]->resetRun(branchClones[ip],estimatorClones[ip]);
MCWalkerConfiguration::iterator wit(W.begin()+wPerNode[ip]), wit_end(W.begin()+wPerNode[ip+1]);
Movers[ip]->initWalkersForPbyP(wit, wit_end);
}
}
// branchEngine->checkParameters(W);
if(BranchInterval<0)
BranchInterval=1;
{
ostringstream o;
// if (useAlternate=="yes") o << " Using Alternate Mover"<<endl;
// else o << " Using Ceperley Mover"<<endl;
o << " BranchInterval = " << BranchInterval << "\n";
o << " Steps per block = " << nSteps << "\n";
o << " Number of blocks = " << nBlocks << "\n";
app_log() << endl << o.str() << endl;
}
// RealType avg_w(0);
// RealType n_w(0);
// RealType logepsilon(0);
// #pragma omp parallel
// {
// int ip=omp_get_thread_num();
//
// for (int step=0; step<nWarmupSteps; ++step)
// {
// avg_w=0;
// n_w=0;
// for (int prestep=0; prestep<myRNWarmupSteps; ++prestep)
// {
// Movers[ip]->advanceWalkers(W.begin()+wPerNode[ip],W.begin()+wPerNode[ip+1],true);
// #pragma omp single
// {
// MCWalkerConfiguration::iterator wit(W.begin()), wit_end(W.end());
// while (wit!=wit_end)
// {
// avg_w += (*wit)->Weight;
// n_w +=1;
// wit++;
// }
// }
// #pragma omp barrier
// }
// #pragma omp single
// {
// avg_w *= 1.0/n_w;
// RealType w_m = avg_w/(1.0-avg_w);
// w_m = std::log(0.5+0.5*w_m);
// if (std::abs(w_m)>0.01)
// logepsilon += w_m;
// }
// #pragma omp barrier
// Movers[ip]->setLogEpsilon(logepsilon);
// }
//
// }
app_log() << " RNDMC Engine Initialization = " << init_timer.elapsed() << " secs " << endl;
}
void VMCSingleOMP::resetRun()
{
////only VMC can overwrite this
if(nTargetPopulation>0)
branchEngine->iParam[SimpleFixedNodeBranch::B_TARGETWALKERS]=static_cast<int>(std::ceil(nTargetPopulation));
makeClones(W,Psi,H);
FairDivideLow(W.getActiveWalkers(),NumThreads,wPerNode);
app_log() << " Initial partition of walkers ";
std::copy(wPerNode.begin(),wPerNode.end(),ostream_iterator<int>(app_log()," "));
app_log() << endl;
if (Movers.empty())
{
Movers.resize(NumThreads,0);
branchClones.resize(NumThreads,0);
estimatorClones.resize(NumThreads,0);
traceClones.resize(NumThreads,0);
Rng.resize(NumThreads,0);
#if !defined(BGP_BUG)
#pragma omp parallel for
#endif
for(int ip=0; ip<NumThreads; ++ip)
{
ostringstream os;
estimatorClones[ip]= new EstimatorManager(*Estimators);//,*hClones[ip]);
estimatorClones[ip]->resetTargetParticleSet(*wClones[ip]);
estimatorClones[ip]->setCollectionMode(false);
traceClones[ip] = Traces->makeClone();
Rng[ip]=new RandomGenerator_t(*(RandomNumberControl::Children[ip]));
hClones[ip]->setRandomGenerator(Rng[ip]);
branchClones[ip] = new BranchEngineType(*branchEngine);
// if(reweight=="yes")
// {
// if (ip== 0) app_log() << " WFMCUpdateAllWithReweight"<<endl;
// Movers[ip]=new WFMCUpdateAllWithReweight(*wClones[ip],*psiClones[ip],*hClones[ip],*Rng[ip],weightLength,Eindex);
// }
// else
// if (reweight=="psi")
// {
// os << " Sampling Psi to increase number of walkers near nodes"<<endl;
// if (QMCDriverMode[QMC_UPDATE_MODE]) Movers[ip]=new VMCUpdatePbyPSamplePsi(*wClones[ip],*psiClones[ip],*hClones[ip],*Rng[ip]);
// else Movers[ip]=new VMCUpdateAllSamplePsi(*wClones[ip],*psiClones[ip],*hClones[ip],*Rng[ip]);
// }
// else
if (QMCDriverMode[QMC_UPDATE_MODE])
{
// if (UseDrift == "rn")
// {
// os <<" PbyP moves with RN, using VMCUpdatePbyPSampleRN"<<endl;
// Movers[ip]=new VMCUpdatePbyPSampleRN(*wClones[ip],*psiClones[ip],*guideClones[ip],*hClones[ip],*Rng[ip]);
// Movers[ip]->setLogEpsilon(logepsilon);
// // Movers[ip]=new VMCUpdatePbyPWithDrift(*wClones[ip],*psiClones[ip],*hClones[ip],*Rng[ip]);
// }
// else
if (UseDrift == "yes")
{
os <<" PbyP moves with drift, using VMCUpdatePbyPWithDriftFast"<<endl;
Movers[ip]=new VMCUpdatePbyPWithDriftFast(*wClones[ip],*psiClones[ip],*hClones[ip],*Rng[ip]);
// Movers[ip]=new VMCUpdatePbyPWithDrift(*wClones[ip],*psiClones[ip],*hClones[ip],*Rng[ip]);
}
else
{
os <<" PbyP moves with |psi^2|, using VMCUpdatePbyP"<<endl;
Movers[ip]=new VMCUpdatePbyP(*wClones[ip],*psiClones[ip],*hClones[ip],*Rng[ip]);
}
//Movers[ip]->resetRun(branchClones[ip],estimatorClones[ip]);
}
else
{
// if (UseDrift == "rn")
// {
// os <<" walker moves with RN, using VMCUpdateAllSampleRN"<<endl;
// Movers[ip]=new VMCUpdateAllSampleRN(*wClones[ip],*psiClones[ip],*guideClones[ip],*hClones[ip],*Rng[ip]);
// Movers[ip]->setLogEpsilon(logepsilon);
// }
// else
if (UseDrift == "yes")
{
os <<" walker moves with drift, using VMCUpdateAllWithDriftFast"<<endl;
Movers[ip]=new VMCUpdateAllWithDrift(*wClones[ip],*psiClones[ip],*hClones[ip],*Rng[ip]);
}
else
{
os <<" walker moves with |psi|^2, using VMCUpdateAll"<<endl;
Movers[ip]=new VMCUpdateAll(*wClones[ip],*psiClones[ip],*hClones[ip],*Rng[ip]);
}
//Movers[ip]->resetRun(branchClones[ip],estimatorClones[ip]);
}
Movers[ip]->nSubSteps=nSubSteps;
if(ip==0)
app_log() << os.str() << endl;
}
}
else
{
#if !defined(BGP_BUG)
#pragma omp parallel for
#endif
for(int ip=0; ip<NumThreads; ++ip)
{
traceClones[ip]->transfer_state_from(*Traces);
}
}
app_log() << " Total Sample Size =" << nTargetSamples << endl;
app_log() << " Walker distribution on root = ";
std::copy(wPerNode.begin(),wPerNode.end(),ostream_iterator<int>(app_log()," "));
app_log() << endl;
//app_log() << " Sample Size per node=" << samples_this_node << endl;
//for (int ip=0; ip<NumThreads; ++ip)
// app_log() << " Sample size for thread " <<ip<<" = " << samples_th[ip] << endl;
app_log().flush();
#if !defined(BGP_BUG)
#pragma omp parallel for
#endif
for(int ip=0; ip<NumThreads; ++ip)
{
//int ip=omp_get_thread_num();
Movers[ip]->put(qmcNode);
Movers[ip]->resetRun(branchClones[ip],estimatorClones[ip],traceClones[ip]);
if (QMCDriverMode[QMC_UPDATE_MODE])
Movers[ip]->initWalkersForPbyP(W.begin()+wPerNode[ip],W.begin()+wPerNode[ip+1]);
else
Movers[ip]->initWalkers(W.begin()+wPerNode[ip],W.begin()+wPerNode[ip+1]);
// if (UseDrift != "rn")
// {
for (int prestep=0; prestep<nWarmupSteps; ++prestep)
Movers[ip]->advanceWalkers(W.begin()+wPerNode[ip],W.begin()+wPerNode[ip+1],true);
if (nWarmupSteps && QMCDriverMode[QMC_UPDATE_MODE])
Movers[ip]->updateWalkers(W.begin()+wPerNode[ip],W.begin()+wPerNode[ip+1]);
// }
}
// //JNKIM: THIS IS BAD AND WRONG
// if (UseDrift == "rn")
// {
// RealType avg_w(0);
// RealType n_w(0);
// #pragma omp parallel
// {
// int ip=omp_get_thread_num();
// for (int step=0; step<nWarmupSteps; ++step)
// {
// avg_w=0;
// n_w=0;
// for (int prestep=0; prestep<myRNWarmupSteps; ++prestep)
// {
// Movers[ip]->advanceWalkers(W.begin()+wPerNode[ip],W.begin()+wPerNode[ip+1],true);
// #pragma omp single
// {
// MCWalkerConfiguration::iterator wit(W.begin()), wit_end(W.end());
// while (wit!=wit_end)
// {
// avg_w += (*wit)->Weight;
// n_w +=1;
// wit++;
// }
// }
// #pragma omp barrier
// }
// #pragma omp single
// {
// avg_w *= 1.0/n_w;
// RealType w_m = avg_w/(1.0-avg_w);
// w_m = std::log(0.5+0.5*w_m);
// if (std::abs(w_m)>0.01)
// logepsilon += w_m;
// }
// #pragma omp barrier
// Movers[ip]->setLogEpsilon(logepsilon);
// }
//
// for (int prestep=0; prestep<nWarmupSteps; ++prestep)
// Movers[ip]->advanceWalkers(W.begin()+wPerNode[ip],W.begin()+wPerNode[ip+1],true);
//
// if (nWarmupSteps && QMCDriverMode[QMC_UPDATE_MODE])
// Movers[ip]->updateWalkers(W.begin()+wPerNode[ip],W.begin()+wPerNode[ip+1]);
// }
// }
for(int ip=0; ip<NumThreads; ++ip)
wClones[ip]->clearEnsemble();
if(nSamplesPerThread)
for(int ip=0; ip<NumThreads; ++ip)
wClones[ip]->setNumSamples(nSamplesPerThread);
nWarmupSteps=0;
//Used to debug and benchmark opnemp
//#pragma omp parallel for
// for(int ip=0; ip<NumThreads; ip++)
// {
// Movers[ip]->benchMark(W.begin()+wPerNode[ip],W.begin()+wPerNode[ip+1],ip);
// }
}
void HDFWalkerIOEngine::readAll(hid_t grp, const char* name, Communicate* comm)
{
int mynode=comm->rank();
int nprocs=comm->size();
vector<int> woffset(nprocs+1,0);
const int RANK = 3;
hsize_t offset[]={1,1,1};
hsize_t gcount[RANK],count[RANK];
hsize_t stride[]={1,1,1};
hid_t dataset = H5Dopen(grp,name);
hid_t dataspace = H5Dget_space(dataset);
int rank_n = H5Sget_simple_extent_ndims(dataspace);
int status_n = H5Sget_simple_extent_dims(dataspace, gcount, NULL);
//assign offsets and size
FairDivideLow(gcount[0],nprocs,woffset);
offset[0]=woffset[mynode]; offset[1] = 0; offset[2] = 0;
count[0]=woffset[mynode+1]-woffset[mynode];
count[1]=gcount[1];
count[2]=gcount[2];
app_log() << " Initial walker distribution: ";
std::copy(woffset.begin(),woffset.end(),ostream_iterator<int>(app_log()," "));
app_log() << endl;
vector<MCWalkerConfiguration::PosType> posIn(count[0]*count[1]);
hid_t memspace = H5Screate_simple(RANK, count, NULL);
herr_t status = H5Sselect_hyperslab(dataspace,H5S_SELECT_SET, offset,NULL,count,NULL);
#if defined(H5_HAVE_PARALLEL)
xfer_plist = H5Pcreate(H5P_DATASET_XFER);
H5Pset_dxpl_mpio(xfer_plist,H5FD_MPIO_COLLECTIVE);
#else
xfer_plist = H5P_DEFAULT;
#endif
hid_t type_id=get_h5_datatype(posIn[0][0]);
status = H5Dread(dataset, type_id, memspace, dataspace, xfer_plist, &(posIn[0][0]));
H5Sclose(dataspace);
H5Sclose(memspace);
H5Dclose(dataset);
#if defined(H5_HAVE_PARALLEL)
H5Pclose(xfer_plist);
#endif
int curWalker = W.getActiveWalkers();
if(curWalker) {
W.createWalkers(count[0]);
} else {
W.resize(count[0],count[1]);
}
MCWalkerConfiguration::iterator it = W.begin()+curWalker;
int ii=0;
for(int iw=0; iw<count[0]; iw++) {
//std::copy(Post_temp[iw],Post_temp[iw+1], (*it)->R.begin());
for(int iat=0; iat < count[1]; iat++,ii++){
(*it)->R(iat) = posIn[ii];
}
++it;
}
}
