static void span(Surface& s,
WFMath::CoordType z,
WFMath::CoordType xStart,
WFMath::CoordType xEnd)
{
assert(xStart <= xEnd);
// quantize and accumulate into the buffer data
unsigned int row = I_ROUND(z),
ixStart = I_ROUND(xStart),
ixEnd = I_ROUND(xEnd);
//std::cout << "span @ z=" << row << ", " << ixStart << " -> " << ixEnd << std::endl;
if (ixStart == ixEnd) {
contribute(s, ixStart, row, ROW_HEIGHT * (xEnd - xStart));
} else {
contribute(s, ixStart, row, ROW_HEIGHT * (ixStart - xStart + 0.5f));
for (unsigned int i=ixStart+1; i < ixEnd; ++i)
contribute(s, i, row, ROW_HEIGHT);
contribute(s, ixEnd, row, ROW_HEIGHT * (xEnd - ixEnd + 0.5f));
}
}
void check_and_compute() {
compute_kernel();
// calculate error
// not being done right now since we are doing a fixed no. of iterations
double *tmp;
tmp = temperature;
temperature = new_temperature;
new_temperature = tmp;
constrainBC();
if (iterations % CKP_FREQ == 0 || iterations > MAX_ITER) {
#ifdef CMK_MEM_CHECKPOINT
contribute(0, 0, CkReduction::concat, CkCallback(CkIndex_Main::report(), mainProxy));
#elif CMK_MESSAGE_LOGGING
if(iterations > MAX_ITER)
contribute(0, 0, CkReduction::concat, CkCallback(CkIndex_Main::report(), mainProxy));
else
AtSync();
#else
contribute(0, 0, CkReduction::concat, CkCallback(CkIndex_Main::report(), mainProxy));
#endif
} else {
doStep();
}
}
void BTest::receive(MyMess *msg) {
delete msg;
if (thisIndex==0) {
if (iter==threshold_msgs) startTime=CmiWallTimer();
//if (iter>0) totalTime += CmiWallTimer() - startTime;
//if (thisIndex==0) CkPrintf("Finished iteration %d (type %d)\n",iter+1,type);
if (iter++<MAX_ITER) {
//CkPrintf("[%2d] totaltime with '%s' = %f\n",CmiMyPe(),conversion[type],totalTime/(iter-1));
//startTime=CmiWallTimer();
send();
//contribute(0, &count, CkReduction::sum_int, cb);
} else {
iter=0;
//CkPrintf("[%2d] totaltime with '%s' = %f\n",CmiMyPe(),conversion[type],(CmiWallTimer()-startTime)/(MAX_ITER-threshold_msgs));
if (type==0) cinst7.beginIteration();
else if (type==1) cinst6.beginIteration();
else if (type==2) cinst5.beginIteration();
else if (type==3) cinst4.beginIteration();
else if (type==4) cinst3.beginIteration();
else if (type==5) cinst2.beginIteration();
else if (type==5) cinst1.beginIteration();
// global barrier
contribute(0, &count, CkReduction::sum_int, cb);
/*
if (++type==MAX_TYPE) {
if (++length==MAX_LENGTH) {
mainProxy.exit();
} else {
type = 0;
//sendall();
contribute(0, &count, CkReduction::sum_int, cball);
}
} else {
totalTime=0;
//startTime = CmiWallTimer();
send();
//contribute(0, &count, CkReduction::sum_int, cb);
}
*/
}
} else {
if (iter++<MAX_ITER) {
iter = 0;
contribute(0, &count, CkReduction::sum_int, cball);
}
}
}
void HybridBaseLB::MigrationDone(int balancing)
{
#if CMK_LBDB_ON
LevelData *lData = levelData[0];
DEBUGF(("[%d] HybridBaseLB::MigrationDone!\n", CkMyPe()));
theLbdb->incStep();
// reset
for (int i=0; i<tree->numLevels(); i++)
levelData[i]->clear();
newObjs.free();
DEBUGF(("[%d] calling ResumeClients.\n", CkMyPe()));
if (balancing && _lb_args.syncResume()) {
// max load of all
CkCallback cb(CkIndex_HybridBaseLB::ResumeClients((CkReductionMsg*)NULL),
thisProxy);
contribute(sizeof(double), &maxLoad, CkReduction::max_double, cb);
}
else
thisProxy[CkMyPe()].ResumeClients(balancing);
maxLoad = 0.0;
#endif
}
void communicate(int iters, bool useTram) {
GroupMeshStreamer<DataItem, Participant, SimpleMeshRouter> *localStreamer;
if (useTram) {
localStreamer = aggregator.ckLocalBranch();
}
int ctr = 0;
for (int i = 0; i < iters; i++) {
for (int j=0; j<CkNumPes(); j++) {
if (useTram) {
localStreamer->insertData(myItem, neighbors[j]);
}
else {
allToAllGroup[neighbors[j]].receive(myItem);
ctr++;
}
}
if (!useTram) {
if (ctr == 1024) {
ctr = 0;
CthYield();
}
}
}
if (useTram) {
localStreamer->done();
}
else {
contribute(CkCallback(CkReductionTarget(Main, allDone), mainProxy));
}
}
void check_and_compute ()
{
// if (--messages_due == 0)
// messages_due = 4;
compute ();
// mainProxy.report();
if (thisIndex < majElements - 1)
{
// printf("DONE WITH index=%d and calling for ind=%d\n",thisIndex,thisIndex+1);
#ifdef PRIOR
opts = new CkEntryOptions ();
opts1 = new CkEntryOptions ();
opts->setPriority (-100);
opts1->setPriority (100);
//printf("-------- Jacobi[%d] sending message to next one at time=%f\n",thisIndex,CkWallTimer());
thisProxy[thisIndex + 1].begin_iteration (1, opts);
for(int i=(thisIndex+1)*7;i<(thisIndex+1)*7+7;i++)
minorProxy[i].begin_iteration(1,opts1);
#else
thisProxy[thisIndex + 1].begin_iteration (1);
for (int i = (thisIndex + 1) * 7; i < (thisIndex + 1) * 7 + 7; i++)
minorProxy[i].begin_iteration (1);
#endif
}
else
{
// printf("CAlling report Jacobi[%d] time=%f!!!!!!!!!!1\n",thisIndex,CkWallTimer());
// else
// mainProxy.report ();
}
if (iterations % ldbTime == 4) AtSync();
else contribute(CkCallback(CkIndex_Main::report(NULL),mainProxy));
}
void Compute::receiveC(float *data, int size, int who) {
int indexY = thisIndex.y;
if(who) {
for(int i=0; i<subBlockDimXy; i++)
for(int k=0; k<blockDimZ; k++)
C[indexY*subBlockDimXy*blockDimZ + i*blockDimZ + k] += data[i*blockDimZ + k];
}
countC++;
if(countC == num_chare_y) {
/*char name[30];
sprintf(name, "%s_%d_%d_%d", "C", thisIndex.x, thisIndex.y, thisIndex.z);
FILE *fp = fopen(name, "w");
for(int i=0; i<subBlockDimXy; i++) {
for(int k=0; k<blockDimZ; k++)
fprintf(fp, "%f ", C[indexY*subBlockDimXy*blockDimZ + i*blockDimZ + k]);
fprintf(fp, "\n");
}
fclose(fp);*/
// counters to keep track of how many messages have been received
countA = 0;
countB = 0;
countC = 0;
contribute(0, 0, CkReduction::concat, CkCallback(CkIndex_Main::done(), mainProxy));
// mainProxy.done();
}
}
void Compute::recvHandle(infiDirectUserHandle shdl, int index, int arr) {
// --- B --- | --- C --- | --- A ---
if(arr == SENDA) {
sHandles[num_chare_x + num_chare_y + index] = shdl;
CkDirect_assocLocalBuffer(&sHandles[num_chare_x + num_chare_y + index], &A[thisIndex.z*subBlockDimXz*blockDimY], sizeof(float)*subBlockDimXz*blockDimY);
countA++;
}
if(arr == SENDB) {
sHandles[index] = shdl;
CkDirect_assocLocalBuffer(&sHandles[index], &B[thisIndex.x*subBlockDimYx*blockDimZ], sizeof(float)*subBlockDimYx*blockDimZ);
countB++;
}
if(arr == SENDC) {
sHandles[num_chare_x + index] = shdl;
CkDirect_assocLocalBuffer(&sHandles[num_chare_x + index], &C[index*subBlockDimXy*blockDimZ], sizeof(float)*subBlockDimXy*blockDimZ);
countC++;
}
if(countA == num_chare_z-1 && countB == num_chare_x-1 && countC == num_chare_y-1) {
countA = 0;
countB = 0;
countC = 0;
contribute(0, 0, CkReduction::concat, CkCallback(CkIndex_Main::setupDone(), mainProxy));
// mainProxy.setupDone();
}
}
/// @brief total feedback quantities for conservation check
///
/// Sums are contributed back to main chare.
/// @param bPreDist Is this before the feedback gets distributed. In
/// this case the "out" quantities need to be summed.
void
TreePiece::massMetalsEnergyCheck(int bPreDist, const CkCallback& cb)
{
double dTotals[5];
for(int j = 0; j < 5; j++)
dTotals[j] = 0.0;
for(unsigned int i = 1; i <= myNumParticles; ++i) {
GravityParticle *p = &myParticles[i];
dTotals[0] += p->mass;
if(p->isGas()) {
dTotals[1] += p->mass*p->fMetals();
dTotals[2] += p->mass*p->fMFracOxygen();
dTotals[3] += p->mass*p->fMFracIron();
dTotals[4] += p->mass*p->fESNrate();
}
if(p->isStar()) {
dTotals[1] += p->mass*p->fStarMetals();
dTotals[2] += p->mass*p->fStarMFracOxygen();
dTotals[3] += p->mass*p->fStarMFracIron();
if(bPreDist) { // sum up the quantities that will be distributed
dTotals[0] += p->fMSN();
dTotals[1] += p->fSNMetals();
dTotals[2] += p->fMOxygenOut();
dTotals[3] += p->fMIronOut();
dTotals[4] += p->fStarESNrate();
}
}
}
contribute(sizeof(dTotals), dTotals, CkReduction::sum_double, cb);
}
void CentralLB::ProcessAtSync()
{
#if CMK_LBDB_ON
if (reduction_started) return; // reducton in progress
CmiAssert(CmiNodeAlive(CkMyPe()));
if (CkMyPe() == cur_ld_balancer) {
start_lb_time = CkWallTimer();
}
#if (defined(_FAULT_MLOG_) || defined(_FAULT_CAUSAL_))
initMlogLBStep(thisgroup);
#endif
// build message
BuildStatsMsg();
#if USE_REDUCTION
// reduction to get total number of objects and comm
// so that processor 0 can pre-allocate load balancing database
int counts[2];
counts[0] = theLbdb->GetObjDataSz();
counts[1] = theLbdb->GetCommDataSz();
CkCallback cb(CkIndex_CentralLB::ReceiveCounts((CkReductionMsg*)NULL),
thisProxy[0]);
contribute(2*sizeof(int), counts, CkReduction::sum_int, cb);
reduction_started = 1;
#else
SendStats();
#endif
#endif
}
void Compute::resetArrays() {
int indexX = thisIndex.x;
int indexY = thisIndex.y;
int indexZ = thisIndex.z;
float tmp;
for(int i=indexZ*subBlockDimXz; i<(indexZ+1)*subBlockDimXz; i++)
for(int j=0; j<blockDimY; j++) {
tmp = (float)drand48();
while(tmp > MAX_LIMIT || tmp < (-1)*MAX_LIMIT)
tmp = (float)drand48();
A[i*blockDimY + j] = tmp;
}
for(int j=indexX*subBlockDimYx; j<(indexX+1)*subBlockDimYx; j++)
for(int k=0; k<blockDimZ; k++) {
tmp = (float)drand48();
while(tmp > MAX_LIMIT || tmp < (-1)*MAX_LIMIT)
tmp = (float)drand48();
B[j*blockDimZ + k] = tmp;
}
for(int i=0; i<blockDimX; i++)
for(int k=0; k<blockDimZ; k++) {
C[i*blockDimZ + k] = 0.0;
#if USE_CKDIRECT
tmpC[i*blockDimZ + k] = 0.0;
#endif
}
contribute(0, 0, CkReduction::concat, CkCallback(CkIndex_Main::resetDone(), mainProxy));
}
void ResumeFromSync ()
{
// printf("Jacobi[%d] calling resumeSync\n",thisIndex);
// if (thisIndex == 0)
// mainProxy.report ();
//CkPrintf("Coming in MAJ MAJ MAJ RRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRR ++++++++\n");
contribute(CkCallback(CkIndex_Main::report(NULL),mainProxy));
}
void getScannedVertexNum() {
CmiUInt8 numScannedVertices = 0;
typedef std::vector<BFSVertex>::iterator Iterator;
for (Iterator it = vertices.begin(); it != vertices.end(); it++)
numScannedVertices += it->getScannedVertexNum();
contribute(sizeof(CmiUInt8), &numScannedVertices, CkReduction::sum_long,
CkCallback(CkReductionTarget(TestDriver, done),
driverProxy));
}
FFTController::FFTController() {
first_time = true;
in_pointer = out_pointer = NULL;
geps = new GSPACE();
// TODO: A group dependency could probably solve this better
contribute(CkCallback(CkReductionTarget(Controller, fftControllerReady), controller_proxy));
}
allToAll() {
iter = 0;
recvCnt = 0;
msgs = new allToAllMsg*[numChares*msgCount];
for(int i = 0; i < msgCount*numChares; i++) {
msgs[i] = new (msgSize) allToAllMsg;
}
// reduction to the mainchare to signal that initialization is complete
contribute(CkCallback(CkReductionTarget(Main,allToAllReady), mainProxy));
}
void PsiCache::reportFTime() {
CkReduction::statisticsElement stats(total_time);
int tuple_size = 2;
CkReduction::tupleElement tuple_reduction[] = {
CkReduction::tupleElement(sizeof(double), &total_time, CkReduction::max_double),
CkReduction::tupleElement(sizeof(CkReduction::statisticsElement), &stats, CkReduction::statistics) };
CkReductionMsg* msg = CkReductionMsg::buildFromTuple(tuple_reduction, tuple_size);
msg->setCallback(CkCallback(CkIndex_Controller::reportFTime(NULL), controller_proxy));
contribute(msg);
}
void CentralLB::ReceiveMigration(LBMigrateMsg *m)
{
storedMigrateMsg = m;
#if CMK_MEM_CHECKPOINT
CkResetInLdb();
#endif
#if (defined(_FAULT_MLOG_) || defined(_FAULT_CAUSAL_))
restoreParallelRecovery(&resumeAfterRestoreParallelRecovery,(void *)this);
#else
CkCallback cb(CkIndex_CentralLB::ProcessReceiveMigration((CkReductionMsg*)NULL),
thisProxy);
contribute(0, NULL, CkReduction::max_int, cb);
#endif
}
Pingping(std::size_t index, uChareSet<Pingping, CProxy_Pingping, CBase_Pingping> *uchareset) :
uChare<Pingping, CProxy_Pingping, CBase_Pingping>(index, uchareset) {
CkPrintf("[uchare=%d, chare=%d,pe=%d]: created \n",
getId(), getuChareSet()->getId(), getuChareSet()->getPe());
pingDone = pongDone = false;
pingCounters.resize(N_uChares);
pingCounters.assign(N_uChares, -1);
pongCounters.resize(N_uChares);
pongCounters.assign(N_uChares, 999);
contribute(CkCallback(CkReductionTarget(Main, start), mainProxy));
}
void Compute::receiveC() {
int indexX = thisIndex.x;
int indexY = thisIndex.y;
int indexZ = thisIndex.z;
// copy C from tmpC to the correct location
for(int j=0; j<num_chare_y; j++) {
if( j != indexY) {
for(int i=0; i<subBlockDimXy; i++)
for(int k=0; k<blockDimZ; k++)
C[indexY*subBlockDimXy*blockDimZ + i*blockDimZ + k] += tmpC[j*subBlockDimXy*blockDimZ + i*blockDimZ + k];
}
}
/*char name[30];
sprintf(name, "%s_%d_%d_%d", "C", thisIndex.x, thisIndex.y, thisIndex.z);
FILE *fp = fopen(name, "w");
for(int i=0; i<subBlockDimXy; i++) {
for(int k=0; k<blockDimZ; k++)
fprintf(fp, "%f ", C[indexY*subBlockDimXy*blockDimZ + i*blockDimZ + k]);
fprintf(fp, "\n");
}
fclose(fp);
CkPrintf("%d_%d_%d\n", thisIndex.x, thisIndex.y, thisIndex.z);
for(int i=0; i<subBlockDimXy; i++) {
for(int k=0; k<blockDimZ; k++)
CkPrintf("%f ", C[indexY*subBlockDimXy*blockDimZ + i*blockDimZ + k]);
CkPrintf("\n");
}*/
// call ready for the buffers
for(int i=0; i<num_chare_x; i++)
if(i != indexX)
CkDirect_ready(&rHandles[i]);
for(int j=0; j<num_chare_y; j++)
if(j != indexY)
CkDirect_ready(&rHandles[num_chare_x + j]);
for(int k=0; k<num_chare_z; k++)
if(k != indexZ)
CkDirect_ready(&rHandles[num_chare_x + num_chare_y + k]);
// counters to keep track of how many messages have been received
countA = 0;
countB = 0;
countC = 0;
contribute(0, 0, CkReduction::concat, CkCallback(CkIndex_Main::done(), mainProxy));
// mainProxy.done();
}
void Workers::complete() {
int size = matrixSize * matrixSize * sizeof(ElementType);
memcpy(C, h_C, size);
#ifdef DEBUG
CkPrintf("[%d] A\n", thisIndex);
for (int i=0; i<matrixSize; i++) {
CkPrintf("[%d] ", thisIndex);
for (int j=0; j<matrixSize; j++) {
CkPrintf("%.2f ", A[i*matrixSize+j]);
}
CkPrintf("\n");
}
CkPrintf("[%d] B\n", thisIndex);
for (int i=0; i<matrixSize; i++) {
CkPrintf("[%d] ", thisIndex);
for (int j=0; j<matrixSize; j++) {
CkPrintf("%.2f ", B[i*matrixSize+j]);
}
CkPrintf("\n");
}
CkPrintf("[%d] C\n", thisIndex);
for (int i=0; i<matrixSize; i++) {
CkPrintf("[%d] ", thisIndex);
for (int j=0; j<matrixSize; j++) {
if(useCublas)
CkPrintf("%.2f ", C[j*matrixSize+i]);
else
CkPrintf("%.2f ", C[i*matrixSize+j]);
}
CkPrintf("\n");
}
CkPrintf("[%d] C-gold\n", thisIndex);
for (int i=0; i<matrixSize; i++) {
CkPrintf("[%d] ", thisIndex);
for (int j=0; j<matrixSize; j++) {
C[i*matrixSize + j] = 0;
for (int k=0; k<matrixSize; k++) {
C[i*matrixSize + j] += A[i*matrixSize +k] * B[k * matrixSize + j];
}
CkPrintf("%.2f ", C[i*matrixSize+j]);
}
CkPrintf("\n");
}
#endif
contribute(CkCallback(CkIndex_Main::finishWork(NULL), mainProxy));
}
void System::ImportFluidData()
{
#endif
assert(ImportFluidData_nRequestedUpdates == 0);
const int nremote = nimport_glb;
const int iremote_end = ptcl_act.size();
const int iremote_beg = iremote_end - nremote;
std::vector< std::pair<int, int> > request_list;
request_list.reserve(nremote);
for (int i = iremote_beg; i < iremote_end; i++)
if (ptcl_act[i]->is_ngb())
request_list.push_back(std::make_pair(ptcl_act[i]->chare(), i));
std::sort(request_list.begin(), request_list.end(), std_pair_first_sort());
const int nrequest = request_list.size();
CkVec< pair<int, int> > sites2request;
sites2request.reserve(nrequest);
request_list.push_back(std::make_pair(-1,-1));
for (int i = 0; i < nrequest; i++)
{
const int iElement = request_list[i].first;
const int iId = request_list[i].second;
sites2request.push_back(std::make_pair(ptcl_act[iId]->id(), iId));
assert(iElement >= 0);
assert(iElement < numElements);
assert(iElement != thisIndex);
if (iElement != request_list[i+1].first && sites2request.size() > 0)
{
ImportFluidData_nRequestedUpdates++;
systemProxy[iElement].ImportFluidData_request(sites2request, thisIndex);
sites2request.clear();
}
}
if (ImportFluidData_nRequestedUpdates == 0)
{
#if 1
ImportFluidData_completeCb.send();
#else
contribute(ImportFluidData_completeCb);
#endif
}
}
// Function that checks whether it must start the following step or wait until other messages are received
void Patch::checkNextStep(){
int i;
double timer;
if (updateFlag && incomingFlag) {
// resetting flags
updateFlag = false;
incomingFlag = false;
stepCount++;
// adding new elements
for (i = 0; i < incomingParticles.length(); i++)
particles.push_back(incomingParticles[i]);
incomingParticles.removeAll();
if (thisIndex.x == 0 && thisIndex.y == 0 && thisIndex.z == 0 && stepCount%NUM_STEPS == 0) {
timer = CmiWallTimer();
CkPrintf("Step %d Benchmark Time %f ms/step, Total Time Elapsed %f s\n", stepCount, ((timer - stepTime)/NUM_STEPS)*1000, timer);
stepTime = timer;
// if (stepCount == 300)
// traceBegin();
// if (stepCount == 400)
// traceEnd();
}
// if (stepCount == 300 && thisIndex.x*patchArrayDimY*patchArrayDimZ + thisIndex.y*patchArrayDimZ + thisIndex.z < 8)
// traceBegin();
// if (stepCount == 301 && thisIndex.x*patchArrayDimY*patchArrayDimZ + thisIndex.y*patchArrayDimZ + thisIndex.z < 8)
// traceEnd();
// checking for next step
if (stepCount >= finalStepCount) {
// CkPrintf("Final number of particles is %d on Patch [%d][%d][%d]\n", particles.length(), thisIndex.x, thisIndex.y, thisIndex.z);
print();
contribute(CkCallback(CkIndex_Main::allDone(), mainProxy));
} else {
if (perform_lb){
AtSync();
LBTurnInstrumentOff();
perform_lb=false;
}
else{
thisProxy(thisIndex.x, thisIndex.y, thisIndex.z).start();
//contribute(CkCallback(CkIndex_Main::lbBarrier(),mainProxy));
}
}
}
}
void
piPart::compute(int ns)
{
int i;
int count=0;
for (i= 0; i<ns; i++) {
double x = CrnDrand();
double y = CrnDrand();
if ((x*x + y*y) <= 1.0) count++;
}
#if ! USE_REDUCTION
CProxy_main mainproxy(mainhandle);
mainproxy.results(count);
#else
contribute(sizeof(int), (void *)&count, CkReduction::sum_int);
#endif
}
void CentralLB::endMigrationDone(int balancing){
DEBUGF(("[%d] CentralLB::endMigrationDone step %d\n",CkMyPe(),step()));
if (balancing && _lb_args.syncResume()) {
CkCallback cb(CkIndex_CentralLB::ResumeClients((CkReductionMsg*)NULL),
thisProxy);
contribute(0, NULL, CkReduction::sum_int, cb);
}
else{
if(CmiNodeAlive(CkMyPe())){
DEBUGF(("[%d] Sending ResumeClients balancing %d \n",CkMyPe(),balancing));
thisProxy [CkMyPe()].ResumeClients(balancing);
}
}
}
void check_and_compute ()
{
// if (--messages_due == 0)
// messages_due = 4;
compute ();
if (iterations % ldbTime == 4/* || iterations == 100*/)
{
// printf("MINOR[%d] itr=%d ----------------------------- ssssssssssssss\n",thisIndex,iterations);
AtSync ();
}
else
// mainProxy.report ();
contribute(CkCallback(CkIndex_Main::report(NULL),mainProxy));
}
Participant() {
int numPes = CkNumPes();
neighbors = new int[numPes];
for (int i = 0; i < numPes; i++) {
neighbors[i] = i;
}
// shuffle to prevent bottlenecks
for (int i = numPes-1; i >= 0; i--) {
int shuffleIndex = rand() % (i+1);
int temp = neighbors[i];
neighbors[i] = neighbors[shuffleIndex];
neighbors[shuffleIndex] = temp;
}
contribute(CkCallback(CkReductionTarget(Main, prepare), mainProxy));
}
/* Default constructor */
Patch::Patch(FileDataMsg* fdmsg) {
LBTurnInstrumentOff();
inbrs = numNbrs;
usesAtSync = CmiTrue;
updateCount = 0;
forceCount = 0;
stepCount = 0;
resumeCount = 0;
updateFlag = false;
incomingFlag = false;
perform_lb = false;
incomingParticles.resize(0);
// setMigratable(CmiFalse);
int i;
// Particle initialization
myNumParts = 0;
for(i=0; i < fdmsg->length; i++) {
particles.push_back(Particle());
particles[myNumParts].charge = fdmsg->charge[i];
particles[myNumParts].mass = fdmsg->mass[i];
particles[myNumParts].x = fdmsg->coords[i].x;
particles[myNumParts].y = fdmsg->coords[i].y;
particles[myNumParts].z = fdmsg->coords[i].z;
particles[myNumParts].vx = 0;
particles[myNumParts].vy = 0;
particles[myNumParts].vz = 0;
particles[myNumParts].fx = 0;
particles[myNumParts].fy = 0;
particles[myNumParts].fz = 0;
particles[myNumParts].id = (thisIndex.x*patchArrayDimX + thisIndex.y) * numParts / (patchArrayDimX*patchArrayDimY) + i;
particles[myNumParts].vdw_type = fdmsg->vdw_type[i];
myNumParts++;
}
delete fdmsg;
contribute(CkCallback(CkIndex_Main::startUpDone(), mainProxy));
}
void run() {
for (int i = 0 ; i < numelements; i++) {
if(thisIndex % 2 == 0 && thisIndex != numelements -1 ) {
myMsg* m = workerarray[thisIndex + 1].sendSmaller(val);
val = m->val;
delete m;
}
barrier();
if (thisIndex % 2 == 1 && thisIndex != numelements -1 ) {
myMsg* m = workerarray[thisIndex + 1].sendSmaller(val);
val = m->val;
delete m;
}
barrier();
}
contribute(CkCallback(CkIndex_Main::done(NULL), mainproxy));
}
void PsiCache::receivePsi(PsiMessage* msg) {
if (msg->spin_index != 0) {
CkAbort("Error: We don't support multiple spins yet!\n");
}
CkAssert(msg->k_index < K);
CkAssert(msg->state_index < L);
CkAssert(msg->size == psi_size);
if(msg->shifted==false){std::copy(msg->psi, msg->psi+psi_size, psis[msg->k_index][msg->state_index]);}
if(msg->shifted==true){std::copy(msg->psi, msg->psi+psi_size, psis_shifted[msg->k_index][msg->state_index]);}
delete msg;
// Once the cache has received all of it's data start the sliding pipeline
// sending of psis to P to start the accumulation of fxf'.
int expected_psis = K*L;
if(qindex == 0)
expected_psis += K*L;
if (++received_psis == expected_psis) {
//CkPrintf("[%d]: Cache filled\n", CkMyPe());
contribute(CkCallback(CkReductionTarget(Controller,cachesFilled), controller_proxy));
}
}
void CentralLB::MigrationDone(int balancing)
{
#if CMK_LBDB_ON
migrates_completed = 0;
migrates_expected = -1;
// clear load stats
if (balancing) theLbdb->ClearLoads();
// Increment to next step
theLbdb->incStep();
DEBUGF(("[%d] Incrementing Step %d \n",CkMyPe(),step()));
// if sync resume, invoke a barrier
#if (defined(_FAULT_MLOG_) || defined(_FAULT_CAUSAL_))
savedBalancing = balancing;
startLoadBalancingMlog(&resumeCentralLbAfterChkpt,(void *)this);
#endif
LBDatabase::Object()->MigrationDone(); // call registered callbacks
LoadbalanceDone(balancing); // callback
#if (!defined(_FAULT_MLOG_) && !defined(_FAULT_CAUSAL_))
// if sync resume invoke a barrier
if (balancing && _lb_args.syncResume()) {
CkCallback cb(CkIndex_CentralLB::ResumeClients((CkReductionMsg*)NULL),
thisProxy);
contribute(0, NULL, CkReduction::sum_int, cb);
}
else{
if(CmiNodeAlive(CkMyPe())){
thisProxy [CkMyPe()].ResumeClients(balancing);
}
}
#if CMK_GRID_QUEUE_AVAILABLE
CmiGridQueueDeregisterAll ();
CpvAccess(CkGridObject) = NULL;
#endif
#endif
#endif
}
