PetscErrorCode restrictMatVecType2(Mat R, Vec f, Vec c) {
TransferOpData *data;
PetscFunctionBegin;
iC(MatShellGetContext( R, (void **)&data));
MPI_Comm comm = data->comm;
Vec tmp = data->tmp;
PetscInt tmpSz;
PetscInt fSz;
iC(VecGetLocalSize(tmp,&tmpSz));
iC(VecGetLocalSize(f,&fSz));
scatterValues(f, tmp, fSz, tmpSz, data->sendSzR,
data->sendOffR, data->recvSzR, data->recvOffR, comm);
restrictMatVecType1(R, tmp, c);
PetscFunctionReturn(0);
}
//---------------------------------------------
double CpxCrvletPrtd::globalenergy()
{
double lclsum = 0;
vector< vector<int> >& c = _nx;
for(int s=0; s<c.size(); s++)
for(int w=0; w<c[s].size(); w++)
if(_owners[s][w]==mpirank())
lclsum += energy(_blocks[s][w]);
double glbsum = 0;
iC( MPI_Reduce((void*)(&lclsum), (void*)(&glbsum), 1, MPI_DOUBLE, MPI_SUM, 0, PETSC_COMM_WORLD) );
return glbsum;
}
// ----------------------------------------------------------------------
int Dense3d::evaluate(const DblNumVec& srcDen, DblNumVec& trgVal)
{
//-----------------------------------
iA(srcDen.m()==srcDOF()*(*_srcPos).n()); iA(trgVal.m()==trgDOF()*(*_trgPos).n());
int dim = this->dim();
int srcDOF = this->srcDOF();
int trgDOF = this->trgDOF();
/* Number of sources */
int numSrc = (*_srcPos).n();
/* Number of targets */
int numTrg = (*_trgPos).n();
DblNumMat inter(trgDOF, numSrc*srcDOF);
for(int i=0; i<numTrg; i++) {
DblNumMat onePosMat(dim, 1, false, (*_trgPos).clmdata(i));
DblNumVec oneValVec(trgDOF, false, trgVal.data()+trgDOF*i);
iC( _knl.kernel((*_srcPos), (*_srcNor), onePosMat, inter) );
iC( dgemv(1.0, inter, srcDen, 0.0, oneValVec) );
}
return 0;
}
// --------------------------------------------------------
LoadReadoutMenu::LoadReadoutMenu( const std::string& aId, swatch::core::ActionableObject& aActionable ) :
swatch::core::Command(aId, aActionable, xdata::UnsignedInteger())
{
::mp7::MP7Controller& lDriver = getActionable<MP7Processor>().driver();
const ::mp7::ReadoutCtrlNode& rc = lDriver.getReadout().getNode< ::mp7::ReadoutCtrlNode>("readout_control");
mBanks = rc.readNumBanks();
mModes = rc.readNumModes();
mCaptures = rc.readNumCaptures();
std::string bankStr, modeStr, capStr;
for( uint32_t iB(0); iB < mBanks; ++iB ) {
bankStr = "bank"+boost::lexical_cast<std::string>(iB)+":";
registerParameter(bankStr+"wordsPerBx", xdata::UnsignedInteger());
}
for( uint32_t iM(0); iM < mModes; ++iM ) {
modeStr = "mode"+boost::lexical_cast<std::string>(iM)+":";
registerParameter(modeStr+"eventSize", xdata::UnsignedInteger());
registerParameter(modeStr+"eventToTrigger", xdata::UnsignedInteger());
registerParameter(modeStr+"eventType", xdata::UnsignedInteger());
registerParameter(modeStr+"tokenDelay", xdata::UnsignedInteger());
for( uint32_t iC(0); iC < mCaptures; ++iC ) {
capStr = modeStr+"capture"+boost::lexical_cast<std::string>(iC)+":";
registerParameter(capStr+"enable", xdata::Boolean());
registerParameter(capStr+"id", xdata::UnsignedInteger());
registerParameter(capStr+"bankId", xdata::UnsignedInteger());
registerParameter(capStr+"length", xdata::UnsignedInteger());
registerParameter(capStr+"delay", xdata::UnsignedInteger());
registerParameter(capStr+"readoutLength", xdata::UnsignedInteger());
}
}
}
PetscErrorCode restrictMatVecType1(Mat R, Vec f, Vec c) {
PROF_MG_RESTRICT_BEGIN
TransferOpData *data;
iC(MatShellGetContext( R, (void **)&data));
unsigned int dof = data->dof;
//Overlap 75% of the independent computation with the first communication and
//25% with the second communication. In the first communication, we exchange
// fine grid ghosts. In the second, we exchange coarse grid ghosts (1/4 of
//fine grid, assuming uniform refinement). So, the
//first comm. is more expensive.
unsigned int fop = 75;
unsigned char* suppressedDOFc = data->suppressedDOFc;
unsigned char* suppressedDOFf = data->suppressedDOFf;
ot::DA * dac = data->dac;
ot::DA * daf = data->daf;
PetscInt cSz;
iC(VecGetLocalSize(c,&cSz));
unsigned int fopCnt = (fop*cSz)/(100*dof);
std::vector<ot::FineTouchedStatus >* fineTouchedFlags = data->fineTouchedFlags;
ot::FineTouchedStatus* fineTouchedFlagsArr;
PetscScalar *farr = NULL;
PetscScalar *carr = NULL;
daf->vecGetBuffer(f, farr, false, false, true, dof);//Read-only
daf->vecGetBuffer<ot::FineTouchedStatus >(*fineTouchedFlags,
fineTouchedFlagsArr, false, false, true, 1);//read-only
if(daf->iAmActive()) {
daf->ReadFromGhostsBegin<PetscScalar>(farr, dof);
//This communication can be avoided if we store it
daf->ReadFromGhostsBegin<ot::FineTouchedStatus>(fineTouchedFlagsArr, 1);
}
VecZeroEntries(c);
dac->vecGetBuffer(c, carr, false, false, false, dof);//Writable
if(dac->iAmActive()) {
//Note: If Coarse is Independent, then the corresponding Fine is also independent.
//Hence, overlapping comm with comp is possible.
//Order of the test condition is important. We want to store the info before checking loopCtr.
unsigned int loopCtr = 0;
if(suppressedDOFc || suppressedDOFf) {
for(dac->init<ot::DA_FLAGS::INDEPENDENT>(), daf->init<ot::DA_FLAGS::WRITABLE>();
( (daf->currWithInfo() == daf->currWithInfo()) &&
(dac->currWithInfo() < dac->end<ot::DA_FLAGS::INDEPENDENT>()) && (loopCtr < fopCnt) );
dac->next<ot::DA_FLAGS::INDEPENDENT>(), loopCtr++) {
INTERGRID_TRANSFER_LOOP_BLOCK(ITLB_SET_VALUE_SUPPRESSED_DOFS);
}//end Independent loop (overlapping with read from fine ghosts)
} else {
for(dac->init<ot::DA_FLAGS::INDEPENDENT>(), daf->init<ot::DA_FLAGS::WRITABLE>();
( (daf->currWithInfo() == daf->currWithInfo()) &&
(dac->currWithInfo() < dac->end<ot::DA_FLAGS::INDEPENDENT>()) && (loopCtr < fopCnt) );
dac->next<ot::DA_FLAGS::INDEPENDENT>(), loopCtr++) {
INTERGRID_TRANSFER_LOOP_BLOCK(ITLB_SET_VALUE_NO_SUPPRESSED_DOFS);
}//end Independent loop (overlapping with read from fine ghosts)
}
}
if(daf->iAmActive()) {
daf->ReadFromGhostsEnd<PetscScalar>(farr);
daf->ReadFromGhostsEnd<ot::FineTouchedStatus>(fineTouchedFlagsArr);
}
if(dac->iAmActive()) {
if(suppressedDOFc || suppressedDOFf) {
for(dac->init<ot::DA_FLAGS::W_DEPENDENT>(), daf->init<ot::DA_FLAGS::WRITABLE>();
dac->curr() < dac->end<ot::DA_FLAGS::W_DEPENDENT>(); dac->next<ot::DA_FLAGS::W_DEPENDENT>()) {
INTERGRID_TRANSFER_LOOP_BLOCK(ITLB_SET_VALUE_SUPPRESSED_DOFS);
}//end dependent loop
} else {
for(dac->init<ot::DA_FLAGS::W_DEPENDENT>(), daf->init<ot::DA_FLAGS::WRITABLE>();
dac->curr() < dac->end<ot::DA_FLAGS::W_DEPENDENT>(); dac->next<ot::DA_FLAGS::W_DEPENDENT>()) {
INTERGRID_TRANSFER_LOOP_BLOCK(ITLB_SET_VALUE_NO_SUPPRESSED_DOFS);
}//end dependent loop
}
}
if(dac->iAmActive()) {
dac->WriteToGhostsBegin<PetscScalar>(carr, dof);
}
if(dac->iAmActive()) {
//Continue Independent loop from where we left off.
if(suppressedDOFc || suppressedDOFf) {
for(dac->init<ot::DA_FLAGS::FROM_STORED>(), daf->init<ot::DA_FLAGS::FROM_STORED>();
dac->curr() < dac->end<ot::DA_FLAGS::INDEPENDENT>(); dac->next<ot::DA_FLAGS::INDEPENDENT>()) {
INTERGRID_TRANSFER_LOOP_BLOCK(ITLB_SET_VALUE_SUPPRESSED_DOFS);
}//end Independent loop (overlapping with write to coarse ghosts)
} else {
for(dac->init<ot::DA_FLAGS::FROM_STORED>(), daf->init<ot::DA_FLAGS::FROM_STORED>();
dac->curr() < dac->end<ot::DA_FLAGS::INDEPENDENT>(); dac->next<ot::DA_FLAGS::INDEPENDENT>()) {
INTERGRID_TRANSFER_LOOP_BLOCK(ITLB_SET_VALUE_NO_SUPPRESSED_DOFS);
}//end Independent loop (overlapping with write to coarse ghosts)
}
}
if(dac->iAmActive()) {
dac->WriteToGhostsEnd<PetscScalar>(carr, dof);
}
daf->vecRestoreBuffer(f, farr, false, false, true, dof);//Read-only
dac->vecRestoreBuffer(c, carr, false, false, false, dof);//Writable
daf->vecRestoreBuffer<ot::FineTouchedStatus >(*fineTouchedFlags,
fineTouchedFlagsArr, false, false, true, 1);//read-only
#ifdef PETSC_USE_LOG
PetscLogFlops(128*dof*(daf->getElementSize()));
#endif
PROF_MG_RESTRICT_END
}//restrict-3
int main(int argc, char **argv)
{
PetscInitialize(&argc, &argv, "elas.opt", help);
int rank;
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
int Ns = 32;
unsigned int dof = 3;
char problemName[PETSC_MAX_PATH_LEN];
char filename[PETSC_MAX_PATH_LEN];
double t0 = 0.0;
double dt = 0.1;
double t1 = 1.0;
double beta = 0.000001;
double gamma = 0.0; // percent of noise added ...
// double dtratio = 1.0;
DA da; // Underlying scalar DA - for scalar properties
DA da3d; // Underlying vector DA - for vector properties
Vec rho; // density - elemental scalar
Vec lambda; // Lame parameter - lambda - elemental scalar
Vec mu; // Lame parameter - mu - elemental scalar
Vec fibers; // Fiber orientations - nodal vector (3-dof)
Vec fibersElemental; // for IO. will be destroyed. - elemental vector (3-dof)
std::vector<Vec> tau; // the scalar activation - nodal scalar
// Initial conditions
Vec initialDisplacement;
Vec initialVelocity;
timeInfo ti;
// get Ns
CHKERRQ ( PetscOptionsGetInt(0,"-Ns",&Ns,0) );
CHKERRQ ( PetscOptionsGetScalar(0,"-t0",&t0,0) );
CHKERRQ ( PetscOptionsGetScalar(0,"-t1",&t1,0) );
CHKERRQ ( PetscOptionsGetScalar(0,"-dt",&dt,0) );
CHKERRQ ( PetscOptionsGetScalar(0,"-beta",&beta,0) );
CHKERRQ ( PetscOptionsGetString(PETSC_NULL,"-pn",problemName,PETSC_MAX_PATH_LEN-1,PETSC_NULL));
if (!rank) {
std::cout << "Problem size is " << Ns+1 << " spatially and NT = " << (int)ceil(1.0/dt) << std::endl;
}
// Time info for timestepping
ti.start = t0;
ti.stop = t1;
ti.step = dt;
// create DA
CHKERRQ ( DACreate3d ( PETSC_COMM_WORLD, DA_NONPERIODIC, DA_STENCIL_BOX,
Ns+1, Ns+1, Ns+1, PETSC_DECIDE, PETSC_DECIDE, PETSC_DECIDE,
1, 1, 0, 0, 0, &da) );
CHKERRQ ( DACreate3d ( PETSC_COMM_WORLD, DA_NONPERIODIC, DA_STENCIL_BOX,
Ns+1, Ns+1, Ns+1, PETSC_DECIDE, PETSC_DECIDE, PETSC_DECIDE,
dof, 1, 0, 0, 0, &da3d) );
elasMass *Mass = new elasMass(feMat::PETSC); // Mass Matrix
elasStiffness *Stiffness = new elasStiffness(feMat::PETSC); // Stiffness matrix
raleighDamping *Damping = new raleighDamping(feMat::PETSC); // Damping Matrix
cardiacDynamic *Force = new cardiacDynamic(feVec::PETSC); // Force Vector
// create vectors
CHKERRQ( DACreateGlobalVector(da, &rho) );
CHKERRQ( DACreateGlobalVector(da, &mu) );
CHKERRQ( DACreateGlobalVector(da, &lambda) );
CHKERRQ( DACreateGlobalVector(da3d, &initialDisplacement) );
CHKERRQ( DACreateGlobalVector(da3d, &initialVelocity) );
// Set initial conditions
CHKERRQ( VecSet ( initialDisplacement, 0.0) );
CHKERRQ( VecSet ( initialVelocity, 0.0) );
VecZeroEntries( mu );
VecZeroEntries( lambda );
VecZeroEntries( rho );
int x, y, z, m, n, p;
int mx,my,mz, xne, yne, zne;
CHKERRQ( DAGetCorners(da, &x, &y, &z, &m, &n, &p) );
CHKERRQ( DAGetInfo(da,0, &mx, &my, &mz, 0,0,0,0,0,0,0) );
if (x+m == mx) {
xne=m-1;
} else {
xne=m;
}
if (y+n == my) {
yne=n-1;
} else {
yne=n;
}
if (z+p == mz) {
zne=p-1;
} else {
zne=p;
}
double acx,acy,acz;
double hx = 1.0/((double)Ns);
// SET MATERIAL PROPERTIES ...
// @todo - Write routines to read/write in Parallel
// allocate for temporary buffers ...
unsigned int elemSize = Ns*Ns*Ns;
// std::cout << "Elem size is " << elemSize << std::endl;
unsigned int nodeSize = (Ns+1)*(Ns+1)*(Ns+1);
unsigned char *tmp_mat = new unsigned char[elemSize];
double *tmp_tau = new double[dof*elemSize];
// generate filenames & read in the raw arrays first ...
std::ifstream fin;
sprintf(filename, "%s.%d.img", problemName, Ns);
fin.open(filename, std::ios::binary); fin.read((char *)tmp_mat, elemSize); fin.close();
// Set Elemental material properties
PetscScalar ***muArray, ***lambdaArray, ***rhoArray;
CHKERRQ(DAVecGetArray(da, mu, &muArray));
CHKERRQ(DAVecGetArray(da, lambda, &lambdaArray));
CHKERRQ(DAVecGetArray(da, rho, &rhoArray));
// assign material properties ...
// myo, tissue
// nu = 0.49, 0.45
// E = 10000, 1000
// rho = 1.0, 0.1
// std::cout << "Setting Elemental properties." << std::endl;
// loop through all elements ...
for (int k=z; k<z+zne; k++) {
for (int j=y; j<y+yne; j++) {
for (int i=x; i<x+xne; i++) {
int indx = k*Ns*Ns + j*Ns + i;
if ( tmp_mat[indx] ) {
muArray[k][j][i] = 344.82; //3355.7;
lambdaArray[k][j][i] = 3103.448;// 164429.53;
rhoArray[k][j][i] = 1.0;
} else {
muArray[k][j][i] = 344.82;
lambdaArray[k][j][i] = 3103.448;
rhoArray[k][j][i] = 1.0;
}
} // end i
} // end j
} // end k
// std::cout << "Finished Elemental loop." << std::endl;
CHKERRQ( DAVecRestoreArray ( da, mu, &muArray ) );
CHKERRQ( DAVecRestoreArray ( da, lambda, &lambdaArray ) );
CHKERRQ( DAVecRestoreArray ( da, rho, &rhoArray ) );
// std::cout << "Finished restoring arrays" << std::endl;
// delete temporary buffers
delete [] tmp_mat;
// Now set the activation ...
unsigned int numSteps = (unsigned int)(ceil(( ti.stop - ti.start)/ti.step));
// tau = (Vec *) new char*[numSteps+1];
// std::cout << "Numsteps is " << numSteps << std::endl;
Vec tauVec, tmpTau;
CHKERRQ( DACreateGlobalVector(da3d, &tmpTau) );
#ifdef __DEBUG__
if (!rank) {
std::cout << x << ", " << y << ", " << z << " + " << xne << ", " << yne << ", " << zne << std::endl;
}
#endif
PetscScalar ***tauArray;
double tauNorm;
for (unsigned int t=0; t<numSteps+1; t++) {
CHKERRQ( DACreateGlobalVector(da3d, &tauVec) );
CHKERRQ( VecSet( tmpTau, 0.0));
CHKERRQ(DAVecGetArray(da3d, tmpTau, &tauArray));
// std::cout << "Setting force vectors" << std::endl;
sprintf(filename, "%s.%d.%.3d.fld", problemName, Ns, t);
// std::cout << "Reading force file " << filename << std::endl;
fin.open(filename); fin.read((char *)tmp_tau, dof*elemSize*sizeof(double)); fin.close();
for (int k = z; k < z + zne ; k++) {
for (int j = y; j < y + yne; j++) {
for (int i = x; i < x + xne; i++) {
int indx = dof*(k*Ns*Ns + j*Ns + i);
tauArray[k][j][dof*i] = tmp_tau[indx];
tauArray[k][j][dof*i+1] = tmp_tau[indx+1];
tauArray[k][j][dof*i+2] = tmp_tau[indx+2];
}
}
}
// std::cout << CYN"\tFinished elemental loop"NRM << std::endl;
CHKERRQ( DAVecRestoreArray ( da3d, tmpTau, &tauArray ) );
// std::cout << "Converting to Nodal Vector" << std::endl;
// VecNorm(tmpTau, NORM_2, &tauNorm);
// tauNorm = tauNorm/pow(Ns,1.5);
// std::cout << "Elemental Norm is " << tauNorm << std::endl;
// std::cout << rank << " Converting to Nodal" << std::endl;
elementToNode(da3d, tmpTau, tauVec);
/*
VecNorm(tauVec, NORM_2, &tauNorm);
tauNorm = tauNorm/pow(Ns,1.5);
std::cout << "Nodal Norm is " << tauNorm << std::endl;
*/
// std::cout << rank << " Done converting to Nodal Vector" << std::endl;
tau.push_back(tauVec);
}
//if (!rank) {
// std::cout << "Finished setting forces" << std::endl;
// }
// CHKERRQ( VecDestroy( tmpTau ) );
delete [] tmp_tau;
// DONE - SET MATERIAL PROPERTIES ...
// Setup Matrices and Force Vector ...
Mass->setProblemDimensions(1.0, 1.0, 1.0);
Mass->setDA(da3d);
Mass->setDof(dof);
Mass->setDensity(rho);
Stiffness->setProblemDimensions(1.0, 1.0, 1.0);
Stiffness->setDA(da3d);
Stiffness->setDof(dof);
Stiffness->setLame(lambda, mu);
Damping->setAlpha(0.0);
Damping->setBeta(0.00075);
Damping->setMassMatrix(Mass);
Damping->setStiffnessMatrix(Stiffness);
Damping->setDA(da3d);
Damping->setDof(dof);
// Force Vector
Force->setProblemDimensions(1.0,1.0,1.0);
Force->setDA(da3d);
// Force->setActivationVec(tau);
// Force->setFiberOrientations(fibers);
Force->setFDynamic(tau);
Force->setTimeInfo(&ti);
// Newmark time stepper ...
newmark *ts = new newmark;
ts->setMassMatrix(Mass);
ts->setDampingMatrix(Damping);
ts->setStiffnessMatrix(Stiffness);
ts->damp(false);
ts->setTimeFrames(1);
ts->setForceVector(Force);
ts->setInitialDisplacement(initialDisplacement);
ts->setInitialVelocity(initialVelocity);
ts->storeVec(true);
ts->setTimeInfo(&ti);
ts->setAdjoint(false); // set if adjoint or forward
ts->init(); // initialize IMPORTANT
if (!rank)
std::cout << RED"Starting Newmark Solve"NRM << std::endl;
ts->solve();// solve
if (!rank)
std::cout << GRN"Done Newmark"NRM << std::endl;
std::vector<Vec> solvec = ts->getSolution();
/* Set very initial guess for the inverse problem*/
/*
PetscRandom rctx;
PetscRandomCreate(PETSC_COMM_WORLD,&rctx);
PetscRandomSetFromOptions(rctx);
VecSetRandom(guess,rctx);
VecNorm(guess,NORM_INFINITY,&norm);
PetscPrintf(0,"guess norm = %g\n",norm);
*/
double errnorm;
double exsolnorm;
Vec guess;
Vec truth;
Vec Err;
concatenateVecs(solvec, guess);
concatenateVecs(tau, truth);
iC(VecNorm(truth, NORM_2, &exsolnorm));
/*
std::cout << "Forward solver solution size is " << solvec.size() << std::endl;
std::cout << "Forward solver solution norm is " << exsolnorm << std::endl;
*/
VecZeroEntries(guess);
// Inverse solver set up
hyperbolicInverse *hyperInv = new hyperbolicInverse;
hyperInv->setForwardInitialConditions(initialDisplacement, initialVelocity);
hyperInv->setTimeStepper(ts); // set the timestepper
hyperInv->setInitialGuess(guess);// set the initial guess
// hyperInv->setInitialGuess(truth);// set the initial guess
hyperInv->setRegularizationParameter(beta); // set the regularization paramter
hyperInv->setObservations(solvec); // set the data for the problem
hyperInv->init(); // initialize the inverse solver
hyperInv->solve(); // solve
hyperInv->getCurrentControl(guess); // get the solution
// see the error in the solution relative to the actual solution
VecDuplicate(truth, &Err);
iC(VecZeroEntries(Err));
iC(VecWAXPY(Err, -1.0, guess, truth));
iC(VecNorm(Err, NORM_2, &errnorm));
PetscPrintf(0,"errr in inverse = %g\n", errnorm/exsolnorm);
PetscFinalize();
}
// --------------------------------------------------------
core::Command::State
LoadReadoutMenu::code(const ::swatch::core::XParameterSet& aParams)
{
::mp7::ReadoutMenu lMenu(mBanks, mModes, mCaptures);
// Bank IDs
std::string bankStr, modeStr, capStr;
for( uint32_t iB(0); iB < mBanks; ++iB ) {
bankStr = "bank"+boost::lexical_cast<std::string>(iB)+":";
const xdata::UnsignedInteger& bxOffset = aParams.get<xdata::UnsignedInteger>(bankStr+"wordsPerBx");
if ( !bxOffset.isNaN()) lMenu.bank(iB).wordsPerBx = bxOffset.value_;
}
for( uint32_t iM(0); iM < mModes; ++iM ) {
modeStr = "mode"+boost::lexical_cast<std::string>(iM)+":";
const xdata::UnsignedInteger& eventSize = aParams.get<xdata::UnsignedInteger>(modeStr+"eventSize");
const xdata::UnsignedInteger& eventToTrigger = aParams.get<xdata::UnsignedInteger>(modeStr+"eventToTrigger");
const xdata::UnsignedInteger& eventType = aParams.get<xdata::UnsignedInteger>(modeStr+"eventType");
const xdata::UnsignedInteger& tokenDelay = aParams.get<xdata::UnsignedInteger>(modeStr+"tokenDelay");
::mp7::ReadoutMenu::Mode& lMode = lMenu.mode(iM);
if ( !eventSize.isNaN()) lMode.eventSize = eventSize.value_;
if ( !eventToTrigger.isNaN()) lMode.eventToTrigger = eventToTrigger.value_;
if ( !eventType.isNaN()) lMode.eventType = eventType.value_;
if ( !tokenDelay.isNaN()) lMode.tokenDelay = tokenDelay.value_;
LOG(swatch::logger::kWarning) << lMode;
for( uint32_t iC(0); iC < mCaptures; ++iC ) {
capStr = modeStr+"capture"+boost::lexical_cast<std::string>(iC)+":";
const xdata::Boolean& enable = aParams.get<xdata::Boolean>(capStr+"enable");
const xdata::UnsignedInteger& id = aParams.get<xdata::UnsignedInteger>(capStr+"id");
const xdata::UnsignedInteger& bankId = aParams.get<xdata::UnsignedInteger>(capStr+"bankId");
const xdata::UnsignedInteger& length = aParams.get<xdata::UnsignedInteger>(capStr+"length");
const xdata::UnsignedInteger& delay = aParams.get<xdata::UnsignedInteger>(capStr+"delay");
const xdata::UnsignedInteger& readoutLength = aParams.get<xdata::UnsignedInteger>(capStr+"readoutLength");
::mp7::ReadoutMenu::Capture& lCapture = lMode[iC];
if ( !enable.isNaN()) lCapture.enable = enable.value_;
if ( !id.isNaN()) lCapture.id = id.value_;
if ( !bankId.isNaN()) lCapture.bankId = bankId.value_;
if ( !length.isNaN()) lCapture.length = length.value_;
if ( !delay.isNaN()) lCapture.delay = delay.value_;
if ( !readoutLength.isNaN()) lCapture.readoutLength = readoutLength.value_;
}
}
::mp7::MP7Controller& driver = getActionable<MP7Processor>().driver();
const ::mp7::ReadoutCtrlNode& rc = driver.getReadout().getNode< ::mp7::ReadoutCtrlNode >("readout_control");
std::map<uint32_t,uint32_t> lEventSizes = driver.computeEventSizes(lMenu);
for( uint32_t iM(0); iM < mModes; ++iM ) {
::mp7::ReadoutMenu::Mode& lMode = lMenu.mode(iM);
if ( lMode.eventSize == 0xfffff ) continue;
lMode.eventSize = lEventSizes.at(iM);
LOG(swatch::logger::kInfo) << "Mode " << iM << " event size set to " << lMode.eventSize;
}
LOG(swatch::logger::kInfo) << lMenu;
rc.configureMenu(lMenu);
return State::kDone;
}
//------------------------------------------------------------------------------------
int fdct3d_inverse(int N1, int N2, int N3, int nbscales, int nbdstz_coarse,
CpxCrvletPrtd& C, CpxNumTnsBlkd& W,
CpxNumTnsBlkd& X)
{
//check the size of c,w, make sure it is okay
time_t tm0, tm1; tm0 = time(NULL);
int b = W.b();
int e = W.e(); int f = W.f(); int g = W.g();
int mpirank; MPI_Comm_rank(MPI_COMM_WORLD, &mpirank);
iC( MPI_Barrier(MPI_COMM_WORLD) ); //iC( PetscPrintf(MPI_COMM_WORLD, "%d inverse 0\n", mpirank) ); iC( MPI_Barrier(MPI_COMM_WORLD) );
//-------------------------------------------
//1. fft on X
X = W;
BolNumTns newtnsexists(e,f,g);
IntNumTns newtnsowners(e,f,g);
fdct3d_partition_cpxnumtnsblkd_z(N1,N2,N3,b, newtnsexists,newtnsowners);
//scatter x to contain z slices
iC( X.scatter(newtnsexists) ); iC( MPI_Barrier(MPI_COMM_WORLD) ); tm1 = time(NULL); //iC( PetscPrintf(MPI_COMM_WORLD, "inv x scatter %f\n", difftime(tm1,tm0)) ); tm0 = tm1;
//shift x's owner to z slices
iC( X.shift(newtnsowners) ); iC( MPI_Barrier(MPI_COMM_WORLD) ); tm1 = time(NULL); //iC( PetscPrintf(MPI_COMM_WORLD, "inv x shift %f\n", difftime(tm1,tm0)) ); tm0 = tm1;
//discard x's nonowners
iC( X.discard() ); iC( MPI_Barrier(MPI_COMM_WORLD) ); tm1 = time(NULL); //iC( PetscPrintf(MPI_COMM_WORLD, "inv x discard %f\n", difftime(tm1,tm0)) ); tm0 = tm1;
//ifft
iC( fdct3d_fft(X) ); iC( MPI_Barrier(MPI_COMM_WORLD) ); tm1 = time(NULL); //iC( PetscPrintf(MPI_COMM_WORLD, "inv x fft %f\n", difftime(tm1,tm0)) ); tm0 = tm1;
//scale x with POU
DblOffVec big1(N1); fdct3d_lowpass(2.0*N1/3, big1);
DblOffVec big2(N2); fdct3d_lowpass(2.0*N2/3, big2);
DblOffVec big3(N3); fdct3d_lowpass(2.0*N3/3, big3);
IntNumTns& Xowners = X.owners();
for(int i=0; i<e; i++) for(int j=0; j<f; j++) for(int k=0; k<g; k++) {
if(Xowners(i,j,k)==mpirank) {
CpxNumTns& Xblk = X.block(i,j,k);
int istt = i*b-N1/2; int jstt = j*b-N2/2; int kstt = k*b-N3/2;
for(int ioff=0; ioff<b; ioff++) for(int joff=0; joff<b; joff++) for(int koff=0; koff<b; koff++) {
double pou = big1(ioff+istt) * big2(joff+jstt)*big3(koff+kstt);
Xblk(ioff, joff, koff) *= sqrt(1-pou*pou);
}
}
}
//-------------------------------------------
//2. compute wedges
int L = nbscales;
//setup c, 1,2,3, 6*np/8 processors are computing. 0 processor contains also the center wedge
vector< vector<bool> > newcrvexists;
vector< vector<int > > newcrvowners;
fdct3d_partition_cpxcrvletprtd(N1,N2,N3, nbscales, nbdstz_coarse, newcrvexists, newcrvowners);
//vector< vector<double> > fxs, fys, fzs;
//vector< vector<int > > nxs, nys, nzs;
//fdct3d_param(N1,N2,N3, nbscales,nbdstz_coarse, fxs,fys,fzs, nxs,nys,nzs); //LEXING: might not be necessary
//find out the required blocks from x for each processor
fdct3d_dependency(N1,N2,N3,b, nbscales,nbdstz_coarse, newcrvowners, newtnsexists);
//expand x according to c's request
iC( X.expand(newtnsexists) ); iC( MPI_Barrier(MPI_COMM_WORLD) ); tm1 = time(NULL); //iC( PetscPrintf(MPI_COMM_WORLD, "inv x expand %f\n", difftime(tm1,tm0)) ); tm0 = tm1;
//compute contribution to x from c
{
int s = 0;
double L1 = 2.0*N1/3.0 / pow2(L-2-s); double L2 = 2.0*N2/3.0 / pow2(L-2-s); double L3 = 2.0*N3/3.0 / pow2(L-2-s);
fdct3d_inverse_center(N1,N2,N3,b, L1,L2,L3, s, C, X);
}
for(int s=1; s<nbscales-1; s++) {
double L1 = 2.0*N1/3.0 / pow2(L-2-s); double L2 = 2.0*N2/3.0 / pow2(L-2-s); double L3 = 2.0*N3/3.0 / pow2(L-2-s);
int nd = nbdstz_coarse * pow2(s/2);
fdct3d_inverse_angles(N1,N2,N3,b, L1,L2,L3, s, nd, C, X);
}
iC( MPI_Barrier(MPI_COMM_WORLD) ); tm1 = time(NULL); //iC( PetscPrintf(MPI_COMM_WORLD, "inv c compute %f\n", difftime(tm1,tm0)) ); tm0 = tm1;
//combine x
iC( X.combine() ); iC( MPI_Barrier(MPI_COMM_WORLD) ); tm1 = time(NULL); //iC( PetscPrintf(MPI_COMM_WORLD, "inv x combine %f\n", difftime(tm1,tm0)) ); tm0 = tm1;
//-------------------------------------------
//3. ifft on X
iC( fdct3d_ifft(X) ); iC( MPI_Barrier(MPI_COMM_WORLD) ); tm1 = time(NULL); //iC( PetscPrintf(MPI_COMM_WORLD, "inv x ifft %f\n", difftime(tm1,tm0)) ); tm0 = tm1;
//done
return 0;
}
int CpxCrvletPrtd::scatter(vector< vector<bool> >& newexists)
{
//LEXING: usually only called once
vector< vector<int> >& c = _nx;
//1. the global vector
vector<int> glblszs(mpisize(), 0);
int glbnum = 0;
for(int s=0; s<c.size(); s++) for(int w=0; w<c[s].size(); w++) {
int pi = _owners[s][w];
glblszs[pi] += _sizes[s][w];
glbnum += _sizes[s][w];
}
vector<int> glbaccs(mpisize(), 0);
int tmp = 0;
for(int pi=0; pi<mpisize(); pi++) {
glbaccs[pi] = tmp;
tmp += glblszs[pi];
}
vector< vector<int> > glbstts(c); //not cleared, but okay
for(int s=0; s<c.size(); s++) for(int w=0; w<c[s].size(); w++) {
int pi = _owners[s][w];
glbstts[s][w] = glbaccs[pi];
glbaccs[pi] += _sizes[s][w];
}
int lclsum = 0;
vector<int> l2gmap;
for(int s=0; s<c.size(); s++) for(int w=0; w<c[s].size(); w++) {
if(newexists[s][w]==true && _exists[s][w]==false) {
lclsum += _sizes[s][w];
for(int g=0; g<_sizes[s][w]; g++)
l2gmap.push_back( glbstts[s][w] + g );
}
}
iA(l2gmap.size()==lclsum);
IS lclis; iC( ISCreateStride(PETSC_COMM_SELF, l2gmap.size(), 0, 1, &lclis) );
IS glbis; iC( ISCreateGeneral(PETSC_COMM_WORLD, l2gmap.size(), &(l2gmap[0]), &glbis) );
l2gmap.clear(); //SAVE SPACE
//2. allocate a global vector, and copy data
Vec glbvec; iC( VecCreateMPI(PETSC_COMM_WORLD, glblszs[mpirank()], PETSC_DETERMINE, &glbvec) );
double* glbarr; iC( VecGetArray(glbvec, &glbarr) );
double* glbptr = glbarr;
for(int s=0; s<c.size(); s++) for(int w=0; w<c[s].size(); w++) {
int pi = _owners[s][w];
if(pi==mpirank()) {
double* tmpptr = (double*)(_blocks[s][w].data());
for(int g=0; g<_sizes[s][w]; g++) {
*glbptr = tmpptr[g]; glbptr++;
}
}
}
iC( VecRestoreArray(glbvec, &glbarr) );
Vec lclvec; iC( VecCreateSeq(PETSC_COMM_SELF, lclsum, &lclvec) );
//3. vec scatter
VecScatter sc; iC( VecScatterCreate(glbvec, glbis, lclvec, lclis, &sc) );
iC( ISDestroy(lclis) ); iC( ISDestroy(glbis) ); //SAVE SPACE
iC( VecScatterBegin(glbvec, lclvec, INSERT_VALUES, SCATTER_FORWARD, sc) );
iC( VecScatterEnd( glbvec, lclvec, INSERT_VALUES, SCATTER_FORWARD, sc) );
iC( VecScatterDestroy(sc) ); //SAVE SPACE
iC( VecDestroy(glbvec) );
//4. store
double* lclarr; iC( VecGetArray(lclvec, &lclarr) );
double* lclptr = lclarr;
for(int s=0; s<c.size(); s++) for(int w=0; w<c[s].size(); w++) {
if(newexists[s][w]==true && _exists[s][w]==false) {
_blocks[s][w].resize(_nx[s][w], _ny[s][w], _nz[s][w]);
double* tmpptr = (double*)(_blocks[s][w].data());
for(int g=0; g<_sizes[s][w]; g++) {
tmpptr[g] = *lclptr; lclptr++;
}
_exists[s][w] = true; //VERY IMPORTANT
}
}
iC( VecRestoreArray(lclvec, &lclarr) );
iC( VecDestroy(lclvec) );
return 0;
}
int CpxCrvletPrtd::combine()
{
//LEXING: usually only called once
vector< vector<int> >& c = _nx;
//1. the global vector
vector<int> glblszs(mpisize(), 0);
int glbnum = 0;
for(int s=0; s<c.size(); s++) for(int w=0; w<c[s].size(); w++) {
int pi = _owners[s][w];
glblszs[pi] += _sizes[s][w];
glbnum += _sizes[s][w];
}
vector<int> glbaccs(mpisize(), 0);
int tmp = 0;
for(int pi=0; pi<mpisize(); pi++) {
glbaccs[pi] = tmp;
tmp += glblszs[pi];
}
vector< vector<int> > glbstts(c);
for(int s=0; s<c.size(); s++) for(int w=0; w<c[s].size(); w++) {
int pi = _owners[s][w];
glbstts[s][w] = glbaccs[pi];
glbaccs[pi] += _sizes[s][w];
}
int lclsum = 0;
vector<int> l2gmap;
for(int s=0; s<c.size(); s++) for(int w=0; w<c[s].size(); w++) {
if(_exists[s][w]==true && _owners[s][w]!=mpirank()) {
lclsum += _sizes[s][w];
for(int g=0; g<_sizes[s][w]; g++)
l2gmap.push_back( glbstts[s][w] + g );
}
}
iA(l2gmap.size()==lclsum);
IS lclis; iC( ISCreateStride(PETSC_COMM_SELF, l2gmap.size(), 0, 1, &lclis) );
IS glbis; iC( ISCreateGeneral(PETSC_COMM_WORLD, l2gmap.size(), &(l2gmap[0]), &glbis) );
l2gmap.clear(); //SAVE SPACE
//2. allocate a global vector and a local vector, put data in local
Vec glbvec; iC( VecCreateMPI(PETSC_COMM_WORLD, glblszs[mpirank()], PETSC_DETERMINE, &glbvec) );
Vec lclvec; iC( VecCreateSeq(PETSC_COMM_SELF, lclsum, &lclvec) );
double* lclarr; iC( VecGetArray(lclvec, &lclarr) );
double* lclptr = lclarr;
for(int s=0; s<c.size(); s++) for(int w=0; w<c[s].size(); w++) {
if(_exists[s][w]==true && _owners[s][w]!=mpirank()) {
double* tmpptr = (double*)(_blocks[s][w].data());
for(int g=0; g<_sizes[s][w]; g++) {
*lclptr = tmpptr[g]; lclptr++;
}
}
}
iC( VecRestoreArray(lclvec, &lclarr) );
//3. vec scatter
VecScatter sc; iC( VecScatterCreate(glbvec, glbis, lclvec, lclis, &sc) );
iC( ISDestroy(lclis) ); iC( ISDestroy(glbis) ); //SAVE SPACE
iC( VecScatterBegin(glbvec, lclvec, ADD_VALUES, SCATTER_REVERSE, sc) );
iC( VecScatterEnd( glbvec, lclvec, ADD_VALUES, SCATTER_REVERSE, sc) );
iC( VecScatterDestroy(sc) ); //SAVE SPACE
iC( VecDestroy(lclvec) );
//4. store
double* glbarr; iC( VecGetArray(glbvec, &glbarr) );
double* glbptr = glbarr;
for(int s=0; s<c.size(); s++) for(int w=0; w<c[s].size(); w++) {
int pi = _owners[s][w];
if(pi==mpirank()) {
double* tmpptr = (double*)(_blocks[s][w].data());
for(int g=0; g<_sizes[s][w]; g++) {
tmpptr[g] += *glbptr; glbptr++; //LEXING: += is very important
}
}
}
iC( VecRestoreArray(glbvec, &glbarr) );
iC( VecDestroy(glbvec) );
//IMPORTANT
for(int s=0; s<c.size(); s++) for(int w=0; w<c[s].size(); w++) {
if(_owners[s][w]!=mpirank()) {
_blocks[s][w].resize(0,0,0);
_exists[s][w] = false;
}
}
return 0;
}
int main(int argc, char ** argv ) {
int size, rank;
bool incCorner = 1;
unsigned int dim=3;
unsigned int maxDepth=29;
bool compressLut=true;
std::vector<ot::TreeNode> balOct;
double mgLoadFac = 2.0;
unsigned int regLev = 2;
PetscInitialize(&argc, &argv, "options", help);
ot::RegisterEvents();
ot::DAMG_Initialize(MPI_COMM_WORLD);
#ifdef PETSC_USE_LOG
PetscClassId classid;
PetscClassIdRegister("Dendro",&classid);
PetscLogEventRegister("matProp",classid, &matPropEvent);
PetscLogEventRegister("ODAmatDiag",classid, &Jac1DiagEvent);
PetscLogEventRegister("ODAmatMult",classid, &Jac1MultEvent);
PetscLogEventRegister("ODAmatDiagFinest",classid, &Jac1FinestDiagEvent);
PetscLogEventRegister("ODAmatMultFinest",classid, &Jac1FinestMultEvent);
PetscLogEventRegister("OMGmatDiag-2",classid, &Jac2DiagEvent);
PetscLogEventRegister("OMGmatMult-2",classid, &Jac2MultEvent);
PetscLogEventRegister("OMGmatDiagFinest-2",classid, &Jac2FinestDiagEvent);
PetscLogEventRegister("OMGmatMultFinest-2",classid, &Jac2FinestMultEvent);
PetscLogEventRegister("OMGmatDiag-3",classid, &Jac3DiagEvent);
PetscLogEventRegister("OMGmatMult-3",classid, &Jac3MultEvent);
PetscLogEventRegister("OMGmatDiagFinest-3",classid, &Jac3FinestDiagEvent);
PetscLogEventRegister("OMGmatMultFinest-3",classid, &Jac3FinestMultEvent);
int stages[1];
PetscLogStageRegister("Solve",&stages[0]);
#endif
MPI_Comm_size(MPI_COMM_WORLD,&size);
MPI_Comm_rank(MPI_COMM_WORLD,&rank);
if(argc > 1) {
regLev = atoi(argv[1]);
}
if(argc > 2) {
maxDepth = atoi(argv[2]);
}
if(argc > 3) {
dim = atoi(argv[3]);
}
if(argc > 4) { incCorner = (bool)(atoi(argv[4]));}
if(argc > 5) { compressLut = (bool)(atoi(argv[5]));}
if(argc > 6) { mgLoadFac = atof(argv[6]); }
#ifdef PETSC_USE_LOG
PetscLogStagePush(stages[0]);
#endif
MPI_Barrier(MPI_COMM_WORLD);
ot::DAMG *damg;
int nlevels = 1; //number of multigrid levels
unsigned int dof = 1; // degrees of freedom per node
createRegularOctree(balOct, regLev, dim, maxDepth, MPI_COMM_WORLD);
PetscInt nlevelsPetscInt = nlevels; //To keep the compilers happy when using 64-bit indices
PetscOptionsGetInt(0, "-nlevels", &nlevelsPetscInt, 0);
nlevels = nlevelsPetscInt;
// Note: The user context for all levels will be set separately later.
MPI_Barrier(MPI_COMM_WORLD);
if(!rank) {
std::cout<<" nlevels initial: "<<nlevels<<std::endl;
}
ot::DAMGCreateAndSetDA(PETSC_COMM_WORLD, nlevels, NULL, &damg,
balOct, dof, mgLoadFac, compressLut, incCorner);
if(!rank) {
std::cout<<" nlevels final: "<<nlevels<<std::endl;
}
MPI_Barrier(MPI_COMM_WORLD);
if(!rank) {
std::cout << "Created DA for all levels."<< std::endl;
}
MPI_Barrier(MPI_COMM_WORLD);
ot::PrintDAMG(damg);
MPI_Barrier(MPI_COMM_WORLD);
for(int i=0;i<nlevels;i++) {
bool isRegOct = isRegularGrid(damg[i]->da);
if(!rank) {
std::cout<<"Level "<<i<<" is regular? "<<isRegOct<<std::endl;
}
}//end for i
SetUserContexts(damg);
if(!rank) {
std::cout << "Set User Contexts all levels."<< std::endl;
}
MPI_Barrier(MPI_COMM_WORLD);
PetscInt jacType = 1;
PetscOptionsGetInt(0,"-jacType",&jacType,0);
PetscInt rhsType = 1;
PetscOptionsGetInt(0,"-rhsType",&rhsType,0);
createLmatType2(LaplacianType2Stencil);
createMmatType2(MassType2Stencil);
if(jacType == 3) {
createLmatType1(LaplacianType1Stencil);
createMmatType1(MassType1Stencil);
}
createShFnMat(ShapeFnStencil);
if(!rank) {
std::cout << "Created Stencils."<< std::endl;
}
//Function handles
PetscErrorCode (*ComputeRHSHandle)(ot::DAMG damg,Vec rhs) = NULL;
PetscErrorCode (*CreateJacobianHandle)(ot::DAMG damg,Mat *B) = NULL;
PetscErrorCode (*ComputeJacobianHandle)(ot::DAMG damg,Mat J, Mat B) = NULL;
if(rhsType == 0) {
ComputeRHSHandle = ComputeRHS0;
} else if (rhsType == 1) {
ComputeRHSHandle = ComputeRHS1;
} else if (rhsType == 2) {
ComputeRHSHandle = ComputeRHS2;
} else if (rhsType == 3) {
ComputeRHSHandle = ComputeRHS3;
} else if (rhsType == 4) {
ComputeRHSHandle = ComputeRHS4;
} else if (rhsType == 5) {
ComputeRHSHandle = ComputeRHS5;
} else if (rhsType == 6) {
ComputeRHSHandle = ComputeRHS6;
} else if (rhsType == 7) {
ComputeRHSHandle = ComputeRHS7;
} else if (rhsType == 8) {
ComputeRHSHandle = ComputeRHS8;
} else {
assert(false);
}
if(jacType == 1) {
CreateJacobianHandle = CreateJacobian1;
ComputeJacobianHandle = ComputeJacobian1;
} else if (jacType == 2) {
CreateJacobianHandle = CreateJacobian2;
ComputeJacobianHandle = ComputeJacobian2;
} else if (jacType == 3) {
CreateJacobianHandle = CreateJacobian3;
ComputeJacobianHandle = ComputeJacobian3;
//Skip the finest and the coarsest levels. For the other levels, J and B
//must be different
for(int i = 1; i < (nlevels-1); i++) {
ot::DAMGCreateJMatrix(damg[i], CreateJacobianHandle);
}
} else {
assert(false);
}
//Global Function Handles for using KSP_Shell (will be used @ the coarsest grid if not all
//processors are active on the coarsest grid)
if (jacType == 1) {
ot::getPrivateMatricesForKSP_Shell = getPrivateMatricesForKSP_Shell_Jac1;
} else if (jacType == 2) {
ot::getPrivateMatricesForKSP_Shell = getPrivateMatricesForKSP_Shell_Jac2;
} else if (jacType == 3) {
ot::getPrivateMatricesForKSP_Shell = getPrivateMatricesForKSP_Shell_Jac3;
} else {
assert(false);
}
ot::DAMGSetKSP(damg, CreateJacobianHandle, ComputeJacobianHandle, ComputeRHSHandle);
if(!rank) {
std::cout<<"Solving u-Lu=f"<<std::endl;
}
iC(ot::DAMGSolve(damg));
destroyLmatType2(LaplacianType2Stencil);
destroyMmatType2(MassType2Stencil);
if(jacType == 3) {
destroyLmatType1(LaplacianType1Stencil);
destroyMmatType1(MassType1Stencil);
}
destroyShFnMat(ShapeFnStencil);
MPI_Barrier(MPI_COMM_WORLD);
DestroyUserContexts(damg);
if (!rank) {
std::cout << GRN << "Destroyed User Contexts." << NRM << std::endl;
}
MPI_Barrier(MPI_COMM_WORLD);
iC(DAMGDestroy(damg));
if (!rank) {
std::cout << GRN << "Destroyed DAMG" << NRM << std::endl;
}
#ifdef PETSC_USE_LOG
PetscLogStagePop();
#endif
balOct.clear();
if (!rank) {
std::cout << GRN << "Finalizing ..." << NRM << std::endl;
}
ot::DAMG_Finalize();
PetscFinalize();
}//end function
