bool KinematicsSolver::getBestSolution(ColumnVector& cartesian,
ColumnVector& currentQ, ColumnVector& bestSolution) {
if (this->solve(cartesian, currentQ)) {
SolutionSet* set = selector->getBest();
ColumnVector soln(6);
for (int i = 0; i < 6; i++) {
soln(i + 1) = set->theta[i];
}
bestSolution = soln;
return true;
}
return false;
}
/**
* Create query solution.
*/
QuerySolution* createQuerySolution() {
std::auto_ptr<QuerySolution> soln(new QuerySolution());
soln->cacheData.reset(new SolutionCacheData());
soln->cacheData->solnType = SolutionCacheData::COLLSCAN_SOLN;
soln->cacheData->tree.reset(new PlanCacheIndexTree());
return soln.release();
}
void CircuitState::zero()
{
uint i;
for(i=0; i < cir->stateLen; i++)
soln(i) = 0;
time = 0;
}
bool
RichardsMultiphaseProblem::updateSolution(NumericVector<Number>& vec_solution, NumericVector<Number>& ghosted_solution)
{
bool updatedSolution = false; // this gets set to true if we needed to enforce the bound at any node
unsigned int sys_num = getNonlinearSystem().number();
// For parallel procs i believe that i have to use local_nodes_begin, rather than just nodes_begin
// _mesh comes from SystemBase (_mesh = getNonlinearSystem().subproblem().mesh(), and subproblem is this object)
MeshBase::node_iterator nit = _mesh.getMesh().local_nodes_begin();
const MeshBase::node_iterator nend = _mesh.getMesh().local_nodes_end();
for ( ; nit != nend; ++nit)
{
const Node & node = *(*nit);
// dofs[0] is the dof number of the bounded variable at this node
// dofs[1] is the dof number of the lower variable at this node
std::vector<unsigned int> dofs(2);
dofs[0] = node.dof_number(sys_num, _bounded_var_num, 0);
dofs[1] = node.dof_number(sys_num, _lower_var_num, 0);
// soln[0] is the value of the bounded variable at this node
// soln[1] is the value of the lower variable at this node
std::vector<Number> soln(2);
vec_solution.get(dofs, soln);
// do the bounding
if (soln[0] < soln[1])
{
/*
dof_id_type nd = node.id();
Moose::out << "nd = " << nd << " dof_bounded = " << dofs[0] << " dof_lower = " << dofs[1] << "\n";
Moose::out << " bounded_value = " << soln[0] << " lower_value = " << soln[1] << "\n";
*/
vec_solution.set(dofs[0], soln[1]); // set the bounded variable equal to the lower value
updatedSolution = true;
}
}
// The above vec_solution.set calls potentially added "set" commands to a queue
// The following actions the queue (doing MPI commands if necessary), so
// vec_solution will actually be modified by this following command
vec_solution.close();
// if any proc updated the solution, all procs will know about it
_communicator.max(updatedSolution);
if (updatedSolution)
{
ghosted_solution = vec_solution;
ghosted_solution.close();
}
return updatedSolution;
}
//==========================================================================================================================
// Test to verify the evaluate method with a vector input.
//==========================================================================================================================
int EvaluateVectorTest()
{
TestUtil testFramework( "PolyFit", "Evaluate", __FILE__, __LINE__ );
std::string failMesg;
int n = 4;
gpstk::PolyFit<double> test(n);
gpstk::Vector<double> soln(3,0.), eval(3,0.);
double indep[6] = {0, 1, 2, 3, 4, 5}, dep[6] = {0, 1, 4, 9, 16, 25};
eval[0] = 6;
eval[1] = 8;
eval[2] = 10;
for(int i=0; i<6; i++)
{
test.Add(dep[i],indep[i]);
}
soln = test.Evaluate(eval);
//std::cout << "Solution is: " << soln << std::endl;
n = 0;
for (int i = 0; i<3; i++)
{
// Using relative error since the soln >> 1
if (fabs(soln[i] - eval[i]*eval[i])/(eval[i]*eval[i]) > eps)
{
n += 1; // Increment the return value for each wrong value
}
}
failMesg = "Was the solution computed correct?";
testFramework.assert(n==0, failMesg, __LINE__);
return testFramework.countFails(); // Return the result of the test.
}
QuerySolution* makeCollectionScan(const CanonicalQuery& query, bool tailable) {
auto_ptr<QuerySolution> soln(new QuerySolution());
soln->filter.reset(query.root()->shallowClone());
// BSONValue, where are you?
soln->filterData = query.getQueryObj();
// Make the (only) node, a collection scan.
CollectionScanNode* csn = new CollectionScanNode();
csn->name = query.ns();
csn->filter = soln->filter.get();
csn->tailable = tailable;
const BSONObj& sortObj = query.getParsed().getSort();
// TODO: We need better checking for this. Should be done in CanonicalQuery and we should
// be able to assume it's correct.
if (!sortObj.isEmpty()) {
BSONElement natural = sortObj.getFieldDotted("$natural");
csn->direction = natural.numberInt() >= 0 ? 1 : -1;
}
// Add this solution to the list of solutions.
soln->root.reset(csn);
return soln.release();
}
// static
QuerySolution* QueryPlannerAnalysis::analyzeDataAccess(const CanonicalQuery& query,
const QueryPlannerParams& params,
QuerySolutionNode* solnRoot) {
auto_ptr<QuerySolution> soln(new QuerySolution());
soln->filterData = query.getQueryObj();
verify(soln->filterData.isOwned());
soln->ns = query.ns();
soln->indexFilterApplied = params.indexFiltersApplied;
solnRoot->computeProperties();
// solnRoot finds all our results. Let's see what transformations we must perform to the
// data.
// If we're answering a query on a sharded system, we need to drop documents that aren't
// logically part of our shard (XXX GREG elaborate more precisely)
if (params.options & QueryPlannerParams::INCLUDE_SHARD_FILTER) {
// XXX TODO: use params.shardKey to do fetch analysis instead of always fetching.
if (!solnRoot->fetched()) {
FetchNode* fetch = new FetchNode();
fetch->children.push_back(solnRoot);
solnRoot = fetch;
}
ShardingFilterNode* sfn = new ShardingFilterNode();
sfn->children.push_back(solnRoot);
solnRoot = sfn;
}
solnRoot = analyzeSort(query, params, solnRoot, &soln->hasSortStage);
// This can happen if we need to create a blocking sort stage and we're not allowed to.
if (NULL == solnRoot) { return NULL; }
// If we can (and should), add the keep mutations stage.
// We cannot keep mutated documents if:
//
// 1. The query requires an index to evaluate the predicate ($text). We can't tell whether
// or not the doc actually satisfies the $text predicate since we can't evaluate a
// text MatchExpression.
//
// 2. The query implies a sort ($geoNear). It would be rather expensive and hacky to merge
// the document at the right place.
//
// 3. There is an index-provided sort. Ditto above comment about merging.
// XXX; do we want some kind of static init for a set of stages we care about & pass that
// set into hasNode?
bool cannotKeepFlagged = hasNode(solnRoot, STAGE_TEXT)
|| hasNode(solnRoot, STAGE_GEO_NEAR_2D)
|| hasNode(solnRoot, STAGE_GEO_NEAR_2DSPHERE)
|| (!query.getParsed().getSort().isEmpty() && !soln->hasSortStage);
// Only these stages can produce flagged results. A stage has to hold state past one call
// to work(...) in order to possibly flag a result.
bool couldProduceFlagged = hasNode(solnRoot, STAGE_GEO_2D)
|| hasNode(solnRoot, STAGE_AND_HASH)
|| hasNode(solnRoot, STAGE_AND_SORTED)
|| hasNode(solnRoot, STAGE_FETCH);
bool shouldAddMutation = !cannotKeepFlagged && couldProduceFlagged;
if (shouldAddMutation && (params.options & QueryPlannerParams::KEEP_MUTATIONS)) {
KeepMutationsNode* keep = new KeepMutationsNode();
// We must run the entire expression tree to make sure the document is still valid.
keep->filter.reset(query.root()->shallowClone());
if (STAGE_SORT == solnRoot->getType()) {
// We want to insert the invalidated results before the sort stage, if there is one.
verify(1 == solnRoot->children.size());
keep->children.push_back(solnRoot->children[0]);
solnRoot->children[0] = keep;
}
else {
keep->children.push_back(solnRoot);
solnRoot = keep;
}
}
// Project the results.
if (NULL != query.getProj()) {
QLOG() << "PROJECTION: fetched status: " << solnRoot->fetched() << endl;
QLOG() << "PROJECTION: Current plan is:\n" << solnRoot->toString() << endl;
if (query.getProj()->requiresDocument()) {
QLOG() << "PROJECTION: claims to require doc adding fetch.\n";
// If the projection requires the entire document, somebody must fetch.
if (!solnRoot->fetched()) {
FetchNode* fetch = new FetchNode();
fetch->children.push_back(solnRoot);
solnRoot = fetch;
}
}
else {
QLOG() << "PROJECTION: requires fields\n";
const vector<string>& fields = query.getProj()->getRequiredFields();
bool covered = true;
for (size_t i = 0; i < fields.size(); ++i) {
if (!solnRoot->hasField(fields[i])) {
QLOG() << "PROJECTION: not covered cuz doesn't have field "
<< fields[i] << endl;
covered = false;
break;
}
}
QLOG() << "PROJECTION: is covered?: = " << covered << endl;
// If any field is missing from the list of fields the projection wants,
// a fetch is required.
if (!covered) {
FetchNode* fetch = new FetchNode();
fetch->children.push_back(solnRoot);
solnRoot = fetch;
}
}
// We now know we have whatever data is required for the projection.
ProjectionNode* projNode = new ProjectionNode();
projNode->children.push_back(solnRoot);
projNode->fullExpression = query.root();
projNode->projection = query.getParsed().getProj();
solnRoot = projNode;
}
else {
// If there's no projection, we must fetch, as the user wants the entire doc.
if (!solnRoot->fetched()) {
FetchNode* fetch = new FetchNode();
fetch->children.push_back(solnRoot);
solnRoot = fetch;
}
}
if (0 != query.getParsed().getSkip()) {
SkipNode* skip = new SkipNode();
skip->skip = query.getParsed().getSkip();
skip->children.push_back(solnRoot);
solnRoot = skip;
}
// When there is both a blocking sort and a limit, the limit will
// be enforced by the blocking sort.
// Otherwise, we need to limit the results in the case of a hard limit
// (ie. limit in raw query is negative)
if (0 != query.getParsed().getNumToReturn() &&
!soln->hasSortStage &&
!query.getParsed().wantMore()) {
LimitNode* limit = new LimitNode();
limit->limit = query.getParsed().getNumToReturn();
limit->children.push_back(solnRoot);
solnRoot = limit;
}
soln->root.reset(solnRoot);
return soln.release();
}
int main(int argc, char *argv[]) {
Teuchos::GlobalMPISession mpiSession(&argc, &argv);
// This little trick lets us print to std::cout only if a (dummy) command-line argument is provided.
int iprint = argc - 1;
Teuchos::RCP<std::ostream> outStream;
Teuchos::oblackholestream bhs; // outputs nothing
if (iprint > 0)
outStream = Teuchos::rcp(&std::cout, false);
else
outStream = Teuchos::rcp(&bhs, false);
int errorFlag = 0;
// *** Example body.
try {
std::string filename = "input.xml";
Teuchos::RCP<Teuchos::ParameterList> parlist = Teuchos::rcp( new Teuchos::ParameterList() );
Teuchos::updateParametersFromXmlFile( filename, parlist.ptr() );
RealT V_th = parlist->get("Thermal Voltage", 0.02585);
RealT lo_Vsrc = parlist->get("Source Voltage Lower Bound", 0.0);
RealT up_Vsrc = parlist->get("Source Voltage Upper Bound", 1.0);
RealT step_Vsrc = parlist->get("Source Voltage Step", 1.e-2);
RealT true_Is = parlist->get("True Saturation Current", 1.e-12);
RealT true_Rs = parlist->get("True Saturation Resistance", 0.25);
RealT init_Is = parlist->get("Initial Saturation Current", 1.e-12);
RealT init_Rs = parlist->get("Initial Saturation Resistance", 0.25);
RealT lo_Is = parlist->get("Saturation Current Lower Bound", 1.e-16);
RealT up_Is = parlist->get("Saturation Current Upper Bound", 1.e-1);
RealT lo_Rs = parlist->get("Saturation Resistance Lower Bound", 1.e-2);
RealT up_Rs = parlist->get("Saturation Resistance Upper Bound", 30.0);
// bool use_lambertw = parlist->get("Use Analytical Solution",true);
bool use_scale = parlist->get("Use Scaling For Epsilon-Active Sets",true);
bool use_sqp = parlist->get("Use SQP", true);
// bool use_linesearch = parlist->get("Use Line Search", true);
// bool datatype = parlist->get("Get Data From Input File",false);
// bool use_adjoint = parlist->get("Use Adjoint Gradient Computation",false);
// int use_hessvec = parlist->get("Use Hessian-Vector Implementation",1); // 0 - FD, 1 - exact, 2 - GN
// bool plot = parlist->get("Generate Plot Data",false);
// RealT noise = parlist->get("Measurement Noise",0.0);
EqualityConstraint_DiodeCircuit<RealT> con(V_th,lo_Vsrc,up_Vsrc,step_Vsrc);
RealT alpha = 1.e-4; // regularization parameter (unused)
int ns = 101; // number of Vsrc components
int nz = 2; // number of optimization variables
Objective_DiodeCircuit<RealT> obj(alpha,ns,nz);
// Initialize iteration vectors.
Teuchos::RCP<std::vector<RealT> > z_rcp = Teuchos::rcp( new std::vector<RealT> (nz, 0.0) );
Teuchos::RCP<std::vector<RealT> > yz_rcp = Teuchos::rcp( new std::vector<RealT> (nz, 0.0) );
Teuchos::RCP<std::vector<RealT> > soln_rcp = Teuchos::rcp( new std::vector<RealT> (nz, 0.0) );
(*z_rcp)[0] = init_Is;
(*z_rcp)[1] = init_Rs;
(*yz_rcp)[0] = init_Is;
(*yz_rcp)[1] = init_Rs;
(*soln_rcp)[0] = true_Is;
(*soln_rcp)[1] = true_Rs;
ROL::StdVector<RealT> z(z_rcp);
ROL::StdVector<RealT> yz(yz_rcp);
ROL::StdVector<RealT> soln(soln_rcp);
Teuchos::RCP<ROL::Vector<RealT> > zp = Teuchos::rcp(&z,false);
Teuchos::RCP<ROL::Vector<RealT> > yzp = Teuchos::rcp(&yz,false);
Teuchos::RCP<std::vector<RealT> > u_rcp = Teuchos::rcp( new std::vector<RealT> (ns, 0.0) );
Teuchos::RCP<std::vector<RealT> > yu_rcp = Teuchos::rcp( new std::vector<RealT> (ns, 0.0) );
std::ifstream input_file("measurements.dat");
RealT temp, temp_scale;
for (int i=0; i<ns; i++) {
input_file >> temp;
input_file >> temp;
temp_scale = pow(10,int(log10(temp)));
(*u_rcp)[i] = temp_scale*(RealT)rand()/(RealT)RAND_MAX;
(*yu_rcp)[i] = temp_scale*(RealT)rand()/(RealT)RAND_MAX;
}
input_file.close();
ROL::StdVector<RealT> u(u_rcp);
ROL::StdVector<RealT> yu(yu_rcp);
Teuchos::RCP<ROL::Vector<RealT> > up = Teuchos::rcp(&u,false);
Teuchos::RCP<ROL::Vector<RealT> > yup = Teuchos::rcp(&yu,false);
Teuchos::RCP<std::vector<RealT> > jv_rcp = Teuchos::rcp( new std::vector<RealT> (ns, 1.0) );
ROL::StdVector<RealT> jv(jv_rcp);
Teuchos::RCP<ROL::Vector<RealT> > jvp = Teuchos::rcp(&jv,false);
ROL::Vector_SimOpt<RealT> x(up,zp);
ROL::Vector_SimOpt<RealT> y(yup,yzp);
// Check derivatives
obj.checkGradient(x,x,y,true,*outStream);
obj.checkHessVec(x,x,y,true,*outStream);
con.checkApplyJacobian(x,y,jv,true,*outStream);
con.checkApplyAdjointJacobian(x,yu,jv,x,true,*outStream);
con.checkApplyAdjointHessian(x,yu,y,x,true,*outStream);
// Check consistency of Jacobians and adjoint Jacobians.
con.checkAdjointConsistencyJacobian_1(jv,yu,u,z,true,*outStream);
con.checkAdjointConsistencyJacobian_2(jv,yz,u,z,true,*outStream);
// Check consistency of solves.
con.checkSolve(u,z,jv,true,*outStream);
con.checkInverseJacobian_1(jv,yu,u,z,true,*outStream);
con.checkInverseAdjointJacobian_1(yu,jv,u,z,true,*outStream);
// Initialize reduced objective function.
Teuchos::RCP<std::vector<RealT> > p_rcp = Teuchos::rcp( new std::vector<RealT> (ns, 0.0) );
ROL::StdVector<RealT> p(p_rcp);
Teuchos::RCP<ROL::Vector<RealT> > pp = Teuchos::rcp(&p,false);
Teuchos::RCP<ROL::Objective_SimOpt<RealT> > pobj = Teuchos::rcp(&obj,false);
Teuchos::RCP<ROL::EqualityConstraint_SimOpt<RealT> > pcon = Teuchos::rcp(&con,false);
ROL::Reduced_Objective_SimOpt<RealT> robj(pobj,pcon,up,pp);
// Check derivatives.
*outStream << "Derivatives of reduced objective" << std::endl;
robj.checkGradient(z,z,yz,true,*outStream);
robj.checkHessVec(z,z,yz,true,*outStream);
// Bound constraints
RealT tol = 1.e-12;
Teuchos::RCP<std::vector<RealT> > g0_rcp = Teuchos::rcp( new std::vector<RealT> (nz, 0.0) );;
ROL::StdVector<RealT> g0p(g0_rcp);
robj.gradient(g0p,z,tol);
*outStream << std::scientific << "Initial gradient = " << (*g0_rcp)[0] << " " << (*g0_rcp)[1] << "\n";
*outStream << std::scientific << "Norm of Gradient = " << g0p.norm() << "\n";
// Define scaling for epsilon-active sets (used in inequality constraints)
RealT scale;
if(use_scale){ scale = 1.0e-2/g0p.norm();}
else{ scale = 1.0;}
*outStream << std::scientific << "Scaling: " << scale << "\n";
/// Define constraints on Is and Rs
BoundConstraint_DiodeCircuit<RealT> bcon(scale,lo_Is,up_Is,lo_Rs,up_Rs);
//bcon.deactivate();
// Optimization
*outStream << "\n Initial guess " << (*z_rcp)[0] << " " << (*z_rcp)[1] << std::endl;
if (!use_sqp){
// Trust Region
ROL::Algorithm<RealT> algo_tr("Trust Region",*parlist);
std::clock_t timer_tr = std::clock();
algo_tr.run(z,robj,bcon,true,*outStream);
*outStream << "\n Solution " << (*z_rcp)[0] << " " << (*z_rcp)[1] << "\n" << std::endl;
*outStream << "Trust-Region required " << (std::clock()-timer_tr)/(RealT)CLOCKS_PER_SEC
<< " seconds.\n";
}
else{
// SQP.
//Teuchos::RCP<std::vector<RealT> > gz_rcp = Teuchos::rcp( new std::vector<RealT> (nz, 0.0) );
//ROL::StdVector<RealT> gz(gz_rcp);
//Teuchos::RCP<ROL::Vector<RealT> > gzp = Teuchos::rcp(&gz,false);
Teuchos::RCP<std::vector<RealT> > gu_rcp = Teuchos::rcp( new std::vector<RealT> (ns, 0.0) );
ROL::StdVector<RealT> gu(gu_rcp);
Teuchos::RCP<ROL::Vector<RealT> > gup = Teuchos::rcp(&gu,false);
//ROL::Vector_SimOpt<RealT> g(gup,gzp);
ROL::Vector_SimOpt<RealT> g(gup,zp);
Teuchos::RCP<std::vector<RealT> > c_rcp = Teuchos::rcp( new std::vector<RealT> (ns, 0.0) );
Teuchos::RCP<std::vector<RealT> > l_rcp = Teuchos::rcp( new std::vector<RealT> (ns, 0.0) );
ROL::StdVector<RealT> c(c_rcp);
ROL::StdVector<RealT> l(l_rcp);
ROL::Algorithm<RealT> algo_cs("Composite Step",*parlist);
//x.zero();
std::clock_t timer_cs = std::clock();
algo_cs.run(x,g,l,c,obj,con,true,*outStream);
*outStream << "\n Solution " << (*z_rcp)[0] << " " << (*z_rcp)[1] << "\n" << std::endl;
*outStream << "Composite Step required " << (std::clock()-timer_cs)/(RealT)CLOCKS_PER_SEC
<< " seconds.\n";
}
soln.axpy(-1.0, z);
*outStream << "Norm of error: " << soln.norm() << std::endl;
if (soln.norm() > 1e4*ROL::ROL_EPSILON) {
errorFlag = 1;
}
}
catch (std::logic_error err) {
*outStream << err.what() << "\n";
errorFlag = -1000;
}; // end try
if (errorFlag != 0)
std::cout << "End Result: TEST FAILED\n";
else
std::cout << "End Result: TEST PASSED\n";
return 0;
}
int
verify_solution(const simple_mesh_description<typename VectorType::GlobalOrdinalType>& mesh,
const VectorType& x, double tolerance, bool verify_whole_domain = false)
{
typedef typename VectorType::GlobalOrdinalType GlobalOrdinal;
typedef typename VectorType::ScalarType Scalar;
int global_nodes_x = mesh.global_box[0][1]+1;
int global_nodes_y = mesh.global_box[1][1]+1;
int global_nodes_z = mesh.global_box[2][1]+1;
Box box;
copy_box(mesh.local_box, box);
//num-owned-nodes in each dimension is num-elems+1
//only if num-elems > 0 in that dimension *and*
//we are at the high end of the global range in that dimension:
if (box[0][1] > box[0][0] && box[0][1] == mesh.global_box[0][1]) ++box[0][1];
if (box[1][1] > box[1][0] && box[1][1] == mesh.global_box[1][1]) ++box[1][1];
if (box[2][1] > box[2][0] && box[2][1] == mesh.global_box[2][1]) ++box[2][1];
std::vector<GlobalOrdinal> rows;
std::vector<Scalar> row_coords;
int roffset = 0;
for(int iz=box[2][0]; iz<box[2][1]; ++iz) {
for(int iy=box[1][0]; iy<box[1][1]; ++iy) {
for(int ix=box[0][0]; ix<box[0][1]; ++ix) {
GlobalOrdinal row_id =
get_id<GlobalOrdinal>(global_nodes_x, global_nodes_y, global_nodes_z,
ix, iy, iz);
Scalar x, y, z;
get_coords(row_id, global_nodes_x, global_nodes_y, global_nodes_z, x, y, z);
bool verify_this_point = false;
if (verify_whole_domain) verify_this_point = true;
else if (std::abs(x - 0.5) < 0.05 && std::abs(y - 0.5) < 0.05 && std::abs(z - 0.5) < 0.05) {
verify_this_point = true;
}
if (verify_this_point) {
rows.push_back(roffset);
row_coords.push_back(x);
row_coords.push_back(y);
row_coords.push_back(z);
}
++roffset;
}
}
}
int return_code = 0;
const int num_terms = 300;
err_info<Scalar> max_error;
max_error.err = 0.0;
for(size_t i=0; i<rows.size(); ++i) {
Scalar computed_soln = x.coefs[rows[i]];
Scalar x = row_coords[i*3];
Scalar y = row_coords[i*3+1];
Scalar z = row_coords[i*3+2];
Scalar analytic_soln = 0.0;
//set exact boundary-conditions:
if (x == 1.0) {
//x==1 is first, we want soln to be 1 even around the edges
//of the x==1 plane where y and/or z may be 0 or 1...
analytic_soln = 1;
}
else if (x == 0.0 || y == 0.0 || z == 0.0) {
analytic_soln = 0;
}
else if (y == 1.0 || z == 1.0) {
analytic_soln = 0;
}
else {
analytic_soln = soln(x, y, z, num_terms, num_terms);
}
#ifdef MINIFE_DEBUG_VERBOSE
std::cout<<"("<<x<<","<<y<<","<<z<<") row "<<rows[i]<<": computed: "<<computed_soln<<", analytic: "<<analytic_soln<<std::endl;
#endif
Scalar err = std::abs(analytic_soln - computed_soln);
if (err > max_error.err) {
max_error.err = err;
max_error.computed = computed_soln;
max_error.analytic = analytic_soln;
max_error.coords[0] = x;
max_error.coords[1] = y;
max_error.coords[2] = z;
}
}
Scalar local_max_err = max_error.err;
Scalar global_max_err = 0;
#ifdef HAVE_MPI
MPI_Allreduce(&local_max_err, &global_max_err, 1, MPI_DOUBLE, MPI_MAX, MPI_COMM_WORLD);
#else
global_max_err = local_max_err;
#endif
if (local_max_err == global_max_err) {
if (max_error.err > tolerance) {
std::cout << "max absolute error is "<<max_error.err<<":"<<std::endl;
std::cout << " at position ("<<max_error.coords[0]<<","<<max_error.coords[1]<<","<<max_error.coords[2]<<"), "<<std::endl;
std::cout << " computed solution: "<<max_error.computed<<", analytic solution: "<<max_error.analytic<<std::endl;
}
else {
std::cout << "solution matches analytic solution to within "<<tolerance<<" or better."<<std::endl;
}
}
if (global_max_err > tolerance) {
return_code = 1;
}
return return_code;
}
int main(int argc, char* argv[])
{
BoxLib::Initialize(argc,argv);
BL_PROFILE_VAR("main()", pmain);
std::cout << std::setprecision(15);
solver_type = BoxLib_C;
bc_type = Periodic;
Real a = 0.0;
Real b = 1.0;
// ---- First use the number of processors to decide how many grids you have.
// ---- We arbitrarily decide to have one grid per MPI process in a uniform
// ---- cubic domain, so we require that the number of processors be N^3.
// ---- This requirement is somewhat arbitrary, but convenient for now.
int nprocs = ParallelDescriptor::NProcs();
// N is the cube root of the number of processors
int N(0);
for(int i(1); i*i*i <= nprocs; ++i) {
if(i*i*i == nprocs) {
N = i;
}
}
if(N == 0) { // not a cube
if(ParallelDescriptor::IOProcessor()) {
std::cerr << "**** Error: nprocs = " << nprocs << " is not currently supported." << std::endl;
}
BoxLib::Error("We require that the number of processors be a perfect cube");
}
if(ParallelDescriptor::IOProcessor()) {
std::cout << "N = " << N << std::endl;
}
// ---- make a box, then a boxarray with maxSize
int domain_hi = (N*maxGrid) - 1;
Box domain(IntVect(0,0,0), IntVect(domain_hi,domain_hi,domain_hi));
BoxArray bs(domain);
bs.maxSize(maxGrid);
// This defines the physical size of the box. Right now the box is [0,1] in each direction.
RealBox real_box;
for (int n = 0; n < BL_SPACEDIM; n++) {
real_box.setLo(n, 0.0);
real_box.setHi(n, 1.0);
}
// This says we are using Cartesian coordinates
int coord = 0;
// This sets the boundary conditions to be periodic or not
int is_per[BL_SPACEDIM];
if (bc_type == Dirichlet || bc_type == Neumann) {
for (int n = 0; n < BL_SPACEDIM; n++) is_per[n] = 0;
}
else {
for (int n = 0; n < BL_SPACEDIM; n++) is_per[n] = 1;
}
// This defines a Geometry object which is useful for writing the plotfiles
Geometry geom(domain,&real_box,coord,is_per);
for ( int n=0; n<BL_SPACEDIM; n++ ) {
dx[n] = ( geom.ProbHi(n) - geom.ProbLo(n) )/domain.length(n);
}
if (ParallelDescriptor::IOProcessor()) {
std::cout << "Domain size : " << N << std::endl;
std::cout << "Max_grid_size : " << maxGrid << std::endl;
std::cout << "Number of grids : " << bs.size() << std::endl;
}
// Allocate and define the right hand side.
MultiFab rhs(bs, Ncomp, 0, Fab_allocate);
setup_rhs(rhs, geom, a, b);
MultiFab alpha(bs, Ncomp, 0, Fab_allocate);
MultiFab beta[BL_SPACEDIM];
for ( int n=0; n<BL_SPACEDIM; ++n ) {
BoxArray bx(bs);
beta[n].define(bx.surroundingNodes(n), Ncomp, 1, Fab_allocate);
}
setup_coeffs(bs, alpha, beta, geom);
MultiFab anaSoln;
if (comp_norm) {
anaSoln.define(bs, Ncomp, 0, Fab_allocate);
compute_analyticSolution(anaSoln);
}
// Allocate the solution array
// Set the number of ghost cells in the solution array.
MultiFab soln(bs, Ncomp, 1, Fab_allocate);
solve(soln, anaSoln, a, b, alpha, beta, rhs, bs, geom, BoxLib_C);
BL_PROFILE_VAR_STOP(pmain);
BoxLib::Finalize();
}
// static
QuerySolution* QueryPlannerAnalysis::analyzeDataAccess(const CanonicalQuery& query,
const QueryPlannerParams& params,
QuerySolutionNode* solnRoot) {
auto_ptr<QuerySolution> soln(new QuerySolution());
soln->filterData = query.getQueryObj();
verify(soln->filterData.isOwned());
soln->ns = query.ns();
soln->indexFilterApplied = params.indexFiltersApplied;
solnRoot->computeProperties();
// solnRoot finds all our results. Let's see what transformations we must perform to the
// data.
// If we're answering a query on a sharded system, we need to drop documents that aren't
// logically part of our shard.
if (params.options & QueryPlannerParams::INCLUDE_SHARD_FILTER) {
// TODO: We could use params.shardKey to do fetch analysis instead of always fetching.
if (!solnRoot->fetched()) {
FetchNode* fetch = new FetchNode();
fetch->children.push_back(solnRoot);
solnRoot = fetch;
}
ShardingFilterNode* sfn = new ShardingFilterNode();
sfn->children.push_back(solnRoot);
solnRoot = sfn;
}
bool hasSortStage = false;
solnRoot = analyzeSort(query, params, solnRoot, &hasSortStage);
// This can happen if we need to create a blocking sort stage and we're not allowed to.
if (NULL == solnRoot) { return NULL; }
// A solution can be blocking if it has a blocking sort stage or
// a hashed AND stage.
bool hasAndHashStage = hasNode(solnRoot, STAGE_AND_HASH);
soln->hasBlockingStage = hasSortStage || hasAndHashStage;
// If we can (and should), add the keep mutations stage.
// We cannot keep mutated documents if:
//
// 1. The query requires an index to evaluate the predicate ($text). We can't tell whether
// or not the doc actually satisfies the $text predicate since we can't evaluate a
// text MatchExpression.
//
// 2. The query implies a sort ($geoNear). It would be rather expensive and hacky to merge
// the document at the right place.
//
// 3. There is an index-provided sort. Ditto above comment about merging.
//
// TODO: do we want some kind of pre-planning step where we look for certain nodes and cache
// them? We do lookups in the tree a few times. This may not matter as most trees are
// shallow in terms of query nodes.
bool cannotKeepFlagged = hasNode(solnRoot, STAGE_TEXT)
|| hasNode(solnRoot, STAGE_GEO_NEAR_2D)
|| hasNode(solnRoot, STAGE_GEO_NEAR_2DSPHERE)
|| (!query.getParsed().getSort().isEmpty() && !hasSortStage);
// Only these stages can produce flagged results. A stage has to hold state past one call
// to work(...) in order to possibly flag a result.
bool couldProduceFlagged = hasNode(solnRoot, STAGE_GEO_2D)
|| hasAndHashStage
|| hasNode(solnRoot, STAGE_AND_SORTED)
|| hasNode(solnRoot, STAGE_FETCH);
bool shouldAddMutation = !cannotKeepFlagged && couldProduceFlagged;
if (shouldAddMutation && (params.options & QueryPlannerParams::KEEP_MUTATIONS)) {
KeepMutationsNode* keep = new KeepMutationsNode();
// We must run the entire expression tree to make sure the document is still valid.
keep->filter.reset(query.root()->shallowClone());
if (STAGE_SORT == solnRoot->getType()) {
// We want to insert the invalidated results before the sort stage, if there is one.
verify(1 == solnRoot->children.size());
keep->children.push_back(solnRoot->children[0]);
solnRoot->children[0] = keep;
}
else {
keep->children.push_back(solnRoot);
solnRoot = keep;
}
}
// Project the results.
if (NULL != query.getProj()) {
QLOG() << "PROJECTION: fetched status: " << solnRoot->fetched() << endl;
QLOG() << "PROJECTION: Current plan is:\n" << solnRoot->toString() << endl;
ProjectionNode::ProjectionType projType = ProjectionNode::DEFAULT;
BSONObj coveredKeyObj;
if (query.getProj()->requiresDocument()) {
QLOG() << "PROJECTION: claims to require doc adding fetch.\n";
// If the projection requires the entire document, somebody must fetch.
if (!solnRoot->fetched()) {
FetchNode* fetch = new FetchNode();
fetch->children.push_back(solnRoot);
solnRoot = fetch;
}
}
else if (!query.getProj()->wantIndexKey()) {
// The only way we're here is if it's a simple projection. That is, we can pick out
// the fields we want to include and they're not dotted. So we want to execute the
// projection in the fast-path simple fashion. Just don't know which fast path yet.
QLOG() << "PROJECTION: requires fields\n";
const vector<string>& fields = query.getProj()->getRequiredFields();
bool covered = true;
for (size_t i = 0; i < fields.size(); ++i) {
if (!solnRoot->hasField(fields[i])) {
QLOG() << "PROJECTION: not covered due to field "
<< fields[i] << endl;
covered = false;
break;
}
}
QLOG() << "PROJECTION: is covered?: = " << covered << endl;
// If any field is missing from the list of fields the projection wants,
// a fetch is required.
if (!covered) {
FetchNode* fetch = new FetchNode();
fetch->children.push_back(solnRoot);
solnRoot = fetch;
// It's simple but we'll have the full document and we should just iterate
// over that.
projType = ProjectionNode::SIMPLE_DOC;
QLOG() << "PROJECTION: not covered, fetching.";
}
else {
if (solnRoot->fetched()) {
// Fetched implies hasObj() so let's run with that.
projType = ProjectionNode::SIMPLE_DOC;
QLOG() << "PROJECTION: covered via FETCH, using SIMPLE_DOC fast path";
}
else {
// If we're here we're not fetched so we're covered. Let's see if we can
// get out of using the default projType. If there's only one leaf
// underneath and it's giving us index data we can use the faster covered
// impl.
vector<QuerySolutionNode*> leafNodes;
getLeafNodes(solnRoot, &leafNodes);
if (1 == leafNodes.size()) {
// Both the IXSCAN and DISTINCT stages provide covered key data.
if (STAGE_IXSCAN == leafNodes[0]->getType()) {
projType = ProjectionNode::COVERED_ONE_INDEX;
IndexScanNode* ixn = static_cast<IndexScanNode*>(leafNodes[0]);
coveredKeyObj = ixn->indexKeyPattern;
QLOG() << "PROJECTION: covered via IXSCAN, using COVERED fast path";
}
else if (STAGE_DISTINCT == leafNodes[0]->getType()) {
projType = ProjectionNode::COVERED_ONE_INDEX;
DistinctNode* dn = static_cast<DistinctNode*>(leafNodes[0]);
coveredKeyObj = dn->indexKeyPattern;
QLOG() << "PROJECTION: covered via DISTINCT, using COVERED fast path";
}
}
}
}
}
// We now know we have whatever data is required for the projection.
ProjectionNode* projNode = new ProjectionNode();
projNode->children.push_back(solnRoot);
projNode->fullExpression = query.root();
projNode->projection = query.getParsed().getProj();
projNode->projType = projType;
projNode->coveredKeyObj = coveredKeyObj;
solnRoot = projNode;
}
else {
// If there's no projection, we must fetch, as the user wants the entire doc.
if (!solnRoot->fetched()) {
FetchNode* fetch = new FetchNode();
fetch->children.push_back(solnRoot);
solnRoot = fetch;
}
}
if (0 != query.getParsed().getSkip()) {
SkipNode* skip = new SkipNode();
skip->skip = query.getParsed().getSkip();
skip->children.push_back(solnRoot);
solnRoot = skip;
}
// When there is both a blocking sort and a limit, the limit will
// be enforced by the blocking sort.
// Otherwise, we need to limit the results in the case of a hard limit
// (ie. limit in raw query is negative)
if (0 != query.getParsed().getNumToReturn() &&
!hasSortStage &&
!query.getParsed().wantMore()) {
LimitNode* limit = new LimitNode();
limit->limit = query.getParsed().getNumToReturn();
limit->children.push_back(solnRoot);
solnRoot = limit;
}
soln->root.reset(solnRoot);
return soln.release();
}
int main(int argc, char* argv[])
{
BoxLib::Initialize(argc,argv);
BL_PROFILE_VAR("main()", pmain);
std::cout << std::setprecision(15);
ParmParse ppmg("mg");
ppmg.query("v", verbose);
ppmg.query("maxorder", maxorder);
ParmParse pp;
{
std::string solver_type_s;
pp.get("solver_type",solver_type_s);
if (solver_type_s == "BoxLib_C") {
solver_type = BoxLib_C;
}
else if (solver_type_s == "BoxLib_C4") {
solver_type = BoxLib_C4;
}
else if (solver_type_s == "BoxLib_F") {
#ifdef USE_F90_SOLVERS
solver_type = BoxLib_F;
#else
BoxLib::Error("Set USE_FORTRAN=TRUE in GNUmakefile");
#endif
}
else if (solver_type_s == "Hypre") {
#ifdef USEHYPRE
solver_type = Hypre;
#else
BoxLib::Error("Set USE_HYPRE=TRUE in GNUmakefile");
#endif
}
else if (solver_type_s == "All") {
solver_type = All;
}
else {
if (ParallelDescriptor::IOProcessor()) {
std::cout << "Don't know this solver type: " << solver_type << std::endl;
}
BoxLib::Error("");
}
}
{
std::string bc_type_s;
pp.get("bc_type",bc_type_s);
if (bc_type_s == "Dirichlet") {
bc_type = Dirichlet;
#ifdef USEHPGMG
domain_boundary_condition = BC_DIRICHLET;
#endif
}
else if (bc_type_s == "Neumann") {
bc_type = Neumann;
#ifdef USEHPGMG
BoxLib::Error("HPGMG does not support Neumann boundary conditions");
#endif
}
else if (bc_type_s == "Periodic") {
bc_type = Periodic;
#ifdef USEHPGMG
domain_boundary_condition = BC_PERIODIC;
#endif
}
else {
if (ParallelDescriptor::IOProcessor()) {
std::cout << "Don't know this boundary type: " << bc_type << std::endl;
}
BoxLib::Error("");
}
}
pp.query("tol_rel", tolerance_rel);
pp.query("tol_abs", tolerance_abs);
pp.query("maxiter", maxiter);
pp.query("plot_rhs" , plot_rhs);
pp.query("plot_beta", plot_beta);
pp.query("plot_soln", plot_soln);
pp.query("plot_asol", plot_asol);
pp.query("plot_err", plot_err);
pp.query("comp_norm", comp_norm);
Real a, b;
pp.get("a", a);
pp.get("b", b);
pp.get("n_cell",n_cell);
pp.get("max_grid_size",max_grid_size);
// Define a single box covering the domain
IntVect dom_lo(D_DECL(0,0,0));
IntVect dom_hi(D_DECL(n_cell-1,n_cell-1,n_cell-1));
Box domain(dom_lo,dom_hi);
// Initialize the boxarray "bs" from the single box "bx"
BoxArray bs(domain);
// Break up boxarray "bs" into chunks no larger than "max_grid_size" along a direction
bs.maxSize(max_grid_size);
// This defines the physical size of the box. Right now the box is [0,1] in each direction.
RealBox real_box;
for (int n = 0; n < BL_SPACEDIM; n++) {
real_box.setLo(n, 0.0);
real_box.setHi(n, 1.0);
}
// This says we are using Cartesian coordinates
int coord = 0;
// This sets the boundary conditions to be periodic or not
Array<int> is_per(BL_SPACEDIM,1);
if (bc_type == Dirichlet || bc_type == Neumann) {
if (ParallelDescriptor::IOProcessor()) {
std::cout << "Using Dirichlet or Neumann boundary conditions." << std::endl;
}
for (int n = 0; n < BL_SPACEDIM; n++) is_per[n] = 0;
}
else {
if (ParallelDescriptor::IOProcessor()) {
std::cout << "Using periodic boundary conditions." << std::endl;
}
for (int n = 0; n < BL_SPACEDIM; n++) is_per[n] = 1;
}
// This defines a Geometry object which is useful for writing the plotfiles
Geometry geom(domain,&real_box,coord,is_per.dataPtr());
for ( int n=0; n<BL_SPACEDIM; n++ ) {
dx[n] = ( geom.ProbHi(n) - geom.ProbLo(n) )/domain.length(n);
}
if (ParallelDescriptor::IOProcessor()) {
std::cout << "Grid resolution : " << n_cell << " (cells)" << std::endl;
std::cout << "Domain size : " << real_box.hi(0) - real_box.lo(0) << " (length unit) " << std::endl;
std::cout << "Max_grid_size : " << max_grid_size << " (cells)" << std::endl;
std::cout << "Number of grids : " << bs.size() << std::endl;
}
// Allocate and define the right hand side.
bool do_4th = (solver_type==BoxLib_C4 || solver_type==All);
int ngr = (do_4th ? 1 : 0);
MultiFab rhs(bs, Ncomp, ngr);
setup_rhs(rhs, geom);
// Set up the Helmholtz operator coefficients.
MultiFab alpha(bs, Ncomp, 0);
PArray<MultiFab> beta(BL_SPACEDIM, PArrayManage);
for ( int n=0; n<BL_SPACEDIM; ++n ) {
BoxArray bx(bs);
beta.set(n, new MultiFab(bx.surroundingNodes(n), Ncomp, 0, Fab_allocate));
}
// The way HPGMG stores face-centered data is completely different than the
// way BoxLib does it, and translating between the two directly via indexing
// magic is a nightmare. Happily, the way this tutorial calculates
// face-centered values is by first calculating cell-centered values and then
// interpolating to the cell faces. HPGMG can do the same thing, so rather
// than converting directly from BoxLib's face-centered data to HPGMG's, just
// give HPGMG the cell-centered data and let it interpolate itself.
MultiFab beta_cc(bs,Ncomp,1); // cell-centered beta
setup_coeffs(bs, alpha, beta, geom, beta_cc);
MultiFab alpha4, beta4;
if (do_4th) {
alpha4.define(bs, Ncomp, 4, Fab_allocate);
beta4.define(bs, Ncomp, 3, Fab_allocate);
setup_coeffs4(bs, alpha4, beta4, geom);
}
MultiFab anaSoln;
if (comp_norm || plot_err || plot_asol) {
anaSoln.define(bs, Ncomp, 0, Fab_allocate);
compute_analyticSolution(anaSoln,Array<Real>(BL_SPACEDIM,0.5));
if (plot_asol) {
writePlotFile("ASOL", anaSoln, geom);
}
}
// Allocate the solution array
// Set the number of ghost cells in the solution array.
MultiFab soln(bs, Ncomp, 1);
MultiFab soln4;
if (do_4th) {
soln4.define(bs, Ncomp, 3, Fab_allocate);
}
MultiFab gphi(bs, BL_SPACEDIM, 0);
#ifdef USEHYPRE
if (solver_type == Hypre || solver_type == All) {
if (ParallelDescriptor::IOProcessor()) {
std::cout << "----------------------------------------" << std::endl;
std::cout << "Solving with Hypre " << std::endl;
}
solve(soln, anaSoln, gphi, a, b, alpha, beta, beta_cc, rhs, bs, geom, Hypre);
}
#endif
if (solver_type == BoxLib_C || solver_type == All) {
if (ParallelDescriptor::IOProcessor()) {
std::cout << "----------------------------------------" << std::endl;
std::cout << "Solving with BoxLib C++ solver " << std::endl;
}
solve(soln, anaSoln, gphi, a, b, alpha, beta, beta_cc, rhs, bs, geom, BoxLib_C);
}
if (solver_type == BoxLib_C4 || solver_type == All) {
if (ParallelDescriptor::IOProcessor()) {
std::cout << "----------------------------------------" << std::endl;
std::cout << "Solving with BoxLib C++ 4th order solver " << std::endl;
}
solve4(soln4, anaSoln, a, b, alpha4, beta4, rhs, bs, geom);
}
#ifdef USE_F90_SOLVERS
if (solver_type == BoxLib_F || solver_type == All) {
if (ParallelDescriptor::IOProcessor()) {
std::cout << "----------------------------------------" << std::endl;
std::cout << "Solving with BoxLib F90 solver " << std::endl;
}
solve(soln, anaSoln, gphi, a, b, alpha, beta, beta_cc, rhs, bs, geom, BoxLib_F);
}
#endif
#ifdef USEHPGMG
if (solver_type == HPGMG || solver_type == All) {
if (ParallelDescriptor::IOProcessor()) {
std::cout << "----------------------------------------" << std::endl;
std::cout << "Solving with HPGMG solver " << std::endl;
}
solve(soln, anaSoln, gphi, a, b, alpha, beta, beta_cc, rhs, bs, geom, HPGMG);
}
#endif
if (ParallelDescriptor::IOProcessor()) {
std::cout << "----------------------------------------" << std::endl;
}
BL_PROFILE_VAR_STOP(pmain);
BoxLib::Finalize();
}
int main(int argc, char**argv)
{
int block_sz = atoi(argv[1]);
int blocks_nr = atoi(argv[2]);
int edge_sz = atoi(argv[3]);
std::vector<Block> bs(blocks_nr, Block(block_sz,edge_sz) );
Edge e(edge_sz,blocks_nr,block_sz);
for (int i=0; i < bs.size(); ++i)
bs[i].fill();
e.fill();
// for (int i=0; i < bs.size(); ++i)
// std::cout << bs[i] << '\n';
// std::cout << e << '\n';
std::vector<Block> bs0 = bs;
Edge e0 = e;
double total_time=0.0;
clock_t st, end;
st = clock();
for (int i=0; i < bs.size(); ++i)
bs[i].fwd();
end = clock();
total_time += double((end - st)) / CLOCKS_PER_SEC;
for (int b=0; b < bs.size(); ++b)
{
for (int br=0; br < bs[b].rows; ++br)
{
double& diag = bs[b](br,br);
RowNO r;
r.v = &diag;
r.sz = bs[b].cols-br;
r.row_no = bs[b].rows*b + br;
st = clock();
elim_col(e,r);
end = clock();
total_time += double((end - st)) / CLOCKS_PER_SEC;
}
}
st = clock();
e.fwd();
e.bwd();
for (int i=0; i < bs.size(); ++i){
bs[i].bwd(e.edge_soln);
}
end = clock();
total_time += double((end - st)) / CLOCKS_PER_SEC;
std::cout << "elapsed " << total_time << '\n';
std::vector<double> soln(bs[0].rows*bs.size() + edge_sz);
for (int i=0; i < bs.size(); ++i)
for (int k=0; k < bs[i].sol.size(); ++k)
soln[i*block_sz+k] = bs[i].sol[k];
for (int k=0; k < e.edge_soln.size(); ++k)
soln[blocks_nr*block_sz+k] = e.edge_soln[k];
// for (int k=0; k < soln.size(); ++k)
// std::cout << soln[k] << '\n';
std::vector<double> Rhs;
for(int i = 0; i < bs0.size(); ++i)
{
for(int r=0; r < bs0[i].rows; ++r)
Rhs.push_back(bs0[i](r,bs0[i].cols-1));
}
for(int i = 0; i < e0.rows; ++i)
{
Rhs.push_back(e0(i,e0.cols-1));
}
std::vector<double> rhs = mult(bs0,e0,soln);
double residue=0.0;
for (int i=0; i < rhs.size(); ++i)
residue += fabs(rhs[i] - Rhs[i]);
std::cout << "\n\nres = " << residue << '\n';
}
int
main (int argc,
char* argv[])
{
BoxLib::Initialize(argc,argv);
std::cout << std::setprecision(10);
if (argc < 2)
{
std::cerr << "usage: " << argv[0] << " inputsfile [options]" << '\n';
exit(-1);
}
ParmParse pp;
int n;
BoxArray bs;
#if BL_SPACEDIM == 2
Box domain(IntVect(0,0),IntVect(11,11));
std::string boxfile("gr.2_small_a") ;
#elif BL_SPACEDIM == 3
Box domain(IntVect(0,0,0),IntVect(11,11,11));
std::string boxfile("grids/gr.3_2x3x4") ;
#endif
pp.query("boxes", boxfile);
std::ifstream ifs(boxfile.c_str(), std::ios::in);
if (!ifs)
{
std::string msg = "problem opening grids file: ";
msg += boxfile.c_str();
BoxLib::Abort(msg.c_str());
}
ifs >> domain;
if (ParallelDescriptor::IOProcessor())
std::cout << "domain: " << domain << std::endl;
bs.readFrom(ifs);
if (ParallelDescriptor::IOProcessor())
std::cout << "grids:\n" << bs << std::endl;
Geometry geom(domain);
const Real* H = geom.CellSize();
int ratio=2; pp.query("ratio", ratio);
// allocate/init soln and rhs
int Ncomp=BL_SPACEDIM;
int Nghost=0;
int Ngrids=bs.size();
MultiFab soln(bs, Ncomp, Nghost, Fab_allocate); soln.setVal(0.0);
MultiFab out(bs, Ncomp, Nghost, Fab_allocate);
MultiFab rhs(bs, Ncomp, Nghost, Fab_allocate); rhs.setVal(0.0);
for(MFIter rhsmfi(rhs); rhsmfi.isValid(); ++rhsmfi)
{
FORT_FILLRHS(rhs[rhsmfi].dataPtr(),
ARLIM(rhs[rhsmfi].loVect()),ARLIM(rhs[rhsmfi].hiVect()),
H,&Ncomp);
}
// Create the boundary object
MCViscBndry vbd(bs,geom);
BCRec phys_bc;
Array<int> lo_bc(BL_SPACEDIM), hi_bc(BL_SPACEDIM);
pp.getarr("lo_bc",lo_bc,0,BL_SPACEDIM);
pp.getarr("hi_bc",hi_bc,0,BL_SPACEDIM);
for (int i = 0; i < BL_SPACEDIM; i++)
{
phys_bc.setLo(i,lo_bc[i]);
phys_bc.setHi(i,hi_bc[i]);
}
// Create the BCRec's interpreted by ViscBndry objects
#if BL_SPACEDIM==2
Array<BCRec> pbcarray(4);
pbcarray[0] = BCRec(D_DECL(REFLECT_ODD,REFLECT_EVEN,EXT_DIR),
D_DECL(EXT_DIR,EXT_DIR,EXT_DIR));
pbcarray[1] = BCRec(D_DECL(REFLECT_EVEN,REFLECT_ODD,EXT_DIR),
D_DECL(EXT_DIR,EXT_DIR,EXT_DIR));
pbcarray[2] = BCRec(D_DECL(EXT_DIR,EXT_DIR,EXT_DIR),
D_DECL(EXT_DIR,EXT_DIR,EXT_DIR));
pbcarray[3] = BCRec(D_DECL(EXT_DIR,EXT_DIR,EXT_DIR),
D_DECL(EXT_DIR,EXT_DIR,EXT_DIR));
#elif BL_SPACEDIM==3
Array<BCRec> pbcarray(12);
#if 1
pbcarray[0] = BCRec(EXT_DIR,EXT_DIR,EXT_DIR,EXT_DIR,EXT_DIR,EXT_DIR);
pbcarray[1] = BCRec(EXT_DIR,EXT_DIR,EXT_DIR,EXT_DIR,EXT_DIR,EXT_DIR);
pbcarray[2] = BCRec(EXT_DIR,EXT_DIR,EXT_DIR,EXT_DIR,EXT_DIR,EXT_DIR);
pbcarray[3] = BCRec(D_DECL(EXT_DIR,EXT_DIR,EXT_DIR),
D_DECL(EXT_DIR,EXT_DIR,EXT_DIR));
pbcarray[4] = BCRec(D_DECL(EXT_DIR,EXT_DIR,EXT_DIR),
D_DECL(EXT_DIR,EXT_DIR,EXT_DIR));
pbcarray[5] = BCRec(D_DECL(EXT_DIR,EXT_DIR,EXT_DIR),
D_DECL(EXT_DIR,EXT_DIR,EXT_DIR));
pbcarray[6] = BCRec(D_DECL(EXT_DIR,EXT_DIR,EXT_DIR),
D_DECL(EXT_DIR,EXT_DIR,EXT_DIR));
pbcarray[7] = BCRec(D_DECL(EXT_DIR,EXT_DIR,EXT_DIR),
D_DECL(EXT_DIR,EXT_DIR,EXT_DIR));
pbcarray[8] = BCRec(D_DECL(EXT_DIR,EXT_DIR,EXT_DIR),
D_DECL(EXT_DIR,EXT_DIR,EXT_DIR));
pbcarray[9] = BCRec(D_DECL(EXT_DIR,EXT_DIR,EXT_DIR),
D_DECL(EXT_DIR,EXT_DIR,EXT_DIR));
pbcarray[10] = BCRec(D_DECL(EXT_DIR,EXT_DIR,EXT_DIR),
D_DECL(EXT_DIR,EXT_DIR,EXT_DIR));
pbcarray[11] = BCRec(D_DECL(EXT_DIR,EXT_DIR,EXT_DIR),
D_DECL(EXT_DIR,EXT_DIR,EXT_DIR));
#else
for (int i = 0; i < 12; i++)
pbcarray[i] = phys_bc;
#endif
#endif
Nghost = 1; // need space for bc info
MultiFab fine(bs,Ncomp,Nghost,Fab_allocate);
for(MFIter finemfi(fine); finemfi.isValid(); ++finemfi)
{
FORT_FILLFINE(fine[finemfi].dataPtr(),
ARLIM(fine[finemfi].loVect()),ARLIM(fine[finemfi].hiVect()),
H,&Ncomp);
}
// Create "background coarse data"
Box crse_bx = Box(domain).coarsen(ratio).grow(1);
BoxArray cba(crse_bx);
cba.maxSize(32);
Real h_crse[BL_SPACEDIM];
for (n=0; n<BL_SPACEDIM; n++) h_crse[n] = H[n]*ratio;
MultiFab crse_mf(cba, Ncomp, 0);
// FArrayBox crse_fab(crse_bx,Ncomp);
for (MFIter mfi(crse_mf); mfi.isValid(); ++mfi)
{
FORT_FILLCRSE(crse_mf[mfi].dataPtr(),
ARLIM(crse_mf[mfi].loVect()),ARLIM(crse_mf[mfi].hiVect()),
h_crse,&Ncomp);
}
// Create coarse boundary register, fill w/data from coarse FAB
int bndry_InRad=0;
int bndry_OutRad=1;
int bndry_Extent=1;
BoxArray cbs = BoxArray(bs).coarsen(ratio);
BndryRegister cbr(cbs,bndry_InRad,bndry_OutRad,bndry_Extent,Ncomp);
for (OrientationIter face; face; ++face)
{
Orientation f = face();
FabSet& bnd_fs(cbr[f]);
bnd_fs.copyFrom(crse_mf, 0, 0, 0, Ncomp);
}
// Interpolate crse data to fine boundary, where applicable
int cbr_Nstart=0;
int fine_Nstart=0;
int bndry_Nstart=0;
vbd.setBndryValues(cbr,cbr_Nstart,fine,fine_Nstart,
bndry_Nstart,Ncomp,ratio,pbcarray);
Nghost = 1; // other variables don't need extra space
DivVis lp(vbd,H);
Real a = 0.0;
Real b[BL_SPACEDIM];
b[0] = 1.0;
b[1] = 1.0;
#if BL_SPACEDIM>2
b[2] = 1.0;
#endif
MultiFab acoefs;
int NcompA = (BL_SPACEDIM == 2 ? 2 : 1);
acoefs.define(bs, NcompA, Nghost, Fab_allocate);
acoefs.setVal(a);
MultiFab bcoefs[BL_SPACEDIM];
for (n=0; n<BL_SPACEDIM; ++n)
{
BoxArray bsC(bs);
bcoefs[n].define(bsC.surroundingNodes(n), 1,
Nghost, Fab_allocate);
#if 1
for(MFIter bmfi(bcoefs[n]); bmfi.isValid(); ++bmfi)
{
FORT_MAKEMU(bcoefs[n][bmfi].dataPtr(),
ARLIM(bcoefs[n][bmfi].loVect()),ARLIM(bcoefs[n][bmfi].hiVect()),H,n);
}
#else
bcoefs[n].setVal(b[n]);
#endif
} // -->> over dimension
lp.setCoefficients(acoefs, bcoefs);
#if 1
lp.maxOrder(4);
#endif
Nghost = 1;
MultiFab tsoln(bs, Ncomp, Nghost, Fab_allocate);
tsoln.setVal(0.0);
#if 1
tsoln.copy(fine);
#endif
#if 0
// testing apply
lp.apply(out,tsoln);
Box subbox = out[0].box();
Real n1 = out[0].norm(subbox,1,0,BL_SPACEDIM)*pow(H[0],BL_SPACEDIM);
ParallelDescriptor::ReduceRealSum(n1);
if (ParallelDescriptor::IOProcessor())
{
cout << "n1 output is "<<n1<<std::endl;
}
out.minus(rhs,0,BL_SPACEDIM,0);
// special to single grid prob
Real n2 = out[0].norm(subbox,1,0,BL_SPACEDIM)*pow(H[0],BL_SPACEDIM);
ParallelDescriptor::ReduceRealSum(n2);
if (ParallelDescriptor::IOProcessor())
{
cout << "n2 difference is "<<n2<<std::endl;
}
#if 0
subbox.grow(-1);
Real n3 = out[0].norm(subbox,0,0,BL_SPACEDIM)*pow(H[0],BL_SPACEDIM);
ParallelDescriptor::ReduceRealMax(n3);
if (ParallelDescriptor::IOProcessor())
{
cout << "n3 difference is "<<n3<<std::endl;
}
#endif
#endif
const IntVect refRatio(D_DECL(2,2,2));
const Real bgVal = 1.0;
#if 1
#ifndef NDEBUG
// testing flux computation
BoxArray xfluxbox(bs);
xfluxbox.surroundingNodes(0);
MultiFab xflux(xfluxbox,Ncomp,Nghost,Fab_allocate);
xflux.setVal(1.e30);
BoxArray yfluxbox(bs);
yfluxbox.surroundingNodes(1);
MultiFab yflux(yfluxbox,Ncomp,Nghost,Fab_allocate);
yflux.setVal(1.e30);
#if BL_SPACEDIM>2
BoxArray zfluxbox(bs);
zfluxbox.surroundingNodes(2);
MultiFab zflux(zfluxbox,Ncomp,Nghost,Fab_allocate);
zflux.setVal(1.e30);
#endif
lp.compFlux(xflux,
yflux,
#if BL_SPACEDIM>2
zflux,
#endif
tsoln);
// Write fluxes
//writeMF(&xflux,"xflux.mfab");
//writeMF(&yflux,"yflux.mfab");
#if BL_SPACEDIM>2
//writeMF(&zflux,"zflux.mfab");
#endif
#endif
#endif
Real tolerance = 1.0e-10; pp.query("tol", tolerance);
Real tolerance_abs = 1.0e-10; pp.query("tol_abs", tolerance_abs);
#if 0
cout << "Bndry Data object:" << std::endl;
cout << lp.bndryData() << std::endl;
#endif
#if 0
bool use_mg_pre = false;
MCCGSolver cg(lp,use_mg_pre);
cg.solve(soln,rhs,tolerance,tolerance_abs);
#else
MCMultiGrid mg(lp);
mg.solve(soln,rhs,tolerance,tolerance_abs);
#endif
#if 0
cout << "MCLinOp object:" << std::endl;
cout << lp << std::endl;
#endif
VisMF::Write(soln,"soln");
#if 0
// apply operator to soln to see if really satisfies eqn
tsoln.copy(soln);
lp.apply(out,tsoln);
soln.copy(out);
// Output "apply" results on soln
VisMF::Write(soln,"apply");
// Compute truncation
for (MFIter smfi(soln); smfi.isValid(); ++smfi)
{
soln[smfi] -= fine[smfi];
}
for( int icomp=0; icomp < BL_SPACEDIM ; icomp++ )
{
Real solnMin = soln.min(icomp);
Real solnMax = soln.max(icomp);
ParallelDescriptor::ReduceRealMin(solnMin);
ParallelDescriptor::ReduceRealMax(solnMax);
if (ParallelDescriptor::IOProcessor())
{
cout << icomp << " "<<solnMin << " " << solnMax <<std::endl;
}
}
// Output truncation
VisMF::Write(soln,"trunc");
#endif
int dumpLp=0; pp.query("dumpLp",dumpLp);
bool write_lp = (dumpLp == 1 ? true : false);
if (write_lp)
std::cout << lp << std::endl;
// Output trunc
ParallelDescriptor::EndParallel();
}
