void setup_coeffs4(BoxArray& bs, MultiFab& alpha, MultiFab& beta, const Geometry& geom)
{
ParmParse pp;
Real sigma, w;
pp.get("sigma", sigma);
pp.get("w", w);
for ( MFIter mfi(alpha); mfi.isValid(); ++mfi ) {
const Box& bx = mfi.validbox();
const int* alo = alpha[mfi].loVect();
const int* ahi = alpha[mfi].hiVect();
const Box& abx = alpha[mfi].box();
FORT_SET_ALPHA(alpha[mfi].dataPtr(),ARLIM(alo),ARLIM(ahi),
abx.loVect(),abx.hiVect(),dx);
const int* clo = beta[mfi].loVect();
const int* chi = beta[mfi].hiVect();
FORT_SET_CC_COEF(beta[mfi].dataPtr(),ARLIM(clo),ARLIM(chi),
bx.loVect(),bx.hiVect(),dx, sigma, w);
}
if (plot_beta == 1) {
writePlotFile("BETA4", beta, geom);
}
}
void setup_coeffs(BoxArray& bs, MultiFab& alpha, PArray<MultiFab>& beta,
const Geometry& geom, MultiFab& cc_coef)
{
BL_PROFILE("setup_coeffs()");
ParmParse pp;
Real sigma, w;
pp.get("sigma", sigma);
pp.get("w", w);
for ( MFIter mfi(alpha); mfi.isValid(); ++mfi ) {
const Box& bx = mfi.validbox();
const int* alo = alpha[mfi].loVect();
const int* ahi = alpha[mfi].hiVect();
FORT_SET_ALPHA(alpha[mfi].dataPtr(),ARLIM(alo),ARLIM(ahi),
bx.loVect(),bx.hiVect(),dx);
const int* clo = cc_coef[mfi].loVect();
const int* chi = cc_coef[mfi].hiVect();
FORT_SET_CC_COEF(cc_coef[mfi].dataPtr(),ARLIM(clo),ARLIM(chi),
bx.loVect(),bx.hiVect(),dx, sigma, w);
}
BoxLib::average_cellcenter_to_face(beta, cc_coef, geom);
if (plot_beta == 1) {
writePlotFile("BETA", cc_coef, geom);
}
}
void setup_rhs(MultiFab& rhs, const Geometry& geom)
{
BL_PROFILE("setup_rhs()");
ParmParse pp;
Real a, b, sigma, w;
pp.get("a", a);
pp.get("b", b);
pp.get("sigma", sigma);
pp.get("w", w);
int ibnd = static_cast<int>(bc_type);
// We test the sum of the RHS to check solvability
Real sum_rhs = 0.;
for ( MFIter mfi(rhs); mfi.isValid(); ++mfi ) {
const int* rlo = rhs[mfi].loVect();
const int* rhi = rhs[mfi].hiVect();
const Box& bx = rhs[mfi].box();
FORT_SET_RHS(rhs[mfi].dataPtr(),ARLIM(rlo),ARLIM(rhi),
bx.loVect(),bx.hiVect(),dx, a, b, sigma, w, ibnd);
sum_rhs += rhs[mfi].sum(0,1);
}
for (int n=0; n < BL_SPACEDIM; n++) {
sum_rhs *= dx[n];
}
ParallelDescriptor::ReduceRealSum(sum_rhs, ParallelDescriptor::IOProcessorNumber());
if (ParallelDescriptor::IOProcessor()) {
std::cout << "Sum of RHS : " << sum_rhs << std::endl;
}
if (plot_rhs == 1) {
writePlotFile("RHS", rhs, geom);
}
}
/******************************************************************************
mergingAndClustering() : Clustering of GMMs and Realignment; Merging of states
inputs :
features - pointer to all feature vectors, allMixtureMeans, allMixtureVars
numstates, numMixEachState, totalNumFeatures,
posterior - posterior probability matrix, mixtureElemCount - Element count in each mixture
outputs : Perform viterbi realignment, clustering and merging of states or GMMs
******************************************************************************/
void ClusteringAndMerging(VECTOR_OF_F_VECTORS *features, VECTOR_OF_F_VECTORS *allMixtureMeans,
VECTOR_OF_F_VECTORS *allMixtureVars, int *numStates, int *numMixEachState,
int totalNumFeatures, float **posterior, float **mixtureElemCount,
int *numElemEachState, float *Pi, float *mixtureWeight)
{
// local Variable declaration
int VQIter = 7, GMMIter = 23, mergeIter=0, maxMergeIter = 16;
int viterbiIter = 5;
int i = 0, j = 0, k = 0,s= 0, flag = 1;
float *T1[MAX_NUM_STATES];
int *T2[MAX_NUM_STATES];
int *newStateSeq;
float *deltaBIC[MAX_NUM_STATES];
for(i = 0; i < (*numStates); i++){
deltaBIC[i] = (float *)calloc((*numStates), sizeof(float));
}
for(i = 0; i < (*numStates); i++){
T1[i] = (float *)calloc(totalNumFeatures, sizeof(float));
T2[i] = (int *)calloc(totalNumFeatures, sizeof(int));
}
// Merge two GMM States or two mixture models
//perform the steps repeatedly till there are no more states to be merged
while(flag && mergeIter < maxMergeIter){
printf("\nClustering and Merging Iteration: %d...........\n", mergeIter);
printf("starting with, States: %d \n", *numStates);
mergeIter++;
for(s = 0; s < *numStates; s++)
printf("state[%d] : %d \n", s, numElemEachState[s]);
for(i = 0; i < viterbiIter; i++){
//perform Viterbi Realignment
printf("performing viterbi realignment....\n");
newStateSeq = viterbi( features , numStates, allMixtureMeans, allMixtureVars, numMixEachState,
totalNumFeatures, posterior, numElemEachState, T1, T2,
Pi, mixtureWeight);
printf("Viterbi realignment has successfully completed...\n");
//find number of elements in each state
//FindNumberOfElemInEachState(newStateSeq, numStates, totalNumFeatures, numElemEachState, Pi);
//perform GMM clustering for all states
for(s = 0; s < (*numStates); s++){
// find all elements present in state s using viterbi_realignment
int count = 0, d = 0; // to count number of features in a state
for(j = 0; j < totalNumFeatures; j++){
if(newStateSeq[j] == s){
for(d = 0; d < DIM; d++){
featuresForClustering[count]->array[d] = features[j]->array[d];
featuresForClustering[count]->numElements = DIM;
}
count++;//increment count
}
}
extern VECTOR_OF_F_VECTORS *mixtureMeans, *mixtureVars;
extern float probScaleFactor;
extern int varianceNormalize, ditherMean;
int numFeatures = count;
//printf("count: %d num: %d\n", count, numElemEachState[s]);
int numMix = numMixEachState[s];
// Cluster using GMM
ComputeGMM(featuresForClustering, numFeatures, mixtureMeans, mixtureVars,
mixtureElemCount[s], numMix , VQIter, GMMIter, probScaleFactor,
ditherMean, varianceNormalize, time(NULL));
int mixCount = 0, k=0;
//store current mean and variance into all means and variance
for(j = 0; j < s; j++)
mixCount += numMixEachState[j];
for(j = mixCount; j < mixCount + numMixEachState[s]; j++){
for(k = 0; k < DIM; k++){
allMixtureMeans[j]->array[k] = mixtureMeans[j-mixCount]->array[k];
allMixtureVars[j]->array[k] = mixtureVars[j-mixCount]->array[k];
}
float mix_wt = mixtureElemCount[s][j-mixCount] / numElemEachState[s];
mixtureWeight[j] = mix_wt;
}
}//GMM Clustering for each state has completed
//calculate posterior probabilities
PrintAllDetails(numStates, numMixEachState, numElemEachState,
mixtureElemCount, allMixtureMeans, mixtureWeight);
ComputePosteriorProb(features, posterior, allMixtureMeans, allMixtureVars, numStates,
numMixEachState, totalNumFeatures, mixtureWeight);
}
// PrintAllDetails(numStates, numMixEachState, numElemEachState,
// mixtureElemCount, allMixtureMeans, mixtureWeight);
// Calculate BIC Value between Each state
CalculateBIC(deltaBIC, features , allMixtureMeans, allMixtureVars, numStates, totalNumFeatures,
newStateSeq, numMixEachState , numElemEachState, posterior);
int min_i =0, min_j = 0;
float min = 99999999;
// find minimum delta BIC states
for(j = 0; j < *numStates; j++){
for(k = j+1; k < *numStates; k++){
if(deltaBIC[j][k] < min){
min = deltaBIC[j][k];
min_i = j;
min_j = k;
}
}
}
printf("min_i: %d min_j: %d minDBIC: %f\n", min_i, min_j, deltaBIC[min_i][min_j]);
// if no two states can be merged set the flag to False to terminate process
if(deltaBIC[min_i][min_j] >= 0){
printf("We need to stop now ..... Final no of states: %d\n", *numStates);
flag = 0;
}
// Merge two states
if(flag)
MergeTwoStates( min_i, min_j, allMixtureMeans, allMixtureVars, numStates, numMixEachState,
numElemEachState, Pi, totalNumFeatures,
features, newStateSeq, mixtureWeight);
/*printf("Means after Merging of two states.................\n");
PrintAllDetails(numStates, numMixEachState, numElemEachState,
mixtureElemCount, allMixtureMeans);*/
for(i = 0; i < *numStates; i++)
printf("mix[%d]: %d\n", i, numMixEachState[i]);
}
//WRITE PLOT_File to plot distribution of each data point
writePlotFile(posterior, totalNumFeatures, numStates);
//WRITE RTTM FILE
writeRTTMFile(newStateSeq, numStates, totalNumFeatures, numElemEachState);
}
int
main (int argc, char* argv[])
{
BoxLib::Initialize(argc,argv);
// What time is it now? We'll use this to compute total run time.
Real strt_time = ParallelDescriptor::second();
std::cout << std::setprecision(15);
// ParmParse is way of reading inputs from the inputs file
ParmParse pp;
int verbose = 0;
pp.query("verbose", verbose);
// We need to get n_cell from the inputs file - this is the number of cells on each side of
// a square (or cubic) domain.
int n_cell;
pp.get("n_cell",n_cell);
int max_grid_size;
pp.get("max_grid_size",max_grid_size);
// Default plot_int to 1, allow us to set it to something else in the inputs file
// If plot_int < 0 then no plot files will be written
int plot_int = 1;
pp.query("plot_int",plot_int);
// Default nsteps to 0, allow us to set it to something else in the inputs file
int nsteps = 0;
pp.query("nsteps",nsteps);
pp.query("do_tiling", do_tiling);
// Define a single box covering the domain
IntVect dom_lo(0,0,0);
IntVect dom_hi(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 [-1,1] in each direction.
RealBox real_box;
for (int n = 0; n < BL_SPACEDIM; n++) {
real_box.setLo(n,-1.0);
real_box.setHi(n, 1.0);
}
// This says we are using Cartesian coordinates
int coord = 0;
// This sets the boundary conditions to be doubly or triply periodic
int is_per[BL_SPACEDIM];
for (int i = 0; i < BL_SPACEDIM; i++) is_per[i] = 1;
// This defines a Geometry object which is useful for writing the plotfiles
Geometry geom(domain,&real_box,coord,is_per);
// This defines the mesh spacing
Real dx[BL_SPACEDIM];
for ( int n=0; n<BL_SPACEDIM; n++ )
dx[n] = ( geom.ProbHi(n) - geom.ProbLo(n) )/domain.length(n);
// Nghost = number of ghost cells for each array
int Nghost = 1;
// Ncomp = number of components for each array
int Ncomp = 1;
pp.query("ncomp", Ncomp);
// Allocate space for the old_phi and new_phi -- we define old_phi and new_phi as
PArray < MultiFab > phis(2, PArrayManage);
phis.set(0, new MultiFab(bs, Ncomp, Nghost));
phis.set(1, new MultiFab(bs, Ncomp, Nghost));
MultiFab* old_phi = &phis[0];
MultiFab* new_phi = &phis[1];
// Initialize both to zero (just because)
old_phi->setVal(0.0);
new_phi->setVal(0.0);
// Initialize phi by calling a Fortran routine.
// MFIter = MultiFab Iterator
#ifdef _OPENMP
#pragma omp parallel
#endif
for ( MFIter mfi(*new_phi,true); mfi.isValid(); ++mfi )
{
const Box& bx = mfi.tilebox();
init_phi(bx.loVect(),bx.hiVect(),
BL_TO_FORTRAN((*new_phi)[mfi]),Ncomp,
dx,geom.ProbLo(),geom.ProbHi());
}
// Call the compute_dt routine to return a time step which we will pass to advance
Real dt = compute_dt(dx[0]);
// Write a plotfile of the initial data if plot_int > 0 (plot_int was defined in the inputs file)
if (plot_int > 0) {
int n = 0;
const std::string& pltfile = BoxLib::Concatenate("plt",n,5);
writePlotFile(pltfile, *new_phi, geom);
}
Real adv_start_time = ParallelDescriptor::second();
for (int n = 1; n <= nsteps; n++)
{
// Swap the pointers so we don't have to allocate and de-allocate data
std::swap(old_phi, new_phi);
// new_phi = old_phi + dt * (something)
advance(old_phi, new_phi, dx, dt, geom);
// Tell the I/O Processor to write out which step we're doing
if (verbose && ParallelDescriptor::IOProcessor())
std::cout << "Advanced step " << n << std::endl;
// Write a plotfile of the current data (plot_int was defined in the inputs file)
if (plot_int > 0 && n%plot_int == 0) {
const std::string& pltfile = BoxLib::Concatenate("plt",n,5);
writePlotFile(pltfile, *new_phi, geom);
}
}
// Call the timer again and compute the maximum difference between the start time and stop time
// over all processors
Real advance_time = ParallelDescriptor::second() - adv_start_time;
Real stop_time = ParallelDescriptor::second() - strt_time;
const int IOProc = ParallelDescriptor::IOProcessorNumber();
ParallelDescriptor::ReduceRealMax(stop_time,IOProc);
ParallelDescriptor::ReduceRealMax(advance_time,IOProc);
ParallelDescriptor::ReduceRealMax(kernel_time,IOProc);
ParallelDescriptor::ReduceRealMax(FB_time,IOProc);
// Tell the I/O Processor to write out the "run time"
if (ParallelDescriptor::IOProcessor()) {
std::cout << "Kernel time = " << kernel_time << std::endl;
std::cout << "FB time = " << FB_time << std::endl;
std::cout << "Advance time = " << advance_time << std::endl;
std::cout << "Total run time = " << stop_time << std::endl;
}
// Say goodbye to MPI, etc...
BoxLib::Finalize();
}
void main_main ()
{
// What time is it now? We'll use this to compute total run time.
Real strt_time = ParallelDescriptor::second();
std::cout << std::setprecision(15);
int n_cell, max_grid_size, nsteps, plot_int, is_periodic[BL_SPACEDIM];
// Boundary conditions
Array<int> lo_bc(BL_SPACEDIM), hi_bc(BL_SPACEDIM);
// inputs parameters
{
// ParmParse is way of reading inputs from the inputs file
ParmParse pp;
// We need to get n_cell from the inputs file - this is the number of cells on each side of
// a square (or cubic) domain.
pp.get("n_cell",n_cell);
// Default nsteps to 0, allow us to set it to something else in the inputs file
pp.get("max_grid_size",max_grid_size);
// Default plot_int to 1, allow us to set it to something else in the inputs file
// If plot_int < 0 then no plot files will be written
plot_int = 1;
pp.query("plot_int",plot_int);
// Default nsteps to 0, allow us to set it to something else in the inputs file
nsteps = 0;
pp.query("nsteps",nsteps);
// Boundary conditions - default is periodic (INT_DIR)
for (int i = 0; i < BL_SPACEDIM; ++i)
{
lo_bc[i] = hi_bc[i] = INT_DIR; // periodic boundaries are interior boundaries
}
pp.queryarr("lo_bc",lo_bc,0,BL_SPACEDIM);
pp.queryarr("hi_bc",hi_bc,0,BL_SPACEDIM);
}
// make BoxArray and Geometry
BoxArray ba;
Geometry geom;
{
IntVect dom_lo(IntVect(D_DECL(0,0,0)));
IntVect dom_hi(IntVect(D_DECL(n_cell-1, n_cell-1, n_cell-1)));
Box domain(dom_lo, dom_hi);
// Initialize the boxarray "ba" from the single box "bx"
ba.define(domain);
// Break up boxarray "ba" into chunks no larger than "max_grid_size" along a direction
ba.maxSize(max_grid_size);
// This defines the physical size of the box. Right now the box is [-1,1] in each direction.
RealBox real_box;
for (int n = 0; n < BL_SPACEDIM; n++) {
real_box.setLo(n,-1.0);
real_box.setHi(n, 1.0);
}
// This says we are using Cartesian coordinates
int coord = 0;
// This sets the boundary conditions to be doubly or triply periodic
int is_periodic[BL_SPACEDIM];
for (int i = 0; i < BL_SPACEDIM; i++)
{
is_periodic[i] = 0;
if (lo_bc[i] == 0 && hi_bc[i] == 0) {
is_periodic[i] = 1;
}
}
// This defines a Geometry object
geom.define(domain,&real_box,coord,is_periodic);
}
// Boundary conditions
PhysBCFunct physbcf;
BCRec bcr(&lo_bc[0], &hi_bc[0]);
physbcf.define(geom, bcr, BndryFunctBase(phifill)); // phifill is a fortran function
// define dx[]
const Real* dx = geom.CellSize();
// Nghost = number of ghost cells for each array
int Nghost = 1;
// Ncomp = number of components for each array
int Ncomp = 1;
// time = starting time in the simulation
Real time = 0.0;
// we allocate two phi multifabs; one will store the old state, the other the new
// we swap the indices each time step to avoid copies of new into old
PArray<MultiFab> phi(2, PArrayManage);
phi.set(0, new MultiFab(ba, Ncomp, Nghost));
phi.set(1, new MultiFab(ba, Ncomp, Nghost));
// Initialize both to zero (just because)
phi[0].setVal(0.0);
phi[1].setVal(0.0);
// Initialize phi[init_index] by calling a Fortran routine.
// MFIter = MultiFab Iterator
int init_index = 0;
for ( MFIter mfi(phi[init_index]); mfi.isValid(); ++mfi )
{
const Box& bx = mfi.validbox();
init_phi(phi[init_index][mfi].dataPtr(),
bx.loVect(), bx.hiVect(), &Nghost,
geom.CellSize(), geom.ProbLo(), geom.ProbHi());
}
// compute the time step
Real dt = 0.9*dx[0]*dx[0] / (2.0*BL_SPACEDIM);
// Write a plotfile of the initial data if plot_int > 0 (plot_int was defined in the inputs file)
if (plot_int > 0)
{
int n = 0;
const std::string& pltfile = BoxLib::Concatenate("plt",n,5);
writePlotFile(pltfile, phi[init_index], geom, time);
}
// build the flux multifabs
PArray<MultiFab> flux(BL_SPACEDIM, PArrayManage);
for (int dir = 0; dir < BL_SPACEDIM; dir++)
{
// flux(dir) has one component, zero ghost cells, and is nodal in direction dir
BoxArray edge_ba = ba;
edge_ba.surroundingNodes(dir);
flux.set(dir, new MultiFab(edge_ba, 1, 0));
}
int old_index = init_index;
for (int n = 1; n <= nsteps; n++, old_index = 1 - old_index)
{
int new_index = 1 - old_index;
// new_phi = old_phi + dt * (something)
advance(phi[old_index], phi[new_index], flux, time, dt, geom, physbcf, bcr);
time = time + dt;
// Tell the I/O Processor to write out which step we're doing
if (ParallelDescriptor::IOProcessor())
std::cout << "Advanced step " << n << std::endl;
// Write a plotfile of the current data (plot_int was defined in the inputs file)
if (plot_int > 0 && n%plot_int == 0)
{
const std::string& pltfile = BoxLib::Concatenate("plt",n,5);
writePlotFile(pltfile, phi[new_index], geom, time);
}
}
// Call the timer again and compute the maximum difference between the start time and stop time
// over all processors
Real stop_time = ParallelDescriptor::second() - strt_time;
const int IOProc = ParallelDescriptor::IOProcessorNumber();
ParallelDescriptor::ReduceRealMax(stop_time,IOProc);
// Tell the I/O Processor to write out the "run time"
if (ParallelDescriptor::IOProcessor()) {
std::cout << "Run time = " << stop_time << std::endl;
}
}
int
main (int argc, char* argv[])
{
BoxLib::Initialize(argc,argv);
// What time is it now? We'll use this to compute total run time.
Real strt_time = ParallelDescriptor::second();
std::cout << std::setprecision(15);
// ParmParse is way of reading inputs from the inputs file
ParmParse pp;
// We need to get n_cell from the inputs file - this is the number of cells on each side of
// a square (or cubic) domain.
int n_cell;
pp.get("n_cell",n_cell);
// Default nsteps to 0, allow us to set it to something else in the inputs file
int max_grid_size;
pp.get("max_grid_size",max_grid_size);
// Default plot_int to 1, allow us to set it to something else in the inputs file
// If plot_int < 0 then no plot files will be written
int plot_int = 1;
pp.query("plot_int",plot_int);
// Default nsteps to 0, allow us to set it to something else in the inputs file
int nsteps = 0;
pp.query("nsteps",nsteps);
// Define a single box covering the domain
#if (BL_SPACEDIM == 2)
IntVect dom_lo(0,0);
IntVect dom_hi(n_cell-1,n_cell-1);
#else
IntVect dom_lo(0,0,0);
IntVect dom_hi(n_cell-1,n_cell-1,n_cell-1);
#endif
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 [-1,1] in each direction.
RealBox real_box;
for (int n = 0; n < BL_SPACEDIM; n++)
{
real_box.setLo(n,-1.0);
real_box.setHi(n, 1.0);
}
// This says we are using Cartesian coordinates
int coord = 0;
// This sets the boundary conditions to be doubly or triply periodic
int is_per[BL_SPACEDIM];
for (int i = 0; i < BL_SPACEDIM; i++) is_per[i] = 1;
// This defines a Geometry object which is useful for writing the plotfiles
Geometry geom(domain,&real_box,coord,is_per);
// This defines the mesh spacing
Real dx[BL_SPACEDIM];
for ( int n=0; n<BL_SPACEDIM; n++ )
dx[n] = ( geom.ProbHi(n) - geom.ProbLo(n) )/domain.length(n);
// Nghost = number of ghost cells for each array
int Nghost = 1;
// Ncomp = number of components for each array
int Ncomp = 1;
// Make sure we can fill the ghost cells from the adjacent grid
if (Nghost > max_grid_size)
std::cout << "NGHOST < MAX_GRID_SIZE -- grids are too small! " << std::endl;
// Allocate space for the old_phi and new_phi -- we define old_phi and new_phi as
// pointers to the MultiFabs
MultiFab* old_phi = new MultiFab(bs, Ncomp, Nghost);
MultiFab* new_phi = new MultiFab(bs, Ncomp, Nghost);
// Initialize both to zero (just because)
old_phi->setVal(0.0);
new_phi->setVal(0.0);
// Initialize the old_phi by calling a Fortran routine.
// MFIter = MultiFab Iterator
for ( MFIter mfi(*new_phi); mfi.isValid(); ++mfi )
{
const Box& bx = mfi.validbox();
FORT_INIT_PHI((*new_phi)[mfi].dataPtr(),
bx.loVect(),bx.hiVect(), &Nghost,
dx,geom.ProbLo(),geom.ProbHi());
}
// Call the compute_dt routine to return a time step which we will pass to advance
Real dt = compute_dt(dx[0]);
// Write a plotfile of the initial data if plot_int > 0 (plot_int was defined in the inputs file)
if (plot_int > 0)
{
int n = 0;
const std::string& pltfile = BoxLib::Concatenate("plt",n,5);
writePlotFile(pltfile, *new_phi, geom);
}
// build the flux multifabs
MultiFab* flux = new MultiFab[BL_SPACEDIM];
for (int dir = 0; dir < BL_SPACEDIM; dir++)
{
BoxArray edge_grids(bs);
// flux(dir) has one component, zero ghost cells, and is nodal in direction dir
edge_grids.surroundingNodes(dir);
flux[dir].define(edge_grids,1,0,Fab_allocate);
}
for (int n = 1; n <= nsteps; n++)
{
// Swap the pointers so we don't have to allocate and de-allocate data
std::swap(old_phi, new_phi);
// new_phi = old_phi + dt * (something)
advance(old_phi, new_phi, flux, dx, dt, geom);
// Tell the I/O Processor to write out which step we're doing
if (ParallelDescriptor::IOProcessor())
std::cout << "Advanced step " << n << std::endl;
// Write a plotfile of the current data (plot_int was defined in the inputs file)
if (plot_int > 0 && n%plot_int == 0)
{
const std::string& pltfile = BoxLib::Concatenate("plt",n,5);
writePlotFile(pltfile, *new_phi, geom);
}
}
// Call the timer again and compute the maximum difference between the start time and stop time
// over all processors
Real stop_time = ParallelDescriptor::second() - strt_time;
const int IOProc = ParallelDescriptor::IOProcessorNumber();
ParallelDescriptor::ReduceRealMax(stop_time,IOProc);
// Tell the I/O Processor to write out the "run time"
if (ParallelDescriptor::IOProcessor())
std::cout << "Run time = " << stop_time << std::endl;
// Say goodbye to MPI, etc...
BoxLib::Finalize();
}
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();
}
void solve_with_HPGMG(MultiFab& soln, MultiFab& gphi, Real a, Real b, MultiFab& alpha, PArray<MultiFab>& beta,
MultiFab& beta_cc, MultiFab& rhs, const BoxArray& bs, const Geometry& geom, int n_cell)
{
BndryData bd(bs, 1, geom);
set_boundary(bd, rhs, 0);
ABecLaplacian abec_operator(bd, dx);
abec_operator.setScalars(a, b);
abec_operator.setCoefficients(alpha, beta);
int minCoarseDim;
if (domain_boundary_condition == BC_PERIODIC)
{
minCoarseDim = 2; // avoid problems with black box calculation of D^{-1} for poisson with periodic BC's on a 1^3 grid
}
else
{
minCoarseDim = 1; // assumes you can drop order on the boundaries
}
level_type level_h;
mg_type MG_h;
int numVectors = 12;
int my_rank = 0, num_ranks = 1;
#ifdef BL_USE_MPI
MPI_Comm_size (MPI_COMM_WORLD, &num_ranks);
MPI_Comm_rank (MPI_COMM_WORLD, &my_rank);
#endif /* BL_USE_MPI */
const double h0 = dx[0];
// Create the geometric structure of the HPGMG grid using the RHS MultiFab as
// a template. This doesn't copy any actual data.
CreateHPGMGLevel(&level_h, rhs, n_cell, max_grid_size, my_rank, num_ranks, domain_boundary_condition, numVectors, h0);
// Set up the coefficients for the linear operator L.
SetupHPGMGCoefficients(a, b, alpha, beta_cc, &level_h);
// Now that the HPGMG grid is built, populate it with RHS data.
ConvertToHPGMGLevel(rhs, n_cell, max_grid_size, &level_h, VECTOR_F);
#ifdef USE_HELMHOLTZ
if (ParallelDescriptor::IOProcessor()) {
std::cout << "Creating Helmholtz (a=" << a << ", b=" << b << ") test problem" << std::endl;;
}
#else
if (ParallelDescriptor::IOProcessor()) {
std::cout << "Creating Poisson (a=" << a << ", b=" << b << ") test problem" << std::endl;;
}
#endif /* USE_HELMHOLTZ */
if (level_h.boundary_condition.type == BC_PERIODIC)
{
double average_value_of_f = mean (&level_h, VECTOR_F);
if (average_value_of_f != 0.0)
{
if (ParallelDescriptor::IOProcessor())
{
std::cerr << "WARNING: Periodic boundary conditions, but f does not sum to zero... mean(f)=" << average_value_of_f << std::endl;
}
//shift_vector(&level_h,VECTOR_F,VECTOR_F,-average_value_of_f);
}
}
//- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
rebuild_operator(&level_h,NULL,a,b); // i.e. calculate Dinv and lambda_max
MGBuild(&MG_h,&level_h,a,b,minCoarseDim,ParallelDescriptor::Communicator()); // build the Multigrid Hierarchy
//- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
if (ParallelDescriptor::IOProcessor())
std::cout << std::endl << std::endl << "===== STARTING SOLVE =====" << std::endl << std::flush;
MGResetTimers (&MG_h);
zero_vector (MG_h.levels[0], VECTOR_U);
#ifdef USE_FCYCLES
FMGSolve (&MG_h, 0, VECTOR_U, VECTOR_F, a, b, tolerance_abs, tolerance_rel);
#else
MGSolve (&MG_h, 0, VECTOR_U, VECTOR_F, a, b, tolerance_abs, tolerance_rel);
#endif /* USE_FCYCLES */
MGPrintTiming (&MG_h, 0); // don't include the error check in the timing results
//- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
if (ParallelDescriptor::IOProcessor())
std::cout << std::endl << std::endl << "===== Performing Richardson error analysis ==========================" << std::endl;
// solve A^h u^h = f^h
// solve A^2h u^2h = f^2h
// solve A^4h u^4h = f^4h
// error analysis...
MGResetTimers(&MG_h);
const double dtol = tolerance_abs;
const double rtol = tolerance_rel;
int l;for(l=0;l<3;l++){
if(l>0)restriction(MG_h.levels[l],VECTOR_F,MG_h.levels[l-1],VECTOR_F,RESTRICT_CELL);
zero_vector(MG_h.levels[l],VECTOR_U);
#ifdef USE_FCYCLES
FMGSolve(&MG_h,l,VECTOR_U,VECTOR_F,a,b,dtol,rtol);
#else
MGSolve(&MG_h,l,VECTOR_U,VECTOR_F,a,b,dtol,rtol);
#endif
}
richardson_error(&MG_h,0,VECTOR_U);
// Now convert solution from HPGMG back to rhs MultiFab.
ConvertFromHPGMGLevel(soln, &level_h, VECTOR_U);
const double norm_from_HPGMG = norm(&level_h, VECTOR_U);
const double mean_from_HPGMG = mean(&level_h, VECTOR_U);
const Real norm0 = soln.norm0();
const Real norm2 = soln.norm2();
if (ParallelDescriptor::IOProcessor()) {
std::cout << "mean from HPGMG: " << mean_from_HPGMG << std::endl;
std::cout << "norm from HPGMG: " << norm_from_HPGMG << std::endl;
std::cout << "norm0 of RHS copied to MF: " << norm0 << std::endl;
std::cout << "norm2 of RHS copied to MF: " << norm2 << std::endl;
}
// Write the MF to disk for comparison with the in-house solver
if (plot_soln)
{
writePlotFile("SOLN-HPGMG", soln, geom);
}
//- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
MGDestroy(&MG_h);
destroy_level(&level_h);
//- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
PArray<MultiFab> grad_phi(BL_SPACEDIM, PArrayManage);
for (int n = 0; n < BL_SPACEDIM; ++n)
grad_phi.set(n, new MultiFab(BoxArray(soln.boxArray()).surroundingNodes(n), 1, 0));
#if (BL_SPACEDIM == 2)
abec_operator.compFlux(grad_phi[0],grad_phi[1],soln);
#elif (BL_SPACEDIM == 3)
abec_operator.compFlux(grad_phi[0],grad_phi[1],grad_phi[2],soln);
#endif
// Average edge-centered gradients to cell centers.
BoxLib::average_face_to_cellcenter(gphi, grad_phi, geom);
}
void solve(MultiFab& soln, const MultiFab& anaSoln, MultiFab& gphi,
Real a, Real b, MultiFab& alpha, PArray<MultiFab>& beta, MultiFab& beta_cc,
MultiFab& rhs, const BoxArray& bs, const Geometry& geom,
solver_t solver)
{
BL_PROFILE("solve()");
std::string ss;
soln.setVal(0.0);
const Real run_strt = ParallelDescriptor::second();
if (solver == BoxLib_C) {
ss = "CPP";
solve_with_Cpp(soln, gphi, a, b, alpha, beta, rhs, bs, geom);
}
#ifdef USE_F90_SOLVERS
else if (solver == BoxLib_F) {
ss = "F90";
solve_with_F90(soln, gphi, a, b, alpha, beta, rhs, bs, geom);
}
#endif
#ifdef USEHYPRE
else if (solver == Hypre) {
ss = "Hyp";
solve_with_hypre(soln, a, b, alpha, beta, rhs, bs, geom);
}
#endif
#ifdef USEHPGMG
else if (solver == HPGMG) {
ss = "HPGMG";
solve_with_HPGMG(soln, gphi, a, b, alpha, beta, beta_cc, rhs, bs, geom, n_cell);
}
#endif
else {
BoxLib::Error("Invalid solver");
}
Real run_time = ParallelDescriptor::second() - run_strt;
ParallelDescriptor::ReduceRealMax(run_time, ParallelDescriptor::IOProcessorNumber());
if (ParallelDescriptor::IOProcessor()) {
std::cout << " Run time : " << run_time << std::endl;
}
if (plot_soln) {
writePlotFile("SOLN-"+ss, soln, geom);
writePlotFile("GPHI-"+ss, gphi, geom);
}
if (plot_err || comp_norm) {
soln.minus(anaSoln, 0, Ncomp, 0); // soln contains errors now
MultiFab& err = soln;
if (plot_err) {
writePlotFile("ERR-"+ss, soln, geom);
}
if (comp_norm) {
Real twoNorm = err.norm2();
Real maxNorm = err.norm0();
err.setVal(1.0);
Real vol = err.norm2();
twoNorm /= vol;
if (ParallelDescriptor::IOProcessor()) {
std::cout << " 2 norm error : " << twoNorm << std::endl;
std::cout << " max norm error: " << maxNorm << std::endl;
}
}
}
}
void solve4(MultiFab& soln, const MultiFab& anaSoln,
Real a, Real b, MultiFab& alpha, MultiFab& beta,
MultiFab& rhs, const BoxArray& bs, const Geometry& geom)
{
std::string ss = "CPP";
soln.setVal(0.0);
const Real run_strt = ParallelDescriptor::second();
BndryData bd(bs, 1, geom);
set_boundary(bd, rhs, 0);
ABec4 abec_operator(bd, dx);
abec_operator.setScalars(a, b);
MultiFab betaca(bs,1,2);
ABec4::cc2ca(beta,betaca,0,0,1);
MultiFab alphaca(bs,1,2);
ABec4::cc2ca(alpha,alphaca,0,0,1);
abec_operator.setCoefficients(alphaca, betaca);
MultiFab rhsca(bs,1,0);
ABec4::cc2ca(rhs,rhsca,0,0,1);
MultiFab out(bs,1,0);
compute_analyticSolution(soln,Array<Real>(BL_SPACEDIM,0.5));
MultiFab solnca(bs,1,2);
solnca.setVal(0);
MultiGrid mg(abec_operator);
mg.setVerbose(verbose);
mg.solve(solnca, rhsca, tolerance_rel, tolerance_abs);
Real run_time = ParallelDescriptor::second() - run_strt;
ParallelDescriptor::ReduceRealMax(run_time, ParallelDescriptor::IOProcessorNumber());
if (ParallelDescriptor::IOProcessor()) {
std::cout << " Run time : " << run_time << std::endl;
}
if (plot_soln) {
writePlotFile("SOLN-"+ss, solnca, geom);
}
if (plot_err || comp_norm) {
soln.minus(anaSoln, 0, Ncomp, 0); // soln contains errors now
MultiFab& err = soln;
if (plot_err) {
writePlotFile("ERR-"+ss, soln, geom);
}
if (comp_norm) {
Real twoNorm = err.norm2();
Real maxNorm = err.norm0();
err.setVal(1.0);
Real vol = err.norm2();
twoNorm /= vol;
if (ParallelDescriptor::IOProcessor()) {
std::cout << " 2 norm error : " << twoNorm << std::endl;
std::cout << " max norm error: " << maxNorm << std::endl;
}
}
}
}
void
SMC::evolve ()
{
int istep = 0;
int last_plt_written = -1;
if (plot_int > 0) {
writePlotFile(istep);
last_plt_written = istep;
}
int init_step = 1;
if (ParallelDescriptor::IOProcessor()) {
std::cout << "\n" << "BEGIN MAIN EVOLUTION LOOP" << "\n" << std::endl;
}
ParallelDescriptor::Barrier();
Real wt0 = ParallelDescriptor::second();
if ( (max_step >= init_step) && (t < stop_time || stop_time < 0.0) )
{
for (istep = init_step; istep <= max_step; ++istep)
{
if (ParallelDescriptor::IOProcessor()) {
std::cout << "Advancing time step " << istep << " time = " << t << "\n";
}
advance(istep);
t = t + dt;
if (ParallelDescriptor::IOProcessor()) {
std::cout << "End of step " << istep << " time = " << t << "\n" << std::endl;
}
if (plot_int > 0 && istep%plot_int == 0) {
writePlotFile(istep);
last_plt_written = istep;
}
if (stop_time >= 0.0 && t >= stop_time) {
break;
}
}
if (istep > max_step) istep = max_step;
if (plot_int > 0 && last_plt_written != istep) {
writePlotFile(istep);
}
}
#ifdef BL_USE_UPCXX
upcxx::barrier();
#else
ParallelDescriptor::Barrier();
#endif
Real wt1 = ParallelDescriptor::second();
Real wt_fb = wt_fb1 + wt_fb2;
Real wt_chem = wt_chem1 + wt_chem2;
ParallelDescriptor::ReduceRealMax(wt_fb , ParallelDescriptor::IOProcessorNumber());
ParallelDescriptor::ReduceRealMax(wt_chem , ParallelDescriptor::IOProcessorNumber());
ParallelDescriptor::ReduceRealMax(wt_hypdiff, ParallelDescriptor::IOProcessorNumber());
if (ParallelDescriptor::IOProcessor()) {
std::cout << "======================================" << std::endl;
std::cout << " Total Time : " << wt1-wt0 << "\n";
std::cout << " Communication time: " << wt_fb << "\n";
std::cout << " Chemistry time: " << wt_chem << "\n";
std::cout << " Hyp-Diff time: " << wt_hypdiff << "\n";
std::cout << "======================================" << std::endl;
}
}
