void CurrentSource::add_initial_dc()
{
if (!b_status)
return;
b_i(0) -= m_i;
b_i(1) += m_i;
}
vector<int> grid_index::points_near_segment(Vec2 a, Vec2 b, mesh2d& idxd, int additional_cells) {
/* this function will return points that are nearby a potential segment (NOT present inside mesh2d)
NOTE: we could make this code specific to the use case required, meaning line-point distance, but it's not really necessary and sounds like premature optimization, on top of requiring more filtering (needs the point nearest to pt_a.
*/
vector<int> results;
set<Vec2i> cells;
Vec2i a_i(a * resolution);
Vec2i b_i(b * resolution);
// msgm(a_i, b_i);
// if (a_i != b_i)
{
int x_lo = min(a_i.x, b_i.x)-additional_cells;
int x_hi = max(a_i.x, b_i.x)+additional_cells;
int y_lo = min(a_i.y, b_i.y)-additional_cells;
int y_hi = max(a_i.y, b_i.y)+additional_cells;
for (int x = x_lo; x <= x_hi; ++x) {
for (int y = y_lo; y <= y_hi; ++y) {
if(x>-1 && y>-1 && x<resolution && y<resolution){
cells.insert({x,y});
}
}
}
}
// else {
// cells.insert(a_i);
// }
// msg(cells);
set<int> already;
for (auto&cell : cells) {
// auto range = bucket_segments.equal_range(cell);
auto range = bucket_points.equal_range(cell);
for (auto it = range.first; it != range.second; ++it) {
// if (already.count(it->second)) {msgm("skipping,",it->second); continue;}
if (already.count(it->second)) {continue;}
// msg(it->second);
already.insert(it->second);
results.push_back(it->second);
}
}
return results;
}
long matrix::mul_ijk(const matrix *a, const matrix *b, matrix *r)
{
int width, height;
a->size(&height, &width);
for (int i = 0; i < height; i++) {
iter_row r_i(r, i);
for (int j = 0; j < height; j++) {
iter_row a_i(a, i);
iter_col b_i(b, j);
double n = 0;
for (int k = 0; k < width; k++) {
n += **a_i * **b_i;
a_i++; b_i++;
}
**r_i = n;
r_i++;
}
}
return (long)width * width * height * 2;
}
long matrix::mul_kij(const matrix *a, const matrix *b, matrix *r)
{
int width, height, h_div, h_mod;
a->size(&height, &width);
h_div = height / 16;
h_mod = height % 16;
for (iter_all i(r); *i; i++)
**i = 0;
for (int k = 0; k < width; k++) {
iter_col a_i(a, k);
for (int i = 0; i < height; i++) {
number n = **a_i; a_i++;
iter_row b_i(b, k);
iter_row r_i(r, i);
for (int j = 0; j < h_div; j++) {
(*r_i)[0] += (*b_i)[0] * n;
(*r_i)[1] += (*b_i)[1] * n;
(*r_i)[2] += (*b_i)[2] * n;
(*r_i)[3] += (*b_i)[3] * n;
(*r_i)[4] += (*b_i)[4] * n;
(*r_i)[5] += (*b_i)[5] * n;
(*r_i)[6] += (*b_i)[6] * n;
(*r_i)[7] += (*b_i)[7] * n;
(*r_i)[8] += (*b_i)[8] * n;
(*r_i)[9] += (*b_i)[9] * n;
(*r_i)[10] += (*b_i)[10] * n;
(*r_i)[11] += (*b_i)[11] * n;
(*r_i)[12] += (*b_i)[12] * n;
(*r_i)[13] += (*b_i)[13] * n;
(*r_i)[14] += (*b_i)[14] * n;
(*r_i)[15] += (*b_i)[15] * n;
r_i += 16; b_i += 16;
}
for (int j = 0; j < h_mod; j++) {
**r_i += **b_i * n;
r_i++; b_i++;
}
}
}
return (long)width * width * height * 2;
}
void grid_index::add(Vec2 a, Vec2 b, int i) {
Vec2i a_i(a * resolution);
Vec2i b_i(b * resolution);
// msgm("adding segment", i, "coords", a, b);
if (a_i != b_i) {
int x_lo = min(a_i.x, b_i.x);
int x_hi = max(a_i.x, b_i.x);
int y_lo = min(a_i.y, b_i.y);
int y_hi = max(a_i.y, b_i.y);
for (int x = x_lo; x <= x_hi; ++x) {
for (int y = y_lo; y <= y_hi; ++y) {
bucket_segments.insert({ {x,y},i});
}
}
}
else {
//msgm("adding", i, )
bucket_segments .insert({ a_i,i});
}
}
void add(Vec2 a, Vec2 b, int i) {
Vec2i a_i(a / resolution);
Vec2i b_i(b / resolution);
if (a_i != b_i) {
int x_lo = min(a_i.x, b_i.x);
int x_hi = max(a_i.x, b_i.x);
int y_lo = min(a_i.y, b_i.y);
int y_hi = max(a_i.y, b_i.y);
for (int x = x_lo; x <= x_hi; ++x) {
for (int y = y_lo; y <= y_hi; ++y) {
bucket_index.insert({ {x,y},i });
}
}
}
else {
bucket_index.insert({ a_i,i });
}
if (i == 0) msg("SAY WAT AGAIN");
}
int grid_index::intersects(Vec2 a, Vec2 b, mesh2d& idxd, vector<pair<int,Vec2>> &results) {
set<Vec2i> cells;
Vec2i a_i(a * resolution);
Vec2i b_i(b * resolution);
// msgm(a_i, b_i);
if (a_i != b_i) {
int x_lo = min(a_i.x, b_i.x);
int x_hi = max(a_i.x, b_i.x);
int y_lo = min(a_i.y, b_i.y);
int y_hi = max(a_i.y, b_i.y);
for (int x = x_lo; x <= x_hi; ++x) {
for (int y = y_lo; y <= y_hi; ++y) {
cells.insert({x,y});
}
}
}
else {
cells.insert(a_i);
}
set<int> already;
for (auto&cell : cells) {
// auto range = bucket_index.equal_range(cell);
auto range = bucket_segments.equal_range(cell);
for (auto it = range.first; it != range.second; ++it) {
// if (it->second > -1 || already.count(it->second)) continue;
if (already.count(it->second)) continue;
already.insert(it->second);
// auto segment = idxd[-(it->second)-1];
auto segment = idxd[it->second];
//pair<Vec2, Vec2> segment = idxd[it->second];
Vec2 itsc_pt;
bool inter = ::intersects(a, b, segment.first, segment.second, itsc_pt);
if (inter) {
results.push_back({ it->second, itsc_pt });
}
}
}
return results.size();
}
void SolveSystem (
Vector<double, int>& x_local,
const MatrixDense<double, int>& K_local,
const Vector<double, int>& b_local,
const int numb_node_i,
const int* list_node_i,
const int* l2i,
int numb_node_p,
const int* list_node_p,
const int* l2p,
const int numb_global_node,
const int numb_l2g,
const int* l2g,
const MPI_Comm& mpi_comm ) {
// -- number of processors
int numb_procs;
// -- process number (process rank)
int proc_numb;
// -- get number of processes
MPI_Comm_size( mpi_comm, &numb_procs );
// -- get current process rank
MPI_Comm_rank( mpi_comm, &proc_numb );
MatrixDense<double,int> Kii;
MatrixDense<double,int> Kip;
MatrixDense<double,int> Kpi;
MatrixDense<double,int> Kpp;
Schur::SplitMatrixToBlock(Kii, Kip, Kpi, Kpp, K_local,
list_node_i, numb_node_i,
list_node_p, numb_node_p);
// Check transpose
//CheckTranspose(Kip, Kpi);
// Reconstruct K
//std::string gmatrix_filename = "../output/cube-125_2/cube-125_g_2.csv";
//ReconstructK(K_local, l2g, numb_global_node, gmatrix_filename, mpi_comm);
// LU factorization of Kii
MatrixDense<double, int> Kii_lu;
Factor::LU(Kii_lu, Kii);
MatrixDense<double, int> Uii_inv,Lii_inv;
Lii_inv.Allocate(numb_node_i, numb_node_i);
Uii_inv.Allocate(numb_node_i, numb_node_i);
Vector<double, int> x,rhs;
rhs.Allocate(numb_node_i);
for(int i = 0;i < numb_node_i;++i)
rhs(i) = 0;
// invert Lii and Uii
for(int i = 0;i < numb_node_i;++i){
if(i > 0) rhs(i - 1) = 0;
rhs(i) = 1;
DirectSolver::Forward(x, Kii_lu, rhs);
for(int j = 0;j < numb_node_i;++j)
Lii_inv(j,i) = x(j);
DirectSolver::Backward(x,Kii_lu, rhs);
for(int j = 0;j < numb_node_i;++j)
Uii_inv(j,i) = x(j);
}
// calculate S_local
MatrixDense<double, int> Lpi,Uip,prod;
Kpi.MatrixMatrixProduct(Lpi, Uii_inv);
Lii_inv.MatrixMatrixProduct(Uip, Kip);
Lpi.MatrixMatrixProduct(prod, Uip);
MatrixDense<double, int> S_local;
Kpp.MatrixMatrixSubstraction(S_local, prod);
// merge all numb_node_p in one list
int length_list_p[numb_procs];
MPI_Allgather(&numb_node_p, 1, MPI_INT, length_list_p, 1, MPI_INT, mpi_comm);
/*std::stringstream out_length_list;
for(int i = 0;i < numb_procs;++i){
out_length_list << length_list_p[i] << " ";
}
out_length_list << "\n";*/
int all_list_global_p[numb_procs][numb_global_node];
for(int i = 0;i < numb_procs;++i){
if(proc_numb == i){
for(int j = 0;j < numb_node_p;++j){
all_list_global_p[i][j] = l2g[ list_node_p[j] ];
}
}
MPI_Bcast(all_list_global_p[i], length_list_p[i], MPI_INT, i, mpi_comm);
}
/*for(int i = 0;i < numb_procs;++i){
for(int j = 0;j < length_list_p[i];++j){
out_length_list << all_list_global_p[i][j] << " ";
}
out_length_list << "\n";
}
iomrg::printf("%s\n", out_length_list.str().c_str());*/
// create array pos_S : from global position to position in S
int pos_S[numb_global_node];
for(int i = 0;i < numb_global_node;++i){
pos_S[i] = -1;
}
std::vector<int> S_indices;
for(int i = 0;i < numb_procs;++i){
for(int j = 0;j < length_list_p[i];++j){
S_indices.push_back(all_list_global_p[i][j]);
}
}
std::sort(S_indices.begin(), S_indices.end());
S_indices.erase(std::unique(S_indices.begin(), S_indices.end()), S_indices.end());
int size_S = S_indices.size();
for(int i = 0;i < size_S;++i){
pos_S[ S_indices[i] ] = i;
}
// Assemble S
MatrixDense<double, int> S;
S.Allocate(size_S, size_S);
S.Initialize(0);
double aux_coef[size_S];
iomrg::printf("size_S = %d\n",size_S);
for(int i = 0;i < numb_procs;++i){
int size_local_S = length_list_p[i];
for(int j = 0;j < size_local_S;++j){
if(proc_numb == i){
for(int k = 0;k < size_local_S;++k){
aux_coef[k] = S_local(j,k);
}
}
MPI_Bcast(aux_coef, size_local_S, MPI_DOUBLE, i, mpi_comm);
int r = pos_S[ all_list_global_p[i][j] ];
for(int k = 0;k < size_local_S;++k){
S(r, pos_S[ all_list_global_p[i][k] ]) += aux_coef[k];
}
}
}
// Separate b_local
Vector<double, int> b_i,b_p;
b_i.Allocate(numb_node_i);
b_p.Allocate(numb_node_p);
for(int i = 0;i < numb_l2g;++i){
if(l2p[i] == -1){
b_i( l2i[i] ) = b_local(i);
}else{
b_p( l2p[i] ) = b_local(i);
}
}
// Calculate y_p_local
Vector<double, int> z;
DirectSolver::Forward(z, Kii_lu, b_i);
Vector<double, int> aux_prod;
Lpi.MatrixVectorProduct(aux_prod, z);
Vector<double, int> y_p_local;
y_p_local.Allocate(numb_node_p);
for(int i = 0;i < numb_node_p;++i){
y_p_local(i) = b_p(i) - aux_prod(i);
}
// Calculate y_p
Vector<double, int> y_p(size_S);
for(int i = 0;i < size_S;++i){
aux_coef[i] = 0;
}
for(int i = 0;i < numb_node_p;++i){
aux_coef[ pos_S[ l2g[ list_node_p[i] ] ] ] += y_p_local(i);
}
MPI_Allreduce(aux_coef, y_p.GetCoef(), size_S, MPI_DOUBLE, MPI_SUM,mpi_comm);
// Calculate x_p
Vector<double, int> x_p;
DirectSolver::SolveLU(x_p, S, y_p);
// Calculate y_i
Vector<double, int> aux_prod_2;
aux_prod_2.Allocate(numb_node_i);
aux_prod_2.Assign(0, numb_node_i - 1, 0);
for(int i = 0;i < numb_node_i;++i){
for(int j = 0;j < numb_node_p;++j){
int id_S = pos_S[ l2g[ list_node_p[j] ] ];
aux_prod_2(i) += Uip(i,j) * x_p(id_S);
}
}
Vector<double, int> y_i;
y_i.Allocate(numb_node_i);
for(int i = 0;i < numb_node_i;++i){
y_i(i) = z(i) - aux_prod_2(i);
}
// Calculate x_i
Vector<double, int> x_i;
DirectSolver::Backward(x_i, Kii_lu, y_i);
// Assemble x_local
for(int i = 0;i < numb_node_i;++i){
x_local(list_node_i[i]) = x_i(i);
}
for(int i = 0;i < numb_node_p;++i){
x_local(list_node_p[i]) = x_p(i);
}
}
