void rcb_rec2D(vector<T> *p_coord,
vector<int> *p_map,
vector<int> *p_part,
int start,
int end,
int dims,
int cur_depth,
int ttl_depth){
// end of partitioning
if(cur_depth == 0) return;
//cout << "-------------------------------------------------------------" << endl;
//cout << " start=" << start << " end=" << end << " cur_depth=" << cur_depth << endl;
tmr_span.start();
// calculate max distance on each axis
double x_span = calc_span(p_coord, start, end, dims, 0);
double y_span = calc_span(p_coord, start, end, dims, 1);
tmr_span.stop_and_add_to_total();;
// choose axis
int axis = -1;
if(x_span >= y_span)
axis = 0;
else
axis = 1;
//cout << " x_span=" << x_span << " y_span=" << y_span << " axis=" << axis << endl;
// find mid-point
tmr_pivot1.start();
T pivot = find_pivot(p_coord, start, end, dims, axis);
tmr_pivot1.stop_and_add_to_total();;
//cout << " pivot=" << pivot <<endl;
// partition into two
int level= cur_depth-1;
int part_index = partition(p_coord, p_map, p_part, start, end, dims, axis, pivot, level);
//cout << " part_index=" << part_index << endl;
// next partitioning
rcb_rec2D(p_coord, p_map, p_part, start, start+part_index, dims, cur_depth-1, ttl_depth);
rcb_rec2D(p_coord, p_map, p_part, start+part_index, end, dims, cur_depth-1, ttl_depth);
}
void part_rcb(int numoflevel, int dims,
int nnode, int nedge, int nbedge, int ncell,
point *partnode, point *partedge, point *partbedge, point *partcell,
int *cell, int *ecell, int *becell, int *edge,
float *x){
// calc coordinate of center of gravity in each cell
vector<float> coord_cell(ncell*dims);
calc_cellcentre(ncell, dims, cell, x, &coord_cell);
// initialize map and partition data
vector<int> map, part;
for(int i=0; i<ncell; i++){
map.push_back(i);
part.push_back(0);
}
// call recursive coordinate bisection algorithm
rcb_rec2D(&coord_cell, &map, &part, 0, ncell, dims, numoflevel, numoflevel);
// Debug Print
//part_rcb_DebugPrint(ncell, dims, &coord_cell, &map, &part);
// output partition data
tmr_out.start();
/*
if(check_partdata(ncell, &map, &part, numoflevel)){
generate_partdata(nnode, nedge, nbedge, ncell,
&map, &part,
partnode, partedge, partbedge, partcell,
cell, ecell, becell);
}else{
cout << "partition data is invalid" << endl;
exit(-1);
}
*/
tmr_out.stop_and_add_to_total();
printf("span =%lf\n", tmr_span.total_time());
printf("pivot =%lf\n", tmr_pivot1.total_time());
printf("part1 =%lf\n", tmr_part1.total_time());
printf("part2 =%lf\n", tmr_part2.total_time());
printf("part3 =%lf\n", tmr_part3.total_time());
printf("out =%lf\n", tmr_out.total_time());
}
int partition(vector<T> *p_in,
vector<int> *p_map,
vector<int> *p_part,
int start,
int end,
int dims,
int axis,
T pivot,
int level){
/*
tmr_part1.start();
// Buffer
vector<T> out1, out2;
vector<int> map1, map2;
int cnt1=0, cnt2=0;
for(int i=start; i < end; i++){
if(p_in->at(i*dims+axis)> pivot){
for(int d_index=0; d_index<dims; d_index++)
out1.push_back(p_in->at(i*dims+d_index));
map1.push_back(p_map->at(i));
cnt1+=1;
}else{
for(int d_index=0; d_index<dims; d_index++)
out2.push_back(p_in->at(i*dims+d_index));
map2.push_back(p_map->at(i));
cnt2+=1;
}
}
tmr_part1.stop_and_add_to_total();
*/
tmr_part1.start();
// Buffer
vector<T> out1, out2;
vector<int> map1, map2;
int cnt1=0, cnt2=0;
int halfcnt = (end-start)/2;
map1.reserve(halfcnt);
map2.reserve(halfcnt);
out1.reserve(halfcnt*dims);
out2.reserve(halfcnt*dims);
T tmpcoord;
for(int i=start; i < end; i++){
tmpcoord = p_in->at(i*dims+axis);
// - balance load when multiple node has same coordinate value
if(tmpcoord == pivot){
// if cnt1 is less than half size, data belong to map1
if(halfcnt>cnt1){
for(int d_index=0; d_index<dims; d_index++)
out1.push_back(p_in->at(i*dims+d_index));
map1.push_back(p_map->at(i));
cnt1+=1;
// if cnt1 is more than half size, data belong to map2
}else{
for(int d_index=0; d_index<dims; d_index++)
out2.push_back(p_in->at(i*dims+d_index));
map2.push_back(p_map->at(i));
cnt2+=1;
}
}else if(tmpcoord > pivot){
for(int d_index=0; d_index<dims; d_index++)
out1.push_back(p_in->at(i*dims+d_index));
map1.push_back(p_map->at(i));
cnt1+=1;
}else{
for(int d_index=0; d_index<dims; d_index++)
out2.push_back(p_in->at(i*dims+d_index));
map2.push_back(p_map->at(i));
cnt2+=1;
}
}
tmr_part1.stop_and_add_to_total();;
//cout << " out1.size =" << out1.size()/dims << " out2.size =" << out2.size()/dims << endl;
//cout << " level =" << level << endl;
// Replace to original coord data <--- can be replaced with this loop to memcpy but may not safe..
tmr_part2.start();
for(int i=0; i < (int)out1.size(); i++){
p_in->at(start*dims+i) = out1.at(i);
}
for(int i=0; i < (int)out2.size(); i++){
p_in->at(start*dims+out1.size()+i) = out2.at(i);
}
tmr_part2.stop_and_add_to_total();;
tmr_part3.start();
for(int i=0; i<cnt2; i++){
p_part->at(start+cnt1+i) |= (1 << level);
}
// Replace to original map/part data
for(int i=0; i < (int)map1.size(); i++)
p_map->at(start+i) = map1.at(i);
for(int i=0; i < (int)map2.size(); i++)
p_map->at(start+map1.size()+i) = map2.at(i);
tmr_part3.stop_and_add_to_total();;
return cnt1;
}
