void parallel_wcc(graph_t &g, unsigned threadnum, vector<vector<uint64_t> > &global_input_tasks, gBenchPerf_multi &perf,
int perf_group)
{
vector<vector<uint64_t> > global_output_tasks(threadnum*threadnum);
bool stop = false;
#pragma omp parallel num_threads(threadnum) shared(stop,global_input_tasks,global_output_tasks,perf)
{
unsigned tid = omp_get_thread_num();
vector<uint64_t> & input_tasks = global_input_tasks[tid];
perf.open(tid, perf_group);
perf.start(tid, perf_group);
while(!stop)
{
#pragma omp barrier
// process local queue
stop = true;
for (unsigned i=0;i<input_tasks.size();i++)
{
uint64_t vid=input_tasks[i];
uint64_t size, begin;
size = g.csr_in_edges_size(vid);
begin = g.csr_in_edges_begin(vid);
for (uint64_t i=0;i<size;i++)
{
uint64_t dest_vid = g.csr_in_edge(begin,i);
if(g.csr_vertex_property(dest_vid).root > g.csr_vertex_property(vid).root) {
__sync_bool_compare_and_swap(&(g.csr_vertex_property(dest_vid).root), g.csr_vertex_property(dest_vid).root, g.csr_vertex_property(vid).root);
global_output_tasks[vertex_distributor(dest_vid,threadnum)+tid*threadnum].push_back(dest_vid);
}
}
size = g.csr_out_edges_size(vid);
begin = g.csr_out_edges_begin(vid);
for (uint64_t i=0;i<size;i++)
{
uint64_t dest_vid = g.csr_out_edge(begin,i);
bool done = false;
while(!done) {
if(g.csr_vertex_property(dest_vid).root > g.csr_vertex_property(vid).root) {
done = __sync_bool_compare_and_swap(&(g.csr_vertex_property(dest_vid).root), g.csr_vertex_property(dest_vid).root, g.csr_vertex_property(vid).root);
if(done) {
global_output_tasks[vertex_distributor(dest_vid,threadnum)+tid*threadnum].push_back(dest_vid);
}
} else {
done = true;
}
}
}
}
#pragma omp barrier
input_tasks.clear();
for (unsigned i=0;i<threadnum;i++)
{
if (global_output_tasks[i*threadnum+tid].size()!=0)
{
stop = false;
input_tasks.insert(input_tasks.end(),
global_output_tasks[i*threadnum+tid].begin(),
global_output_tasks[i*threadnum+tid].end());
global_output_tasks[i*threadnum+tid].clear();
}
}
#pragma omp barrier
}
perf.stop(tid, perf_group);
}
}
