KOKKOS_INLINE_FUNCTION
void operator() ( const team_member & dev) const {
Kokkos::View<int**,Kokkos::MemoryUnmanaged> l_histogram(dev.team_shmem(),TEAM_SIZE,TEAM_SIZE);
Kokkos::View<int*,Kokkos::MemoryUnmanaged> l_data(dev.team_shmem(),chunk_size+1);
const int i = dev.league_rank() * chunk_size;
for(int j = dev.team_rank(); j<chunk_size+1; j+=dev.team_size())
l_data(j) = data(i+j);
for(int k = dev.team_rank(); k < TEAM_SIZE; k+=dev.team_size())
for(int l = 0; l < TEAM_SIZE; l++)
l_histogram(k,l) = 0;
dev.team_barrier();
for(int j = 0; j<chunk_size; j++) {
for(int k = dev.team_rank(); k < TEAM_SIZE; k+=dev.team_size())
for(int l = 0; l < TEAM_SIZE; l++) {
if((l_data(j) == k) && (l_data(j+1)==l))
l_histogram(k,l)++;
}
}
for(int k = dev.team_rank(); k < TEAM_SIZE; k+=dev.team_size())
for(int l = 0; l < TEAM_SIZE; l++) {
Kokkos::atomic_fetch_add(&histogram(k,l),l_histogram(k,l));
}
dev.team_barrier();
}
KOKKOS_INLINE_FUNCTION
void operator()(const team_member & teamMember) const{
int ii = teamMember.league_rank() * teamMember.team_size() + teamMember.team_rank() + color_set_begin;
if(ii >= color_set_end) return;
#ifdef REORDER
int crow = ii;//colors_ind(ii);
#else
int crow = colors_ind(ii);
#endif
int row_begin = A.graph.row_map(crow);
int row_end = A.graph.row_map(crow+1);
bool am_i_the_diagonal = false;
double diagonal = 1;
double sum = 0;
Kokkos::parallel_reduce(
Kokkos::ThreadVectorRange(teamMember, row_end - row_begin),
[&] (int i, double & valueToUpdate) {
int adjind = i + row_begin;
int colIndex = A.graph.entries(adjind);
double val = A.values(adjind);
if(colIndex == crow){
diagonal = val;
am_i_the_diagonal = true;
}
else{
valueToUpdate += val * xv(colIndex);
}
}, sum);
if(am_i_the_diagonal){
xv(crow) = (rv(crow) - sum)/diagonal;
}
}
// This is a reduction operator which now takes as first argument the
// TeamPolicy member_type. Every member of the team contributes to the
// total sum.
// It is helpful to think of this operator as a parallel region for a team
// (i.e. every team member is active and will execute the code).
KOKKOS_INLINE_FUNCTION
void operator() ( const team_member & thread, int& sum) const {
sum+=1;
// The TeamPolicy<>::member_type provides functions to query the multi
// dimensional index of a thread as well as the number of thread-teams and the size
// of each team.
printf("Hello World: %i %i // %i %i\n",thread.league_rank(),thread.team_rank(),thread.league_size(),thread.team_size());
}