The greedy policy is better because other thread which finishes its work does not need to worry about the sync point, since the only the last thread reaches the sync point is responsible for it. If in the first approach, the spawn threads instead tries to find some other work to do, it may not respond quickly when the last thread finishes to deal with the sync point. Also, I think the block id A is needed as there might be other cilk_spawn outstanding, so the id A is a unique identifier for this spawn work.
The greedy policy is better because other thread which finishes its work does not need to worry about the sync point, since the only the last thread reaches the sync point is responsible for it. If in the first approach, the spawn threads instead tries to find some other work to do, it may not respond quickly when the last thread finishes to deal with the sync point. Also, I think the block id A is needed as there might be other cilk_spawn outstanding, so the id A is a unique identifier for this spawn work.
This comment was marked helpful 0 times.