Parallelizing internal mutate of elements in vector at specific indices

[jkbch]

Hello, I am trying to parallelize this code:

for i in indices {
    self.rankers[i].update(post_id, score);
}

self.rankers: Vec<Ranker> is a vector of the data structure Ranker which calls the function/method
pub fn update(&mut self, id: Id, score: Score) which mutate self in Ranker.

So far i have gotten this to work:

self.rankers
    .par_iter_mut()
    .enumerate()
    .filter(|(i, _)| indices.contains(i))
    .for_each(|(_, ranker)| ranker.update(post_id, score));

But self.rankers is a very large vector so it feels wastefull to have to filter the whole vector when I only need to update a small amout of indices. Is it possible to parallelize the code without filtering the whole vector?

[cuviper]

Maybe par_chunks_mut would be better for you, and then you can serially apply relevant indices to each parallel chunk. That could be imbalanced though if the indices might be clustered in any chunk.

If the indices are sorted, another possibility is to write a custom split performing indices.split_at(midpoint) paired with rankers.split_at_mut at the least of the right indices split. This would have the advantage of parallelizing well for any distribution of indices.