How to return an unknown amount of data from a parallel job? (C# Jobs + Burst)

Question

Say we have a parallel job (implementing IJobParallelFor) which requires a large array (>100,000 items) as an input to do some complicated processing. After said job is finished, I'd like to return a portion of those items that have passed certain checks.

Specifically, I'd like to check if >100,000 Vector3's satisfy a certain criteria. If they do, I'd like to add them to a list so that I can continue operating on that data in the main thread (or in other jobs).

The issue is that we can't use a NativeList to call .Add() because that data structure doesn't support parallel writing. It is also not viable to use a NativeArray because it requires its size to be initialized ahead of time, and we don't know how many items we are going to return.

What is a good way to solve this problem?

Answer 1

Use NativeQueue for your results, then obtain a NativeQueue<T>.ParallelWriter for it using results.AsParallelWriter(). It exposes a method Enqueue(T) to add elements to the queue.

You can then later read from that queue by copying it to an array using results.ToArray() or by processing the queue using a while(results.TryDequeue(out result)) { loop.

The NativeList also offers a parallel writer, by the way, but that one does not offer operations which exceed the initial capacity.

Answer 2

You could use a NativeArray<bool> IsValid to indicate which items require some more processing.

Items can then be aggregated in one place (e.g. in another IJob in a loop); or the flag can be used in the followup method in the next job to quit early.

This has the advantage of fixed-size, low synchronization overhead. It requires a tiny bit more RAM. Cache locality is sometimes better, sometimes worse.