IJobParallelFor

Description

Interface that represents a job that performs the same independent operation for each element of a native container or for a fixed number of iterations.

This job type mainly exists for backwards compatibility. New code should use IJobFor instead, which exposes the same functionality through a more consistent API. This job type has the following options to schedule work:

• IJobForExtensions.RunByRef runs the job on the main thread and blocks until it is complete.
• IJobForExtensions.ScheduleByRef schedules the job to run asynchronously, distributing the work across multiple worker threads. This scheduling option can give the best performance, but race conditions can occur if the job writes to shared or global data without proper synchronization.

Execute(int index) is executed once for each index from 0 to the provided length.

RunByRef guarantees that the the job's Execute(int index) method is invoked sequentially. ScheduleByRef doesn't invoke the job's Execute method sequentially because it's called from multiple worker threads in parallel to each other.

Each iteration must be independent from other iterations and the safety system enforces this rule for you. The indices have no guaranteed order and are executed on multiple cores in parallel.

Unity automatically splits the work into chunks of no less than the provided batchSize, and schedules an appropriate number of jobs based on the number of worker threads, the length of the array and the batch size. You should choose the batch size based on the amount of work performed in the job. If the batch size is too large, work may not be distributed evenly across the available worker threads. If it's too small, the overhead of fetching new work items may dominate the time it takes to process them. A simple job, for example adding a couple of Vector3 to each other should have a batch size of 32 to 128. However, if the work performed has a large overhead then it's best practice to use a small batch size, for example, a batch size of 1. Work stealing is performed using atomic operations.

You can use the returned JobHandle to check that the job has completed, or pass it to other jobs as a dependency. When you pass a JobHandle as a dependency, it ensures that the jobs are executed one after another on the worker threads.

Additional resources: IJobForExtensions

                    struct ApplyVelocityJobParallelFor : IJobParallelFor
{
    // Jobs declare all data that will be accessed in the job
    // By declaring it as read only, multiple jobs are allowed to access the data in parallel
    [ReadOnly]
    public NativeArray<Vector3> velocity;

    // By default containers are assumed to be read & write
    public NativeArray<Vector3> position;

    // Delta time must be copied to the job since jobs generally don't have concept of a frame.
    // The main thread waits for the job same frame or next frame, but the job should do work deterministically
    // independent on when the job happens to run on the worker threads.
    public float deltaTime;

    // The code actually running on the job
    public void Execute(int i)
    {
        // Move the positions based on delta time and velocity
        position[i] = position[i] + velocity[i] * deltaTime;
    }
}

public void ApplyVelocity_ParallelFor()
{
    var position = new NativeArray<Vector3>(500, Allocator.Persistent);

    var velocity = new NativeArray<Vector3>(500, Allocator.Persistent);
    for (var i = 0; i < velocity.Length; i++)
        velocity[i] = new Vector3(0, 10, 0);

    // Initialize the job data
    var job = new ApplyVelocityJobParallelFor()
    {
        deltaTime = Time.deltaTime,
        position = position,
        velocity = velocity
    };

    // If this job required a previous job to complete before it could safely begin execution,
    // we'd use its handle here. For this simple case, there are no job dependencies,
    // so a default JobHandle is sufficient.
    JobHandle dependencyJobHandle = default;

    JobHandle velocityJobHandle = default;
    // Schedule job to run on multiple worker threads, attempting to distribute work items evenly between them.
    // First parameter is how many for-each iterations to perform.
    // The second parameter is the batch size,
    //   essentially the no-overhead innerloop that just invokes Execute(i) in a loop.
    //   When there is a lot of work in each iteration then a value of 1 can be sensible.
    //   When there is very little work values of 32 or 64 can make sense.
    // The third parameter is a JobHandle to use for this job's dependencies.
    //   Dependencies are used to ensure that a job executes on worker threads after
    //   the dependency has completed execution.
    velocityJobHandle = job.ScheduleByRef(position.Length,
        64, dependencyJobHandle);

    // Ensure the job has completed.
    // It is not recommended to Complete a job immediately,
    // since that reduces the chance of having other jobs run in parallel with this one.
    // You optimally want to schedule a job early in a frame and then wait for it later in the frame.
    // Ideally this job's JobHandle would be passed as a dependency to another job that consumes the
    // output of this one. If the output of this job must be read from the main thread, you should call
    // Complete() on this job handle just before reading it.
    velocityJobHandle.Complete();

    // Once the job is complete, the expected processing should have occurred.
    for (var i = 0; i < velocity.Length; i++)
        Assert.AreEqual(Time.deltaTime * velocity[i].y, position[i].y);

    // Native arrays must be disposed manually.
    position.Dispose();
    velocity.Dispose();
}

Public Methods