Gain some exposure to different task scheduling styles.
Code To Investigate
Static Scheduler-like ray tracer.
Recursively divide the given region into 4 tasks for each of the four quadrants until you get below the threshold (see the IntPredicate areaThreshold instance variable).
When areaThreshold predicate test method returns false, call the
rayTraceBaseCase(context, xMin, yMix, xMax, yMax) method to render the section.
renderMyTasksUntilEmptyThenStealWorkFromOthers method will be called on one of the four quadrants, but it will be given the sections that make up all four quadrants. You can use the
id to find the particular quadrant being called.
This method should repeatedly draw from the tail of its own double-ended queue to render its own tasks using the
Section.render method. If it finishes all of its tasks, and there are still unfinished tasks in the other quadrants, then it should draw tasks from the head of the other quadrants' deques.
As you can see in the
WorkStealingRayTracer.rayTrace method, the parallelism comes from the fact that the
renderMyTasksUntilEmptyThenStealWorkFromOthers method is called multiple times in parallel. You don't have to worry about spawning new tasks. Also, because ConcurrentLinkedDeque is thread-safe, all operations are atomic, so you you shouldn't need to use isolated. However, you should avoid reading and then writing.
private void renderMyTasksUntilEmptyThenStealWorkFromOthers(RayTraceTaskContext taskContext, Deque<Section> myTaskDeque, List<Deque<Section>> otherTaskDeques, int id) (sequential implementation only)