Task dependency and task scheduling

In ALF programming model, task dependency is used to make sure multiple tasks can be run in a specific order when the order is critical. Some common dependency scenarios are listed here:

• Data dependency: where the output of one task is the input of another task
• Resource conflict: where the tasks share some common resources such as temporary buffers
• Timing: where the tasks have to follow a predetermined order

After you have created a task, you can use the function alf_task_depends_on to specify the task's dependency with an existing task. The ALF runtime considers a task's dependency and the number of requested accelerators for scheduling.

The ALF framework does not detect circular dependency. For a task that depends on other tasks, you must define the dependencies before any work block is added to the task. If a circular dependency is present in the task definitions, the application hangs.

A task that depends on other tasks cannot be processed until all the dependent tasks finish. Tasks are created in immediate mode. After a task has been created and its dependencies are satisfied, the task can be scheduled to run.

For an example of how to use task dependency, see Task dependency example.