HPX - High Performance ParalleX

PrevUpHomeNext

Executor Parameters and Executor Parameter Traits

Executors as described in the previous section add a powerful customization capability to any facility which exposes management of parallel execution. However, sometimes it is necessary to be able to customize certain parameters of the execution as well. In HPX we introduce the notion of execution parameters and execution parameter traits. At this point, the only parameter which can be customized is the size of the chunks of work executed on a single HPX-thread (such as the number of loop iterations combined to run as a single task).

An executor parameter object is responsible for exposing the calculation of the size of the chunks scheduled. It abstracts the (potentiall platform-specific) algorithms of determining those chunks sizes.

The way executor parameters are implemented is aligned with the way executors are implemented. All functionalities of concrete executior parameter types are exposed and accessible through a corresponding hpx::parallel::executor_parameter_traits type.

With executor_parameter_traits, clients access all types of executor parameters uniformly:

std::size_t chunk_size =
    executor_parameter_traits<my_parameter_t>::get_chunk_size(my_parameter,
        my_executor, [](){ return 0; }, num_tasks);

This call synchronously retrieves the size of a single chunk of loop iterations (or similar) to combine for execution on a single HPX-thread if the overall number of tasks to schedule is given by num_tasks. The lambda function exposes a means of test-probing the execution of a single iteration for performance measurment purposes (the execution parameter type might dynamically determine the execution time of one or more tasks in order to calculate the chunk size, see hpx::parallel::auto_chunk_size for an example of such a executor parameter type).

Other functions in the interface exist to discover whether a executor parameter type should be invoked once (i.e. returns a static chunk size, see hpx::parallel::static_chunk_size) or whether it should be invoked for each scheduled chunk of work (i.e. it returns a variable chunk size, for an example, see hpx::parallel::guided_chunk_size).

Though this interface appears to require executor parameter type authors to implement all different basic operations, there is really none required. In practice, all operations have sensible defaults. However, some executor parameter types will naturally specialize all operations for maximum efficiency.

In HPX we have implemented the following executor parameter types:


PrevUpHomeNext