Exploring Task Parallelism for Heterogeneous Systems Using Multicore Task Management API.

Current trends in multicore platform design indicate that heterogeneous systems are here to stay. Such systems include processors with specialized accelerators supporting different instruction sets and different types of memory spaces among several other features. These features increase the programming effort to port applications to target platforms. We need effective programming strategies that can exploit the rich feature set of such heterogeneous multicore architectures and yet not require increased learning curve to apply these strategies. To distribute workload effectively across such systems that have different cores running at different speed, we have explored task-based programming models in this paper. This model allows decomposition of a problem into a set of tasks for simultaneous execution. We present a task-based approach that employs the Multicore Association's (MCA) Task Management API (MTAPI), a robust, cross-platform, scalable API that avoids unnecessary synchronization thus offering a tiered and flexible approach and distributing workload efficiently across processors of varying types. For evaluation purposes, we use an NVIDIA Jetson TK1 board (ARM + GPU) as our test bed. As applications, we employ codes from benchmark suites such as Rodinia and BOTS.