Due to physical constraints the performance of single processors has reached its limits, and all major hardware vendors switched to multi-core architectures. In addition, there is a trend towards heterogeneous parallel systems comprised of conventional multi-core CPUs, GPUs, and other types of accelerators. As a consequence, the development of applications that can exploit the potential of emerging parallel architectures and at the same time are portable between different types of systems is becoming more and more challenging. In this paper we discuss recent research efforts of the European PEPPHER project in software development for future parallel architectures. We present a high-level compositional approach to parallel software development in concert with an intelligent task-based runtime system. Such an approach can significantly enhance programmability of future parallel systems, while ensuring efficiency and facilitating performance portability across a range of different architectures.

Top