On June 10, 2014, Jun.-Prof. Dr. Christian Plessl will give a lecture on "Enabling Heterogeneous and Custom Computing".
Abstract:
The use of heterogeneous computing resources has become more widespread over the whole computing system spectrum from embedded to general-purpose to high-performance computing. The driving force behind this development is the demand for more computational power and more efficient energy usage. This need is satisfied by augmenting the system with accelerators that are massively parallel and optionally also customizable. The rapid adoption of accelerators, in particular GPUs, is testament to the benefits provided by heterogeneous computing. While the hardware architectures for accelerators are progressing at a rapid pace, the efficient use and operation of these accelerators still poses challenges for application developers and system administrators because the programming and execution models of accelerators differ significantly from general-purpose CPUs.
In this talk, I will present our work on development tool flows for simplifying the development process. For optimizing the efficient use of these resources we propose a runtime system that performs task scheduling and load-balancing considering all computing resources to meet a user-configurable scheduling goal. Finally, I will present recent work on enabling the transparent use of heterogeneous resources using a novel runtime system that is tightly coupled with a just-in-time compilation system.