Tackling the Software Stack for Heterogeneous Multi-Processors: a balancing act between work-load and memory organisation

Prof. Dr. Sven-Bodo Scholz, Heriot-Watt University, Edinburgh

2 Dec 2013, 14:00–15:30; Location: S4|10-1

In today's economic climate, time to market, or more generally time to solution, is critical. At the same time, mainstream hardware does not only become massively parallel, it also becomes much more diverse: Graphics Processors (GPUs), large vector machines such as Intel's Xeon-Phi, and even programmable hardware in the form of FPGAs find an increasing presence. While this is very exciting from a computer science perspective, for the application programmer it typically just constitutes an unpleasant hurdle.

We try to overcome this hurdle by working on techniques that lead to tools that combine high-productivity and high-performance. We compile architecture agnostic code into target specific high-performance code. Our tools deliver performance close to hand optimised code for various architectures including SMPs, GPGPUs, and prototypical many-core machines without requiring any program changes or annotations.

In this talk, I present some of the key insights from looking at the different compilation technologies that are required when targeting different multi-core systems. I will mainly focus on the role of memory organisation for achieving performance portability. In that context, I present one of our latest developments, where we use type-guided code generation in order to change memory layouts for improved code vectorisations.

Category: CE Seminar


Technische Universität Darmstadt

Graduate School CE
Dolivostraße 15
D-64293 Darmstadt

Phone+49 6151/16-24401
Fax -24404

to assistants' office

Open BSc/MSc Theses

Show a list of open BSc/MSc topics at GSC CE.

 Print |  Impressum |  Sitemap |  Search |  Contact |  Privacy Policy
zum Seitenanfangzum Seitenanfang