Daniele Paolo Scarpazza

Systematic Methodology for Real-Time Cost-Effective Mapping of Dynamic Concurrent Task-Based Systems on Heterogenous Platforms

Systematic Methodology for Real-Time Cost-Effective Mapping of Dynamic Concurrent Task-Based Systems on Heterogeneous Platformsgives an overview of the state-of-the-art in system-level design trade-off explorations for concurrent tasks running on embedded heterogeneous multiple processors. The targeted application domain covers complex embedded real-time multi-media and communication applications. Many of these applications are concurrent in the sense that multiple subsystems can be running simultaneously. Also, these applications are so dynamic at run-time that the designs based on the worst case execution times are inefficient in terms of resource allocation (e.g., energy budgets). A novel systematical approach is clearly necessary in the area of system-level design for the embedded systems where those concurrent and dynamic applications are mapped. This material is mainly based on research at IMEC and its international university network partners in this area in the period 1997-2006. In order to deal with the concurrent and dynamic behaviors in an energy-performance optimal way, we have adopted a hierarchical system model (i.e., the gray-box model) that can both exhibit the sufficient detail of the applications for design-time analysis and hide unnecessary detail for a low-overhead run-time management. We have also developed a well-balanced design-time/run-time combined task scheduling methodology to explore the trade-off space at design-time and efficiently handle the system adaptations at run-time. Moreover, we have identified the connection between task-level memory/communication management and task scheduling and illustrated how to perform the task-level memory/communication management in order to obtain the design constraints that enable the this connection. A fast approach is also shown to estimate at the system-level, the energy and performance characterization of applications executing on the target platform processors.

Software simultaneous multi-threading, a technique to exploit task-level parallelism to improve instruction- and data-level parallelism

The search for energy efficiency in the design of embedded systems is leading toward CPUs with higher instruction-level and data-level parallelism. Unfortunately, individual applications do not have sufficient parallelism to keep all these CPU resources busy. Since embedded systems often consist of multiple tasks, task-level parallelism can be used for the purpose. Simultaneous multi-threading (SMT) proved a valuable technique to do so in high-performance systems, but it cannot be afforded in system with tight energy budgets. Moreover, it does not exploit data-level parallel hardware, and does not exploit the available information on threads. We propose software-SMT (SW-SMT), a technique to exploit task-level parallelism to improve the utilization of both instruction-level and data-level parallel hardware, thereby improving performance. The technique performs simultaneous compilation of multiple threads at design-time, and it includes a run-time selection of the most efficient mixes. We have applied the technique to two major blocks of a SDR (software-defined radio) application, achieving energy gains up to 46% on different ILP and DLP architectures. We show that the potentials of SW-SMT increase with SIMD datapath size and VLIW issue width.