System-level Mapping of Dataflow Applications onto MPSoCs

This book provides two novel methodologies for the system-level mapping of dataflow applications onto multi-processor systems-on-chip (MPSoCs) in order to meet the complex demands of application mapping in accordance with given design constraints. Nowadays, different variants of these MPSoCs exist with varying architectures regarding their field of application. Thus, new design methodologies are imperative to keep track with technological innovations of modern and future multi-/many-core platforms. Extensive experiments to investigate the benefits of the two application mapping methodologies are presented and supported by examples and figures.

The first part of this book presents a design-time (static) methodology to determine throughput-optimal mappings of dataflow applications on multi-core architectures. While for these systems various techniques like simulation and analytical methods exist for performance evaluation, this work addresses the challenge of how to capture the effect of design decisions on the real target platform focusing on the effects of scheduling overhead and compiler optimization. The second part presents hybrid application mapping on many-core architectures. This kind of mapping approach is termed hybrid since it unifies a design-time exploration to derive sets of mapping constraints that specify optimal mappings and a run-time determination of a concrete mapping depending on the pre-analyzed mapping constraints.