In the past, research on Multiprocessor Systems-on-Chip (MPSoC) has focused mainly on increasing the available processing power on a chip, while less effort was put into specific system-level performance analysis, or into behavior prediction. This paper introduces PAM-SoC, a light-weight performance predictor for MPSoC system-level performance. Being based on Pamela, a static performance predictor for parallel applications, PAM-SoC can compute its prediction in seconds for cases when cycle-accurate simulation takes tens of minutes. The paper includes a set of PAM-SoC validation experiments, as well as two sets of experiments to show how PAM-SoC can be used for either application tuning or MPSoC platform tuning in early system design phases.
Ana Lucia Varbanescu, Henk J. Sips, Arjan J. C. va