Multiprocessor Performance Analysis

This work contributes accurate performance analysis techniques for the embedded multiprocessor systems that process streams of video and audio data in real time. Our performance analysis method works in the context of the run-time optimization of visual quality and power consumption under real-time constraints. It guides the optimization by dynamically predicting the data-dependent throughput of the system. An accurate analysis is not trivial when two factors are combined: parallelism and data-dependent execution delays. We achieve good accuracy for an important class of applications and hardware platforms, including networks-on-chip. Our performance analysis is realizable in practice for the run-time quality and power management. We also propose a methodology to minimize the required on-chip communication channel capacity and describe an overall trajectory for implementing applications on embedded multiprocessors. Two case studies are presented in detail: the JPEG image decoding and the MPEG-4 arbitrarily-shaped video decoding. This book should be especially useful to embedded system researchers and engineers working on video and audio processing applications.