Application analysis with integrated identification of complex instructions for configurable processors

Abstract

An extensible and configurable processor is a programmable platform offering the possibility to customize the instruction set and/or underlying microarchitecture. Efficient application analysis can identify the application parameters and instruction extensions that would influence processor performance. An application characterization flow is presented and demonstrated on the Wavelet/Scalar Quantization image compression application. In this context, novel application metrics are identified as the percentage cover, maximum cycle gain for each basic block and candidate-induced application speedup due to possible complex instructions. Furthermore, evaluating the instruction candidates during application analysis is proposed in order to establish a link with subsequent design space exploration steps. © Springer-Verlag 2004.