On the choice of models of computation for writing executablespecifications of system level designs

摘要

System level designs are typically heterogeneous, thus combiningdifferent technologies. In order to create executable specifications atsuch a level, a hardware description language, a programming language,or a combination of both is used. However, the resulting description maynot be efficient because the ability of each language to capture thesystem's features is limited by its semantics. This paper presents amethodology for analyzing the efficiency of a set of models ofcomputation to build executable specifications. We defined a set of“behavioral primitives” and evaluated how efficiently theyare captured by different models of computation. A debugger/profilertool was developed. The analysis of the data produced by our tool allowsus to derive criteria to choose the most adequate model of computationfor each primitive. The analysis of the basic block behavioral primitiveis detailed in order to illustrate the method