Key points are not available for this paper at this time.
The increasing usage of application-specific instruction set processors (ASIPs) in audio and video telecommunications has made strong demands on the rapid availability of dedicated compilers. A rule-driven approach to code generation may have benefits over model-based approaches as the user is not confined to the capabilities supported by the model. However, the sole use of transformation rules may or may not be sufficient in optimization abilities depending on the target architecture. This paper outlines experiences with a rule-driven code generation approach for two applications in audio and video processing. The first is a controller for the VideoPhone codec at SGS-Thomson Microelectronics. The second is a VLIW (very large instruction word) processor for high-fidelity and MPEG audio at Thomson Consumer Electronic Components. The experience has shown that a rule-driven approach to compilation is applicable to both the controller and VLIW architectures; however, is limited in optimization abilities for the latter.
Liem et al. (Tue,) studied this question.