Fast graph-based instruction selection for multi-output instructions

摘要: A multi-output instruction (MOI) is an that produces multiple outputs to its destination locations. Such inherently parallel instructions are becoming more and popular in embedded processors, due the advances application-specific architectures. In order provide high-level programmability thus guarantee widespread acceptance, sophisticated compiler support for these programmable cores necessary. However, traditional tree-based approaches selection, although very fast, fail exploit MOIs mainly because of fundamental limitation tree representation. fact, generate optimal code with requires a general graph-based formulation selection problem, which at least NP-complete. this paper we present new methodology automatically from simple set descriptions, selectors can effectively utilize all provided including MOIs. Our experimental results using benchmarks on target processor various up two demonstrate our generated quickly many application level, therefore highly desirable both architecture exploration as generators after fixed. Copyright © 2010 John Wiley & Sons, Ltd. (This revised expanded version published Fifth IEEE/ACM International Conference Hardware/Software Codesign System Synthesis (CODES+ISSS), September 2007 1.)