Fast identification algorithm for application-specific instruction-set extensions

Customizable and extensible processors can efficiently meet the growing demand of application-specific IC device designs in performance and flexibility. Due to the increasing complexity of software applications, it is essential to automatically decide operations to be carried out in custom function units from high-level application code. This paper addresses efficient techniques for identifying application-specific instruction candidates. New pruning criterions are proposed and combined with the latest work cited in this paper to reduce the search space, resulting in a fast algorithm for enumerating all valid candidates corresponding to given micro-architectural constraints. Experimental results show that, the latest algorithm is improved in runtime by up to 50% for the case of single-output constraint and up to 18% for the case of multiple-output constraint.