Exploring the algorithmic design space using high level synthesis

HYPER is a high level synthesis system, targeted at numerically intensive applications. By shifting the emphasis from the traditional high level synthesis tasks (such as scheduling and assignment) to the domain of transformations, new venues for high level synthesis are opened. One of the most exciting among them, with potentially the largest impact on the quality of the design, is design and selection of the algorithms for a given application. After a brief overview of the HYPER system, the authors concentrate on the exploration of the algorithmic design space. They show how HYPER can improve the performance or cost of real life applications with orders of magnitude by guiding and conducting a proper algorithmic design selection process