Design Exploration For High-performance Pipelines

Abstract

Exploration plays an important role in the design of high-performance pipelines. We propose an exploration strategy for varying three design parameters by using a performance-constrained component selection and pipelining algorithm on different “architectures”. The architecture is specified manually by using a mix of behavioral and structural constructs, while the component selection and pipelining is performed automatically using our algorithms. Results on two industrial-strength DSP systems, indicate the effectiveness of our strategy in exploring a large design space within a matter of seconds.