Branch Optimisation Techniques for Hardware Compilation

Abstract

This paper explores using information about program branch probabilities to optimise reconfigurable designs. The basic premise is to promote utilization by dedicating more resources to branches which execute more frequently. A hardware compilation system has been developed for producing designs which are optimised for different branch probabilities. We propose an analytical queueing network performance model to determine the best design from observed branch probability information. The branch optimisation space is characterized in an experimental study for Xilinx Virtex FPGAs of two complex applications: video feature extraction and progressive refinement radiosity. For designs of equal performance, branch-optimised designs require 24% and 27.5% less area. For designs of equal area, branch optimised designs run upto 3 times faster. Our analytical performance model is shown to be highly accurate with relative error between 0.12 and 1.1 x 10− 4.