Optimization of instruction fetch mechanisms for high issue rates

Abstract

Solutions to the problems of instruction fetch mechanisms are presented. The most general scheme, the collapsing buffer, achieves near perfect performance and consistently aligns instructions in excess of 90% of the time, over a wide range of issue rates. The performance boost provided by compiler optimization techniques is also investigated and results show that compiler optimization can significantly enhance performance across all schemes.