Aligning single path loops to reduce the number of capacity cache misses

Abstract

In this paper we address the problem of improving the instruction cache performance for single-path code. The properties of single-path code allow us to align single-path loops within the cache in order to reduce the number of cache misses during the loop execution. We propose an algorithm that categorizes loops in a simple way so that the loops can be aligned and NOP instructions can be inserted to support this loop alignment. Our experimental results show the predictability for cache misses in single-path loops and demonstrate the benefit of the single-path loop alignment.