Accelerating Function Merging

The importance of file compression in hardware limited devices has long been known; compressing files can free up valuable storage and memory space on SoC devices, smartphones, personal computers and web servers. In 2019, a new technique paved the way for binary file compression through an additional step in compilation by merging similar functions, borrowing a technique from BioInformatics known as sequence alignment. In its infancy and later upgrades, this technique provides far greater code reduction than its counterparts but falls short due to its increased compile-time requirements. In this paper, we present a vastly superior search strategy that achieves improvements in code reduction, memory usage, and compilation times when compared to the current state-of-the-art. Applied to the SPEC 2006 suite, we show that our approach delivers average compilation times below that of a baseline which performs no function merging. Consequently, we show that function merging can serve as a notable speedup in compilation by reducing the effort of the code generator.