Branchless Code Generation for Modern Processor Architectures

Abstract

Compilers apply transformations to the code they compile in order to make it run faster without changing its behavior. This process is called code optimization. Modern compilers apply many different passes of code optimization to ensure maximum runtime performance and efficiency, at the rather negligible expense of larger compilation times. This study focuses on a particular optimization, called branchless optimization, which eliminates code branches by utilizing different data transformation techniques that have the same effect. Such techniques are explored on their implementation on the LLVM IR and MIPS and partly ARM assembly, and ranked based on their runtime efficiency. Moreover, the stages of implementing the optimization transformation are explored, as well as different instruction set features that some CPU architectures provide that can be used to increase the efficiency of the optimization. © 2021 ACM.