Making Your Code Faster by Taming Branches

Abstract

Most software code contains conditional branches. In code, they appear in if-then-else clauses, loops, and switch-case constructs. When encountering a conditional branch, the processor checks a condition and may jump to a new code path, if the branch is taken, or continue with the following instructions. Unfortunately, the processor may only know whether a jump is necessary after executing all instructions before the jump. For better performance, modern processors predict the branch and execute the following instructions speculatively. It is a powerful optimization. There are some limitations to speculative execution, however. For example, the processor must discard the work done after the misprediction and start anew when the branch is mispredicted. Thankfully, processors are good at recognizing patterns and avoiding mispredictions, when possible. Nevertheless, some branches are intrinsically hard to predict and they may cause a performance bottleneck.

Publication
InfoQ (online) 2020
Date