GitHub Achieves 45 GiB/s Case-Folding Speed in Code Search Using Branch-Free Logic

GitHub engineers have developed a high-performance technique to case-fold source code during search operations, achieving speeds exceeding 45 GiB/s on a single CPU core. The approach relies on a branch-free loop design combined with byte-space arithmetic, eliminating costly conditional checks that typically slow down text processing. Case-folding is essential for case-insensitive code search, requiring every byte of indexed source code to be transformed efficiently. By avoiding early exits in the processing loop, the implementation maintains consistent throughput close to the theoretical memory bandwidth limit. GitHub detailed the technique on its engineering blog, highlighting it as a key optimization powering its code search infrastructure.