Don't stop early: Case-folding source code at memory speed

How a branch-free loop and byte-space arithmetic let GitHub case-fold every byte of code search at >45 GiB/s on a single core.

Raccontata dagithub.blog