Re:Binary Vector Search at 350GB/S Using ARM Neon
by MarekDlugos on Friday July 18, 2025 @03:11PM
(#44608641)
re: optimization for 1024b vectors — do you pad shorter ones, or fallback to a more general kernel?
[ Parent
| on HN ]
Re:Binary Vector Search at 350GB/S Using ARM Neon
by marekgalovic on Friday July 18, 2025 @03:24PM
(#44608792)
We do a projection of the original vectors so that it matches one of our optimized kernel. This generally gives us better recall vs. simple padding since all bits are utilized.