Binary Vector Search at 350GB/S Using ARM Neon

Binary Vector Search at 350GB/S Using ARM Neon(topk.io)

7 points by Equiet 304 days ago | 2 comments

MarekDlugos 304 days ago |

re: optimization for 1024b vectors — do you pad shorter ones, or fallback to a more general kernel?

marekgalovic 304 days ago | |

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.

MarekDlugos 304 days ago |

re: optimization for 1024b vectors — do you pad shorter ones, or fallback to a more general kernel?

marekgalovic 304 days ago | |

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.