QuadRank: Engineering a High Throughput Rank

TL;DR

QuadRank introduces high-throughput, space-efficient rank data structures for DNA and binary alphabets, achieving near-cache-miss-free performance and significant speedups in bioinformatics applications.

Contribution

It extends recent inlining techniques to the DNA alphabet, enabling single cache miss queries and efficient batch processing with prefetching.

Findings

BiRank and QuadRank are 1.5x and 2x faster than similar methods without inlining.

Prefetching yields an additional 2x speedup, limited by RAM bandwidth.

QuadRank with prefetching outperforms state-of-the-art FM-indexes, reducing size and increasing speed.

Abstract

Given a text, a query $\mathsf{rank}(q, c)$ counts the number of occurrences of character $c$ among the first $q$ characters of the text. Space-efficient methods to answer these rank queries form an important building block in many succinct data structures. For example, the FM-index is a widely used data structure that uses rank queries to locate all occurrences of a pattern in a text. In bioinformatics applications, the goal is usually to process a given input as fast as possible. Thus, data structures should have high throughput when used with many threads. Contributions. For the binary alphabet, we develop BiRank with 3.28% space overhead. It merges the central ideas of two recent papers: (1) we interleave (inline) offsets in each cache line of the underlying bit vector [Laws et al., 2024], reducing cache-misses, and (2) these offsets are to the middle of each block so that only half of them need popcounting [Gottlieb and Reinert, 2025]. In QuadRank (14.4% space overhead), we extend these techniques to the $\sigma=4$ (DNA) alphabet. Both data structures require only a single cache miss per query, making them highly suitable for high-throughput and memory-bound settings. To enable efficient batch-processing, we support prefetching the cache lines required to answer upcoming queries. Results. BiRank and QuadRank are around $1.5\times$ and $2\times$ faster than similar-overhead methods that do not use inlining. Prefetching gives an additional $2\times$ speedup, at which point the dual-channel DDR4 RAM bandwidth becomes a hard limit on the total throughput. With prefetching, both methods outperform all other methods apart from SPIDER [Laws et al., 2024] by $2\times$. When using QuadRank with prefetching in a toy count-only FM-index, QuadFm, this results in a smaller size and up to $4\times$ speedup over Genedex, a state-of-the-art batching FM-index implementation.