Parallel Construction of Suffix Arrays in Rust

See https://docs.rs/libsufr/latest/libsufr/ for API documentation.

This is a Rust library crate for creating suffix arrays.

The basic ideas are as follow:

• Read the input file as u8 (unsigned 8-bit integer values).
• Select the suffixes, which are normally 0 to the length of the text but there is the option to skip suffixes that don't start with A/C/G/T if the input is DNA. Note: Following the C++ implementation, we use 32-bit integers if the input text length is less than 2^32 and 64-bit integers, otherwise.
• Create partitions by randomly choosing suffixes and copying the suffixes to the highest possible partition bounded by any pivot.
• Sort the partitions in parallel.
• Merge the partitions. Because the values fall into nonoverlapping ranges, these subarrays can be appended in order to produce the final SA.
• Produce a binary-encoded output file with the suffix/LCP arrays and other metadata.

Some advantages to this algorithm:

• The various partitioned subarrays can be processed independently by separate threads, and no thread will ever have to merge the entire input.
• Suffix comparisons are made faster by caching LCPs.
• Using u8 for the input text and 32-bits (when possible) for SA/LCP results in lower memory usage.

See the repository for documentation: https://github.com/TravisWheelerLab/sufr

See Also

Use cargo install sufr for a CLI.

Authors

• Ken Youens-Clark kyclark@arizona.edu
• Jack Roddy jroddy@pharmacy.arizona.edu
• Travis Wheeler twheeler@arizona.edu