Fast Sorting Algorithms using AVX-512 on Intel Knights Landing
Max Planck Computing and Data Facility (MPCDF)
arXiv:1704.08579 [cs.MS], (24 Apr 2017)
@article{bramas2017fast,
title={Fast Sorting Algorithms using AVX-512 on Intel Knights Landing},
author={Bramas, Berenger},
year={2017},
month={apr},
archivePrefix={"arXiv"},
primaryClass={cs.MS}
}
This paper describes fast sorting techniques using the recent AVX-512 instruction set. Our implementations benefit from the latest possibilities offered by AVX-512 to vectorize a two-parts hybrid algorithm: we sort the small arrays using a branch- free Bitonic variant, and we provide a vectorized partitioning kernel which is the main component of the well-known Quicksort. Our algorithm sorts in-place and is straightforward to implement thanks to the new instructions. Meanwhile, we also show how an algorithm can be adapted and implemented with AVX-512. We report a performance study on the Intel KNL where our approach is faster than the GNU C++ sort algorithm for any size in both integer and double floating-point arithmetics by a factor of 4 in average.
May 9, 2017 by hgpu
