This repository contains the in-place MSD radix sort implementation of the publication
@inproceedings{polychroniou2014comprehensive,
title={A comprehensive study of main-memory partitioning and its application to large-scale comparison-and radix-sort},
author={Polychroniou, Orestis and Ross, Kenneth A},
booktitle={Proceedings of the 2014 ACM SIGMOD international conference on Management of data},
pages={755--766},
year={2014}
}from Orestis Polychroniou et al. The source code has been retrieved from Orestis Polychroniou's personal web page www.cs.columbia.edu/~orestis.
This repository adds support for CMake and provides a library for the in-place MSD radix sort algorithm.
Limitations:
- All CPUs should have the same number of cores. Other configurations will not adapt properly.
- LSB methods can run for up to 8 sockets due to implementing up to 8-way range partition.
- MSB methods must use exactly 64 threads due to implementing only 64-way range partition.
- MSB methods need a large fudge factor or a small block, otherwise the algorithm will be unable to distribute blocks across NUMA regions correctly.
- MSB methods are implemented to run on full 32-bit data while LSB methods can skip bits.
- Zipfian distributions are implemented for 32 bit data only and not for 64-bit data.