Parallel O(sqrt(n)) Overhead LSD Radix Sort
===========================================

This repository holds supplementary material for the paper
"Parallel O(sqrt(n)) Overhead LSD Radix Sort" by R. Clausecker
and F. Schintke.  It is made available under a 2-clause BSD
license.

You can find the paper at https://arxiv.org/abs/2607.05302

The benchmarks sort 32 bit integers in four passes.

To compile:

    cc -O3 -fopenmp -DNDEBUG -o benchmark *.c

add -mavx2 when compiling on recent x86, -faltivec when
compiling on powerpc*.

The benchmark's output follows the conventions of the Go
benchmark framework and can be analysed with the benchstat
tool.

The subdirectory generic contains a largely untested
type-generic implementation.  While complete, it is likely
not functional as is.  Unfortunately I ran out of time due
to other commitments and was unable to finish it.

The results directly holds the raw data on which the results
in the paper were based.  The zen4 and volterra result sets
did not make the cut due to lack of space and correctness
issues (no frequency-scaling driver was loaded on the volterra
board, leaving it to run on the lowest supported CPU clock).

Code  (c) 2025, 2026 Robert Clausecker <clausecker@zib.de>
Paper (c) 2025, 2026 Robert Clausecker <clausecker@zib.de>,
                     Florian Schintke <schintke@zib.de>