cormackikkert / nautygpu

README file for nauty-gpu

This code contains nauty 2.7R4, parallelised with GPU acceleration.

HOW TO USE

make 

Then you can run dreadnaut as before. See nauty's user manual, nug27.pdf.

EXAMPLE

Here is an example of a use case (and how dreadnaut was interfaced for benchmarking). For additional graphs to test see https://pallini.di.uniroma1.it/Graphs.html

./dreadnaut

< example_graph.dre
i
q

(takes about ~1 second)

VERSIONING

CUDA version: 11.8 GPU: NVIDIA GA106 (RTX 3060)

ACKNOWLEDGEMENTS

All code is based on the original nauty code, written by:

Brendan McKay: Australian National University; [email protected] Adolfo Piperno: University of Rome "Sapienza"; [email protected] Gunnar Brinkmann: University of Ghent; [email protected] Magma Administration: University of Sydney; [email protected] Patric Ostergard: Aalto Univerity; [email protected]

See COPYRIGHT, for each file's author. I only edited .cu files (and their respective header files). In these .cu files I only added the GPU specific code.

The most interesting edits are in naugraph.cu, this is where the refine function resides.

CORRECTNESS

I have tested the refinement procedure heavily, as that is the only code I edited. I've run lots of graphs, and algorithmically confirmed the outputted partition is indeed equitable, and has the same number of colours as the original c version of dreadnaut. However dreadnaut is a huge program, with many different code flows. I have neglected the longcode aspect of the refinement routine, as it pertains to the global search routine. If there are any errors, it is likely because of this.