CS4395 Independent Study Presentation
Title: High-Performance Motif Counting on GPUs
Presenter: Cameron Bradley
Advisor: Martin Burtscher
Date/Time: Friday, August 5th @ 2:00 p.m.
Location: https://txstate.zoom.us/my/martin.burtscher (no password required)
Abstract:
Motif counting counts the number occurrences of a motif (small subgraph) in a given input graph. It is an important computational step in a multitude of fields. However, it tends to be very time intensive for even moderately sized inputs. As a result, speeding up motif counting has the potential to accelerate many real-world applications. I have devised several optimizations to substantially decrease the execution time. Specifically, my GPU implementation focuses on the use of shared memory and warp-level parallelization. With these optimizations, my code is, averaged over many inputs, over 8 times faster than the baseline GPU code and 32 times faster than parallel CPU code.
Deadline: Aug. 8, 2022, midnight