Faculty Mentor(s)
Dr. Omofolakunmi Olagbemi, Computer Science
Document Type
Poster
Event Date
4-11-2025
Abstract
Mathematical integration is a common process in a wide array of fields such as medical imaging, statistical analysis, orbital mechanics, and simulation modeling. Since analytical solutions to many integration problems are impossible to obtain, they are instead numerically approximated. There exists a variety of numerical integration software used to estimate such integrals; one such software is DCUHRE which employs the global adaptive algorithm over a hyperrectangular region to estimate a given function. However, as the integral dimension increases, the number of region evaluation points increases exponentially. While DCUHRE was written to accommodate parallel execution on multiple processors (subregion-level parallelization), this level of parallelization is significantly less than what can be achieved using GPUs, thus resulting in significantly longer execution times in higher dimensions. Our parallel solution ParDCUHRE implements evaluation-point-level parallelization and achieves speedups well into the triple digits with selected integrals (with dimensionality as high as 25) while yielding results comparable to the original DCUHRE.
Recommended Citation
Repository citation: Shaw, Tobias and Worden, Peter, "ParDcuhre: A Scalable Parallel Solution for Multivariate Integration with CUDA" (2025). 24th Annual A. Paul and Carol C. Schaap Celebration of Undergraduate Research and Creative Activity (2025). Paper 19.
https://digitalcommons.hope.edu/curca_24/19
April 11, 2025. Copyright © 2025 Hope College, Holland, Michigan.
Comments
This research was supported by the Brookstra Faculty Development Fund, private donors through the Office of the Dean for Natural and Applied Sciences and the Computer Science department at Hope College.
Title on poster differs from abstract booklet. Poster title: ParDCUHRE - A CUDA Parallelization of DCUHRE