Login
Talks & Seminars
Title: Irregular Algorithms on GPUs
Dr. Rupesh Nasre, University of Texas at Austin
Date & Time: February 13, 2013 14:15
Venue: Conference Room, 01st Floor, C Block, Department of Computer Science and Engineering, Kanwal Rekhi Building
Abstract:
I would present parallelization techniques for irregular algorithms on graphics processing units (GPUs). Programs with unpredictable control-flow or data-access patterns are termed irregular. Graph algorithms such as the shortest paths computation, approximate n-body simulation and mesh refinement are a few examples of irregular algorithms. Unlike regular algorithms like matrix multiplication, irregular algorithms are relatively difficult to parallelize, since static analysis techniques fail to capture the unstructured access patterns that are input dependent. The challenge of scalable parallelization of such irregular algorithms is further exacerbated in the presence of massive multithreading such as that on the GPUs. In fact, until recently, it was unclear how to parallelize irregular algorithms. Therefore, we devised a set of parallelization principles for irregular algorithms on GPUs. Extending these principles, I would illustrate how to parallelize several morph algorithms wherein the underlying graph itself gets modified, e.g., in mesh refinement. Finally, I would briefly discuss how to synthesize such GPU programs from a higher level specification.
Speaker Profile:
Rupesh Nasre is a postdoctoral researcher in the Institute for Computational Engineering and Sciences at the University of Texas at Austin. He received the Ph.D. degree from Indian Institute of Science, Bangalore. His research interests include program parallelization and static program analysis. He is the recipient of the NVIDIA special prize in CodeForScience, a winner of the Yahoo! University Hack Day and holds five US patents.
List of Talks

Webmail

Username:
Password:
Faculty CSE IT
Forgot Password
    [+] Sitemap     Feedback