Skip to main content.
Advanced search >
<< Back to previous page Print

<< Wednesday, May 01, 2013 >>

Remind me

Tell a friend

Add to my Google calendar (bCal)

Download to my calendar

Bookmark and ShareShare

A Framework for Low-Communication 1-D FFT: Scientific Computing and Matrix Computations Seminar

Seminar: Scientific Computing | May 1 | 12:10-1 p.m. | 380 Soda Hall

Peter Tang, Intel Corporation

Electrical Engineering and Computer Sciences (EECS)

In state-of-the-art high-performance computing on distributed-memory systems, communication often represents a significant part of the overall execution time, and quite likely consumes a major share of the total energy used. For distributed 1-D FFT, every industry-standard implementation performs three all-to-all internode data exchanges which make up the bulk of communication. We present here a mathematical framework for deriving a family of easy-to-implement single-all-to-all 1-D FFT algorithms. Furthermore, our framework allows tradeoff between accuracy and performance. Depending on the problem size and the computer system used, implementations at comparable accuracy based on our new approach can outperform leading FFT libraries by as much as twofold, higher still if reduced accuracy is acceptable., 510-516-4321