WAKE FOREST UNIVERSITY
DEPARTMENT OF COMPUTER SCIENCE

Dr. Nikos Pitsianis
Departments of Electrical and Computer Engineering and Computer Science
Duke University

Thursday, December 6, 2007
4:00 p.m., Manchester Hall, Room 241

Fast Fourier Transforms on Graphics Processing Units

A Graphics Processing Unit (GPU) is a specialized computer architecture that supports massively parallel fine-grained pixel operations, originally designed for fast rendering of three-dimensional scenes onto two dimensional displays.
Fast Fourier Transform (FFT) algorithms are critical components in many scientific computing application fields, especially image processing and reconstruction.
 It is only natural to utilize GPUs to perform FFTs in addition to image rendering.
This talk will present decompositions of high-dimensional FFTs into kernels consisting of two-dimensional element-wise array operations.  Thus, we exploit and utilize unique GPU architectural features by appropriately adapting the mathematical structures associated with FFT algorithms in order to derive highly efficient GPU implementations.
This work is supported by DARPA under FANTOM and STAPBOY projects.
 
              Refreshments at 3:30  p.m. in Manchester Hall, Room 241

Return to Homepage