FFT in the Exascale: Opportunities and Challenges
The Fast Fourier Transform (FFT) is still a critical part of main applications run on supercomputers. Unfortunately, in its present form, it does not appear to be an exascale algorithm and does not have replacement algorithms in all the areas in which it is used. The aim of this BoF is to discuss the current state of the art on FFT applications and implementations, and in particular future prospects for the FFT.
There are many variations of methods to use for implementing Fourier Transforms, and it is not always clear what will work best on a given distributed memory parallel platform. To allow for comparisons of different Fourier transform libraries, in a previous venue, the FFT developers discussed the possibility of having a common test case study and what would be a good way to benchmark FFT. It was an opportunity to learn about the different available packages from developers and be able to differentiate between them for future efficient utilization. Each developer had given a brief summary of his/her software. In this session speakers will highlight existing or potential bottlenecks in the different FFT packages and propose how to overcome them. The BoF organizers will query the audience on their interest in FFT requirements. The end of the BoF will be open to audience questions. The targeted audience includes those who seek performance in a software utilizing an FFT library.
|Daisuke Takahashi||University of Tsukuba|
|Cris Cecka||Nvidia and Stanford University|
|Hari Subramoni||Ohio State University|
|Peter Steinbach||Max Planck Institute of Molecular Cell Biology and Genetics and Scionics Computer Innovation GmbH|