There are CFD/MHD codes that employ FFTs in 3D for a particular set of boundary conditions  -- good question whether they will continue to scale on new architectures.

Aleks


From: cps-employees-bounces@lists.anl.gov <cps-employees-bounces@lists.anl.gov> on behalf of Keceli, Murat <kecelim@anl.gov>
Sent: Friday, December 20, 2019 6:03 PM
To: Habib, Salman <habib@anl.gov>; cps-all@anl.gov <cps-all@anl.gov>
Subject: Re: FFTs
 
Periodic electronic structure calculations with DFT (density functional theory) require FFTs. It's indeed the scaling bottleneck after the diagonilazation step, so there has been some work on parallel FFTs.

https://link.springer.com/chapter/10.1007%2F978-3-540-92859-1_25
https://www.nersc.gov/assets/Events/AndrewCanningNERSCBrownbag.pdf
https://www.sciencedirect.com/science/article/pii/S0010465519301213

Happy holidays,
Murat



From: cps-employees-bounces@lists.anl.gov <cps-employees-bounces@lists.anl.gov> on behalf of Habib, Salman <habib@anl.gov>
Sent: Friday, December 20, 2019 4:57 PM
To: cps-all@anl.gov <cps-all@anl.gov>
Subject: FFTs
 
Hi Everyone,

As a parting shot, is there any interest in parallel FFTs in
applications of interest to CPS?

There's a bunch of activity in this space and I am interested in
collecting all possible use cases.

Thanks!

Salman

_______________________________________________
cps-employees mailing list
cps-employees@lists.anl.gov
https://lists.anl.gov/mailman/listinfo/cps-employees