Scalable Parallel 3d FFTs for Electronic Structure Codes
Andrew Canning (Lawrence Berkeley National Laboratory)
Abstract:
First-principles methods based on Density Functional Theory (DFT)
where the wavefunctions are expanded in plane waves (Fourier components) are the most widely used approach
for electronic structure calculations in materials science.
The scaling of this method depends critically on having an efficient parallel 3d FFT that minimizes communications and calculations. We present an implementation and performance data
of a parallel 3d FFT specifically designed for electronic structure calculations that scales to thousands of processors on leading parallel and vector computer platforms (IBM SP, Cray XT, NEC SX).
keywords: FFT, Materials Science, Parallelization
Keywords:
Large Scale Simulations in CS&E (earth, environment, finance, geoscience, engineering, ...), Parallel and Distributed Computing
