Improving the performance of GWW
A dCSE Project
EPCC, The University of Edinburgh, James Clerk Maxwell Building,
Mayfield Road, Edinburgh, EH9 3JZ, UK
In this report, we present the results of investigation into improving the performance of GWW, part of the Quantum Espresso suite of software for ab initio simulation. In particular, the 3D Fourier Transform was found to be a significant bottleneck to application scaling. Several alternative methods for the FFT transpose were implemented, and the performance of these was studied on HECToR (Phase 2a and 2b). Speedups of up to 400% (on 128 cores of HECToR Phase 2a) were demonstrated for the 3D FFT in isolation, which delivered benefits of in the range of 4-36% in full application benchmarks. A checkpoint and restart mechanism was also added to help jobs complete in under the 12 hour queue limit on HECToR.