Optimising CASTEP on Intel's Knight's Landing Platform Technical Report for eCSE 11-17

CASTEP is a widely-used, UK-developed software package based on density functional theory, and capable of predicting the properties of materials from “first-principles”; that is, by solving quantum mechanical equations to determine what the behaviour is, without the need for adjustable parameters. CASTEP was designed from the beginning to run well on conventional parallel HPC machines, but in recent years a number of new computer architectures have emerged which do not follow the conventional trends for CPUs. One such architecture is Intel’s Knights Landing (KNL). Knights Landing’s theoretical performance is very high, but much of its performance is delivered by long vector instructions on many low-power cores, and its performance profile differs considerably from that of a conventional CPU. In this work we profiled and analysed CASTEP’s performance on KNL, with particular attention to its vectorisation, and optimised the computational bottlenecks. The effects of hyperthreading and the KNL memory mode were also investigated. Substantial performance gains were realised in several key subroutines, and CASTEP performance on KNL was improved considerably.1.0This work was funded under the embedded CSE programme of the ARCHER UK National Supercomputing Service (http://www.archer.ac.uk)