BLonD++: performance analysis and optimizations for enabling complex, accurate and fast beam dynamics studies

This paper focuses on the performance analysis and optimization for enabling efficient implementations of next generation beam dynamics simulations. Nowadays large worldwide research centers, e.g. CERN, Fermilab etc. are continuously investing in resources and infrastructures for progressing knowledge in the fields of particle physics, thus requiring careful studies and planing for the upcoming upgrades of the synchrotrons and the design of future machines. Consequently, there is an emerging need for simulations that incorporate a collection of complex physics phenomena, produce extremely accurate predictions while keeping the computing resources and run-time to a minimum. A variety of simulator suites have been developed, however, they have been reported to lack in simulation speed, features and ease-of-use. In this paper we introduce the Beam Longitudinal Dynamics (BLonD) simulator suite from a computer engineering perspective. We analyze its performance to understand its current bottlenecks and enhance it further in an attempt to make complex, accurate and fast beam dynamics simulations possible. We show that through careful and targeted analysis and code tuning, the proposed BLonD++ implementation delivers significant gains in terms of performance, i.e. up-to 23X single-core speedup and scalability, thus enabling the deployment of even more complex simulations than the current state-of-art.