high performance computing on graphics processing units: hgpu.org

This thesis reports a study for a new real-time trigger for the NA62 experiment based on Graphical Processing Units (GPUs). The NA62 experiment was devised to study with unprecedented precision the ultra-rare decay K+->pi+ nu nu_bar, a process mediated by Flavour-Changing Neutral Currents (FCNC) whose exceptional theoretical cleanliness provides a unique probe to test the Standard Model. The use of a high-rate kaon beam will result in an event rate of about 15 MHz, so high that it will be impossible to store data on disk without an efficient selection. The experiment therefore devised three trigger levels, allowing to reduce the data rate fed to the readout PC farm down to ~10 kHz. For this thesis I developed an online trigger algorithm that uses data fed by the RICH (Ring Imaging CHerenkov counter) detector in real-time to allow a rejection of the dominant background K+-> pi+ pi0 based on kinematical constraints. As a starting point for the development of this algorithm, I verified the feasibility of such a trigger through Montecarlo simulations. I measured the reconstruction resolution, achieved by the RICH detector alone, of the kinematical variables used for the event selection. After that, I analysed the background rejection power and the signal efficiency of several kinematical constraints, and I designed an actual trigger algorithm. The necessity of running the algorithm in real-time, with a maximum latency of 1 ms per event, drove the choice of exploiting the parallel computing power of GPUs. A parallelized algorithm was therefore developed, that can fit up to 4 Cherenkov rings per event. Moreover, a large number of events are processed concurrently. No parallelized and seedless multi-ring fitting algorithm existed before. The developed algorithm consists of a pattern recognition stage, to assign the hits to up to 4 ring candidates, and of a robust single-ring fit routine. The program was tested on GPUs, and its performance and execution latency proved to be compatible with the requirements. This work proves that alternative trigger designs are possible for the NA62 experiment, and represents a starting point for the introduction of flexible GPU-based real-time triggers in High Energy Physics.