KM3NeT [1] will house two next-generation underwater neutrino experiments. Two detectors are being constructed at two deep sites in the Mediterranean Sea, one of the coast of Sicily and the other south of Toulon in Southern France. Nikhef [2] is involved in the design and construction of the detectors and in the JPP software framework used to reconstruct signals from neutrinos.
Neutrino’s traversing the earth sometimes interact with material in the ground or atmosphere to produce a muon or electron. The KM3NeT detectors search for the signature of light emitted by these muons and electrons as they traverse the its volumes.
One of the main algorithms that is used to detect such electrons and muons assumes many possible directions they might originate from. It first transforms the orientation and position of the detectors’ sensors, and then searches for signatures of emitted light that match electrons and muons.
The current implementation of this algorithm is not optimized to use modern CPU features. This project proposes to benchmark the performance of the algorithm and to investigate opportunities for improvements by introducing vector instructions and optimization of data structures.
A clear picture of performance bottlenecks in the algorithm and a implementation of a vectorized version.