Particle Flow Reconstruction on Heterogeneous Architecture for CMS

The higher instantaneous luminosity expected during the upcoming years of LHC Run 3 operations will impose challenges for CMS event reconstruction, and this will be amplified in the HL-LHC era, where luminosity and pileup rates are expected to be significantly higher. One of the approaches for CMS to cope with this challenge is to utilize heterogeneous computing architectures in order to accelerate event reconstruction. In this presentation, we discuss an effort to utilize GPU accelerators for particle flow (PF) reconstruction. The PF algorithm, used for the vast majority of CMS data analyses for event reconstruction, provides a comprehensive list of final-state particle candidates and enables efficient identification and mitigation methods for simultaneous proton-proton collisions (pileup). The PF algorithm consists of multiple steps, and the clustering of calorimeter hits is one of its most time-consuming steps. As a first step toward accelerated PF reconstruction, a GPU version of PF clustering for hadronic showers has been developed. The cluster outputs and computational performance of the GPU-accelerated algorithm are compared to those of the CPU-based implementation.