ACCELERATING CLUE WITH INTEL ONEAPI SYCL

Parallel particle reconstruction algorithms are necessary to sustain the increasing event output size and frequency of high granularity calorimeters (HGCAL) like the ones installed in the LHC at CERN. Employing hardware accelerators like GPUs or FPGAs and their respective SDKs has resulted in parallel applications capable of handling the required throughput demand. However, it brings a new software development challenge when a combination of different vendors and architecture hardware accelerators are used to run the programs. Performance portability frameworks like OpenCL, Alpaka, among others have been recently developed to enable heterogeneous applications to run on multiple hardware. Selecting any of them requires exploring the benefits and drawbacks of each technology to make an informed decision. In this work, the performance portability framework SYCL with DPC++ was used to re-implement the energy clustering algorithm (CLUE) with the purpose to evaluate the potential benefits of the framework, efficiency in contrast to native implementations and generate a SYCL porting knowledge base useful for future projects