Experience in SYCL/oneAPI for event recostruction at the CMS experiment

The CMS software framework (CMSSW) has been recently extended to integrate heterogeneous computing with the aim of performing part of the physics reconstruction with GPUs to face the computational challenge that will come with the increase of luminosity during the High-Luminosity phase of the Large Hadron Collider (HL-LHC). To avoid writing a different implementation of the code for each backend, a performance portability library is used: Alpaka has been chosen as the compatibility layer for Run-3. With the idea of exploring new solutions, SYCL has been considered since it would allow to target also Intel GPUs. SYCL is a cross-platform abstraction C++ programming model for heterogeneous computing. It allows developers to reuse code across different hardware and also perform custom tuning for a specific accelerator. The SYCL implementation used in this work is the Data Parallel C++ library (DPC++) in the Intel oneAPI Toolkit. This work shows the performance of physics reconstruction algorithms written in SYCL on different hardware. Strengths and weaknesses of this heterogeneous programming model will also be presented.