Abstraction of the CMS track reconstruction software for heterogeneous computing: application to the CMS silicon strip clustering algorithm

The reconstruction of charged particles in the CMS tracking detector, during both the online and the offline collision event reconstruction, is crucial for the success of the CMS physics program. However, at increasing instantaneous luminosity, the required computing resources for an efficient and timely track reconstruction grow almost exponentially. For this reason, the CMS track reconstruction software needs to maximally exploit parallel computing on multi- and many-core CPUs as well as GPUs (heterogeneous computing). In this context, the ability to abstract the CMS track reconstruction software for heterogeneous computing allows not to depend on specific choices of computing architectures, at the present time as well as in the future. This project focused on the algorithm used to cluster charge deposits in the CMS silicon strip tracker, for both the online and the offline event reconstruction, with the goal to enable its execution on any CPU or GPU architecture, in view of the future usage of the Alpaka library.