The need for effective and portable parallel programming models has increased due to the complexity of high-performance computing systems, especially with the introduction of many-core processors like GPUs. Creating applications for various multi-core and many-core architectures frequently necessitates maintaining separate codebases, which increases development effort. In this paper, we evaluate Kokkos, a high-level C++ framework that allows for performance portability across architectures, by running five benchmark applications on a CPU and comparing their performance with OpenMP-based counterparts. The obtained results and corresponding observations are discussed in detail.
Paunović et al. (Wed,) studied this question.