Radio interferometric imaging with current and future instruments requires the handling of large datasets. GPUs provide an effective means of dealing with this data deluge. However, to fully leverage the computational power available on these GPUs, a tightly coupled iterative approach involving performance engineering and software engineering is required.

In the case of the high-performance gridder, we have utilized the Kokkos software framework to develop high-level C++ code that can be deployed across a wide range of software environments, benefiting from compile-time optimizations. In this presentation, we will explore the role of performance engineering in the context of the development and testing of the high-performance gridder. We will demonstrate the performance improvements achieved through optimizing already highly performant code, which has been tested on a variety of GPUs spanning different architectures and hardware generations