POSTER P504: Accelerated Briggs Weighting Function on NVIDIA GPUs with CUDA

Theme: GPU implementations for core astronomical libraries

pretalx

Radio interferometric imaging requires weighting of the measured visibilities to produce images suitable for scientific analysis. Natural and Uniform weighting schemes are implemented to either minimise the effects  of sidelobes (Uniform) or to decrease the system noise in the data (Natural). The Briggs Weighting function provides a compromise between these two weighting schemes, using a robustness parameter. In this work we present an implementation of the Briggs Weighting function with CUDA on NVIDIA GPUs, achieving at least 3x speedup on a GeForce RTX 3090 Ti compared to DDFacet on synthetic data with 16 threads on an Intel i9 processor.