$\texttt{cunuSHT}$: GPU Accelerated Spherical Harmonic Transforms on Arbitrary Pixelizations

• DOI: 10.48550/arxiv.2406.14542
• Publication Date: 2024-06-20
• Authors: Sebastian Belkner; Adriaan J. Duiven...; Julien Carron; Nathanaël Schaeffer; Martin Reinecke
• Keywords: Harmonic

We present $\texttt{cunusht}$, a general-purpose Python package that wraps highly efficient CUDA implementation of the nonuniform spin-$0$ spherical harmonic transform. The method is applicable to arbitrary pixelization schemes, including schemes constructed from equally-spaced iso-latitude rings as well completely ones. algorithm has an asymptotic scaling $\mathrm{O}{(\ell_{\rm max}^3)}$ for maximum multipole $\ell_{\rm max}$ and achieves machine precision accuracy. While $\texttt{cunusht}$ developed applications in cosmology mind, it various other interpolation problems on sphere. outperform fastest available CPU by factor up 5 with max}>4\cdot10^3$ when comparing single modern GPU 32-core CPU. This performance achieved utilizing double Fourier sphere combination fast transform avoiding transfers between host device. For scenarios without availability, existing libraries. publicly includes tests, documentation, demonstrations.