GPU accelerated multilevel fast physical optics algorithm for radiation from non-planar apertures

Acceleration of the multilevel physical optics (MLPO) algorithm using the single instruction multiple threads (SIMT) computing scheme, as implemented on a CUDA-architecture based GPU accelerator is presented. The MLPO algorithm already reduces the computing complexity (CC) of the physical optics based radiation pattern calculation algorithm to O(N²logN), as compared to the O(N³) or O(N⁴) CC of the straight-forward implementation (depending on the nature of the problem). Here, N=kR, R being the radius of the smallest sphere containing the radiating aperture and k - the wavenumber. By implementing on a GPU, the MLPO algorithm is accelerated tenfold, allowing fast and efficient evaluation of physical optics integrals. Performance improving optimizations of the parallel code are described. Finally, a timing model that provides estimates of the GPU computation time is presented and compared to the actual GPU run-time measurements.