Paralelização do algoritmo MOZYME no MOPAC

Abstract

This article explores the parallelization of the MOZYME algorithm, a linear scaling technique imple mented in the MOPAC quantum chemistry software. Parallelization aims to enable calculations on larger molecular systems, such as biomolecules. The focus of this work is on the parallelization of two crucial MOZYME subroutines: density for MOZYME and diagg1. The parallelization tool chosen was OpenMP, which allows the workload to be distributed among multiple processing cores. The Intel® VTune™ Profiler was used to analyze performance, identifying the density for MOZYME and diagg1 subroutines as those that demand the most computing resources. To evaluate the effectiveness of pa rallelization, a benchmark was carried out using an AMD Ryzen Threadripper PRO 5995WX processor with 64 physical cores and 128 threads. The tests covered several molecular systems, varying in size and complexity, with different cut-off radii for atomic interactions. The results show a speedup of up to 3x in total MOPAC time in a MOZYME calculation. The density and diagg1 subroutines achieved speedups of up to 30x and 6.7x, respectively.