Paralelizando o MOPAC usando CUDA e bibliotecas de Matrizes Esparsas

Abstract

This work describes the implementation of parallel algorithms whose main goal is to accelerate the implementation of numerical calculations existing in quantum chemistry programs. These programs use some methods whose order of complexity varies from O(n3) and O(n5), where n is the parameter related to the amount of atoms in a molecule. This becomes a limiting factor when one wants to work with molecular systems containing thousands of atoms, such as proteins, DNA and polysaccharides. It is explored both the parallelism provided by graphics cards and the CUDA programming model are also used libraries for manipulating sparse matrices, which are common in these calculations. The results show gains of more than 100% for test instances.