Shared memory parallelization of the conjugate gradient linear system solver of a FEM topology optimization eode

This paper aims to study and improve the runtime of a topology optimization scientific code named OPTRUSS (OPTimization of TRUSS structures), developed at the Romanian National Institute for Aerospace Research "Elie Carafoli". OPTRUSS is a FORTRAN code that optimizes the stiffness of truss structure, minimizing compliance, following the SIMP algorithm. The physical problem solved is relevant because of the linear system´s assembled stiffness sparse matrix - emphasis on sparse. The primary metric for performance is the average Wall Time per iteration of the Conjugate Gradient algorithm, and we will focus on evaluating the speedup of the proposed OpenMP and CUDA parallelizations.