Performance Acceleration of Multifrontal Solver Using GPU

초록

In this research, an efficient way to accelerate the performance of a finite element analysis code by accelerating the performance of a linear equation solver using GPU has been studied. The multifrontal solver is known as the most efficient direct linear equation solver for finite element analysis, so it was used as the linear equation solver for acceleration of the performance of the finte element analysis code. By using GPU acceleration for the computation of partial factorization of dense matrices, the performance of the multifrontal solver could be accelerated considerably. The performance of partial factorization could be improved further by developing a modified partial factorization function compared to existing libraries.