In this paper, we propose an algorithm for solving random sparse linear equations on the NVIDIA Volta architecture to speed up the transient analysis of electronic circuits. In transient analysis, it is necessary to linearize the circuit equations and solve random sparse simultaneous linear equations. The extended vectorized LU decomposition method is effective for solving such linear equations. The extended vectorized LU decomposition method is proposed for vector computers and can extract high parallelism. Therefore, it is possible to speed up equation solving by utilizing the high parallelism of the GPU. In this study, we propose and evaluate two types of optimization methods for the extended vectorized LU decomposition method that are tailored to the characteristics of the NVIDIA Volta Architecture. The first optimization is data transfer, and the second is warp divergence. As a result of the evaluation, the speedup ratio of the proposed method is about 1.43 times faster than the maximum.
Tominaga et al. (Sat,) studied this question.