Lower-upper (LU) factorization is widely used in many scientific computations. It is one of the most critical modules in circuit simulators, such as the Simulation Program With Integrated Circuit Emphasis. To exploit the emerging graphics process unit (GPU) computing platforms, several GPU-based sparse LU solvers have been recently proposed. In this paper, efficient algorithms are presented to enhance the ability of GPU-based LU solvers to achieve higher parallelism as well as to exploit the dynamic parallelism feature in the state-of-the-art GPUs. Also, rigorous performance comparisons of the proposed algorithms with GLU as well as KLU, for both the single-precision and double-precision cases, are presented.

Additional Metadata
Keywords Circuit simulation, graphics processing unit (GPU), lower-upper (LU) factorization, multicore, parallel simulation, Simulation Program With Integrated Circuit Emphasis (SPICE), sparse matrices
Persistent URL dx.doi.org/10.1109/TVLSI.2018.2858014
Journal IEEE Transactions on Very Large Scale Integration (VLSI) Systems
Citation
Lee, W. (Wai-Kong), Achar, R, & Nakhla, M.S. (2018). Dynamic GPU Parallel Sparse LU Factorization for Fast Circuit Simulation. IEEE Transactions on Very Large Scale Integration (VLSI) Systems. doi:10.1109/TVLSI.2018.2858014