Efficient Matrix Inversion via Gauss-Jordan Elimination and ItsParallelization
Enrique S. Quintana, Gregorio Quintana, Xiaobai Sun, Robert A. Geijn · 1998
We present a new parallel matrix inversion algorithm and report its implementation on parallel computers with distributed memory. The algorithm features natural load balance, simple programming and easy performance optimization, while maintaining the same arithmetic cost and numerical properties of the conventional inversion algorithm. Our analysis and experiments on a Cray T3E report near-peak performance for the new approach.