Solving linear systems on a vector computer (Q579836)

This paper considers some variants of the well-known Gaussian elimination algorithm for solving linear systems of equations on a vector computer. The author emphasizes three essential choices: normalization of the diagonals, ordering of the calculation and pivoting strategy. Then it is proposed an algorithm which is considered as optimal for the Cyber 205. Comparative experiments in FORTRAN 77 with FORTRAN 200 extensions for the Cyber 205 are reported. Several algorithms are carefully described in the appendix. Similar analysis can be helpful in selecting the optimal algorithm for other supercomputers, depending on machine architecture and programming language used.