Parallel algorithm for solving systems of linear equations with dynamically changed length of operands

Peculiarity of the most direct algorithms for solving system of linear equations is the use of divisions for the elimination of unknowns. Division requires great time for its execution, when the multiprecision arithmetic is used because in this case it is realized by special programs. In this paper a parallel algorithm is described implementing the elimination procedure without divisions. Some results are presented of execution time, speedup, and accuracy. Measurements were taken for a system supporting massively parallel high-accuracy SIMD-computations by dynamically changed length of operands.<<ETX>>