Parallel conjugate gradient-like algorithms for solving sparse nonsymmetric linear systems on a vector multiprocessor (Q1124269)

Two parallel modifications of the preconditioned conjugate gradient method for systems with a nonsymmetric matrix are presented. The algorithms are the conjugate gradient squared algorithm and the generalized minimal residual algorithm. In the algorithms preconditionings by a diagonal matrix and by an incomplete LU factorization are considered. A compressed-matrix representation storage scheme developed at the University of Texas in Austin is adopted. The global and local preconditioners in the incomplete LU factorization are used. The algorithms were implemented on the vector multiprocessor IBM 3090-600E VF and tested for two sets of test matrices.