The expanded use of parallel computers heightens the need for numerical software which meets the joint requirements of ease of use, transportability across different architectures, state-of-the-art methods, and high megaflop rates. This study concerns the PCG package which is currently being developed to solve sparse systems of linear equations by preconditioned gradient iterative methods. Here we briefly describe the basic design and some of the features of the package. Issues involved in designing the software to be transportable across different architectures are discussed. Finally some preliminary timings for a model test problem are given to verify feasibility.
展开▼
