Mapping Large Wavefront Algorithms onto Processor Square Meshes of Fixed Size