2. Lis: a Library of Iterative Solvers for Linear

Construction of a software infrastructure for highly parallel computing environments requires precisely prediction of future hardware technologies, and design of scalable and portable software for these technologies. The Scalable Software Infrastructure for Scientific Computing (SSI). Project was initiated in November 2002, as a national project in Japan, for the purpose of constructing a scalable software infrastructure [1], [2], [3]. Based on the policies, we have used various types of parallel computers, and carefully designed our libraries on them, to maintain portability and usability. The architectures included shared-memory parallel computers, distributed-memory parallel computers, and vector supercomputers. Since 2006, the SSI project has been selected for a joint research with the Earth Simulator Center to port our libraries on massively parallel vector computing environments. Since 2003, we have signed a contract with the IBM Watson Research Center on the joint study of library implementation on massively parallel environments with tens of thousands of processors. The results of the SSI project will be evaluated on larger computers in the near future. In the SSI project, we have studied object-oriented implementation of libraries, autotuning mechanisms, and languages for the implemented libraries. The results were applied to a modular iterative solver library Lis and a fast Fourier transform library FFTSS. The libraries were written in C and equipped with Fortran interfaces. We have also developed a simple interface for library collections SILC, with an extension to scripting language