The proposed project will pursue the theory and practice of implementing algorithms for dense linear algebra operations on SMP and multicore systems. It will develop algorithms, abstractions, APIs, implementations, software architectures, analyses, and simulators. The successful completion of the project will improve the understanding of the subject, will provide tools for implementing scalable libraries and pedagogical tools, and will make libraries available to the scientific computing community. In addition, it is expected to provide important insights into the design and use of current and multicore processors
The intellectual merit of the proposed work lies with the knowledge that will be gained regarding the parallel implementation of linear algebra libraries and applications in general on SMP systems and multicore processors. In addition, insights are expected to affect future design of architectures with large numbers of cores on a single chip.
The broader impact lies with the tools, in form of libraries, that will be made available to the computational science community, and the pedagogical benefits. In the body of this proposal and in the letters of support from industry, evidence is presented that the proposed work is: (1) of scientific importance, (2) based on sound preliminary results, (3) achievable, (4) of benefit to the scientific computing community, (5) of interest to industry, and (6) of pedagogical value.
