Many different fields of science and engineering rely on linear algebra computations for solving problems and creating simulations. These computations are expensive and can often become the most time-consuming part of the simulations. Writing computer programs for doing scientific computation is a difficult task and optimizing those programs for better performance is even more difficult. A large number of software libraries are available for solving matrix algebra related problems. It is good to have many choices. However, finding the suitable software package for solving a particular linear algebra problem can itself become a major challenge. Finding the appropriate routines, integrating them to a larger application and optimizing them is a complicated process that requires expertise in computer programming, numerical linear algebra, mathematical software, compilers, and computer architecture. We have been studying ways to ease the process of creating and using high-performance matrix algebra software. The Lighthouse Taxonomy is the product of our attempt to face the daunting challenges of high-performance numerical linear algebra computation. Like a lighthouse that helps sailors navigate their ships in the dark seas, the Lighthouse guides the numerical linear algebra practitioners through the dark seas of numerical software development. Lighthouse is an open-source web application that currently serves as a guide to the dense linear system solver routines from one of the most widely used numerical linear algebra package known as LAPACK. We have been working on expanding the Lighthouse framework for the production of matrix algebra software by adding support for sparse matrix algebra computations and integrating high-performance parallel numerical libraries. We are particularly interested in a scientific toolkit called PETSc because of its efficiency, unique features and widespread popularity. In this thesis, we explain the process of integrating the sparse linear solvers from PETSc to the Lighthouse Taxonomy and how this could be beneficial to the scientific computing community.
展开▼
