Future linear-algebra libraries

J. Dongarra
1996 IEEE Computational Science & Engineering  
he ultimate development of fully mature, parallel scalable libraries will necessarily depend on breakthroughs in many supporting techologes. Scalable library development cannot wait, however, until all the enabling technologes are in place for two reasons: The need for such libraries for existing and near-term parallel architectures is immediate, and progress in all the supporting technologies depends on feedback from concurrent efforts in library development. Traditional libraries We in the
more » ... ear-algebra community have long recognized that we needed something to help us develop our algorithms into software libraries. Several years ago, as a community effort we put together a de facto standard for identifjmg the basic operations required in our algorithms and software. Our hope was that many manufacturers would implement the standard on their machines, and that we could then draw on the power of that implementation in a rather portable way. We began with BLAS operations (basic linear-algebra subprograms) designed for basic matrix computation. Since message passing is critical in a parallel system, we worked on developing message-passing standards. Both the PVM (parallel virtual machine) and the MPI (message-passing interface) have helped establish standards and promote portable software, critical for software library work. User interfaces As computer architectures and programming paradigms become increasingly complex, we want to hide this complexity from users as 38 1070-9924/96/$5.00 0 1996 IEEE
doi:10.1109/99.503308 fatcat:nx2binbb4vdcteaiqugkt735nu