A SLAP (Sparse Linear Algebra Package) for the masses

Conference · Sun Dec 11 23:00:00 EST 1988

OSTI ID:6525265

Seager, M K

A Sparse Linear Algebra Package (SLAP), written in FORTRAN77, for the iterative solution of large sparse symmetric and non-symmetric linear systems is presented. SLAP Version 2.0 consists of three levels of routines: ''high level'', ''core'' and ''utility.'' The ''core'' routines implement the following preconditioned iterative methods: iterative refinement, conjugate gradient, conjugate gradient on the normal equations, bi-conjugate gradient, bi-conjugate gradient squared, orthomin and generalized minimum residual. All of these methods do not require the data structure of the matrix being solved nor of the preconditioning matrix, but do require the ''user'' to supply a matrix vector product and preconditioning routines. The ''high level'' routines assume one of two specific data structures and provide the required ''user routines.'' The preconditioners supported are diagonal scaling and incomplete factorization. One of the SLAP data structures allows for the vectorization of the matrix multiply and the backsolve of the incomplete factorization operations on machines with hardware gather/scatter capabilities. We present results for SLAP on the Cray Y/MP and Alliant FX/8 machines. 10 refs., 3 figs., 10 tabs.

• Research Organization: Lawrence Livermore National Lab., CA (USA)
• DOE Contract Number: W-7405-ENG-48
• OSTI ID: 6525265
• Report Number(s): UCRL-100195; CONF-8810280-1; ON: DE89006209
• Country of Publication: United States
• Language: English

Related Subjects

99 GENERAL AND MISCELLANEOUS
990210* -- Supercomputers-- (1987-1989)
990230 -- Mathematics & Mathematical Models-- (1987-1989)
ALGORITHMS
COMPUTER ARCHITECTURE
COMPUTERS
DIFFERENTIAL EQUATIONS
DIGITAL COMPUTERS
EQUATIONS
FACTORIZATION
FORTRAN
ITERATIVE METHODS
MATHEMATICAL LOGIC
MATRICES
PROGRAMMING LANGUAGES
SUPERCOMPUTERS