Next: ../STGIRONS/pcg_c++

Up: Presentation

Previous: ../STGIRONS/bookcover

Templates for Sparse Matrix Computations

• Alternate to software
• Describe the basic features of the algorithms
• Language not so important
• Communication between disciplines
• Clear documentation
• Test cases
• Provide understanding of the results
• Allow customization
• Serve as a pedagogical role
• Retain the delicate numerical details
• Describe parallel opportunities

Building Blocks
for Iterative Solution of Linear Systems

• Some users want algorithms which are fast for their particular application even if not reliable as general methods.

• They want the tools to customize iterative
  methods for solving the linear systems that arise in their specific problem.

• Offer reusable software templates
  • Algorithm
    • Jacobi, Gauss-Seidel, SOR
    • CG, Precond-CG, CG-Squared, Bi-CG
    • Bi-CG-Stab, QMR, GMRES, Chebychev
  • Matlab script
  • Fortran code in a 2-D array
  • Preconditioners
  • Convergence tests
  • Performance Summary
  • Data Structures
  • Hints on parallel implementation