- Next: ../STGIRONS/oo_eg
- Up: Presentation
- Previous: ../STGIRONS/pcg_c++
OO Sparse Matrix Libraries
- motivated by ``Building Blocks for Iterative Solution of Linear
Systems''
- based on Level 3 Sparse BLAS interface
- support distributed formats: compressed row / column
- implementing parallel Level 3 Sparse BLAS
- inital development of conventional sparse matrix classes
- compressed column (Harwell/Boeing)
- compressed row
- jagged diagonal storage
- compressed diagonal storage
- block compressed row/col storage
- skyline storage
OO Sparse Matrix Libraries (contd.)
- Iterative algorithms
- Stationary Methods (SOR),
- Conjugate Gradient (CG),
- Generalized Minimal Residual (GMRES),
- Minimum Residual (MINRES),
- Quasi-Minimal Residual (QMR),
- Chebyshev Iteration (Cheb),
- Conjugate Gradient Squared (CGS),
- Biconjugate Gradient (BiCG),
- Biconjugate Gradient Stabilized (Bi-CGSTAB)
- Preconditioners are user-defined
- some basic stationary preconditioners provided
- application specific
- more difficult to parallelize
- Current Development
- Phase I: development of conventional sparse
matrix classes:
- Phase II: development of shared memory sparse matrix
- Phase III: development of distributed memory sparse
matrix
objects with Level 3 Sparse BLAS
Slide 83 of SIAM Tutorial, Jack Dongarra UT/ORNL.
(Updated 01/31/95)