Vectorized Sparse Examination.
Abstract
Vectorizable sparse equation solution algorithms are classified by the matrix structure which they favor. The state-of-the-art for solution of relatively dense systems is then reviewed. A hybrid vector construct is defined for the increasingly common structure of both moderate local matrix density and global matrix regularity. Estimates are made of CRAY-1 speedup achievable with this construct. A finite difference matrix is studied as an example. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 01, 1979
- Accession Number
- ADA074091
Entities
People
- Donald Albert Calahan
Organizations
- University of Michigan