The storage format is the underlying data structure in which
the sparse matrix is represented. Available formats are:
For further information regarding these formats, see
"A Revised Proposal for a Sparse BLAS Toolkit", by S. Carney,
M. Heroux, G. Li, R. Pozo, K. Remington and K. Wu.
The BLAS operation can be either:
a sparse matrix by dense matrix multiplication of the form:
or a sparse matrix triangular solve of the form:
In each case, A is a sparse matrix, B and C are dense matrices or
vectors, and and
are scalar parameters.
For triangular solves, diagonal scaling on the left
() and/or on the right
() may be specified.
Identity scaling is the default.
The column dimension of B and C determines whether the operation is matrix-vector (Level 2) or matrix-matrix (Level 3). Note that the Level 3 code can be used to do a Level 2 operation, but not vice versa.
Certain special case values of the scalar parameters
and can be chosen to yield simplified code.
The allowed special cases are: 1 for , and
0 or 1 for .
Return to NIST Sparse BLAS Home Page ...
Last updated: July 25, 1996 by
KAR.
Return to the SourceService request page...