What you need to know to use these subprograms

The next three letters in the naming convention, YYY, indicate the data structure of the sparse matrix, as presented in Table 4-2.

Table 4-2 Sparse BLAS Naming Convention—Matrix Form

YYY

Point Entry

COO

Coordinate

 

 

CSC

Compressed sparse column

 

 

MSC

Modified sparse column

 

 

CSR

Compressed sparse row

 

 

MSR

Modified sparse row

 

 

DIA

Sparse diagonal

 

 

ELL

Ellpack-Itpack

 

 

JAD

Jagged diagonal

 

 

SKY

(Triangular) Skyline

 

 

YYY

Block Entry

 

 

BCO

Block coordinate

 

 

BSC

Block compressed sparse column

 

 

BMC

Block modified sparse column

 

 

BSR

Block compressed sparse row

 

 

BMR

Block modified sparse row

 

 

BDI

Block sparse diagonal

 

 

BEL

Block Ellpack-Itpack

 

 

VBR

Variable block row

 

 

The table below lists the final two characters in the naming convention, ZZ, indicating the type of operation.

Sparse BLAS Naming Convention—Operation

ZZSubroutine Operation

MM Matrix-Matrix product

SM

Solution of a triangular system with mulitple right-hand-sides

Sparse matrix storage formats

Each Sparse BLAS subroutine handles a specific sparse matrix storage format. These formats are classified into two major groups: point entry and block entry data structures, as presented in Table 4-2. Point entry formats store the nonzero entries and keep track of their location in the matrix. Block entry formats represent sparse matrices whose nonzero entries are dense blocks.

424HP MLIB User’s Guide