Function Description

The naming rules of the Sparse BLAS Level 2 or 3 (CSR format) functions are as follows:

kml_sparse_<character><data><operation>
kml_sparse_<character><data><mtype><operation> (only for one-based indexing of matrices)
kml_csparse_<character><data><mtype><operation> (only for zero-based indexing of matrices)

<character> indicates the data type. Possible values are s (single-precision real floating-point), d (double-precision real floating-point), c (single-precision complex number), and z (double-precision complex number).

<data> indicates the storage type. Possible values are coo, csr, dia, sky, csc, and bsr.

<mtype> indicates the matrix type. Possible values are ge (general matrix), sy (symmetric matrix), and tr (triangular matrix).

<operation> indicates the matrix operation.

**Table 1** Simple interfaces (only for one-based indexing)
Function Family	Data Type	Description
kml_sparse_?csrgemv	s, d, c, z	Computes the product of a matrix and a vector. The matrix is a sparse general matrix stored in the CSR format (3-array variation) with one-based indexing.
kml_sparse_?csrsymv	s, d, c, z	Computes the product of a matrix and a vector. The matrix is a sparse symmetric matrix stored in the CSR format (3-array variation) with one-based indexing.
kml_sparse_?csrtrsv	s, d, c, z	Solves a system of equations for a triangular matrix. The matrix is a sparse symmetric matrix stored in the CSR format (3-array variation) with one-based indexing.

**Table 2** Simple interfaces (only for zero-based indexing)
Function Family	Data Type	Description
kml_csparse_?csrgemv	s, d, c, z	Computes the product of a matrix and a vector. The matrix is a sparse general matrix stored in the CSR format (3-array variation) with zero-based indexing.
kml_csparse_?csrsymv	s, d, c, z	Computes the product of a matrix and a vector. The matrix is a sparse symmetric matrix stored in the CSR format (3-array variation) with zero-based indexing.
kml_csparse_?csrtrsv	s, d, c, z	Solves a linear system of equations for a triangular matrix. The matrix is a sparse symmetric matrix stored in the CSR format (3-array variation) with zero-based indexing.

**Table 3** Common interfaces (for one- and zero-based indexing)
Function Family	Data Type	Description
kml_sparse_?csrmv	s, d, c, z	Computes the product of a matrix and a vector. The matrix is a sparse matrix stored in the CSR format.
kml_sparse_?csrsv	s, d, c, z	Solves a system of linear equations for a sparse matrix that is stored in the CSR format.
kml_sparse_?csrmm	s, d, c, z	Computes the product of a sparse matrix and a dense matrix that are stored in the CSR format.
kml_sparse_?cscmv	s, d, c, z	Computes the product of a matrix and a vector. The matrix is a sparse matrix stored in the CSC format.
kml_sparse_?cscsv	s, d, c, z	Solves a system of linear equations for a sparse matrix that is stored in the CSC format.
kml_sparse_?cscmm	s, d, c, z	Computes the product of a sparse matrix and a dense matrix that are stored in the CSC format.
kml_sparse_?csrsm	s, d, c, z	Solves a system of sparse linear equations with multiple right-hand side (RHS) terms. The matrix is stored in the CSR format.
kml_sparse_?cscsm	s, d, c, z	Solves a system of sparse linear equations with multiple RHS terms. The matrix is stored in the CSC format.

**Table 4** Interfaces for sparse matrix operations
Function Family	Data Type	Description
kml_sparse_?csradd	s, d, c, z	Computes the sum of two sparse matrices that are stored in the CSR format (3-array variation) with one-based indexing. The result is stored in a sparse matrix.
kml_sparse_?csrmultcsr	s, d, c, z	Computes the product of two sparse matrices that are stored in the CSR format (3-array variation) with one-based indexing. The result is stored in a sparse matrix.
kml_sparse_?csrmultd	s, d, c, z	Computes the product of two sparse matrices that are stored in the CSR format (3-array variation) with one-based indexing. The result is stored in a dense matrix.

matdescra Parameter Values

Parameter matdescra is a char pointer and contains six elements (only the first four are used).

Parameter	Value
Element 1	G: general matrix S: symmetric matrix (A = A') H: Hermitian matrix (A = conjg(A')) T: triangular matrix D: diagonal matrix
Element 2	L: lower triangle U: upper triangle
Element 3	N: non-unit matrix U: unit matrix
Element 4	F: one-based indexing C: zero-based indexing

Table 5 lists the possible values of matdescra.

**Table 5** Possible values of **matdescra**
Element 1	Element 2	Element 3	Element 4
G	0 (ignored)	0 (ignored)	F or C
S or H	L	N	F or C
S or H	L	U	F or C
S or H	U	N	F or C
S or H	U	U	F or C
T	L	N	F or C
T	L	U	F or C
T	U	N	F or C
T	U	U	F or C
D	0 (ignored)	N	F or C
D	0 (ignored)	U	F or C

Parent topic: Sparse BLAS Level 2 and Level 3 Functions