Porting MKL’s Traditional SPARSE BLAS APIs to AOCL-Sparse - 5.2 English - 68552

Table 11.4 shows the mapping between MKL’s traditional Sparse BLAS and AOCL-Sparse’s APIs. Unlike the MKL’s I/E APIs, the traditional APIs work with pointers, without utilizing any abstract data structures. Some of the APIs of interest are listed in the table with the corresponding AOCL-Sparse API. Additionally, Table 11.5 lists the mappings of various types (value of pointers) within MKL’s traditional interface with AOCL-Sparse types. This section describes the porting process with a few illustrations.

Table 11.4 Mapping MKL’s traditional APIs to AOCL-Sparse#

MKL Old	AOCL-Sparse	Comments
mkl_?csrcsc	aoclsparse_convert_csr()
mkl_cspblas_dcsrgemv mkl_?csrmv mkl_?csrgemv mkl_?csrsymv	aoclsparse_?mv()
mkl_?csrtrsv mkl_?csrsv	aoclsparse_?trsv()
mkl_dcsrmultcsr	aoclsparse_sp2m()
mkl_dcsrmm	aoclsparse_?csrmm()	General, symmetric and Hermitian matrices supported
mkl_?csrsm	aoclsparse_?trsm()
mkl_dcsradd	aoclsparse_?add()
mkl_sparse_destroy	aoclsparse_destroy()

Table 11.5 Mapping MKL Parameters to AOCL-Sparse#

MKL		AOCL-Sparse	Comments
transa	N or n	op = aoclsparse_operation_none
transa	T or t	op = aoclsparse_operation_transpose
transa	C or c	op = aoclsparse_operation_conjugate_transpose
matdescra[b]	G	mat_type = aoclsparse_matrix_type_general	b is 0/1 based on zero/one-based indexing
matdescra[b]	S	mat_type = aoclsparse_matrix_type_symmetric
matdescra[b]	H	mat_type = aoclsparse_matrix_type_hermitian
matdescra[b]	T	mat_type = aoclsparse_matrix_type_triangular
matdescra[b+1] *uplo	L	fill = aoclsparse_fill_mode_lower	matdescra[b+1] is S or H or T
matdescra[b+1] *uplo	U	fill = aoclsparse_fill_mode_upper	matdescra[b+1] is S or H or T
matdescra[b+2]	N	diag = aoclsparse_diag_type_non_unit	matdescra[b+1] is S or H or T
matdescra[b+2]	U	diag = aoclsparse_diag_type_unit	matdescra[b+1] is S or H or T
matdescra[b+3]	C	base = aoclsparse_index_base_zero
matdescra[b+3]	F	base = aoclsparse_index_base_one