What you need to know to use these subprograms
Consider, for example, the 5 x 4 matrix:
Table 4-9 | 5 x 4 Matrix | | | |
| 11 | 12 | 0 | 14 |
| 21 | 22 | 0 | 0 |
| 0 | 0 | 33 | 34 |
| 0 | 0 | 0 | 44 |
| 51 | 52 | 0 | 54 |
This matrix could be represented in DIA format (using ndiag = 6) as:
Table 4-10 DIA Format Matrix
idiag= | -4 | -3 | -1 | 0 | 1 | 3 |
val= | - | - | 21 | 11 | 12 | 14 |
| - | - | 0 | 22 | 0 | - |
| - | 0 | 0 | 33 | 34 | 0 |
| 51 | 52 | 54 | 44 | - | - |
ELL- Ellpack-Itpack.Given a sparse m-by-kmatrix A with maxnz nonzero elements in any row, the ELL format stores the nonzero entries of A row by row. Two arrays are required for the ELL representation:
•val( lda,*) - Two dimensional lda-by-maxnzscalar array where lda is greater or equal to m. The first entries in row val( i, :) consist of nonzero elements in row i of A.
•indx( lda, *) - Two dimensional lda-by-maxnzinteger array where row indx( i, :) stores column indices for row i of A: indx( i, j) corresponds to the column index of val( i, j). If a row has t nonzero elements with t less than maxnz, then indx( i, t+1) is set to a negative value.
Chapter 4 Sparse BLAS Operations 429