What you need to know to use these subprograms

This matrix could be represented in BCO format using mb=2, kb=3, and lb=2.

Table 4-17 BCO Format Matrix

bindx=

1

1

2

2

bjndx=

1

3

2

3

val(1:2, 1:2, 1)=

11

12

 

 

 

21

22

 

 

val(1:2, 1:2, 2)=

15

16

 

 

 

25

26

 

 

val(1:2, 1:2, 3)=

33

0

 

 

 

43

44

 

 

val(1:2, 1:2, 4)=

35

36

 

 

 

0

46

 

 

BSC - Block compressed sparse column. Given a sparse block matrix A formed by mb-by-kbsquare blocks of size lb-by-lbeach, the block compressed sparse column format represents the bnnz nonzero block entries using the same variables as in the CSC format. Each nonzero dense block is stored in column major order. Four arrays are required for the BSC representation:

•val( lb, lb, *) - Scalar matrix of dimension lb-by-lb-by-maxnnzcontaining the nonzero lb-by-lbblocks, where maxnnz = max(bnnz, bpntre(k)-1).

•bindx(*) - Integer array of length maxnnz containing block column indices such that bindx(i) corresponds to the block column index of val(:, :, i).

•bpntrb(*) - Integer array of length kb such that bpntrb(j) points to location val(:, :, j) of the first nonzero block in block column j.

•bpntre(*) - Integer array of length kb such that pntre(j)-1points to location val(:, :, j) of the last nonzero block in block column j.

Chapter 4 Sparse BLAS Operations 433