HP MLIB User's Guide Vol. 2 7th Ed.

992 HP MLIB User’s Guide
mlib_METIS_NodeND Computes fill reducing orderings of sparse matrices
Name mlib_METIS_NodeND
Computes fill reducing orderings of sparse matrices
Purpose This function computes fill reducing orderings of sparse matrices using the
multilevel nested dissection algorithm. It provides the functionality of the
onmetis program.
Usage VECLIB:
void mlib_METIS_NodeND (int *n, int *xadj, int *adjncy, int *numflag,
int *options, int *perm, int *iperm)
VECLIB8:
void mlib_METIS_NodeND (long long *n, long long *xadj, long long
*adjncy, long long *numflag, long long *options, long long *perm, long
long *iperm)
Arguments n The number of vertices in the graph.
xadj,
adjncy
The adjacency structure of the graph.
numflag Used to indicate which numbering scheme is used for
the adjacency structure of the graph. numflag can take
the following two values:
0—C-Style numbering is assumed that starts from 0
1—Fortran-style numbering is assumed that starts
from 1
options This is an array of 8 integers that is used to pass
parameters for the various phases of the algorithm. If
options[0]=0 then default values are used. If
options[0]=1 then the remaining 7 elements of options
are interpreted as follows:
options[1] Determines matching type. Possible values
are:
1—Random Matching (RM)
2—Heavy-Edge Matching (HEM)
3—Sorted Heavy-Edge Matching (SHEM)(Default)
Experiments have shown that all three matching
schemes perform quite well. In general, SHEM is faster
and RM is slower, but feel free to experiment with the
other matching schemes.