This page summarizes options relative to graph build for graph partitioning with ParMetis, Scotch, sparse matrix ordering with ParMetis, Scotch, and graph coloring.
|CHECK_GRAPH||Level of error checking for graph input: 0 = no checking, 1 = on-processor checking, 2 = full checking. (CHECK_GRAPH==2 is very slow and should be used only during debugging).|
|GRAPH_SYMMETRIZE||How to symmetrize the graph:
NONE = graph is symmetric and no symmetrization is needed
TRANSPOSE = if M is adjacency matrix of the input graph, output will be the graph representation of M+MT
BIPARTITE = graph is symmetrized in a bipartite way : [[ 0 M ][Mt 0]]
|GRAPH_SYM_WEIGHT||How edge weights are handled during symmetrization:
ADD = weights of each arc are added
MAX = only the heaviest arc weight is kept
|GRAPH_BUILD_TYPE||Type of input, do allow some optimizations in the build process:
NORMAL = graph is generic, no optimization can be performed
FAST = graph global IDs are in the interval [0,n-1], with IDs [0,a] on process 0, IDs [a+1, b] on process 1, IDs [b+1, c] on process 2, etc.
FAST_NO_DUP = graph global IDs are in the interval [0,n-1] with IDs [0,a] on process 0, IDs [a+1, b] on process 1, IDs [b+1, c] on process 2, etc., and there are no duplicate edges and no need of symmetrization.
See GRAPH_FAST_BUILD_BASE below to allow IDs to that are one-based instead of zero-based.
|GRAPH_FAST_BUILD_BASE||When using GRAPH_BUILD_TYPE is FAST or FAST_NO_DUP, IDs specified are in the range [GRAPH_FAST_BUILD_BASE, n-1+GRAPH_FAST_BUILD_BASE]. This parameter has no effect when GRAPH_BUILD_TYPE is NORMAL.|
|CHECK_GRAPH = 1|
|GRAPH_SYMMETRIZE = NONE|
|GRAPH_SYM_WEIGHT = ADD|
|GRAPH_BUILD_TYPE = NORMAL|
|GRAPH_FAST_BUILD_BASE = 0|
|Required Query Functions:|
ZOLTAN_EDGE_LIST_MULTI_F N or ZOLTAN_EDGE_LIST_FN