1gmap(1) Scotch user's manual gmap(1)
2
6 gmap, gpart - compute static mappings and partitions sequentially
7
9 gmap [options] [gfile] [tfile] [mfile] [lfile]
10 gpart [options] [nparts/pwght] [gfile] [mfile] [lfile]
12
15 The gmap program computes, in a sequential way, a static mapping of a
16 source graph onto a target graph.
17 The gpart program is a simplified interface to gmap, which performs
19 graph partitioning instead of static mapping. Consequently, the desired
20 number of parts has to be provided, in lieu of the target architecture.
21 When using the program for graph clustering, the number of parts turns
22 into maximum cluster weight.
23 The -b and -c options allow the user to set preferences on the behavior
25 of the mapping strategy which is used by default. The -m option allows
26 the user to define a custom mapping strategy.
27 The -q option turns the programs into graph clustering programs. In
29 this case, gmap only accepts variable-sized target architectures.
30 Source graph file gfile can only be a centralized graph file. For gmap,
32 the target architecture file tfile describes either algorithmically-
33 coded topologies such as meshes and hypercubes, or decomposition-de‐
34 fined architectures created by means of the amk_grf(1) program. The re‐
35 sulting mapping is stored in file mfile. Eventual logging information
36 (such as the one produced by option -v) is sent to file lfile. When
37 file names are not specified, data is read from standard input and
38 written to standard output. Standard streams can also be explicitely
39 represented by a dash '-'.
40 When the proper libraries have been included at compile time, gmap and
42 gpart can directly handle compressed graphs, both as input and output.
43 A stream is treated as compressed whenever its name is postfixed with a
44 compressed file extension, such as in 'brol.grf.bz2' or '-.gz'. The
45 compression formats which can be supported are the bzip2 format
46 ('.bz2'), the gzip format ('.gz'), and the lzma format ('.lzma').
47
49 -bval Set maximum load imbalance ratio for graph partitioning or
50 static mapping. When programs are used as clustering tools, this
51 parameter sets the maximum load imbalance ratio for recursive
52 bipartitions. Exclusive with the -m option.
53 -copt Choose default mapping strategy according to one or several op‐
55 tions among:
56 b enforce load balance as much as possible.
58 q privilege quality over speed (default).
60 s privilege speed over quality.
62 t enforce safety.
64 It is exclusive with the -m option.
66 -h Display some help.
68 -mstrat
70 Use sequential mapping strategy strat (see Scotch user's manual
71 for more information).
72 -q (for gpart)
74 -qpwght
76 (for gmap) Use the programs as graph clustering tools instead of
77 static mapping or graph partitioning tools. For gpart, the num‐
78 ber of parts will become the maximum cluster weight. For gmap,
79 this number pwght has to be passed after the option.
80 -V Display program version and copyright.
82 -vverb Set verbose mode to verb. It is a set of one of more characters
84 which can be:
85 m mapping information.
87 s strategy information.
89 t timing information.
91
93 Target architectures represent graphs onto which source graphs are
94 mapped. In order to speed-up the obtainment of target architecture
95 topological properties during the computation of mappings, some classi‐
96 cal topologies are algorithmically coded into the mapper itself. These
97 topologies are consequently simply defined by their code name, followed
98 by their dimensional parameters:
99 cmplt dim
101 unweighted complete graph of size dim.
102 cmpltw dim w0 w1 ... wdim-1
104 weighted complete graph of size size and of respective loads w0,
105 w1, ..., wdim-1.
106 hcub dim
108 hypercube of dimension dim.
109 leaf hgt n0 w0 ... nhgt-1 whgt-1
111 tree-leaf graph of height hgt with (n0 times n1 times ...
112 nhgt-1) vertices, with inter-cluster link weights of w0, w1, ...
113 whgt-1.
114 mesh2D dimX dimY
116 2D mesh of dimX times dimY nodes.
117 mesh3D dimX dimY dimZ
119 23 mesh of dimX times dimY times dimZ nodes.
120 torus2D dimX dimY
122 2D torus of dimX times dimY nodes.
123 torus3D dimX dimY dimZ
125 3D torus of dimX times dimY times dimZ nodes.
126 Other target topologies can be created from their source graph descrip‐
128 tion by using the amk_grf(1) command. In this case, the target descrip‐
129 tion will begin with the code name deco.
130
132 Mappings are represented by as many lines as there are vertices in the
133 source graph. Each of these lines is made of two figures: the number of
134 the vertex (or its label if source graph vertices are labeled) and the
135 index of the target vertex to which it has been assigned. Target vertex
136 indices range from 0 to the number of vertices in the target architec‐
137 ture (that is, the number of parts) minus one.
138 This block of lines is always preceded by the number of such lines. In
140 most cases, since full mappings are requested, the number of lines is
141 equal to the number of vertices in the source graph.
142
144 Run gpart to compute a partition into 7 parts of graph 'brol.grf' and
145 save the resulting ordering to file 'brol.map'.
146 $ gpart 7 brol.grf brol.map
148 Run gmap to compute a partition, into 3 parts of respective weights 1,
150 2 and 4, of graph 'brol.grf' and save the resulting mapping to file
151 'brol.map'. The dash '-' standard file name is used so that the target
152 architecture description is read from the standard input, through the
153 pipe, as provided by the 'echo' shell command.
154 $ echo "cmpltw 3 1 2 4" | gmap brol.grf - brol.map
156
159 amk_grf(1), acpl(1), gmtst(1), dgmap(1).
160 Scotch user's manual.
162
164 Francois Pellegrini <francois.pellegrini@labri.fr>
165 23 November 2019 gmap(1)