1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
|
README
************************************************************************
* This MATLAB interface to MUMPS is part of the MUMPS package *
* (see ../LICENSE for the conditions of use) *
* Up-to-date copies of MUMPS can be obtained from the web *
* page http://mumps-solver.org *
* *
* THIS MATERIAL IS PROVIDED AS IS, WITH ABSOLUTELY NO WARRANTY *
* EXPRESSED OR IMPLIED. ANY USE IS AT YOUR OWN RISK. *
* *
* More information is available in the main MUMPS userguide and in: *
* *
* [2006] Aurelia Fevre, Jean-Yves L'Excellent and Stephane Pralet *
* MATLAB and Scilab interfaces to MUMPS. LIP Report RR2006-06. *
* Also available as INRIA and ENSEEIHT-IRIT Technical Reports. *
* *
************************************************************************
************************************************************************
COMPATIBILITY WITH OCTAVE:
Thanks to the Octave MEX compatibility, it is pretty straightforward
to generate an Octave interface for MUMPS. Please refer to the comments
inside the make.inc file for instructions on how to do it. Everything
said below applies for both cases where a MATLAB or an Octave interface
is needed.
Thanks to Carlo De Falco from "Politecnico di Milano" for support
provided on the usage of Octave.
************************************************************************
CONTENT OF DIRECTORY:
README
Makefile
make.inc
initmumps.m
mumps.m
other *.m files: examples of usage
mumpsmex.c : MATLAB CMEX-file to let you use sequential MUMPS
in double precision from MATLAB.
USAGE:
see example below and MUMPS documentation
INSTALLATION:
You need
1-
to have compiled/linked a sequential version of MUMPS with both double
precision and double complex arithmetics ("make d" and "make z",
or "make all"). The code must be position-independent (with gfortran,
please add the option -fPIC in both FC, CC, and FL of the main
Makefile.inc). Note that this also applies to other external
libraries, such as METIS, SCOTCH, BLAS, etc.
2-
to specify an adequate BLAS library.
Unless you compile with the option -DINTSIZE64, MUMPS expects 32-bit
integers by default (see the MUMPS_INT datatype in mumps_c_types.h).
In that case, you should specify a BLAS library relying on 32-bit
integers. Otherwise, an error at execution time is likely to occur
(e.g., "segmentation fault in idamax"). If you use a shared library,
make sure that Matlab will not override your default BLAS library
(One way to do that is to issue LD_PRELOAD=my_blas_library.so matlab
instead matlab)
2-
to edit make.inc.
Modify paths for orderings and BLAS. You should also
give the path to the runtime libraries of your FORTRAN 90
compiler. Some commented examples are provided.
You can use something like
nm -o /opt/intel/compiler80/lib/*.a | grep <name of missing compiler symbol>
to help finding which libraries should be added
3-
to run the "make" command
4- We advise you to run the 4 examples
simple_example.m, multiplerhs_example.m, sparserhs_example.m and
schur_example.m
and to check that everything runs smoothly.
******************************************************************************
LIMITATION:
This interface enables you to call MUMPS from MATLAB only
in sequential for double precision and double complex versions.
For example it does not support:
- other versions (single precision arithmetic, parallel version...)
- elemental format for the input matrix
******************************************************************************
%Example of using MUMPS in matlab
% initialization of a matlab MUMPS structure
id = initmumps;
% here JOB = -1, the call to MUMPS will initialize C and fortran MUMPS structure
id = dmumps(id);
% load a sparse matrix
load lhr01;
mat = Problem.A;
% JOB = 6 means analysis+factorization+solve
id.JOB = 6;
id.ICNTL(6) = 0;
% we set the rigth hand side
id.RHS = ones(size(mat,1),1);
%call to mumps
id = dmumps(id,mat);
% we see that there is a memory problem in INFO(1) and INFO(2)
id.INFOG(1)
id.INFOG(2)
% we activate the numerical maximum transversal
id.ICNTL(6) = 6;
id = dmumps(id,mat);
norm(mat*id.SOL - ones(size(mat,1),1),'inf')
% solution OK
% destroy mumps instance
id.JOB = -2;
id = dmumps(id)
|
