SDPA-C 7.3.X Install Manual

[Copyright (C) 2004-2013 SDPA Project]

This text explains how to install SDPA-C.
If you have questions on the installation,
please contact us from
http://sdpa.sourceforge.net/contact.html

The user-manual of SDPA is available for SDPA-C, and can be downloaded at
https://sourceforge.net/projects/sdpa/files/sdpa/sdpa.7.1.1.manual.20080618.pdf/download

--- index ------------------------------------

1. Install the SDPA-C
2. Compile callable-library example  [optional]
3. Compile Matlab Interface          [optional]
4. Difference from SDPA
5. If you have some trouble

----------------------------------------------


1. Install the SDPA-C

(1) Check your environment

Requirement: wget, git
Recommendation: gcc (>= 4.2), g++ (>= 4.2), gfortran (>=4.2)

We have checked the compilation on RedHat Linux (version 6)
and Debian (wheezy).
Other C/C++/Fortran compilers than gcc/g++/gfortran may be
available, but we have not tested them.

(2) Download the SDPA from the SDPA Homepage

http://sdpa.sourceforge.net/download.html

You can get the sdpa-c_7.3.X.tar.gz, then unpack the file
$ tar xzf sdpa-c_7.3.X.tar.gz
You will find the directory 'sdpa-c'.
$ cd sdpa-c

(3) Edit make.inc

At least you should check 'sdpac_dir'.
In the default setting, sdpac_dir is $(HOME)/sdpa-c.
In other words, the default setting assumes
you unpacked the downloaded file in your home directory.

(4) Compile SDPA-C

$ make all
will compile SuiteSparse, MUMPS, OpenBLAS, then SDPA-C.

If you want compile them step by step,
$ make SuiteSparse
$ make MUMPS
$ make OpenBLAS
$ make sdpa-c


(5) Test SDPA-C
Try the following command to execute sdpa
   $ cd $HOME/sdpa-c
   $ ./sdpa-c
Then you will see message from SDPA and can check command line options.
To solve input dat-s and write its result,
   $ ./sdpa example1.dat-s example1.result

If you specify the number of threads with the SDPA-C,
set OMP_NUM_THREADS environment variable(ex. export OMP_NUM_THREADS=4).

----------------------------------------------    
2. Compile callable-library example (optional)

To compile callable-library example, use 'make' command 
in 'libexample' directory
   $ cd $HOME/sdpa-c/libexample
   $ make

The details of callable-library can be found in SDPA user manual.

----------------------------------------------    
3. Compile Matlab Interface (optional)

To compile the Matlab interface, use 'make matlab' command.
   $ cd $HOME/sdpa-c
   $ make matlab

If you use Octave instead of Matlab, then
   $ cd $HOME/sdpa-c
   $ make octave

Since the usage of this Matlab Interface as SDPA-M 6.2.0,
SDPA-M 6.2.0 manual is very useful.
(https://sourceforge.net/projects/sdpa/files/sdpa-m/sdpamManual.pdf/download)
In addition, the file 'mex/CommandList.txt' summarizes
each command.

	
When you have trouble with SDPA-M(Matlab environment),
the following command will display useful messages.

$ matlab -Dgdb
[On gdb environment]
(gdb) run -nodisplay
[On Matlab environment]
>> (Call SDPA-M by sdpam or sedumiwrap)
If Matlab aborts, try the following command in gdb environment.
(gdb) where
The command 'where' will display what happened.
To exit gdb environment,
(gdb) quit

In some case, the segmentation fault may be avoided by
$ export LD_PRELOAD=/usr/lib/gcc/x86_64-linux-gnu/4.4/libgfortran.so:$LD_PRELOAD
$ export LD_PRELOAD=/usr/lib/gcc/x86_64-linux-gnu/4.4/libstdc++.so:$LD_PRELOAD
$ export LD_PRELOAD=/usr/lib/gcc/x86_64-linux-gnu/4.4/libgcc_s.so:$LD_PRELOAD
$ matlab
In particular, when the gcc version is higher than the version Matlab
assumes, Matlab sometimes can not load appropriate shared libraries.
The command above load the libraries before Matlab launches.

-------------------------------------------------------
4. Deference from SDPA

(1) Since SDPA-C employs the completion method, which extensively
uses the structural sparsity, the SDPA dense format is not acceptable.
More precisely, example1.dat-s is available while example1.dat is not.

(2) In the callable library, some function mainly related to the dense
input file are not available.
In addition, the function SDPA::terminate() is renamed as
SDPA::finalize().
In libexample directory, the difference in example1,2,5.cpp from SDPA
is only terminate() to finalize().
However, the example3,4,6.cpp in SDPA contained the functions
which are not available in SDPA-C. We removed these examples
from the SDPA-C package.

(3) param.sdpa is renamed as param.sdpaC.

(4) The matlab interface only contains the mex functions of SDPA-C.
If you want to read the SDPA sparse format, you should employ
such routines from SDPA. We suggest you add the paths of both SDPA
and SDPA-C to the Matlab search path.

-------------------------------------------------------
5. If you have some trouble

If some bugs are found or you encounter some trouble
when you are compiling SDPA, please let us know from
http://sdpa.sourceforge.net/contact.html

In particular, attaching error messages and/or 'config.log' file
is very useful to let us understand your situation.