[Help-glpk] [Fwd: Re: [Fwd: Problem with GLPK when trying to find value functions]]

[Andrew Makhorin]

-------- Forwarded Message --------
From: Sergey Kuznetsov <address@hidden>
To: Robbie Morrison <address@hidden>
Cc: GLPK help <address@hidden>
Subject: Re: [Help-glpk] [Fwd: Problem with GLPK when trying to find
value functions]
Date: Thu, 26 Jul 2012 17:07:33 +0000

Hello Robbie,


I ran my binary under valgrind, here is summary:
==12805== HEAP SUMMARY:
==12805==     in use at exit: 0 bytes in 0 blocks
==12805==   total heap usage: 30 allocs, 30 frees, 1,971 bytes allocated
==12805== 
==12805== All heap blocks were freed -- no leaks are possible
==12805== 
==12805== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 11 from
6)
--12805-- 
--12805-- used_suppression:     11 U1004-ARM-_dl_relocate_object
==12805== 
==12805== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 11 from
6)


and still the same problem as before running under valgrind:
*** glibc detected *** ./test: free(): invalid pointer: 0x0828f350 ***
======= Backtrace: =========
/lib/i386-linux-gnu/i686/cmov/libc.so.6(+0x6e3f1)[0xb74773f1]
/lib/i386-linux-gnu/i686/cmov/libc.so.6(+0x6fc58)[0xb7478c58]
/lib/i386-linux-gnu/i686/cmov/libc.so.6(cfree+0x6d)[0xb747bd0d]
/usr/lib/libglpk.so.0(glp_free+0x141)[0xb76ed6b1]
======= Memory map: ========
....


>One other thought: GLPK is not officially thread safe:
>do you call GLPK concurrently?
At first, I ran GLPK in omp parallel for, it didn't enough memory for
big instances, that's why I'm using in a loop:
for each right-hand side{
   call GLPK;
   store optimal value;  
}
Thanks,


Sergey
 

On Thu, Jul 26, 2012 at 4:41 PM, Robbie Morrison <address@hidden>
wrote:
        
        Hello Sergey
        
        ------------------------------------------------------------
        To:           address@hidden
        Subject:      Problem with GLPK when trying to find value
        functions
        
        From:         Sergey Kuznetsov <address@hidden>
        Date:         Wed, 25 Jul 2012 19:19:39 +0000
        
        ------------------------------------------------------------
        
        > Hello,
        >
        > I'm using GLPK to find value functions, it's a family
        > of linear programs with right-hand side as parameter.
        >
        > For instance, I'm constructing 3D integer lattice and
        > at each site (point) I need to find optimal solution
        > using site coordinates as right-hand side (in this case
        > it's 3D vector).
        >
        > I wrote a function that solves the linear program using
        > GLPK. At first it crashed on the second call. I
        > replaced glp_delete_prob() on gpl_erase_prob() and it
        > worked well for [0;3]^3 lattice (and [0;3]^2
        > lattice). The problems began [0;4]^2 (2D) lattice. I
        > have segmentation fault and it's unusual because in
        > [0;3]^3 I'm solving more programs.
        >
        > OS : Linux Debian, i686;  Compiler: GCC  4.7.1
        > Test instances: 3D: 4 columns, 3 rows; 2D: 3 columns, 2 rows;
         all the
        > entries (matrix, objective function coeff, right-hand side)
        are
        > integers.
        >
        > Clearly, the problems begin when I'm setting span of
        > the lattice to 4 (5, 6, etc.).
        >
        > If anybody has any idea how to fix this, please help
        > me.  Thanks
        >
        > Sergey Kuznetsov
        
        
        Run your program under 'valgrind':
        
          http://en.wikipedia.org/wiki/Valgrind
          http://valgrind.org
        
        It is dead easy to do so:
        
          $ valgrind [valgrind-options] myprog [my-options]
        
        'valgrind' should pick up initialized variables and out
        of bounds array access.
        
        If not already on your system, 'valgrind' is available
        via debian package 'valgrind' (or my system).
        
        One other thought: GLPK is not officially thread safe:
        do you call GLPK concurrently?
        
        HTH, Robbie
        ---
        Robbie Morrison
        PhD student -- policy-oriented energy system simulation
        Technical University of Berlin (TU-Berlin), Germany
        University email (redirected) : address@hidden
        Webmail (preferred)           : address@hidden
        [from Webmail client]