XPRSmipoptimize - received signal SIGFPE, Arithmetic exception.

I would like to add that the version of XPRESS that I am running was installed from xp9.0.2_linux_x86_64_setup.tar

Just to clariy: the problem only occurs when you solve from your program? If you solve the exported model file with the console optimizer then there is no exception?

Can you share the exported model file? If you don't want to share it here, you may send it to danieljunglas(at)fico(dot)com. If the variable/constraint names contain sensitive information, you may scramble these names by passing option `s` to XPRSwriteprob().

The problem occurs only from the program, IVE on Windows seems to solve just fine. The console optimizer on the same machine does however not run because of a missing library libncurses.so.5 (libncurses.so.6 is installed). This is a recent XPRESS upgrade as 8.5 was previously installed.

$ file /opt/xpressmp/bin/optimizer
 
/opt/xpressmp/bin/optimizer: ELF 64-bit LSB executable, x86-64, version 1 (SYSV), dynamically linked, interpreter /lib64/ld-linux-x86-64.so.2, for GNU/Linux 2.6.32, BuildID[sha1]=cbe6d2c2dd85a0ed171495954b8cec9b1f20c54a, stripped

And this is the beginning of xphostid

$ /opt/xpressmp/bin/xphostid
 
== FICO Xpress host ID tool ==
 
== (c) Fair Isaac Corporation 2005-2022. All rights reserved ==
 
When applying for a license you must send the entire text printed below to your vendor:
 
 
 
<xphostid version="3.2.5" xpress="9.0" build="linux_x86_64">

I attached the relevant matrix file.

Thank you very much.

Also, in my original message, when I said:

"XPRSloadmip, XPRSchgobjsense and XPRSmipoptimize return 0, so no error is detected in those steps."

I meant:

XPRSloadmip, XPRSchgobjsense and XPRSwriteprob return 0, so no error is detected in those steps.

XPRSmipoptimize never actually returns anything as the function call does not complete and is interrupted by the exception mentioned in my message.

Well, I linked the missing .so's, so that it references the .so.6 when attempting to load the .so.5 both for libncurses.so.5 and libtinfo.so.5 and now the optimizer command works.

This is the output:

[xpress jpsierra] readprob /tmp/problema_antes_mipoptimize.mps.lp
 
Reading Problem problema_antes_mipoptimize.mps
Problem Statistics
          32 (      0 spare) rows
         158 (      0 spare) structural columns
         242 (      0 spare) non-zero elements
MIP Entity Statistics
           5 entities        0 sets        0 set members
[xpress jpsierra] mipoptimize
FICO Xpress v9.0.2, Hyper, solve started 9:03:26, Mar 30, 2023
Heap usage: 409KB (peak 601KB, 130KB system)
Maximizing MILP problema_antes_mipoptimize.mps using 1 thread and up to 125GB memory, with default controls
Original problem has:
        32 rows          158 cols          242 elements         5 entities
Presolved problem has:
        25 rows           85 cols          145 elements         5 entities
LP relaxation tightened
Presolve finished in 0 seconds
Heap usage: 461KB (peak 601KB, 130KB system)
 
Coefficient range                    original                 solved
  Coefficients   [min,max] : [ 1.34e-02,  1.43e+02] / [ 6.25e-02,  2.00e+00]
  RHS and bounds [min,max] : [ 4.37e-10,  8.21e+07] / [ 1.66e-09,  4.12e+02]
  Objective      [min,max] : [ 1.00e-02,  4.36e+05] / [ 3.20e+00,  1.45e+06]
Autoscaling applied standard scaling
 
Will try to keep branch and bound tree memory usage below 118.9GB
 
   Its         Obj Value      S   Ninf  Nneg   Sum Dual Inf  Time
     0       1240117.979      D     11     0        .000000     0
    25      -4868744.855      D      0     0        .000000     0
Optimal solution found
Dual solved problem
  25 simplex iterations in 0.00 seconds at time 0
 
Final objective                       : -4.868744854550569e+06
  Max primal violation      (abs/rel) :       0.0 /       0.0
  Max dual violation        (abs/rel) :       0.0 /       0.0
  Max complementarity viol. (abs/rel) :       0.0 /       0.0
 
Starting root cutting & heuristics
 
 Its Type    BestSoln    BestBound   Sols    Add    Del     Gap     GInf   Time
c        -4951199.318 -4868744.855      1                  1.67%       0      0
d        -4947818.410 -4868744.855      2                  1.60%       0      0
   1  K  -4947818.410 -4946084.252      2     27      0    0.04%       5      0
 *** Search completed ***
Uncrunching matrix
Final MIP objective                   : -4.947818410385440e+06
Final MIP bound                       : -4.947813541640585e+06
  Solution time / primaldual integral :      0.00s/ 68.381709%
  Number of solutions found / nodes   :         2 /         1
  Max primal violation      (abs/rel) : 1.990e-13 / 2.842e-14
  Max integer violation     (abs    ) :       0.0

By default the console optimizer runs with floating point exceptions disabled, this is probably why you don't get an exception there.

Could you try to explicitly disable floating point exceptions in your Pascal program? You would have to do something equivalent to feenableexcept(), see for example here.

If this fixes the problem, you could try to figure out which of the exceptions is thrown (inexact, invalid, overflow, underflow, divbyzero) and then potentially only disable that exception.

So far I have only been able to produce an "inexact" exception here with your model and that exception is expected in certain places.

I'm trying to do what you suggested, I'll keep you updated. Thank you.

With this line it still crashes:

SetExceptionMask(GetExceptionMask + [exZeroDivide]);

But with this one, it does not:

SetExceptionMask(GetExceptionMask + [exZeroDivide,exInvalidOp]);

So apparently it's the invalidOp that's causing the problem.

Took the code from this page

I managed to reproduce the "invalid operation" exception here. We are investigating. For the moment, the only workaround I see is to mask that exception :-(

I found the problem. It was an invalid floating point speculation. This will be fixed in the next release. Until then it is safe to ignore the exception.

Thank you very much!