Description

There is a bug in the Intel 9.1.XXX compilers (e.g. 9.1.043) that
causes some floating point computations to be incorrect - not by a
little, but just plain wrong.  This causes problems with many models.
This example is taken from a model sent by Michael Ferris that was our
first reported instance of this problem.

Contributor: Steve Dirkse

Small Model of Type : GAMS

Category : GAMS Test library

Main file : compile2.gms

$title 'test for data errors caused by C compiler bug' (COMPILE2,SEQ=289)

$onText
There is a bug in the Intel 9.1.XXX compilers (e.g. 9.1.043) that
causes some floating point computations to be incorrect - not by a
little, but just plain wrong.  This causes problems with many models.
This example is taken from a model sent by Michael Ferris that was our
first reported instance of this problem.

Contributor: Steve Dirkse
$offText

sets
i        /o1*o10/
j        /j1*j6/
k        /k1*k3/;

table JK(j,k)
         k1      k2      k3
j1       1
j2       1
j3               1
j4               1
j5                       1
j6                       1;

table
Bmax(i,j)
         j1      j2      j3      j4      j5      j6
o1*o10   50      30      40      35      30      45
;

*forbidden assignment
parameter FA(i,j);
FA(i,j) = no;
FA('o1','j2') = yes;
FA('o3','j6') = yes;


parameter lmin(i);
lmin(i) = smin(k,smax(j$(JK(j,k) and (not FA(i,j))),Bmax(i,j)));

execute_unload 'lmin', lmin;

if {(lmin('o1') = 40),
  display "OK - this is working fine";
elseif (lmin('o1') = 0),
  abort$1 "expected error in lmin: check compiler!";
else
  abort$1 "unexpected error in lmin - what's going on?";
};