Created attachment 672144 [details]
contains all routines, compile script, run script; also contains 32 bit executable

Description of problem: results of calculation displayed after redirect to a file are different than the results displayed when stdout goes directly to a terminal window


Version-Release number of selected component (if applicable): 
Fedora 3.6.10-2.fc17.i686, gcc (GCC) 4.7.2 20120921 (Red Hat 4.7.2-2) 


How reproducible: run enclosed script; should also happen if you run
enclosed 32 bit executable


Steps to Reproduce:
1. run script lusolve.bug.run in terminal window
2. look at results on the screen; they should be the same
3. notice that they are not
  
Actual results:
   1:1  7.0016212 +- 0.24727210E-02
   1:2  -0.22314226E-02 +- 0.27870338E-02
   1:4  -0.72836803E-04 +- 0.12898492E-03
   1:5  -0.99971462E-02 +- 0.58309151E-05
   1:6  1.0006572 +- 0.96526108E-03
    average error in rhs   0.52919594E-03
   1:1  7.0016298 +- 0.24756978E-02
   1:2  -0.22461086E-02 +- 0.27903942E-02
   1:4  -0.73604322E-04 +- 0.12914048E-03
   1:5  -0.99971117E-02 +- 0.58379455E-05
   1:6  1.0006628 +- 0.96642534E-03
    average error in rhs   0.52983098E-03


Expected results:
   1:1  7.0016212 +- 0.24727210E-02
   1:2  -0.22314226E-02 +- 0.27870338E-02
   1:4  -0.72836803E-04 +- 0.12898492E-03
   1:5  -0.99971462E-02 +- 0.58309151E-05
   1:6  1.0006572 +- 0.96526108E-03
    average error in rhs   0.52919594E-03
   1:1  7.0016212 +- 0.24727210E-02
   1:2  -0.22314226E-02 +- 0.27870338E-02
   1:4  -0.72836803E-04 +- 0.12898492E-03
   1:5  -0.99971462E-02 +- 0.58309151E-05
   1:6  1.0006572 +- 0.96526108E-03
    average error in rhs   0.52919594E-03


Additional info: does not occur in double precision version of routine 
                 does not occur on x86_64 machine running same version of Fedora and same version of gcc
                 may be a Linux redirect problem rather than a gcc problem (cannot tell; very peculiar bug)