You don't want to use that option, which corresponds to /Qtrapuv (under Data > Initialize stack variables to an unusual value). It does not do anything useful, but if you want to turn it on in Debug, that's where. It certainly does not set anything to a NaN, despite the name in the project file. (The original description of that option said it set variables to NaN, but that wasn't the implementation.)

I suggest instead you set Floating Point > Floating-Point Exception Handling to "Underflow gives 0.0; trap on other IEEE exceptions (/fpe0)". This will most likely cause the program to stop when it would otherwise produce a "floating invalid", giving you a better chance of seeing where it comes from.

Most likely, though, you have some other error that is revealed by optimization, such as use of an uninitialized variable. But give that a try.

Steve,

I was able to cause the crashto appear and disappear in either debug or release mode by turning /Qtrapuvon and off. This made me even more confused and made me think there was some un-initialized variable. The change you suggested in the settingsfor the underflow handling alsocaused the debug versionto crash. Still, I was not able to identify the specificline of code where theprogram was blowing up.

I figured out the problem using CVF. An error in the logic was causing an exponential overflow. I still do not understand why /Qtrapuv affected the behavior of the program (it should have crashed with or without it). It is also disappointing that the latest Intel compiler +MS debugger never gave me the actual instruction where the program was blowing up.

Thanks,

Gabriel

I'd like to see the program in the mode where it was failing. I think the traceback did show you the point of error, but it was inside a math library routine. Setting /Qtrapuv does change the code generator behavior somewhat, so it may have helped show the error.

I believe if you look at this line:

HAZ_Table.exe 0102E505 _CALC_HAZ_CAV 361 HAZ_Table.f

361 in HAZ_Table.f

you will find it calling an EXP function with an invalid number or similar.