Different results between compaq and intel fortran

[tigergrace]

Hello,

Recently, I convert my code fromcompaq fortran to intel fortran, after convertion, i found some calculate results are different from previous version(compaq), does someone know there are some conifiguration in inter fortran? Just like the single precision or something else.

Thanks in advance!

[anthonyrichards]

Quoting - tigergrace

Hello,

Recently, I convert my code fromcompaq fortran to intel fortran, after convertion, i found some calculate results are different from previous version(compaq), does someone know there are some conifiguration in inter fortran? Just like the single precision or something else.

Thanks in advance!

How 'different' are the results?
I recently helped a poster who fell foul of the difference between the CVF and IVF default behaviours with local variables in subprograms called more than once. The CVF default is to automatically SAVE local variables, whereas IVF does not, so if you rely on local variables keeping their values between calls and think that this might becontributing to yourproblem, add SAVE to your subprograms and recompile. Local variables will then be saved and at least this possible source of different behaviours will vanish. Otherwise, define your locals in a module and USE the module.

[TimP]

In addition to default SAVE behavior, CVF had double precision evaluation of single precision expressions. ifort has some similar options, such as /arch:ia32 (for the ia32 compiler only).
Also, you would need options to obey parentheses in both compilers; /fltconsistency for CVF, /fp:source or /assume:protect_parens for ifort. /fp:source would disable some optimizations which might produce different numerical results from CVF.