Re: Optimizer Problem

[langton2 at llnl dot gov]

Quoting Stefanos Folias <sf@math.bu.edu>:
> I have been using gfortran on Mac running Tiger.  I have been wrestling
> with a horrible discrepancy between the use of optimizers and not.  I
> have tried omitting certain flags to determine what may be causing the
> problem to no avail.  The discrepancy is a function of one of the
> dimension of the array, i.e., there is a threshold length where one
> type of result occurs below and another occurs above.  However, the
> result shouldn't at all be dependent on the length of the array.  With
> no optimizers, the same result occurs, and matches the result as
> generated by f95 on a server running unix.  Thus, the optimizers have
> an inherent problem.

Perhaps, but optimization sometimes reveals bugs that otherwise would  
have gone undetected.  In any case, could you post the code with which  
you are having problems (preferably a reduced testcase)?  Also, what  
version of gfortran are you using?  Also what optimization flags are  
you using?  (If one of them is --fast-math, then that could be your  
problem . . .)


-Asher