Optimizer Problem

Stefanos Folias sf@math.bu.edu
Tue Jul 4 21:53:00 GMT 2006


Yesterday I reported a problem I was having with optimizers on  
gfortran.  The bug seems to generate different output with optimizers  
on, but only if the array size is sufficiently large. The problem  
persisted through a previous update(march?).  I have updated again  
(yesterday) to the latest version

gcc version 4.2.0 20060512 (experimental)

and the problem still occurs.  I obtain the same output for -O, -O2, - 
O3.  In the cases of  no optimization and --fast-math I find the same  
output, which differs from -O, if one dimension of the array  
increases in length through a critical value.

Are there other optimizers that I could try?  --fast-math didn't  
improve the speed very much, especially compared to -O (4-5 fold).

It will take me a little time to find a suitable reduced test case,  
especially since, at one point, I realized the discrepancy could not  
only be generated by different array sizes, but also on the inclusion  
of an irrelevant variable declaration!!!

In the case that the bug is my problem, what's the best way to  
determine this.  I occasionally use an older ftnchk, disregarding  
obsolete warnings.  What might be more appropriate?

Thanks

Stefanos



More information about the Fortran mailing list