Re: g95 vs. gfortran

[Victor Zhorin <vzhorin at uchicago dot edu>]

I was using 4.2 version and after the update to the most recent version the
problem became less severe. There is still a visible difference in results
under certain test scenarios, but it in percents range rather than simply
obviously wrong.

Yes, the algorithm could be unstable, not sure why though.

Thank you for you help.

On 1/21/09 12:29 PM, "Tobias Burnus" <burnus@net-b.de> wrote:

> Hi Victor,
> 
> Victor Zhorin wrote:
>> I have a simple linear programming subroutine that compiles and executes
>> under both gfortran and g95. However, the results produced are dramatically
>> different with g95 working well and gfortran not working at all.
>>   
> 
> First question: What version of gfortran are you using (gfortran -v)?
> For instance, gfortran 4.0.x had tons of bugs, 4.1.x is mostly usable
> but had still several bugs (esp. for small x). Later versions generally
> work well - though you might be unlucky to hit a bug in gfortran. The
> latest stable version is 4.3.2 (4.3.3 will be released presumably this
> week) and the latest developer version (4.4.0) will also be release soon
> (within two weeks?). Thus: Can you try with a recent 4.4.0 build? Binary
> builds are available at http://gcc.gnu.org/wiki/GFortranBinaries
> 
> If it still doesn't work there can be either of the following problems
> (or a combination of those):
> 
> (a) Bug in the program, which only works by chance with another compiler
> (b) Unstable algorithm
> (c) Bug in gfortran
> 
> It might help if you give us the (least optimizing) options you have
> tried. And using the option -finit-local-zero cannot harm.
> 
> Tobias

-- 
Victor V