This is the mail archive of the fortran@gcc.gnu.org mailing list for the GNU Fortran project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

RE: -ffast-math and gfortran


> HP floating-point guide for HP Integrity servers
> http://h21007.www2.hp.com/dspp/tech/tech_TechDocumentDetailPage_IDX/1,1701,8359,00.html

I cannot access this page currently, because my X Window implementation
(Debian/testing) doesn't work at the moment; I'll look into it as soon
as that works again, because:

> Very good web page on floating point from HP. Includes a flow diagram on how
> to decide what floating point options to use for a given problem.

Yep, that would be my ultimate goal.  However, that's more like a summer
project (I can normally get 6 weeks off during the summer season).

> It has the best explanation of the cx-limited-range option I've seen. It
> really shows what the problems are. If you really want to use this option,
> the compiler needs a #pragma to shut it off for some parts of the code.

If you cannot be sure that the divisors of complex division do not always
have a norm less than the square root of FLT_MAX and more than the square
root of FLT_MIN, then yes, you have to be careful.

> It would be nice to have a flow diagram like this in the gfortran
> documentation.

I agree.  What's missing is a guide for the user that details what
s/he has to know about his/her program to be sure that
-funsafe-math-optimizations won't generate surprises.

Cheers,

-- 
Toon Moene - e-mail: toon@moene.indiv.nluug.nl - phone: +31 346 214290
Saturnushof 14, 3738 XG  Maartensdijk, The Netherlands
A maintainer of GNU Fortran 95: http://gcc.gnu.org/fortran/
Gfortran At Work: http://mysite.verizon.net/serveall/moene.pdf


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]