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: Optimisation prevents overflow?


On Thu, Jul 19, 2007 at 11:43:14PM +0200, Davide Mancusi wrote:
> Daniel Franke ha scritto:
> > I'd guess, that's the (in)famous extended precision, 80 vs. 64 bits (iirc). 
> 
> I'm going from memory here and I might be wrong, but wasn't this 
> something that concerned (affected?!) only x86 processors? I'm asking 
> because my OP stemmed from an incosistent behaviour of g77 on a 
> Sun/SPARC machine and a Linux/x86 machine, which I suspect might be 
> related to the minimal example I posted here.
> 
> Is there any way to turn off this extended precision in g77 (or gfortran)?

This does only affect x86 processors, as far as I know.  And going from
18 to 273 is an unlikely result for your program even with the extra
precision.

The -ffloat-store is the traditional way of turning it off, and amounts
to a sort of overkill but works.  Also, reportedly the "-mfpmath=sse"
reportedly works for this, but I haven't tried it.

(To GFortran people: This was discussed in the floating-point math BOF
at the Summit yesterday.  Geert Bosch says it works fine; a few people
were slightly dubious.)

- Brooks


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