This is the mail archive of the
libc-alpha@sources.redhat.com
mailing list for the glibc project.
Re: glibc conditioning
Hi,
After I turned on 64-bit mode, I did several hours of numerical
crunching and indeed sometimes here and there there are some digits
(mostly beyond 6th) do not match results from Sparc Solaris. If I turn
on 80-bit, results are way off. I'm not sure if the difference comes
from the "non-IEEE conforming" FPU at 64bit mode, or comes from the
difference in the math library implementation.
Another problem we have in porting to Linux is the endian conversion
thing. The binary data (int and double) we wrote on Sparc Solaris of
course cannot be read out as is on Linux. What is the most
"professional" way of handling these things?
Jinsong
> Paul Eggert <eggert@twinsun.com> writes:
>
> > That may work for your particular example, but it does not guarantee
> > consistency for all applications. When the x86 FPU operates in 64-bit
> > mode, it does not conform to the IEEE floating point standard. In
> > some cases, behavior will still disagree with your other platforms.
>
> And most importantly: the math library will produce wrong results.
>
> --
> ---------------. ,-. 1325 Chesapeake Terrace
> Ulrich Drepper \ ,-------------------' \ Sunnyvale, CA 94089 USA
> Red Hat `--' drepper at redhat.com `------------------------