This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH/gfortran] Fix handling of subnormal numbers
Steve Kargl wrote:
> I just ran paranoia from netlib (spara.f and dpara.f).
> It found no problems with the subnormal stuff I just
> committed.
That's no surprise, I believe that paranoia leaves very little work for the
constant folders, but it's been a while since I looked at its code.
One thing that we should remember about nearest in the context of implementing
proper dealing of subnormals is that while NEAREST(0.,1.) works correctly now,
NEAREST(NEAREST(0.,1.), 1.) will not be correctly rounded, and the code which
checks for a shrinking exponent doesn't make much sense for subnormals. I
don't think it's important to fix this before we match target arithmetic in
other cases, because otherwise all the effort we put into this will probably
be lost in successive constant folding.
> I'll, however, think about rth's truncation.
Please.
Thanks,
- Tobi