This is the mail archive of the mailing list for the GCC 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: [PATCH/gfortran] Chop bits of subnormal numbers

On Mon, May 09, 2005 at 05:20:38PM +0200, Tobias Schl?ter wrote:
> Steve Kargl wrote:
> > I've updated the patch to chop the bits on the absolute
> > value, and then set the sign if need be.
> > 
> > Bubblestrapped and reg. tested onn i386-*-freebsd.
> > 
> > OK for mainline?
> This is ok, but I'd prefer if a testcase went along with it.  Something like
>  if (tiny(x)/2. /= tiny(x)/2. - (nearest(tiny(x),1.) - tiny(x)))/2.) &
>      call abort
> should do the trick.  (The idea is to change the digit that gets chopped off
> when we go from normalized to subnormal numbers, I'm not sure if I got it
> right, though).  I'll have a go at fixing NEAREST for subnormals now.

I've come up with the following test case.

! { dg-do run }
! { dg-options "-Wno-underflow" }
! Check that the chopping of bits of subnormal numbers works.
program chop
  real x
  x = 1.
  if (tiny(x)/2. /= tiny(x)/2. - (nearest(tiny(x),1.) - tiny(x))/2.) then
    call abort
  end if
end program chop

I usually do not do FP comparison as in the IF() statement above.
Tobi, do you see any problems with the testcase.

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