Mon Apr 9 12:03:00 GMT 2001
> > I think the gcc.c-torture/execute/20010114-2.c test is invalid, because it
> > will fail if you have excess precision in intermediate calculations, or
> > if you just have a float implementation with more than the minimal
> > precision required.
> I'm not sure. On PPC it gets calculated with extra precision as well and it
> works as only exactly representable numbberss are used, but I have no problem
> restricting the test to some platforms.
The rintf founction relies on rounding error to do its work. When more
than 23 bit precision comes into play, the argument is returned unchanged.
More information about the Gcc-bugs