Re: Comparisons and decorations, part 2
Arnold Neumaier wrote:
> Dan Zuras Intervals wrote:
> > I suggest, for the sake of assured computing, that we
> > ALWAYS deal with decorated intervals at a formal level.
> > Then we can leave it to compiler optimizers to know
> > when it is safe to strip them bare & run without the
> > decorations.
>
> A compiler must be very intelligaent to know that without directives
> from the user, since it must understand some mathematical theory.
Even if there *was* such a smart compiler, *it* would need some basis
regarding available machine primitives to know what code to generate.
So the standard must provide sufficient guidance for this to be possible.
The same considerations apply when defining library function interfaces.
Finally, we have to consider that initial interval support is likely to
come from external function packages used with compilers that are not
directly aware of Interval Arithmetic.
> But expert users must be able to strip the decorations explicitly,
> without having to prove to the compiler that it is allowed.
Most definitely!
(My predilection for assembler programming shows through. I understand
compilers only in terms of the code they generate from a given source
program!)
Michel.
---Sent: 2010-09-28 20:35:12 UTC