(j3.2006) decimal floats

[Andy Vaught]

On Sat, 12 Jan 2008, Andy Vaught wrote:

> On Fri, 11 Jan 2008, Robert Corbett wrote:
> 
> > 
> > I agree that implicit conversions should be banned.  There should be
> > functions for explicit conversions.  I suspect that most people do not
> > understand how expensive radix conversions are, especially for large
> > exponents.
> > 
> > Some people on the committee think arithmetic should be done in the
> > format used to represent values in memory.  That could create problems
> > for programs that use IEEE 754r's storage format such as binary16 and
> > decimal32.
> 
>   The parsing and printing of binary numbers to and from ascii is a prime
> example of radix conversion.
> 
>   The fortran philosophy of the past, however, is that you can even
> interoperate reals and integers without explicit conversion.  Or even
> adding reals to logicals on some implementations.  If people are concerned
> about speed, they're going to figure out not to do a conversion pretty
> quick, and, IMO, not use decimal arithmetic at all.

  Having given some more thought to this, I think the real insanity is a
chip that does floating point in two different radices.  If someone wants
to make a chip that does decimal floating point, fortran 90 will support
that perfectly-- radix(0.0) is 10 and everything else works as it should.  
If a chipmaker uses two radices, it means that they don't think that
decimal floating point will stand on its own and are hedging their bets.

      Andy