[DGD] Native floats

[Felix A. Croes]

Shentino <shentino at gmail.com> wrote:

> I remembered something on Noah's site about Dworkin stating a plan to
> use native (IEEE) floats.
>
> Just a bit curious as to the schedule, if any.  It would be nice to
> use the FPU someday...especially for physics nuts like me :P

There is no schedule -- it is merely on my TODO list.  Also, let
me clear up some possible confusion.  DGD floats are like IEEE
doubles, but with the mantissa truncated to 36 bits.  That is not
going to change, even if the underlying implementation were to use
actual IEEE doubles for floating point computations.

Any resulting speedup is likely to be minor, compared to the
interpreter overhead, which is considerable.  DGD floats might suffice
for some prototype work, but serious physics simulations would benefit
enormously from some tailored kfuns employing SSE/VMX/Altivec code.

Regards,
Dworkin