processors and FP (was: OQO)

Tim Rowledge tim at sumeru.stanford.edu
Thu May 16 04:19:21 UTC 2002


Jecel Assumpcao Jr wrote:

> > > However, ARM10 devices do had fp - in fact array FP - hardware. An
> > > 800MHz ARM1020FE based machine would be very nice to have.
> > > Supposedly Lucent has produced sample units, but I haven't heard of
> > > anyone shipping yet.
> 
> It would take a lot of effort to make Squeak take advantage of this
> hardware. Unless you just wanted to speed up selected parts, like
> balloon3D. The result of a more general implementation would be worth
> it, in my opinion.
Well, I can;t see any way to take advantage in a general way of array FP
- but for anything involving arrays of flaots (like everything to do
with games, flashy 3D UIs, all those eyecandy things) it's wonderful.
For simple float arithmetic, its just the same as usual
> 
> > > PS Oh yes, ARM11 is already on the way.
> 
> How about the Jazelle bytecode accelerator which will be standard (I
> think) on ARM11 (http://www.arm.com/armtech/jazelle?OpenDocument) ?
> Would that help Squeak?
Yuck. I doubt it to be honest. Even for java it doesn't really do very
much. Last time I read the documentation it pretty much claimed that it
could run jave faster than a simple interpreter vm but not as fast as a
hotspot-like system. They semed to be aiming art embedded situations
where a couple of times faster was valued and lots of ram was
impossible; cellphones etc.

tim




More information about the Squeak-dev mailing list