???? Hardware-accellerated python ????

[Alex Martelli]

"Armin Steinhoff" <Armin at Steinhoff_de> wrote in message
news:8mr18c$12s9 at drn.newsguy.com...
> In article <DW4k5.118428$1h3.1883630 at news20.bellglobal.com>, "Olivier
says...
> >
> >If the dictionary lookups are hash-based, there isn't very much
improvement
> >to be done, as hash tables are usually O(1).  I'd say the best
accelerator
> >for Python is RAM.  You can buy more RAM (at your local computer hardware
> >store) and add it to your computer and EVERYTHING will go faster!
> >
> >The nature of an interpreted language (and many other features of python)
> >makes it hard to put it on a chip (or to compile).  At least, that's how
I
> >understand it.  Maybe I'm wrong?
>
> You are wrong ... tokens processed as machine code ist just faster.

It's a bit more complex than you make it out, Armin.  A general-purpose CPU
(like other similarly general-purpose pieces of hardware -- RAM, caches, &c)
has a huge market: it can amortize huge investments in technology and design
to shrink it down to the best that's technologically feasible at any given
time.

A special-purpose CPU designed/built to similar economic constraints cannot
hope to match the design-rules/process-learning-curve/etc of the general
purpose component.  So the "ist just faster" need not apply -- more often
than
a naive intuition would expect, the solution relying on general-purpose tech
is going to beat or leapfrog the one relying on special-purpose hardware.

E.g., one may be able, starting a special-purpose hardware design project
when the fastest general-purpose CPU on the mass-market runs at 133MHz,
to produce a special-purpose CPU that (for that single specific task only)
is
twice as fast... but when it comes out on the market, the general-purpose
CPU's have leap-frogged to 300MHz, and so easily beat the special-purpose
one (that's equivalent to a 266-MHz general-purpose one).  And the special
purpose project just doesn't have a wide-enough market to finance enormous
continuing investments to try to catch up with mainstream technology.

You need a *huge* 'leverage' factor to defeat this Moore-curve effect; or a
"niche" for your special-purpose hardware that is actually pretty large
(video
& advanced-graphics processing, maybe). Most of the time, neither applies.


Alex