[Gllug] Small rant

Nix nix at esperi.org.uk
Fri Nov 11 22:35:52 UTC 2005


On Fri, 11 Nov 2005, Richard Jones suggested tentatively:
> On Fri, Nov 11, 2005 at 11:25:26AM +0000, Nix wrote:
>> OK, you've piqued my curiosity. I know nothing about the i432, but...
>> how could they have made something worse than the segmented joy which
>> was the 80286? What on earth did they do?
> 
> Lots more segments - it really was designed so that you could put
> every object in a typical program (of the day) into a separate
> segment.

Wow-eee. Techniques from the past :)

> As ever Wikipedia has the gory details.  One interesting point seems
> to have been that the compiler was very poor:
> 
> http://en.wikipedia.org/wiki/Intel_i432

1981?! Ye gods. Early experimental stuff or what. :)

> Oh and of course it wasn't backwards compatible with the x86.  You'd
> have thunk that Intel would have learned their lesson the first time,
> but obviously not: http://www.sun.com/emrkt/itanic/

The Itanium too was meant to be the Great Architecture for the 2000s,
supplanting the x86, was fiercely complex, the engineers couldn't
translate it into efficient silicon, it's slow and expensive, and it has
been bogged down by requiring compilers to do too much. Let's make the
same mistakes over and over again!

It also has instructions of weird lengths and peculiar semantics (three
instructions per bundle, each IIRC 41 bits long, with 5 bits allowing
the instructions to be parallel-processed, but not in all ways you might
think possible as you'd need 8 bits to specify that).

But it doesn't lack for registers or cache, so they've learned
something. :)


The 432's compiler strikes me from that description as terrible;
Itanium's problem is that the compiler needs to be Einstein, but the
432's seems to have been a gumby.

The hardware object type checking seems reminiscent of Lisp Machine
design (although of course this predated the Lisp Machine!)

-- 
`Heinlein is quite competent at putting together sentences, but usually
he also puts together a plot to go with them.' --- Russ Allbery
-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug




More information about the GLLUG mailing list