microcode, compilers, and supercomputer architecture

From: jpero_at_cgocable.net <(jpero_at_cgocable.net)>
Date: Sun Apr 4 20:04:28 1999

Date sent: Mon, 5 Apr 1999 00:57:38 +0100 (BST)
Send reply to: classiccmp_at_u.washington.edu
From: ard_at_p850ug1.demon.co.uk (Tony Duell)
To: "Discussion re-collecting of classic computers" <classiccmp_at_u.washington.edu>
Subject: Re: microcode, compilers, and supercomputer architecture

> >
> > > It may be that compilable languages are defined for the purpose of
> > > providing for microcode but, that would mean that the sequence of
> > > microinstructions is generally not predictable from the source code of
> > > the program thus translated.
> >
> > Are you trying to claim that a microcode compiler is non-deterministic?
> > This seems like a dubious proposition, given that there are generally no
> > calls to a PRNG.
>
> The problem with most compilers is that they're too clever ;-). And you
> don't always know what they're going to do in some circumstances (unless
> you wrote them yourself ;-)). I find half the time I spend with CAD tools
> is figuring out how to get them to do what I want - something I could do a
> lot quicker and more accurately by hand...
>
Snip!

> The answer (on the PERQ) _is_ defined. By default, AMUX will be set so R0
> will appear on those lines. But there's actually an obscure instruction to
> set the AMUX as you want it. The point of that? Tracing pattern-related
> ALU faults.Had that instruction not existed (and it's not mentioned in
> some versions of the manual), you can bet I'd have been patching the
> binaries...
>
> Or : Should the microcode assembler stick your code all
> over the control store and put in (free) jumps as necessary? Or should it
> maintain the code as you typed it? Does it matter? Do you want to know
> what it's done? Often, it makes a heck of a difference...
>
> >
> > The only deliberately non-deterministic development tool I've ever used
> > was a Xilinx FPGA fitter. Somehow it seemed reminiscent of the bogosort
> > algorithm.
>
> Don't get me started on the Xilinx tools. Suffice it to say, I spent
> enough time undoing the damage that they did to my designs. I hate
> computers that think they can design better than I can. They are almost
> always wrong ;-)
>
> -tony
>
Beauiful!

That what I want to avoid in that homebrew computer by doing on
the wire to insure it's optimal operation and design.

Wizard
Received on Sun Apr 04 1999 - 20:04:28 BST