🚀 go-pugleaf

ttn@cs.tut.fi (Tero Nieminen) writes:
> In article <1992Nov6.205045.26293@netcom.com>
> abell@netcom.com (Steven T. Abell) writes:

>   Unfortunately, you can't just brush off an obnoxious architecture by saying
>   "The compiler people will worry about this." Let's stay real, folks.
>   Underlying models matter, and the Intel model is about as bad as they come.
>   If this is what we have to live with, I suppose we'll manage somehow,
>   but if you think it doesn't affect you, you're wrong.

> Actually what do you mean by the intel model? And how is it inferior to
> Motorola 68k architecture?

> Not by speed anyway.

The Intel CPU programming model involves 12+ registers, no two of which have
the same capability. Of the 12+, 6+ are special-purpose to the point of
being useless as programming assets. Using the remaining 6 involves strategic
and tactical planning that will break your mind if you're trying to be
efficient. Declaring register variables in a C program becomes largely a
waste of time. A *very* large number of machine instructions are devoted
to picking up and putting down locals. The superior speed of Intel CPUs
is *not* an indicator of a superior CPU, just an indicator of better silicon
technology and greater design capacity.

Don't get me wrong: the 68K model has its own significant problems, but
outright wretchedness isn't one of them. Try writing some large procedures in
68K assembler, then do the same for an i86. Now, look at what you did from a
compiler-writer's point of view. Do you even want to *think* about codifying
your thoughts during your i86 efforts?

The Intel programming model is literally a bastard outgrowth of the 8080 model,
which was designed back when getting any kind of CPU at all on one chip was
still a miracle. If IBM hadn't chosen the 8088 for the original PC because
Motorola was late (as usual) with the 68008, the currently-touted Intel CPUs
*simply wouldn't exist* today, largely because they are too damned hard to
program.

Steve abell@netcom.com