[Info-vax] CPU architectures, was: Re: problem with 64-bit pointers in C

[already5chosen at yahoo.com]

On Tuesday, February 13, 2018 at 11:35:37 AM UTC+2, Jan-Erik Soderholm wrote:
> Den 2018-02-13 kl. 09:27, skrev Simon Clubley:
> > On 2018-02-13, Jan-Erik Soderholm <jan-erik.soderholm at telia.com> wrote:
> >> Den 2018-02-13 kl. 07:36, skrev Simon Clubley:
> >>>
> >>> Those code size changes from VAX to Alpha were not helped by the lousy
> >>> code density on Alpha when compared to VAX.
> >>
> >> But that was not a mistake in the Alpha part, that is a known and
> >> predictable difference between CISC and RISC. You get higher speed
> >> instead due to a "cleaner" architecture.
> >>
> >> And as the hardware and memory development has been, code density
> >> has hardly been an issue as such.
> >>
> > 
> > I am currently operating under somewhat different memory constraints
> > for my Macro-32 and Macro-64 code. :-)
> > 
> 
> But you refered to the "lousy code density on Alpha".
> 
> Your specific needs are probably not a valid benchmark to
> evalute processor architectures against. :-)
> 
> >>>
> >>> Macro-32 is a much nicer assembly language than Macro-64 is.
> >>>
> >>
> >> For the same reason, a CISC environment does more with fewer
> >> instructions. Also a know effect from RISC, to get higher speed.
> >>
> > 
> > This is somewhat architecture specific, even on RISC. ARM is a much
> > nicer architecture to write assembly language for than Alpha.
> 
> ARM was to a higher degree designed for assembled programming.
> Then in later days ARM has grown higher up into areas where other
> processorn (such as Alpha) had made its living. And in the same
> way, Alpha was designed no as much for assembler programming,
> but for efficient compiler designs and high speed.
> 
> > 
> >> And at the same time, you where expected to switch from assembler
> >> to higher lever languages. And if I'm not wrong, it is easier to
> >> design a good compiler on a cleaner architecture (like Alpha).
> >>
> > 
> > Alpha is cleaner than x86 (and hence writing compilers for Alpha is
> > easier) but VAX was also a clean architecture when it came to
> > writing compilers.
> > 
> 
> Of course it helps to write assembler on a very-CISC architecture,
> since each instruction does more. But with the expense of speed
> (at that time).
> 
> Today, even x86 has got some steorids to get up in speed in a way
> that was not predicable 15-20 years ago.

It was easily predictable 15-20 years ago.
Pentium Pro is 22 y.o. and it clearly showed the way to anybody willing to see.
Exactly 20 years ago is when Intel released 2nd-generation Pentium-II that demonstrated that fast x86 is not just possible technically, but could be manufactured cheaply and does not have to consume a lot of power.
And 18.5 years ago AMD demonstrated that it's possible to design fast x86 without multi-billion budget.

Make it 25 years instead of 15-20 and you would have a point.

> 
> > Don't confuse a simple architecture with a clean architecture.
> >
> 
> I was saying that the Alpha code density is "lousy" by-design
> and not by some mistake.
> 
> 
> > Simon.
> >