[Info-vax] RealWorldTech on Poulson

[Neil Rieck]

On Jul 5, 1:04 pm, koeh... at eisner.nospam.encompasserve.org (Bob
Koehler) wrote:
> In article <4e0e88f8$0$6590$c3e8da3$b1356... at news.astraweb.com>, JF Mezei <jfmezei.spam... at vaxination.ca> writes:
>
>
>
> > Consider that a true change in architecture to really make it out of
> > order, would likely require changes to all compilers and recompiling of
> > software with new compilers, and it would also require that the new
> > chips support the old executables with the EPIC mentality. Me thinks
> > this would be way too big of a change to be acceptable.
>
>    All the important stuff _is_ in the compilers.  Has been for decades.
>    We've seen, and I wouldn't have any touble seeing more, compilers
>    with chip generation knowledge.
>
>    I was impressed just how far the DEC C compiler for Alpha would go,
>    back on the first release.  I've been impressed just how much better
>    some chip vendor's compilers are over gnu (I've seen perfomance
>    double).
>
>    This really shouldn't be suprizing.  Back in VAX days, Ultrix-32
>    offered a choice of the standard UNIX cc, or vcc, which was based
>    on the VAX C compiler for VMS.  vcc knew VAX better and the
>    performance difference was real.  With RISC and VLIW it's not unusual
>    to have generation specific qualifiers for compilers to generate the
>    best code for different generations of the same chip.
>
>    My only problem was not having generation qualifiers on VAX
>    compilers.  All the NOPs inserted to make 6000 series systems optimal
>    might not have been the right thing on my MV II.  Generating MOVCx
>    certainly wasn't.

Yep, doesn't one alternate definition of RISC mean "Relegate Important
Stuff to Complier"?

NSR