[rescue] SS10/20 death

Dave McGuire mcguire at neurotica.com
Tue Feb 5 19:20:47 CST 2002


On February 5, dave at cca.org wrote:
> >> Do you think the SPARC's register windows were a good idea? (For a
> >> general purpose unix box, that is.)
> 
> >  Tremendously good, yes.
> 
> That's good to hear. I thought the extra work of spilling to the stack
> and especially the added complexity of context switches might negate
> the wins, at least for general unix/c codes. It kind of looks more like
> a custom OS embedded app type design.

  Register spilling does cause complications and performance
problems...which is why one strives to avoid it whenever possible.
Deep call stacks are best avoided on SPARCs...if you're within the
depth of the register windows, you're in good shape and things fly.
If you spill, you slow down quite a bit.  It can be looked at in the
same manner as caching but at a lower level...there's a bit of very
high-speed local storage that, if your data access patterns are
tailored to take advantage of it, will be used more heavily, reducing
access to the next level of larger/slower storage.

> Have you ever enocuntered anyone using the windows for anything
> other than function calls?

  Not yet, though it might be interesting for context switching and
possibly interprocess communication stuff.

> SPARC's register windows always seemed like one of the weird &
> abandoned experiments in the crazy early days of the "RISC hype" before
> a few years of actually producing machines made it obvious that the
> wins of "RISC" amounted to: load/store arch & fixed length instructions.
> (Which is why I still think Cray's machines qualify, all the way back
> to the CDC-6600.)

  It's not abandoned at all.  All SPARCs, including UltraSPARCs, use
register windowing.  I've heard that several other (now
non-mainstream) processors have but I can't think of any off the top
of my head.

  And yes, many early computer architectures qualify as RISC by today's
definitions.  That's why I always preach to people about the
importance of studying the TRUE beginnings of computers.

     -Dave

-- 
Dave McGuire
St. Petersburg, FL         "Less talk.  More synthohol." --Lt. Worf



More information about the rescue mailing list