[Gllug] Re: Dual core AMDs

Nix nix at esperi.org.uk
Tue Jul 12 10:14:53 UTC 2005

On Tue, 12 Jul 2005, Chris Bell moaned:
>    That I do understand, but I thought there was a big difference between
> the pipeline and an on-chip cache. Both the Strong ARM and the Crusoe had a
> cache large enough to house most of a programming language as well as data,

What does that mean?

- Most of the language runtime, well, that really depends on the language.
  C++ and C have small runtimes --- if you ignore glibc, which might fit
  in 2Mb caches, I suppose; as for Java, no chance. Ada, maybe.
- Most of the compiler, no chance.
- Most of the language spec text (!) no chance except for a very few
  languages. (I really doubt this is what you mean; there's little call for
  holding language specs in L1 cache ;) )

>    Some CISC chips are only efficient when they are doing work which
> involves steady progress through a long pipeline. Any looping or jumping to
> other locations often involves clearing the pipeline and is expensive, even
> when the new data is held in the on-chip cache.

Much of this is true of *all* chips. Even RISC chips need long pipelines
these days, relative RAM/CPU speeds being what they are. (In fact,
they're even more dependent on pipelines and sensitive to pipeline
stalls than are CISC chips.)

`But of course, GR is the very best relativity for the masses.'
 --- Wayne Throop
Gllug mailing list  -  Gllug at gllug.org.uk

More information about the GLLUG mailing list