On Thu, Dec 07, 2000 at 10:54:09AM +0000, Piers Cawley wrote:
> "Bradley M. Kuhn" <[EMAIL PROTECTED]> writes:
> > Dan Sugalski <[EMAIL PROTECTED]> wrote:
> > > 2) GCC produces slow code on all platforms where there's an
> > > alternative.
> >
> > This is likely going to change with the release of GCC 3.0, based on
> > current benchmarks of test releases of GCC 3.0.
I can beat the optimiser on 2.96 on ARM for the loop in runops_standard
It used 7 instructions, I can do it in 5. I am disappointed that in such
a tiny function which doesn't even use all the cpu registers it fails to
spot two obvious things it could do better. I'd forgive it if it used 6
instructions. Norcroft on arm produces much better code, and I am told
that the gcc maintainers never except to reach the code quality it
produces. Oops. Digression. [Not sure why the gcc maintainers should
be so pessimistic]
> Err, are you seriously suggesting that we implement Perl 6 using a
> language which currently only has one 'efficient compiler' (to native
> code), when said compiler isn't actually there yet?
>
> I think that that would be a 'courageous' decision.
Making decisions now that make it hard to use anything other than 1 compiler
are as wise as decisions that make it hard to use anything other than one
implementation language.
[I'm not saying that we should use Java (or anything else for that
matter), just that I'd feel we'll have a more portable design if we
don't constrain our choices now]
Nicholas Clark
