Re: Learning embedded systems
From: Darin Johnson (darin__at__usa_._net)
Date: Mon, 07 Mar 2005 00:00:19 GMT
Paul Keinanen <firstname.lastname@example.org> writes:
> >Each variant of an instruction set would need a new variant of the compiler,
> Is this really a problem, since each variant of an instruction set
> would require a new variant of the code generator anyway :-).
Sort of, but not really. Compilers don't do so well here as I've
noticed. Ie, the Metrowerks PowerPC compiler allowed you to tell it
that you were on an MPC860, but it still insisted on unrolling loops
so that they'd no longer fit into the instruction cache. I'm assuming
it essentially just had one code generator that didn't change much,
and telling it what cpu you head merely enabled different
> >The compiler writers aren't
> >spending their time worried about optimizing accesses to memory mapped
> These are perfect candidates for in-line assembly.
But I thought you were arguing that assembly wouldn't be necessary
in the future, and wouldn't that include inline-assembly also?
> I have nothing against assembly programming as such, I have written
> quite a lot of assembly work in the 1970's, but the problem with most
> assembly programs is that they are hard to read and hence hard to
Of course. But aren't we just talking about small fragments of code
here? Inline-assembler, context switches, bootstrapping, etc?
-- Darin Johnson "You used to be big." "I am big. It's the pictures that got small."