If we want a specific taylord cpu, i think it's better to code the
function you want in C as golden model. Then it's much easy to see
common point and to design specific ALU.

I have see RISC cpu for complex Flash interface using a depth of one
for stack. So it's not a true "complete" cpu, but it's very small and
very fast.

A lot of optimisation for cpu must be balance. Cache are good but
performance depend on the cache line length or the set number.
Pipeline raise the maximum clock speed but raise also the latency,
which could be a killer depending on the code. Predicat are good to
avoid branch penalty but complexify the pipeline and could became hard
to use for superscalar core.

Computer Architecture: A Quantitative Approach" by Hennessy and
Patterson give numbers on software for workstation/server (the "big"
computer engineering) not for specfic application like ours.

I think that HDL for the cpu and the code that will run on it must be
written in the same time to optimise both correclty.

Regards,
Nicolas Boulay
_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)

Reply via email to