> Yea, I agree... One consequence is that it would allow a lot of the
> common ISA code to be fleshed out if you decided to optimize it.
>
> But I guess "overly conservative" isnt the all the way right word for
> that single class solution (i dont have a better word though!).
>
> What you would gain in immediate programmability (theISA->func()), you
> would probably definitely lose in performance.

You can't make an ISA independent CPU model.  It's just not possible.
It must use virtual functions, templates, or #defines.  Virtual
functions are out because of performance.  We don't really use
templates for the ISA, but it would be nice if we did.  We mostly use
#defines (THE_ISA) and it would be nice to move more to templates, but
I think we'll have to keep using #defines and the solution is to then
use build tricks to wrap stuff in namespaces so we can compile the
same file multiple times and not have symbol conflicts.

   Nate
_______________________________________________
m5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users

Reply via email to