Re: [f-cpu] Come back of loadcons

[Michael Riepe <michael@stud.uni-hannover.de>]

On Sun, Aug 04, 2002 at 10:50:50PM +0200, cedric wrote:
> Because no body continu the discussion about loadcons, I don't know if we take 
> a decision. For memory, Yann say that the shifted loadcons cost a lot in the 
> CDP and that's a reason to not implement it. For Michael "it only cost a mux" 
> in the CDP that currently always exist.
> 	From Thomas point of view it's better because of scalability reason and I 
> agree with him. If I resume all this different argument I can only go to the 
> conclusion that we must implement this loadcons.

There is a new argument: When we have `cshiftl', the old loadcons[x].4
(and higher suffixes) and the new `shifting loadcons' are no longer
required. Building large constants may take a little longer with cshift,
but it is possible to populate the whole register, no matter how wide
it is - and we need only two opcodes (loadcons.[0-3] and loadconsx.[0-3]).

The only remaining argument against the original loadcons is that it
requires a partial write (that is, a feedback from the register read
port and a mux inside the CDP). But that's true for the shifting
loadcons as well, unless we use a `shadow register' implementation.

I think we should implement cshift and drop the shifting loadcons
variant, as well as loadcons[x].[4-15].

Any objections?

-- 
 Michael "Tired" Riepe <Michael.Riepe@stud.uni-hannover.de>
 "All I wanna do is have a little fun before I die"
*************************************************************
To unsubscribe, send an e-mail to majordomo@seul.org with
unsubscribe f-cpu       in the body. http://f-cpu.seul.org/