RE: Paramset Proposal

> -----Original Message-----

> From: geoffrey.coram@analog.com [mailto:geoffrey.coram@analog.com]

> Sent: Friday, May 14, 2004 8:19 AM

> To: Kevin Cameron

> Subject: Re: Paramset Proposal

>

> Kevin Cameron wrote:

> > I just had a look at the paramset proposal on the website,

> > and I'm a bit concerned about the mechanism for selecting

> > between multiple sets of the same name: the proposal appears

> > rule-based but doesn't seem to have any rules for when multiple

> > matches are valid.

>

> The first valid match is the one selected, just as in Spice.

>

> I think this is consistent with SV/V2001, but I can't find

> the explicit statement of this in the LRMs. The best I found

> was this from 1364-2005-d2.pdf:

>

> 13.5.1 Default configuration from library map file

> With no configuration, the libraries are searched according

> to the library declaration order in the library map file.

> This means all instances of module adder shall use aLib.adder

> (since aLib is the first library speci ed which contains a cell

> named adder), and all instances of module foo shall use

> rtlLib.foo (since rtlLib is the first library which contains foo).

>

> -Geoffrey

I don't see much similarity between paramsets and libraries myself.

I could equally well point at discipline resolution as a mechanism

for picking paramsets.

The implication of just picking the first match would be that if you

have a paramset that matches for "corner" and a different one matching

for "l & w" and another which matches both (the best match), then which

one you get depends on the order of declaration? I would prefer to
always

get the best match.

That kind of methodology also works against doing modular/incremental

compilation, since you can't just tack on extra paramsets if their

source order is important.

Kev.

Kevin Cameron, CPU Technology, CA 94588, Tel.: (925) 225 4862

 
Received on Fri May 14 09:39:16 2004