Re: Note on NaN


Subject: Re: Note on NaN
From: Kevin Cameron x3251 (dkc@galaxy.nsc.com)
Date: Thu Nov 01 2001 - 09:13:49 PST


>
> It might be best to propagate both
> state and value. NaN's should be avoided, since it generally
> contaminates all associated math. In the past, we usually
> trapped the NaN interrupt and had a service routine
> return a large signed number as a limiter. I think of NaN as
> an overflow, not an unknown.
>
> --Kim

That makes sense in an all-analog simulation where NaN would
probably result from division by zero, however AMS 'X' values
are not necessarily out-of-range i.e. it is actually a 1,0 or
'Z', and will be somewhere between Vdd and Vss.

Users should take responsibility for handling 'X' states in
their code, having Verilog-AMS discard the fact that a value
is unknown would lead to code making false assumptions and
result in non-working Silicon.

Kev.
 



This archive was generated by hypermail 2b28 : Thu Nov 01 2001 - 09:15:13 PST