RE: [sv-ec] A possible ambiguity in circular solve/before constraints

From: Ryan, Ray <Ray_Ryan_at_.....>
Date: Fri Jun 30 2006 - 17:22:42 PDT
 Bill,

A clarification might be appropriate for the sixth paragraph
in 13.4.9 (that talks about solve/before not causing
randomize to fail).  The intent of this statement was related
to the solution space, and that the effect of a solve-before
constraint is to influence the distribution of selected
solutions. A (valid) solve-before constraint does not alter
the solution space. So, if there is a solution without the
solver-before constraint, there must be a solution with the
solve-before constraint (assuming there isn't a cyclic error
in the solve-before constraint). 

The LRM is clear that a circular solve/before constraint is
not allowed. It's probably up to the implementation how to
report/deal with the coding error. Reporting the error and causing
randomize to fail seems appropriate to me. The same would be
the case if the solve/before constraint references a non-random 
variable.

- Ray Ryan

> -----Original Message-----
> From: owner-sv-ec@server.eda-stds.org 
> [mailto:owner-sv-ec@server.eda-stds.org] On Behalf Of William Paulsen
> Sent: Thursday, June 29, 2006 8:39 AM
> To: sv-ec@server.eda-stds.org
> Subject: [sv-ec] A possible ambiguity in circular 
> solve/before constraints
> 
> 
> I think there might be an ambiguity in the LRM Section 13.4.9 
> when there's a circular dependency in solve/before 
> constraints.  Should a randomize call fail if there's a 
> circular dependency?  Or should randomize just ignore all the 
> solve/befores, and just consider the other constraints?
> 
> The sixth paragraph in 13.4.9 seems to say that randomize 
> should ignore circular solve/befores:
> 
> "Variable ordering can be used to force selected corner cases 
> to occur more frequently than they would otherwise. However, 
> a "solve...before..." constraint does not change the solution 
> space and, therefore, cannot cause the solver to fail."
> 
> But the fifth bullet of paragraph 9 seems to say that randomize should
> fail:
> 
> - There must be no circular dependencies in the ordering, 
> such as "solve a before b" combined with "solve b before a".
> 
> 
> Thanks,
> Bill Paulsen
> 
> 
Received on Fri Jun 30 17:22:46 2006

This archive was generated by hypermail 2.1.8 : Fri Jun 30 2006 - 17:23:05 PDT