Re: [sv-bc] Mantis 1345: 10.4: "illegal" unique if/case issues

Stu,

In the following example does o get i or 3? 

typedef bit [1:0] T ; 
module test ( input T i, output T o ) ; 
always_comb 
   unique case (i) 
     ( o = 0 ) : ; 
     ( o = 1 ) : ; 
     ( o = 2 ) : ; 
     ( o = 3 ) : ; 
   endcase 
endmodule 

For a related Verilog example, see

   http://www.boyd.com/1364_btf/report/full_pr/566.html

-- Brad

-----Original Message-----
From: owner-sv-bc@eda.org [mailto:owner-sv-bc@eda.org] On Behalf Of
Stuart Sutherland
Sent: Thursday, March 16, 2006 3:37 PM
To: sv-bc@eda.org
Subject: RE: [sv-bc] Mantis 1345: 10.4: "illegal" unique if/case issues

Brad,

The behavior I suggested would allow--actually require--every simulator
to
do the same thing for every "pathological case".

a) Look for static overlap of decision select expressions at compile
time.

b) When entering a decision statement, analyze how many branch select
expressions are true at that instant, and warn if multiple or no
selections
are true.

c) Execute the first matching branch (same behavior as today, without a
'unique' modifier.

In my mind, the only possible divergence between tools would be due to
event
ordering for when the decision statement is entered relative to other
events
in the active event queue at that moment in time.  That might lead to a
difference in what decision select expressions are true at that moment
in
time.  But, that risk exists today, and always has existed, and affects
which branch is the first matching branch.

Stu
~~~~~~~~~~~~~~~~~~~~~~~~~
Stuart Sutherland
stuart@sutherland-hdl.com
+1-503-692-0898
  

> -----Original Message-----
> From: owner-sv-bc@eda.org [mailto:owner-sv-bc@eda.org] On 
> Behalf Of Brad Pierce
> Sent: Thursday, March 16, 2006 2:34 PM
> To: sv-bc@eda.org
> Subject: Re: [sv-bc] Mantis 1345: 10.4: "illegal" unique 
> if/case issues
> 
> >Let's not make a mountain out of a molehill!
> 
> Tool implementers don't have the luxury of only handling typical test
> cases correctly.  We have to handle even the most apparently
> pathological test cases correctly.
> 
> -- Brad
> 
> 
> 
>