RE: [sv-ac] RE: AC 296

Hello John,

I think that your proposal of handling ended vs. troggered in assertions is
interesting. But if we do allow triggered in sequences, such that it is true
throughout the time step and allows sync between inequivalent events, what
is the validity of $past(s.triggered)? Is it now synced to the clock like
$past(s.ended) and thus synthesizable? For any event e, is it then possible
to use e.triggered within a sequence? Since the assertions are evaluated in
the observed region and e could be emitted in the reactive region, how would
that behave?

It seems to me that allowing .triggered inside sequences may open up a lot
of other issues that we may not have the time to resolve.

ended as an event - I think this is hidden under @(s), i.e., we do not need
@(s.ended) even though its meaning would be equivalent. Disallowing this
would be a limitation, it is potentially quite useful to trigger sampling of
covergroups using @(s) (do not need a cover poperty to instantiate the
sequence.)

Allowing actual arguments in s(...).triggered is OK, for reasons that you
give. The system can do the instantiation under the hood.

Best regards,

ed

> -----Original Message-----
> From: owner-sv-ac@eda.org [mailto:owner-sv-ac@eda.org] On
> Behalf Of John Havlicek
> Sent: Sunday, November 14, 2004 4:30 AM
> To: Manisha_Kulshrestha@mentorg.com
> Cc: sv-ac@eda.org
> Subject: Re: [sv-ac] RE: AC 296
>
> Manisha and All:
>
> After thinking more about these questions I have a few comments.
>
> First, from the point of view of the assertions, there has
> been from a long time ago a desire to ensure synthesizability
> of the checkers.
> Therefore, synchronizing constructs that could detect
> simultaneity of "inequivalent" events were disallowed. This
> consideration led to the distinction between "ended" and
> "matched" in the assertions, where "ended" is used to mean
> that the synchronization is known to be between equivalent
> clock events, while "matched" is used when it is not so
> known. When "matched" is used, simultaneity of distinct
> clock events is not detected.
>
> If "triggered" is allowed within the assertion code, then it
> seems to lie between the above partitioning of the
> synchronization methods.
> The semantics of "triggered" seems closest to "ended". "triggered"
> carries no latching mechanism like "matched" to guarantee
> synchronization between inequivalent clock events. However,
> "triggered" can opportunistically synchronize between
> inequivalent events that happen to be simultaneous. This
> could be useful, and preserving the syntactic distinction
> between "triggered" and "ended"
> would be needed because of their distinct semantics.
> Assertions using "triggered" would not be synthesizable
> unless substitution of "ended"
> for "triggered" were legal.
>
> I can see an argument for disallowing "ended" in event
> expressions because "ended" implies an extra guarantee that
> the synchronization is between equivalent events, while the
> formation of an event expression must, of essence, allow the
> creation of an inequivalent event.
>
> I believe that at module level (or analogous levels), "triggered"
> should be allowed with arguments. The user can always define
> a wrapper sequence at module level that instantiates a
> sequence with arguments, and so disallowing this is just an
> inconvenience. I am not sure if there are difficulties that
> arise from allowing "triggered" on instances of sequences in
> other contexts. Perhaps the rules that govern the definition
> of a concurrent in a procedural context might be helpful.
>
> Best regards,
>
> John H.
>
> >
> > Hi All,
> >
> > As we discussed in the meeting. I think we need to discuss the
> > following things before I can update my proposal:
> >
> > 1. Is .ended allowed in event expressions (i.e. in wait
> statement) in
> > the procedural code. My impression is it is allowed based on the
> > discussion few weeks ago. If it is allowed then why not make it
> > available till the time the event wakes up to get executed.
> >
> > 2. Is .triggered allowed to be used to form sequence expressions ?
> > Here we need to consider .triggered property of any event not just
> > sequence events. Current syntax allows it and LRM does not
> prohibit it.
> >
> > 3. Joseph raised this question: can .triggered be invoked
> on sequences
> > which have formal arguments ?
> >
> > 4. In what region .triggered goes high ? Currently LRM says that it
> > goes high in Observe region. My understanding is that it should go
> > high at the same time .ended goes high. Here is a statement
> from LRM
> > section 8.11:
> >
> > The triggered status of a sequence is set during the Observe region
> > and persists through the remainder of the time-step (i.e., until
> > simulation time advances).
> >
> > From this, it looks like the only difference in .ended and
> .triggered
> > is that .triggered defines when it goes down but .ended
> does not define it.
> > I think it will be good idea if we clearly define in the LRM when
> > .ended
> >
> > goes down even if we restrict its usage to sequence
> expressions only.
> >
> >
> > Thanks.
> > Manisha
> >
> >
> >
> > -----Original Message-----
> > From: Eduard Cerny [mailto:Eduard.Cerny@synopsys.com]
> > Sent: Wednesday, November 10, 2004 1:27 PM
> > To: sv-ac@eda.org; Kulshrestha, Manisha
> > Cc: Ed Cerny
> > Subject: AC 296
> >
> > Hello Manisha,
> >
> > I agree with you and think that it is a good idea to summarize the
> > sequence methods in a se[arate section. However I would like to
> > suggest a some changes to your proposal. In particular, ended and
> > matched should not be allowed outside expressions in
> sequences. They
> > are more like boolean events valid only at the clock tick.
> We should
> > discuss it tomorrow at the meeting.
> >
> > Best regards,
> >
> > ed
> >
> >
> > ---
> > In the paragraph on ended you say:
> >
> > "The ended status of the sequence is set in the Observe region and
> > persists through the remainder of the time step (i.e. until
> simulation
> > time advances). This method can be used to detect the endpoint of a
> > sequence used in another sequence."
> >
> > I think that extending the duration of the ended in not
> necessary as
> > it cannot be used anywhere but in another sequence which does also
> > evaluate in the Observed region. (Clearly, when evaluating
> sequences
> > the evaluation must be ordered by mutual dependency and the
> dependency
> > graph must not have
> > cycles.) Perhaps you would agree if the text is changed as follows:
> >
> > "The ended status of the sequence is set in the Observed
> region. This
> > method can only be used to detect the endpoint of a sequence in
> > another sequence.
> > There must be no circular dependency between sequences
> induced by the
> > use of ended."
> >
> > ---
> > In the paragraph on triggered :
> >
> > "The value of method triggered evaluates to true if the
> given sequence
> > has reached its endpoint at that particular point in time and false
> > otherwise.
> > The triggered status of the sequence is set in the Observe
> region and
> > persists through the remainder of the time-step. This
> method can be
> > used anywhere that the triggered property of events can be used
> > (section 13.5.4).
> > It is an error to pass a local variable as an actual argument to a
> > sequence instance on which method triggered is applied.
> This prevents
> > local variables from flowing out of method triggered. (See section
> > 17.8)"
> >
> > I propose that you add:
> >
> > "... This method can be used anywhere that the triggered
> property of
> > events can be used (section 13.5.4), including the boolean
> expressions
> > in disable iff statements of properties. There should be no
> circular
> > dependency between properties and sequences induced by the use of
> > sequence triggered in disable if clauses. It is an error ... "
> >
> > ---
> > The paragraph starting with: "The methods ended and triggered ..."
> >
> > Change as
> >
> > "The method triggered can be used in wait statements or boolean
> > expressions outside of property context. However, it is an error to
> > invoke those methods on sequences which treat their formal
> arguments
> > as local variables. A sequence treats its formal argument
> as a local
> > variable if the formal argument is used as an lvalue in
> > operator_assignment or inc_or_dec_expression in
> sequence_match_item."
> >
> > ---
> > The paragraph starting with: "Unlike ended and triggered,
> matched uses
> > ..."
> >
> > Change to:
> >
> > "Unlike ended and triggered, matched provides means for
> > synchronization between two clocks by making available the
> result of
> > the source sequence match to the evaluation of the destination
> > sequence at the first clock tick of the destination
> sequence after the
> > source sequence match. This method is used to detect the
> endpoint of a
> > sequence in a multi-clocked sequence. If the source and destination
> > clocks are the same, again the result is transferred to the first
> > subsequent clock tick. Like ended, matched can only be used
> in boolean expressions in sequences."
> >
> >
> >
> >
> >
>
Received on Mon Nov 15 06:31:54 2004