RE: [sv-ac] FW: #805

From: Bassam Tabbara <Bassam.Tabbara_at_.....>
Date: Fri Jun 09 2006 - 16:24:01 PDT
Hi Manisha,
 
Well that kinda explains why I thought I saw this fix (to 29) before and
surprised I now find it's missing -- may be we should keep earlier
proposals as an RCS of sorts :). 
 
Ok, I think there is some mixup here -- The "outstanding" discussion I
think revolves around *cover* construct (meaning whether cover(disabled)
returns True or False),  there were no issues with assert. For the
coverage chapter, despite the naming and the "covered" all over the
place :), the coverable entity is an *assertion*, clearly stated in
chapter, this is merely a count of all the different results -- so it's
hard to say we're not collecting this. In fact the formula (Lisa's
original email) has this data ....
 
As far as the proposal, I think not adding this would be inconsistent,
not just because the callbacks exist but more simply as I say above
because the result ("disabled") exists, that is a state so no escaping
it. BTW, I recall we had a similar discussion about vacuous and counting
it.
 
Thx.
-Bassam.
 

________________________________

From: Kulshrestha, Manisha [mailto:Manisha_Kulshrestha@mentor.com] 
Sent: Friday, June 09, 2006 3:57 PM
To: Bassam Tabbara; Lisa Piper; Eduard Cerny; sv-ac@verilog.org
Subject: RE: [sv-ac] FW: #805


Hi Bassam,
 
So far my understanding is that we do not want to consider disabling of
a property as a coverage event. This is the reason we do not want to
provide coverage counting for disabled evaluations. This is the reason I
skipped vpi routines for them. I believe providing vpi routines for
counting disabled evaluations will not be consistent with rest of the
proposal. 
 
Thanks.
Manisha


________________________________

From: Bassam Tabbara [mailto:Bassam.Tabbara@synopsys.com] 
Sent: Friday, June 09, 2006 3:43 PM
To: Kulshrestha, Manisha; Lisa Piper; Eduard Cerny; sv-ac@verilog.org
Subject: RE: [sv-ac] FW: #805


Hi Lisa/Manisha, here's my opinion
 
Lisa:
- Yep seems missing, good catch -- adding it removes the need for the
sentence "The coverage counters ..."
- Let's keep separate, we don't want to go back to associating disabled
with vacuous and/or success 
 
Manisha:
 - You point out something that surely needs to be added, I missed this
and thought we had this coverage chapter fix already between 805 and
1361 but I am mistaken. 
 
** I fail to see why tools are being discussed -- this is the LRM and we
should add a basis set of access from VPI to any and all language
(syntax/execution semantics) no picking and choosing ... this is indeed
an omission. Actually another correction on p. 481
vpiAssertVacuousSuccessCovered is missing from the list but used later. 
 
Thx.
-Bassam.
 

________________________________

From: owner-sv-ac@verilog.org [mailto:owner-sv-ac@verilog.org] On Behalf
Of Kulshrestha, Manisha
Sent: Friday, June 09, 2006 3:26 PM
To: Lisa Piper; Eduard Cerny; sv-ac@verilog.org
Subject: RE: [sv-ac] FW: #805


Hi Lisa,
 
This is just a formula which I have modified as disabled is not
considered a success or failure any more. This does not mean that tools
need to keep a counter for disabled evaluations or report them in any
form. If you look at that section in the LRM, there are specific vpi
functions for getting different counts and I have not added any new
function for disabled so there is no need to keep this information in
the tool (atleast LRM does not require the tools to do so). I hope this
clarifies.
 
Thanks
Manisha
 

________________________________

From: Lisa Piper [mailto:piper@cadence.com] 
Sent: Friday, June 09, 2006 3:16 PM
To: Kulshrestha, Manisha; Eduard Cerny; sv-ac@verilog.org
Subject: RE: [sv-ac] FW: #805



Manisha,

 

Is there an implication here that a "disabled" counter also exists?  It
was not added to the list of counters in 17.13.3 - it was stated that
disabled evaluations do not affect the other counters. But in 29.4.3,
the section that talks about accessing coverage counts via vpi calls, it
states: in progress = attempts - (successes + vacuous success + disabled
+ failures).  Isn't this implying that a disabled counter exists since
it is not meaningful otherwise?  

 

I question the value of requiring a disabled counter.  I think the
disabled counts should be included with the vacuous counts. If a tool
vendor wants to add separate counters for the two, they can choose to do
so, but it should not be required.

 

Lisa

 

ps - sorry I did not pick up on that earlier!

 

________________________________

From: owner-sv-ac@verilog.org [mailto:owner-sv-ac@verilog.org] On Behalf
Of Kulshrestha, Manisha
Sent: Friday, June 09, 2006 2:35 PM
To: Eduard Cerny; sv-ac@verilog.org
Subject: RE: [sv-ac] FW: #805

 

Hi All,

 

I have uploaded a new version of the proposal on mantis. This includes
suggestions by Lisa (done differently than suggested) and corrections
from Ed ( I have made succeeded to succeeded nonvacuously). 

Ed, I have made the corrections but I am not sure about usefulness of
those callbacks. I feel callback on each disable evaluation when disable
is already true, is going to be unnecessary and expensive. Probably
Bassam can suggest something. Anyone else ?

 

Thanks.

Manisha

 

________________________________

From: Eduard Cerny [mailto:Eduard.Cerny@synopsys.com] 
Sent: Thursday, June 08, 2006 7:39 AM
To: Kulshrestha, Manisha; sv-ac@verilog.org
Subject: RE: [sv-ac] FW: #805

Manisha, 

 

the last statement in the document you sent says:

 

- cbAssertionSuccess, cbAssertionVacuousSuccess,
cbAssertionDisabledEvaluation and cbAssertionFailure: cb_time is the
time when the assertion succeeded or failed.

 

Should it say that in the case of disabled evaluation it is the time
when it was disabled? Rather than just succeeded or failed? Is it useful
to report disabled time likely to be the start time in many cases?)
Also, under succeeded, does it cover vacuous success? 

 

Best regards,

ed

	 

	
________________________________


	From: owner-sv-ac@verilog.org [mailto:owner-sv-ac@verilog.org]
On Behalf Of Kulshrestha, Manisha
	Sent: Tuesday, June 06, 2006 11:50 AM
	To: sv-ac@verilog.org
	Subject: [sv-ac] FW: #805

	
	
	

	-----Original Message-----
	From: Kulshrestha, Manisha
	Sent: Thu 6/1/2006 12:17 PM
	To: 'sv-ac@eda.org'
	Subject: #805
	
	
	Hi All,
	
	I have incorporated most of the feedback in my updated proposal
(attached here). Please send your feedback. The original proposals are
still on mantis in case you want to compare. The main change in this
document is that disabled is not a success and thus very few changes are
needed in the LRM.
	
	Thanks.
	Manisha
	
	805: disable iff condition should produce vacuous match
	Lisa:           I agree with this too - failure counters do not
make
	                        sense for coverage.
	                        failure counters do not make sense for
coverage.
	Joseph:         yes
	Doron:          I think that disabled should not count as a
success 
	                        in coverage. we need to change is the
report of the 
	                        number of failures in coverage
	Bassam:         yes
	Dmitry:         I don't think the failure should be reported for
coverage at all.
	Surrendra:      yes
	Ed:                     No success with disable to be reported.
	Dmitry:         I agree with the definition of the vacuous
success.
	                        According to our discussion about the
property
	                        coverage definition, there is no meaning
in disabled
	                        coverage success, since it should not
count as a
	                        coverage event at all. 
	                        Therefore the suggestions concerning
	                        Clause 17.13.3, page 288,
	                        Clause 29.4.3, page 482, Clause 29.4.2,
page 481,
	                        and Annex I are not relevant.
	                        I agree with the proposal concerning
Clause 28.4.2.
	Volkan:         yes
	
	
Received on Fri Jun 9 16:23:40 2006

This archive was generated by hypermail 2.1.8 : Fri Jun 09 2006 - 16:23:47 PDT