Phone Conference P1450.1 Working Doc Subgroup

Thurs Nov 21, 10:00 am PST

 

Attendees:

Bruce Kaufman

Tony Taylor

Greg Maston

Doug Sprague

Peter Wohl

Jason Doege

 

Documents

p1450.1 D15 draft 10/7.

D15 review-resolution document 10/15.

Bruce's email, enclosed below these minutes.

 

Agenda

 

Agenda as posted, with addition of Bruce's email.


Meeting

 

Clause 6.9, scancells definition

Discussed the description added in this new section. Biggest question

about the constructs presented here is how to 'uniquify' names when a

scanchain is referenced with a instance name. Greg identified that at

the moment the actual process used is not required to be defined in

the spec because nothing actually references these

names. Unfortunately that moment did not last even the entire meeting

as this issue was revisited when scancellgroups was discussed later.

 

Tony raised a concern about referencing 'instances' of chains because

of multiple instances of modules, vs. redefining chains due to

"reconfigurable scan". These are two distinct usage models of this

data and the solutions are different: instantiating chains has a

shorthand reference to the previously-defined chain to be

instantiated; reconfiguring chains requires repeating the scancell

names in all configurations, at this time. There is no shorthand that

'references' a scanchain under a reconfigured definition at this time.

 

Clause 10.1: new definitions

No concerns raised about the existing definitions added by

Tony. However, Tony identifed a question about whether default

attribute values should be defined, and the Working Group felt this

was appropriate. Tony will add the following proposed defaults:

for integers: - usage test - init 0 - no enum reference

- integer values no bounds

for signalvariables: - defaults as for signal/group attributes

- initial value 'indeterminate'

AI[1] to Tony

 

LockStep

Reviewed the complete LockStep proposal now that Greg's example has

been partitioned into the intended 3-parallel-executions and everyone

can now understand what's happening there. No issues were raised with

this proposal although it was identified that it's complex, but so is

the issue it's addressing.

 

Bruce's email

Clarification proposed by Bruce (at end of these minutes) in regards

to highlighting that LockStep pattern synchonization processes padding

and shift-data-alignment differently than defined in dot-zero, was

accepted by the WG.

 

Jason raised a concern about pre-padding interactions of core

data. Pre-padding scan data (aligning scan data as part of the pattern

data itself) is unnecessary because STIL provides alignement

constructs, and may cause problems if that pattern data is then

re-used in a core environment and integrated into serial chains with

other data. Greg identified that the context of application of a STIL

data does affect the constructs present in that data, and having

constructs present that do not fit the application can invalidate the

usability of that data. Core test does have special requirements

(another one is that all operations in the patterns be performed

through only macro and procedure calls), and it is vital that these

issues be identified by the application contexts.

 

Tony proposed several grammer/clarifications to this section, all

accepted.

 

Clause 15.5

Reviewed the example created by Tony to present the need for the

ActiveScanChains statement. WG accepted this example with one request

to add a short comment on the Vector statements between Shifts, to

identify that these vectors are changing the scan-access to various

chains. Tony AI[2].

 

Review Resolutions Discussions

All outstanding Review Resolutions were reviewed. The following

review resolutions refer to issues/questions about the XRef statement

and failure-feedback mechanisms. These issues are under much

discussion at this point and are listed here to highlight the scope of

this discussion: DS-3, GR-8, GR-14, GR-21. [AI12] to address

fail-feedback and Xref issues.

 

DM-3: open for Jason to confirm and close.

 

DS-1:, DS-2: Doug and Greg to review and close, AI[5,6]

 

DS-3: WG to review section 17.4, in particular the second part which

is how to reference hierarchy of calls through the X statement. Tied

to [AI12].

 

DS-4: To close, recommended usage doc avaiable. Tony to put on web

(AI[3]).

 

GM-3: Greg to review and close, AI[7].

 

GM-9: Request was to move the Variables clause to the back of the

document. Discussion in the Working Group proposed to move the

Enviroment clause forward, and Greg accepted this as resolution. Tony

to reorder clauses, AI[4].

 

GR-8: Jason has been tagged to orchestrate some fail feedback

discussions. This is not happening at the moment due to real

work. This is related to XRef issues as well, [AI12].

 

GR-9: Expect that this issue has been clarified in the current doc,

close.

 

GR-11: LockStep has been clarified in the new doc; issue closed.

 

GR-12: WG feels that the definitions are sufficient, and requires

concrete issues if there are problems here. Close.

 

GR-13: Tony to review and identify disconnects, [AI9]

 

GR-14: X STMT - open issue [AI12]

 

GR-16: TONY to finish and close, [AI8]

 

GR-18: TONY to finish and close, [AI8]

 

GR-21: X STMT - open issue [AI12]

 

GR-22: Tony took [AI10] to review the issues here.

 

GR-23: See email in Appendix 1. Response: no, we do not need to

reserve all 'keywords', just those that start statements.

 

RK-1: done, close

 

RK-2: done, close

 

RK-3: scancell-groups concept. Working Group agrees to pull p1450.6

proposed syntax into .1. However this opens the issue of requiring a

definition for hierarchical scan cells referenced through the

instancing mechanism, in order to support referencing these cells in a

scangroup. Tony [AI11] to pull syntax in, ALL [AI13] to identify

instanced-scancell naming constructs.

 

TT-2: closed

 

WG-2: \e Greg [AI14] to complete.


Next meeting

next phone meeting Dec 5.

 

Meeting adjourned at 11:31 PST.

 

AIs

[AI1] Tony - define defaults for Integer and SignalVariables

[AI2] Tony - add comments to vectors in ex. in 15.5.

[AI3] Tony - put dot1 recommended usage doc on web.

[AI4] Tony - pull the Environment clause forward in the doc.

[AI5] Greg - review reference in DS-1 and fix if necc.

[AI6] Doug - review reference in DS-2 and fix if necc.

[AI7] Greg - review new words and accept or identify fix.

[AI8] Tony - finish and close GR-16,18.

[AI9] Tony - review the two contexts of the Fix statement.

[AI10]Tony - review issues in GR-22 and identify concerns.

[AI11]Tony - pull scangroup constructs from p1450.6 to here.

[AI12] ALL - Fail Feedback and Xref closure.

[AI13] ALL - consider proposals for hierarchical scancell naming.

[AI14]Greg - describe \e.

 


>From: "bruce_kaufman" <bruce_kaufman@teseda.com>

>To: "Tony Taylor" <Tony.Taylor@synopsys.COM>

>Subject: My Action Item

>Date: Thu, 07 Nov 2002 11:24:33 -0800

>

>Hi Tony,

>Here's my suggestion for a clarification. I added a parenthetical remark

>at the end of the 3rd paragraph in Section 13.3

>Feel free to modify.

>Bruce

>(3rd para with addition)

>-- For all shift blocks in parallel, the length of the scan data is

>defined as the longest data length of all signals with a "#" across all

>shift blocks. To facilitate this determination, it is required of the

>integration process that the signal/group definitions be adjusted such

>that the ScanIn/ScanOut-integer attribute defines the length to which each

>signal/group is to be padded, and this value shall be used rather than

>padding to the longest scan chain length of an individual pattern. (Note

>that determination of padding for LockStep is slightly different from the

>calculation of padding for Procedure and Macro Substitution (see IEEE Std

>1450-1999 Section 24.5 "Procedure and Macro Data substitution") which

>states that padding is based on the length of the longest chain.)