Thurs Nov 7, 10:00 am PST
Attendees:
Documents
Agenda
Review the following changes made in D15:
1. ScanStructures->ScanChainGroups added to draft (clause 15). Pattern->ActiveScanChains
added to draft (clause 17).
2. The keyword ScanModule has been totally removed from the document
(clause 15 and Annex L3).
3. The ScanCells block in clause 15 has been modified to specify semi-colon
separators when a block is used. This resolves GM-1, GM-2, and GM-5.
4. Clause 6.3 has been eliminated and the scan-cell-list capabilities
are defined in clause 15.
5. Table 3 has been updated. See question re: variables and signals.
6. Added example to clause 6.2.
7. Added Greg's new LockStep stuff (clause 13.4 + Patern clause)
ScanStructures & ScanChainGroups
Started with a review of the changes to ScanStructures, starting at pg
37. Tony briefly covered the changed syntax.
Peter had an issue: Pg 39 15.2: duplicated intro paragraph. Tony will
fix - AI[1].
Discussion about the application of the ScanChainGroups, and the
ActiveScanChain statement. Discussion highlighted the need for an
example in the document. Tony took AI[2] on this.
Douglas identified some concerns between .6 constructs (and
references) and these constructs. He will tag some .6 members to
validate that there is concensus between these two areas.
'Scan Module' and hierarchical scan descriptions
Previously the document had used the term "scan module". This has been
removed, and examples of 'hierarchical scan application' cleaned
up. Tony started with a review of example on page 85/Annex L.3.
Douglas was confused that the bubbles in Figure 10 (without driving
triangles) were meant to represent inverters. Tony AI[3] to add
footnote to Figure 10 about the use of bubbles to represent inverters
in this diagram.
It was identified that the example in L.3 used single periods between
the chainname and instance name reference, yet the description under
bullet 5, pg 38 would indicate that single-colons should be present
here. Change example in L.3 to use SINGLE COLONS instead of
periods. Tony AI[4].
Move information from bullet item 5, pg 38, up to a new section under
clause 6 (re-insert 6.3), to define 'scancell names' or
something. Also add optional braces around (SCANSTRUCT ::) Tony AI[5].
Namespace table
Table 3: Discussed briefly the generality of the signals/variables
namespace, and the encompassing of many different constructs
here. Also discussed the last entry in this table at this time,
apparently to support a ScanCellGroup concept, which is associated
with some concerns that Douglas previously identified. Tony will add
this to the issues list as well so it doesn't get lost from this
review; AI[6]
section 6.2: boolean expressions
Clarifications look good; no questions from group. Tony identified
that currently we don't support '==' operations on wfc-list constructs
because there are too many forms of wfc-lists and Greg doesn't want to
define the words around what a canonical form of a wfc-expression
would be. Also by *not* supporting wfc-lists we enforce using
enumerated values for this purpose which makes the code more
clear/readable/dare-i-say-self-documenting...
LockStep
In 5 short minutes. Due to lack of time, Tony overviewed the additions
on LockStep. Jim made a request to align the text in the examples on
pg 32 (code lines 380-388 and 394-399); interpreting these constructs
(there are supposed to be 3 parallel definitions here) is complicated
unless the examples are in the right columns (yes, Tony AI[7]).
Jim asked about why the scan-data-padding/length-alignment process
here uses the ScanIn/ScanOut attribute value on signals instead of the
data length as used with dot-zero. It was fortunate that this was a
phone conference because Greg would have knocked someone out with his
hand-waving response. Response from Greg: it is perceived that the
attributes across all LockStep'ed signals will be easily available
during pattern processing, while having the data inside each of the
LockStep'ed patterns, available all at the same time, may not be
possible (consider an environment that uses a pass-per-pattern to
'interleave' all the LockStep'ed data together). By supporting
specification of the data length at the attribute level, each pattern
can be resolved independently of the other data present, (but still
dependent on the attribute values across all LockStep'ed
patterns). Greg is sitting down again.
Jim asked for perhaps some large, bold, bright red or possibly yellow
with a green background, wording here to indicate that in LockStep
there is a different data alignment process (Jim actually asked for a
parenthetical expression. The minute taker added the emphasis).
Ended this meeting on LockStep discussions. Again.
Next meeting
next phone meeting Nov 21.
Meeting adjourned at 11:00 PST.
Action Items
[AI1] Tony - remove one of the first paragraphs of 15.2
[AI2] Tony - add example of ActiveScanChain usage
[AI3] Tony - footnote on fig 10 about use of bubbles
[AI4] Tony - correct example L.3 to use single-colons.
[AI5] Tony - move item 5, pg 38 to a new subclause under 6.
[AI6] Tony - add ScanCellGroups questions to the dot1 issues list
[AI7] Tony - align the 3-parallel LockStep code examples