Subject: [sv-cc] Minutes for 10/29/03 SV-CC Meeting
From: Duncan, Ralph (ralph_duncan@mentorg.com)
Date: Thu Oct 30 2003 - 16:38:02 PST
Minutes of 10/29 SV-CC Meeting.
ATTENDEES
. Swapnajit Mittra
. Andrzej Litwiniuk
. Joao Geada
. Ghassan Khoory
. Francoise Martinolle
. Michael Rohleder
. Doug Warmke
. Clint Olsen
. Bassam Tabbara
. John Stickley
. Ralph Duncan
. Abhinash (multiple spellings in prev. minutes)
REGULAR PROCEDURES:
Swapnajit called for accepting the previous minutes; Joao moved for
acceptance;
Ralph seconded.
A. NOVA DONATION DISCUSSION
Presumed typos were discussed by John, Bassam and Doug.
BASS: Donation's writer section intended to provide a shell for callback
mechanism on value changes, responding to user requests; intend to
capture and load data for various tools.
SWAP: Asked if the donation was not backwards compatible with some data
representations.
BASS: No, just an API; sidesteps representation issues.
JOHN: Wondered if the data could be any format.
MICH: Questioned whether the user could alternate between VCD file
generation
and writing to internal memory, noted signals selected for dumping
by VCD commands. Would one use VCD commands to select 50 signals
for
dumping? Must one generate a dump file?
BASS: Register callback with anything you want to dump and use write
commands. Noted there are various read mechanisms and that API
use
constitutes asking questions, does not necessarily involve
generating
a dump file.
JOHN If have these and $dump commands, how channel data to user and not
to file: desirable to intercept data stream whether headed for
memory or file.
BASS: Donation does not address this.
DOUG: Requests examples that show (a) reading thousands of signals (b)
for
many milliseconds, and (c) involving complex types. Examples
should
address performance concerns and might separately address c.
DOUG: Expresses concern that the donation expects the DB to organized in
a
given way, esp. for reader functionality (e.g., organized in
signal
fashion vs. time fashion).
BASS: API shaped by desire for simplicity and symmetry, rather than
assume
specific DB organization.
DOUG, BASSAM, JOHN and others discussed relative performance of donation
API with different kinds of DB organization.
FRAN, JOHN, ANDR?/JOAO discussed performance issues involving large
numbers
of calls and accesses: whether getting one value at a time means
making 1000s
of calls, whether that would matter and whether the implementor
would
aggregate data loads required by many calls for individual data
items
DOUG, FRAN, BASSAM and others discussed the performance merits of
read-oriented
callback-based schemes with the 'loop' near the DB, rather than
near user code,
whether one of the donation modes could use such a scheme.
DOUG: Asked for examples to address read-side performance issues by
providing
examples for reading a value change DB organized by time and one
organized
by signal id.
DOUG, JOHN and others discussed the benefits of a retargetable write API
that
could handle different DB formats; they pointed out that the
donation's
dependence on using VPI handles would prevent writing out values
not found
in the simulator DB.
BASS: Stated that their intent was not to restrict access to existing
data types.
B. ACTIONS
. Doug is slated to broadcast a revised donation.
. Francois argued for the LRM explicitly stating the effects of
disabling on
'ref' arguments and is slated to talk with SV-BC.
. Bassam is preparing a second revision of Nova's donation to address
issue
raised during the meeting and in related emails.
Ralph Duncan, 10/29/2003
This archive was generated by hypermail 2b28 : Thu Oct 30 2003 - 16:58:11 PST