Subject: [sv-cc] SV-CC Meeting Minutes for September 24, 2003
From: Swapnajit Mittra (mittra@juno.com)
Date: Fri Sep 26 2003 - 09:07:30 PDT
[Fwd-ed from Ralph]
Minutes of 09/24 SV-CC Meeting.
ATTENDEES
. Swapnajit Mittra
. Andrzej Litwiniuk
. Joao Geada
. Francoise Martinolle
. Michael Rohleder
. Doug Warmke
. Ralph Duncan
REGULAR PROCEDURES:
Swapnajit called for accepting the previous minutes; Joao moved for
acceptance;
Ralph seconded.
A. DONATION CHANGES:
Citing time pressures for streamlining the process, Swapnajit called for
donation
proposals to be offered in LRM form and discussed in terms of LRM changes.
B. SPECIAL WORKING GROUPS:
Swapnajit stated that three working groups would address the three donations
in
parallel, working separately to reflect that the proposed donations are
technically disjoint.
C. BUG TRACKING
Bug tracking facilities, in the form of MANTIS freeware is envisioned for
SV-CC
use at the eda.org website.
. response to Francoise question: only for SV committee use.
. response to Joao question: Joao will maintain VPI spec. compliance in
'champion' role.
D. Andrzej PROPOSAL & DAVID SMITH CORRECTION
Doug surmised that D. Smith did not have major objections to the proposal.
Joao requested an
Andrzej action item to address Smith remarks. Swapnajit asked Doug to find
and re-broadcast
the relevant D. Smith email and asked Andrzej/Joao to address D.Smith's
objections.
E. DOUG'S DPI TASK PROPOSAL
DOUG: summarized rationale for import 'task' capability and nomenclature,
emphasizing the need
to allow native and DPI tasks/functions to be equivalent, while
preserving the restriction
that neither flavor of function can enable tasks.
JOAO: raised concerns about such C functions being thread-safe, e.g., WRT
using statics and globals.
DOUG: summarized required LRM changes that import tasks would entail.
ANDRZ:expressed performance concerns and skepticism on import task
necessity.
DOUG: contested the notion of innate tradeoff between this functionality and
performance;
presented C testbench motivation for import task feaure.
MICHL:recounted seeing example of the kind of C 'task' testbench usage Doug
described.
ANDRJ:argued that SV (task) functionality ought not be replicated through C
to support such testbenches.
DOUG: enumerated prospective syntax, fn-vs-task semantics and requirement
for C 'tasks' to be
VOID functions.
ANDRJ:asked if a master-slave relationship between SV and C were assumed.
JOAO: responded that System Verilog initialization must occur before any
'initial' block execution;
asked whether LRM would state C code constraints (i.e., for
thread-safety).
ANDRJ:asked whether a capability for specifying stack size would be
available.
JOAO: described thread schemes that involve constraints on user stack size
(e.g., 15K).
DOUG: continued the discussion of thread schemes and unpredicatability of C
portion stack consumption.
ANDRJ:mentioned stack management issues, e.g., single heap usage, lock-out
vs. copying a global stack.
JOAO: asked whether we should involve users in explicit stack sharing or
preallocation.
ANDRJ:voiced concerns that the proposed functionality would not allow
satisfactory performance.
MICHL:suggested that thread synchronization was a key performance issue.
DOUG: argued against including detailed C constraint information in the SV
LRM.
MICHL:questioned whether a C import 'task' returned if it called an SV task
that never finished.
DOUG: stated that such an import task would not finish and return.
JOAO: pointed to existing restrictions (e.g., forking count) and contended
that moving a threaded
model into a new environment would involve substantial C code
constraints for thread safety.
DOUG: posited that the principal, envisioned use would not involve reentrant
C code.
JOAO: suggested one would likely have multiple tasks running in parallel.
<discussion called at this point>
F. EPILOGUE
Swapnajit suggested that the above discussion continue through emails.
Given the intensity of the discussion, Ralph stiffled his remarks about
LRM section 26.4.1.2 errata and resolved to broadcast it in a relevant
email.
Ralph Duncan, 09/24/2003
________________________________________________________________
The best thing to hit the internet in years - Juno SpeedBand!
Surf the web up to FIVE TIMES FASTER!
Only $14.95/ month - visit www.juno.com to sign up today!
This archive was generated by hypermail 2b28 : Fri Sep 26 2003 - 09:10:11 PDT