Re: udp scheduling semantics

From: Michael McNamara (mac@verisity.com)
Date: Mon Sep 15 2003 - 16:58:29 PDT

  • Next message: Stephen Williams: "Re: udp scheduling semantics"

    Precedence: bulk

    -- On Sep 15 2003 at 14:06, Stephen Williams sent a message:
    > To: mac@verisity.com, sharp@cadence.com, etf@boyd.com, cliffc@sunburst-design.com
    > Subject: "Re: udp scheduling semantics"
    > Michael McNamara wrote:
    >
    > > VCS did not schedule sequential UDPs using the NBA queue up through
    > > version 3 of VCS ( We implemented UDPs long before I implemented non
    > > blocking assigns, and we did not rewrite UDPs under my tenure). It is
    > > possible that Seq UDPs were re-written later in a manner that was
    > > visible to the user in terms of scheduleing semantics, but I find that
    > > very hard to believe. Customers really do not like to have tested
    > > designs get different answers.
    >
    > People *are* seeing differences between simulators when UDPs are
    > included in the designs. Some users I've talked to have written
    > UDP devices out of their libraries because they don't schedule
    > consistently from simulator to simulator.
    >

    This is quite distressing; and do please continue to drive this
    issue.

    Again, a significant example that shows that two different simulators
    are getting different results with the same test; AND the fact that
    the standard does not tell one which is right, or specifically leave
    this matter undefined, is pretty much the definition of an errata.

    This archive was generated by hypermail 2.1.4 : Mon Sep 15 2003 - 16:10:05 PDT and
    sponsored by Boyd Technology, Inc.