Subject: Re: Meeting minutes for 1/8/03
From: Kevin Cameron x3251 (Kevin.Cameron@nsc.com)
Date: Wed Jan 08 2003 - 11:18:38 PST
>
> SV-CC Meeting Minutes for January 8, 2003:
>
...
> Direct access issues:::::::::::::::::::::::::::::::::::::::::::::::
>
> [This was a frenzied discussion that has not been captured well in these
> notes - sorry. JA]
>
> Andrzej discusses the direct pointer / access function issue.
> He notes that direct access can be a problem, but we must depend on smart
> users. There is no efficient way to do otherwise. This is especially true
> with arrays embedded within structures.
>
> Joao wants us all to decide now on this, as it is a fundamental issue. He
> reminds us that we decided earlier that we would have binary portability,
> and this can still be true even with direct access. And due to binary
> compatibility we cannot use macros to solve this issue.
>
> Kevin wants handle-based access, and notes that handles can be pointers for
> efficiency.
>
> Joao says a structure that contains an array must be laid out with the array
> embedded inline for System Verilog. This is why a handle can't be placed
> in-line.
Layout rules for C data have only tentatively been agreed on. There are no
rules on how SV data lays out in memory, it does not have to be contiguous or
ordered.
NB: Net data is legacy code in the simulators, C data is new, so while it
makes sense to force the layout of C data, net data layout is already
implemented and hard to change.
Kev.
> Joao - there are performance issues with very large arrays, but this can't
> be helped. In fact, even large structures will have performance issues.
>
> [... there was much more back and forth discussion on what is possible / not
> possible, and what the potential performance issues are]
>
> Doug and John assert that anyone passing a large structure/array back and
> forth will do so only if they are changing large amounts of data, like with
> an ethernet packet. Michael notes as a counter-example that he has seen
> ethernet packets passed around where only one small part changed.
>
> Joao thinks what we can stay with direct access and possibly change things
> later to address performance issues.
>
> A consensus was not built during the discussion, and Swanajit closes this
> topic for now. More emails need to be exchanged to bring this to a close.
> He wants to set up for a poll for Monday.
>
>
> John Stickley's queuable
> request::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
>
> John is withdrawing his request for the queuable attribute for now. He may
> propose it for the next API release.
>
>
> Joao's assertion
> status::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
> ::::::::::::
>
> His goal is to finish the updates and send out the spec by the end of this
> week. He says he still needs an update from Francoise. Then Francoise says
> the defines he needs are being addressed.
>
>
> Other misc
> issues::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
> :::::::::::::::::
>
> Joao is working with Dave Smith from sv-ec. Strings and pointers are being
> addressed. Externs have some conflicts that must be resolved.
>
> Francoise says that the unicodes are being withdrawn.
>
> Andrzej will combine all his proposals into a single document and send it
> out within the next day or two.
>
> Swapnajt says that for the next face-to-face we are still targeting 1/23,
> but this is tentative. Francoise is checking to see if Cadence can be the
> host.
>
>
> Polls:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
> ::::::::::::::::::::::::::::::::::::
>
> Poll on Joao's assertion for 1/14
> Poll for Andrzej proposal for 1/13
>
>
This archive was generated by hypermail 2b28 : Wed Jan 08 2003 - 11:20:12 PST