Phil Keslin (philk++at++cthulhu.engr.sgi.com)
Thu, 18 Mar 1999 20:46:57 -0800
- Phil
Angus Dorbie wrote:
>
> Forgot to cc info-performer with my reply.
>
> This is the correct behaviour. All that the other processes need is the
> address returned by the allocations and they can read & write. The
> allocations are made from a shared arena.
>
> Cheers,Angus.
>
> Acosta, Mark W wrote:
> >
> > I'm seeing some unexpected behavior with my Performer app and shared memory.
> > Maybe someone here can enlighten me.
> >
> > In my application, I occasionally allocate chunks of shared memory (c api,
> > no Performer) and load some data into them. I also need to be able to access
> > this shared memory in the cull and draw phase of the application. If I'm
> > running in multiprocess mode, my assumption was that I would have to attach
> > to the shared memory segments in all app and cull processes since they were
> > created after Performer forks everything off. Well, a long time ago when I
> > originally did this, I could swear that this was the case. However, recently
> > while I was rewriting some stuff, I found that if I create shared memory in
> > the app process, that the cull and draw processes could access it without
> > explicitly attaching to it. This doesn't seem right to me. A forked process
> > will inherit attached shared memory segments from the parent but I wouldn't
> > expect the children to inherit any shared memory segments created after the
> > fork. I've tried this under IRIX 6.5 and 6.4 on an Onyx2 and several
> > Octanes. Can anyone confirm that this is the correct behavior?
-- Phil Keslin <philk++at++engr.sgi.com>
This archive was generated by hypermail 2.0b2 on Thu Mar 18 1999 - 20:47:00 PST