[sldev] SSE2 / Ouch

[Dirk Moerenhout]

I just thought of the fact that static there isn't the point. It's
outside scope of a class so it's there to limit the var to the file. I
should really polish up my C++ :-)

It doesn't make things easier as I still don't see how a simple change
can solve the issue. You need the correct arch to be enabled or the
class will not be compiled with the correct code. What if you make
descendant classes and overload the functions instead?

I'm starting to remember why I'm not fond of OO :)

Dirk aka Blakar Ogre

On 7/9/07, Dirk Moerenhout <blakar at gmail.com> wrote:
> Well I already noted in a previous mail I wasn't convinced that it
> would be safe to do it this way as everything you include gets handled
> with the same architecture. So you really need to go over all your
> includes in there and look what exactly you are doing and what the
> global impact will be.
>
> I guess the issue is that we have this:
> static LLV4Matrix4      sJointMat[32];
>
> That initialisation will probably use SSE.
>
> You can't just move that elsewhere because then no SSE code would be
> generated (as there's  #if LL_VECTORIZE in there).
>
> Tricky :) I suppose you can drop the static but it'll likely have
> performance impact and I'm not sure it's enough to stop MSVC from
> bleeding SSE instructions over to unexpected areas.
>
> Dirk aka Blakar Ogre
>
>
> On 7/9/07, Nicholaz Beresford <nicholaz at blueflash.cc> wrote:
> >
> > Dunno if it's true, but if it is (and this guy sounds like
> > he knows what he's talking about), it's a definite "OUCH!"
> > and explains a lot of reports about crashing on startup.
> >
> > https://jira.secondlife.com/browse/VWR-1610
> >
> >
> >
> >
> > Nick
> > _______________________________________________
> > Click here to unsubscribe or manage your list subscription:
> > /index.html
> >
>