Hi Robert,
> It's a bit of long shot, but you could have a look at what the
> osgUtil::Optimizer is doing with the data, perhaps OSG-2.x is being
> more conservative with collapsing "duplicate" state.
Thanks, precious hint. I've just finished diff'ing the log messages emitted
by osgUtil::Optimizer, and I've found the problem: duplicate state
attributes aren't being shared because their data variance is marked as
UNSPECIFIED, whereas the optimizer expects it to be STATIC in order to
perform state sharing. The UNSPECIFIED state didn't even exist in OSG 1.2.
:-)
The obvious fix in my application is to add a STATIC_OBJECT_DETECTION flag
to the Optimizer, but it's not enough. It seems that
StateSet::computeDataVariance() only changes attributes' variance from
UNSPECIFIED to DYNAMIC if they have any callbacks; but if they have no
callbacks, then their variance is left unchanged. Is this a bug?
In StateSet::computeDataVariance(), changing the code:
if (itr->second.first->getDataVariance()==UNSPECIFIED &&
(itr->second.first->getUpdateCallback() ||
itr->second.first->getEventCallback()))
{
itr->second.first->setDataVariance(DYNAMIC);
}
to:
if (itr->second.first->getDataVariance()==UNSPECIFIED &&
(itr->second.first->getUpdateCallback() ||
itr->second.first->getEventCallback()))
{
itr->second.first->setDataVariance(DYNAMIC);
}
else // <----- ADDED this block
{
itr->second.first->setDataVariance(STATIC);
}
(and similarly for texture attributes) seems to bring memory usage back to
normal. Is this an acceptable fix?
BTW, performance is still slightly lower than expected. I think this has to
do with extended thread safety as you suggested, because my app is already
CPU-bound. Not a great issue though, at least if compared to running out of
memory.
Cheers,
Marco
_______________________________________________
osg-users mailing list
[email protected]
http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
