Hi Andrey, There isn't any way for us to answer your question about why the performance is not what you are expecting given such limited information about your actual scene graph.
The best I can do is suggest general things like have you compiled a release build and tested this? Debug builds are many times slower than release builds. Also on the data front, how many vertices, primitives, separate geometry objects, nodes etc. do you have in your scene graph? Robert. On 6 January 2012 10:28, Andrey Ibe <[email protected]> wrote: > Hi, > > i finally got to cast some rays with my ray-casting application and i found > out the perforamce is terribly low. i am not sure how fast the intersections > should be counted, but i am pretty sure that an image 160x120 pixels (which > is 19200 points) should be computed in less than 5.5 seconds ... which is the > time i am getting when measuring. > > i literally copied the part of the osgintersection example that is using the > intersectorGroup. i am using the intersectionGroup and intersectionVisitor. > > Code: > osgUtil::LineSegmentIntersector* intersector = new > osgUtil::LineSegmentIntersector(osgUtil::LineSegmentIntersector::WINDOW, > osg::Vec3(x, y, 0), osg::Vec3(x, y, 1)); > intersectorGroup->addIntersector(intersector); > > this way i add intersectors to the group for each pixel of my window. > this way i try to compute the intersections and measure time > Code: > osgUtil::IntersectionVisitor intersectionVisitor(intersectorGroup.get()); > intersectionVisitor.setUseKdTreeWhenAvailable(true); > > osg::Timer_t startTick = osg::Timer::instance()->tick(); > > osg::Camera *cam = _wm->getSceneView()->getCamera(); > cam->accept(intersectionVisitor); > > osg::Timer_t endTick = osg::Timer::instance()->tick(); > std::cout << "Frame in " << osg::Timer::instance()->delta_s(startTick, > endTick) << " sec." << std::endl; > > > the time varies from 4.9 to 6 seconds for 160x120 pixel image, for 320x240 > the time increases 4 times to roughly 22 seconds ... though i am using a > notebook it is NOT any 86286 computer :) this way i won't be able to compute > full screen app window in reasonable time :( > > i tried the built-in KdTree support by calling > Code: > osgDB::Registry::instance()->setBuildKdTreesHint(osgDB::ReaderWriter::Options::BUILD_KDTREES); > > as the first command in my main() function. > As i didn't record any changes in the computing time, i even tried running > the KdTreeBuilder on my loaded model "manually" > Code: > osg::KdTreeBuilder *kdtb = new osg::KdTreeBuilder; > loadedModel->accept(*kdtb); > > still with NO preformance change. And by NO i mean NOthing, i am getting the > same runing time. i also tried limiting the amount o intersections by setting > the intersection limit for each intersector calling > Code: > intersector->setIntersectionLimit(osgUtil::Intersector::LIMIT_NEAREST); > > in the loop that creates one for every pixel of the window (the first code > snippet in this post). only the LIMIT_ONE option decreased the computing time > by roughly 0.3 seconds (which is still unacceptable), but the result i got > was not accurate. > > what am i doing wrong? why are the Kd-Trees not working properly? the > documentation says the performance should improve from tens to thousands > times. then why not for me? > > Thank you for any tips that may help me. > > Andrey > > ------------------ > Read this topic online here: > http://forum.openscenegraph.org/viewtopic.php?p=44655#44655 > > > > > > _______________________________________________ > osg-users mailing list > [email protected] > http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org _______________________________________________ osg-users mailing list [email protected] http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
