Hi JS Just little FYI , you can build texture atlases that repeat the caveat being they can only repeat in one direction so you can build vertical and horizontal sets :)
I would also say the 1k verts per primitive set is lowish I would say you can look at 10k or more on modern hardware ( assuming you model(s) that many) Gordon Tomlinson 3D Technology System Engineering Consultant Overwatch® An Operating Unit of Textron Systems __________________________________________________________ "WARNING: Documents that can be viewed, printed or retrieved from this E-Mail may contain technical data whose export is restricted by the Arms Export Control Act (Title 22, U.S.C., Sec 2751, et seq,) or the Export Administration Act of 1979, as amended, Title 50, U.S.C., App. 2401 et seq. and which may not be exported, released or disclosed to non-U.S. persons (i.e. persons who are not U.S. citizens or lawful permanent residents ["green card" holders]) inside or outside the United States, without first obtaining an export license. Violations of these export laws are subject to severe civil, criminal and administrative penalties." -----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of Jean-Sébastien Guay Sent: Monday, January 24, 2011 9:36 AM To: [email protected] Subject: Re: [osg-users] Optimizing scene structure and geometry Hi Tim and Robert, OK, a few great suggestions, and a possible answer to why the work I've done hasn't sped things up that much. First, I've been focusing mostly on automatic optimizations that I can code up in some processing tool since we already do that now (we use osgconv to convert all our models to ive, so in the past week I've been writing my own tool that uses parts of osgconv, the osgUtil::Optimizer and some slightly modified classes to give better results on our data). My goal was to be minimally intrusive on my artist's work. But I think I'm at a point now that I'll have to get him to modify some things... For the number of transforms, I doubt I'll be able to do much to lower it, since it's a central part of our infrastructure that physics objects are attached to transforms (with the physics object's transformation being copied to the transform node each step). We might be able to get some other kind of relationship, but not in the short term. So I'll have to work on some other areas. Reducing the number of statesets is a worthy goal, I've already tried (building texture atlases mostly). I've gone from over 2500 statesets to 1345 in the stats I gave in my OP. The next steps will be to get my modeler to remove most parts of models set to render both sides (cull face off) and as Tim suggested removing textures set to REPEAT since those are two things that defeat atlas building sometimes. Using the INDEX_MESH | VERTEX_POSTTRANSFORM | VERTEX_PRETRANSFORM optimizers is a good idea, up until now I was compiling my tool against the same version of OSG as our simulator uses (2.8.3) though, so I couldn't use them. I'll see if I can backport them to that version and see if they give better results. About the state graph number. I guess that's pretty much proportional to the number of unique statesets in the graph? I'm at 1345 now, I guess if I get that number down it should reduce the number of state graphs? Tim mentions keeping a figure of about 1000 verts per primitive set in mind. That's good info. It's really hard to find numbers that you can relate to the stats you see in the OSG stats display. Thanks a lot, I'll keep working on it and keep you updated. J-S -- ______________________________________________________ Jean-Sebastien Guay [email protected] http://www.cm-labs.com/ http://whitestar02.webhop.org/ _______________________________________________ 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
