Hi Wojtek, On Wed, Nov 3, 2010 at 8:54 AM, Wojciech Lewandowski <[email protected]> wrote: > I agree. But I thought it was the intent. Thats, how original code was doing > it. I only made sure it does not use unbound counter to cut down number of > vertices and relaxed the limits. In my opinion we could simply fix it by > using 32 bit indices.
Using 32 bit indicies means that the index array could end up be extremely large. It just isn't sensible to let it get so unbounded. > Maybe. I am unsure about instancing though. Function should work with > instanced geometry. Spliting the pack into sections, renders some of the > geometries out of order. Its rather unlikely we see the artifacts due to > rather large chunks but theoretically it may happen. On the other hand I > remember I did a search and not found drawQuads called with more than one > instance so maybe solution would be to get rid of instance count parameter > and do what you propose. As I wrote before the alternative solution could > use 32 bit indices but limit the arrays to the GL max vertex array and GL > index array limits. I may try doing one of above when I come back next week. > I have no access to code now. 32bit indices is not a viable solution. What to do about the instancing problem? Yep that's an awkward one. Perhaps we change the API so it doesn't support instancing and report an error when one attempts to use instancing in this combination. This code path is all about a fall back for OpenGL ES. Robert. _______________________________________________ osg-submissions mailing list [email protected] http://lists.openscenegraph.org/listinfo.cgi/osg-submissions-openscenegraph.org
