Hi serkan, hi J.P., Here is a link to a simple example solving your problem: http://www.cg.informatik.uni-siegen.de/data/Downloads/svt/osgComputeDemoApp.zip We render a rotating cow into a texture then run a cuda-kernel and display the result in the same scene.
You need to set the usage flags of your render targets to GL_TARGET, e.g. : Target0->setUsage( osgCompute::GL_TARGET_COMPUTE_SOURCE ). With this flag you tell the texture that it is used as a FBO. Unfortunately, OpenSceneGraph does not call any function of osg::Texture when it renders into a texture. In other words something like texture->applyAsFBO() is missing. So there is no way to get notified when this happens. The only way to deal with this is to copy the texture memory each time a mapping function is called (GPU->GPU which is still fast). However, for this the user needs to setup the usage flag. Additionally, with older drivers/CUDA versions (we are not sure about this) you have to call MRTTexture->unmap() after your kernel has finished. If you do not call unmap(), the MRTTexture is still mapped in the CUDA context and OpenGL cannot render into it successfully. Hopefully this helps, Jens ------------------ Read this topic online here: http://forum.openscenegraph.org/viewtopic.php?p=36373#36373 _______________________________________________ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
