up (in order my post edits appear on the ML)
mp3butcher wrote: > Hello Robert, > I achieve a basevertex example and debug with nvidia gl debugger. > https://github.com/mp3butcher/OpenSceneGraph/tree/master/examples/osgSharedVAO > It appears that glBindVertexArrayObject seams to have a cost 500ns !? when > big bufferobjects are involved in the vas > Sure bos are remapped at each call... > > I added a check not to call glBindVertexArrayObject when currentvaobound > doesn't change and achieve the same perf as my rejected submission (a near 0 > draw submission overhead ) > > Code: > void VertexArrayState::bindVertexArrayObject() const { > if(_state->getCurrentBoundVAO()!=_vertexArrayObject){ > _state->setCurrentBoundVAO(_vertexArrayObject); > _ext->glBindVertexArray (_vertexArrayObject); > } > } > > void VertexArrayState::unbindVertexArrayObject() const { > no overhead here if(_state->getCurrentBoundVAO()!=0){ > _state->setCurrentBoundVAO(0); > _ext->glBindVertexArray (0); > } > } > ////and inlined mutable property get/setBoundVAO in osg/State > > > > It's not very elegant but it do the job. > > Bench with a DrawArrays low res sparse buildings scene: > (on a NVGTX660OC) > 1VAO/Geom: DrawCost 4 > shared VAOs: DrawCost 2 > > Bench with a DrawElementShort BigCity scene: > https://drive.google.com/file/d/0BxIH-jcsgYDdTG5ha21HZE1jX1E/view > 1VAO/Geom: DrawCost 9 > shared VAOs: DrawCost 4 > > > Any toughts about it? > > Cheers > > > robertosfield wrote: > > Hi Julien, > > > > On 6 February 2017 at 10:33, Julien Valentin <> wrote: > > > > > I remember your rejection, but it's not what I asked. > > > I would like your sentiment about what the book extract says about > > > efficient draw submission and basevertex based drawcall.... > > > > > > > Personally I have plenty of more pressing things to work on the OSG > > than chase after a possible small benefit. For OSG-3.6 my focus will > > be completing my work on fixed function -> shader pipeline and fixing > > bugs. > > > > Please remember these potential gains are in the wider context of real > > application with a wider range of different potential bottlenecks, > > typically the CPU and GPU bottlenecks be such that you may not see > > expected benefits. > > > > The right way to optimize graphics applications is benchmark test > > cases that most closely follow the needs that your > > application/OS/hardware with require when dealing with real datasets. > > Narrow benchmarks won't typically reveal what bottlenecks the full > > application and full data will see so you have to be very careful > > about extrapolating possible bottlenecks. With the full application > > and data tests you get a real picture of the what are the biggest > > bottlenecks and these are ones that you will see the most benefit from > > optimizing. > > > > Given this context I would be sceptical about how much value will see > > in sharing VAO and using base vertex. With anything that adds > > complexity to the API and implementation you need to make sure there > > is real value in added them to justify this complexity. > > > > Robert. > > _______________________________________________ > > osg-users mailing list > > > > http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org > > > > ------------------ > > Post generated by Mail2Forum > ------------------ Read this topic online here: http://forum.openscenegraph.org/viewtopic.php?p=70270#70270 _______________________________________________ osg-users mailing list [email protected] http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
