Hi Loinel, Are yo testing with release build?
On 6/12/07, Lionel Lagarde <[EMAIL PROTECTED]> wrote:
Hi, With the sources retrieved this morning, compiled with MS Visual 2005 express, the performances are the same. I'll bench more precisely the application to find the problem. Just a little question... I've tried the different threading models and I can't get in the cull task overlap the draw task in the statistics. I'm missing something ? Should I set an environment variable to specify the number of processors or something like that ? Robert Osfield wrote: > Hi Loinel, > > I see no performance drop with the SVN version of the OSG. Please > try upgrading. > > Robert. > > On 6/11/07, Lionel Lagarde <[EMAIL PROTECTED]> wrote: >> My samples.tar.bz2 is refused by the mail list server. >> >> The non compressed size is 60Mo. >> >> If someone want to retrieve the files, they can be downloaded at: >> http://perso.orange.fr/lionel.lagarde/samples.tar.bz2 >> >> >> Lionel Lagarde wrote: >> > I've built some test files. >> > >> > The attached archive contains: >> > - withoutshader.osg: a 20x20x20 3D grid of cubes >> > - withshader_0.osg: the same grid but with a basic shader in the root >> > node (no uniforms) >> > - withshader_1.osg: the same grid but with a basic shader in the root >> > node (1 uniforms) >> > - withshader_10.osg: the same grid but with a basic shader in the root >> > node (10 uniforms) >> > - withshader_20.osg: the same grid but with a basic shader in the root >> > node (20 uniforms) >> > - withshader_30.osg: the same grid but with a basic shader in the root >> > node (30 uniforms) >> > - withshader_100.osg: the same grid but with a basic shader in the >> > root node (100 uniforms) >> > >> > >> > The uniform are defined in the root node (with the shader). >> > >> > On my machine (Core2 duo 6400 + GeForce 7900GTX + OSG 1.2): >> > - withoutshader.osg: 25.7 fps >> > - withshader_0.osg: 23.2 fps >> > - withshader_1.osg: 23.1 fps >> > - withshader_10.osg: 22.3 fps >> > - withshader_20.osg: 21.2 fps >> > - withshader_30.osg: 20.1 fps >> > - withshader_100.osg: 11.7 fps >> > >> > >> > Lionel Lagarde wrote: >> >> Hi Robert, >> >> >> >> The uniform values are not changing per drawables. They are fixed in >> >> the root >> >> state set (it contains also the shader program). >> >> >> >> None of the children of the root node define uniforms (nor shader) >> >> >> >> But, per-drawable, the PerContextProgram::apply function is called >> >> for each >> >> uniform registered in the root node. >> >> >> >> >> >> >> >> Robert Osfield wrote: >> >>> Hi Loinel, >> >>> >> >>> High number of uniforms will be the bottleneck, changing uniforms in >> >>> OpenGL is relatively expensive. If you shader parameter various >> >>> changing per drawable then use vertex attributes rather than >> uniforms. >> >>> >> >>> Robert. >> >>> >> >>> On 6/11/07, Lionel Lagarde <[EMAIL PROTECTED]> wrote: >> >>>> It's seems that nobody has this problem. >> >>>> >> >>>> High number of drawables + high number of uniforms = bad >> performances >> >>>> >> >>>> >> >>>> >> >>>> >> >>>> Lionel Lagarde wrote: >> >>>> > Hi, >> >>>> > >> >>>> > Problem description: >> >>>> > We have a scene with a lot of drawables (>1500). The stateset >> of the >> >>>> > root of the >> >>>> > scene graph contains a shader program. >> >>>> > >> >>>> > When we increase the number of uniform variables used by the >> shader, >> >>>> > the application performance drop down. >> >>>> > >> >>>> > A little debugging session after, it appears that for each >> drawable >> >>>> > that has a stateset >> >>>> > defined, the uniform variables are applied. >> >>>> > >> >>>> > They are not really applied, (their modifiedCount has not >> >>>> changed) but >> >>>> > the >> >>>> > osg::Program::PerContextProgram (line 163) do a "find" on a >> std::map >> >>>> > (std::string -> GLint) >> >>>> > to retrieve the uniform location. >> >>>> > >> >>>> > It seems that it is this access to the std::map that cause the >> >>>> > performance drop. >> >>>> > >> >>>> > Does someone has encountered this problem ? >> >>>> > Is there a way to disable the uniform application for each >> >>>> drawables ? >> >>>> > >> >>>> > _______________________________________________ >> >>>> > osg-users mailing list >> >>>> > [email protected] >> >>>> > http://openscenegraph.net/mailman/listinfo/osg-users >> >>>> > http://www.openscenegraph.org/ >> >>>> > >> >>>> >> >>>> _______________________________________________ >> >>>> osg-users mailing list >> >>>> [email protected] >> >>>> http://openscenegraph.net/mailman/listinfo/osg-users >> >>>> http://www.openscenegraph.org/ >> >>>> >> >>> _______________________________________________ >> >>> osg-users mailing list >> >>> [email protected] >> >>> http://openscenegraph.net/mailman/listinfo/osg-users >> >>> http://www.openscenegraph.org/ >> >>> >> >> >> >> _______________________________________________ >> >> osg-users mailing list >> >> [email protected] >> >> http://openscenegraph.net/mailman/listinfo/osg-users >> >> http://www.openscenegraph.org/ >> >> >> > >> > _______________________________________________ >> > osg-users mailing list >> > [email protected] >> > http://openscenegraph.net/mailman/listinfo/osg-users >> > http://www.openscenegraph.org/ >> > >> >> _______________________________________________ >> osg-users mailing list >> [email protected] >> http://openscenegraph.net/mailman/listinfo/osg-users >> http://www.openscenegraph.org/ >> > _______________________________________________ > osg-users mailing list > [email protected] > http://openscenegraph.net/mailman/listinfo/osg-users > http://www.openscenegraph.org/ > _______________________________________________ osg-users mailing list [email protected] http://openscenegraph.net/mailman/listinfo/osg-users http://www.openscenegraph.org/
_______________________________________________ osg-users mailing list [email protected] http://openscenegraph.net/mailman/listinfo/osg-users http://www.openscenegraph.org/
