Re: [osg-users] Segfault in VTP, but looks like OSG: SOLVED
Thanks Robert, for include/osg/Config. Starting with OSG 2.6, the VTP can now check OSG_USE_FLOAT_MATRIX to detect whether OSG was built with float matrices or not. That makes it easy to handle both cases, so the user doesn't need it built one way or the other. It is not so surprising that the VTP prefers float matrices: 1. It is pointless to use the extra RAM (and memory bandwidth, and CPU time) for double matrices, when they aren't needed. 2. The rest of VTP's stack uses single matrices, so having OSG use doubles does not even gain theoretical precision improvement. It would have to be all doubles, top to bottom. 3. OpenGL itself is single-precision, so at best doubles would affect temporary, intermediate computations. 4. In many years, i have never encountered any rare situation that would be improved with double matrices. I'm certain they exist (among the many unusual uses of OSG) but since they are less common, it would make far more sense for single matrices to be the default. The VTP uses double-precision for GIS data, and single-precision for the 3d scenegraph. A design which requires the scenegraph to be double-precision is arguably.. an odd choice. -Ben -Original Message- From: Robert Osfield Sent: Saturday, September 13, 2008 2:29 AM Hi Pascal, I wonder if you could add something to the VTP build system to detect problems with Matrixd being used for Matrix. The other thing one might be able to do is adapt VTP so that it can handle Matrxf and Matrixd versions of Matrix. With OSG 2.6 onward there now is an include/osg/Config which includes details of the which version of Matrix is used, perhaps this might be of some help. As a general note, I've always been surprised by VTP using float Matrices, as GIS related app I would have expect double Matrices as it solves many of the precision problems associated with real world data. Robert. On Sat, Sep 13, 2008 at 9:40 AM, Pascal Rheinert [EMAIL PROTECTED] wrote: Hi, I got exactly the same error when trying to start VTP And I solved it now: The reason is (as expected) a discrepency between what OSG does by default and what VTP expects concerning the Matrix ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Segfault in VTP, but looks like OSG: SOLVED
Ben Discoe wrote: Thanks Robert, for include/osg/Config. Starting with OSG 2.6, the VTP can now check OSG_USE_FLOAT_MATRIX to detect whether OSG was built with float matrices or not. That makes it easy to handle both cases, so the user doesn't need it built one way or the other. It is not so surprising that the VTP prefers float matrices: 1. It is pointless to use the extra RAM (and memory bandwidth, and CPU time) for double matrices, when they aren't needed. 2. The rest of VTP's stack uses single matrices, so having OSG use doubles does not even gain theoretical precision improvement. It would have to be all doubles, top to bottom. 3. OpenGL itself is single-precision, so at best doubles would affect temporary, intermediate computations. As Robert states, the double matrix accumulation means that precision errors are eliminated on their way to the openGL model/view matrices. So if you are close to something, you get really good precision. If you are far away, you don't but it also doesn't matter. 4. In many years, i have never encountered any rare situation that would be improved with double matrices. I'm certain they exist (among the many unusual uses of OSG) but since they are less common, it would make far more sense for single matrices to be the default. Anecdotally, in one of our apps we could not draw city blocks on a float-based whole earth database because the buildings tended to stack on top of each other. Resolution was something awful (I remember it being as high as 100 meters). Double matrices fix that problem completely without anything else special. -Paul The VTP uses double-precision for GIS data, and single-precision for the 3d scenegraph. A design which requires the scenegraph to be double-precision is arguably.. an odd choice. -Ben -Original Message- From: Robert Osfield Sent: Saturday, September 13, 2008 2:29 AM Hi Pascal, I wonder if you could add something to the VTP build system to detect problems with Matrixd being used for Matrix. The other thing one might be able to do is adapt VTP so that it can handle Matrxf and Matrixd versions of Matrix. With OSG 2.6 onward there now is an include/osg/Config which includes details of the which version of Matrix is used, perhaps this might be of some help. As a general note, I've always been surprised by VTP using float Matrices, as GIS related app I would have expect double Matrices as it solves many of the precision problems associated with real world data. Robert. On Sat, Sep 13, 2008 at 9:40 AM, Pascal Rheinert [EMAIL PROTECTED] wrote: Hi, I got exactly the same error when trying to start VTP And I solved it now: The reason is (as expected) a discrepency between what OSG does by default and what VTP expects concerning the Matrix ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Segfault in VTP, but looks like OSG: SOLVED
Hi Ben, On Mon, Nov 10, 2008 at 8:11 AM, Ben Discoe [EMAIL PROTECTED] wrote: Thanks Robert, for include/osg/Config. Starting with OSG 2.6, the VTP can now check OSG_USE_FLOAT_MATRIX to detect whether OSG was built with float matrices or not. That makes it easy to handle both cases, so the user doesn't need it built one way or the other. Kudos to Matthias Froehilch on this useful feature :-) It is not so surprising that the VTP prefers float matrices: 1. It is pointless to use the extra RAM (and memory bandwidth, and CPU time) for double matrices, when they aren't needed. Um a small number of double Matrix is unlikely to ever be a memory issue. If you running on a 16k ZX Spectrum then I might say that this is a valid point, but these days with multi-gigabyte machines as standard. 2. The rest of VTP's stack uses single matrices, so having OSG use doubles does not even gain theoretical precision improvement. It would have to be all doubles, top to bottom. Perhaps it's time to change the VTP stack so it supports double matrices 3. OpenGL itself is single-precision, so at best doubles would affect temporary, intermediate computations. This is the key point to using double Matrices in the OSG, all the Camera and scene graph Transform nodes are accumulated in double matrices so that as far as possible the intermediate computation all cancel out as far as possible prior to being passed to OpenGL. This strategy works extremely well - allowing the OSG to work with full gecentric databases and still zoom into ground level and not see any jitter. 4. In many years, i have never encountered any rare situation that would be improved with double matrices. I'm certain they exist (among the many unusual uses of OSG) but since they are less common, it would make far more sense for single matrices to be the default. I guess you've never built a whole earth database yet... This is what VirtualPlanetBuilder does, and scales very well to multi-terrabyte databases, the use of double Matrices is central to the OSG being able to cope with the precision issues seamlessly. The VTP uses double-precision for GIS data, and single-precision for the 3d scenegraph. A design which requires the scenegraph to be double-precision is arguably.. an odd choice. Wow, like the year 2000 never ticked over. There are many OSG users that absolutely rely upon the use of double Matrices. All the vis-sim, the GIS apps, any app that does paged databases. All I can say is that you are out of touch with the needs of all these users. Robert. ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
[osg-users] Segfault in VTP, but looks like OSG: SOLVED
Hi, I got exactly the same error when trying to start VTP (vterrain.org) on my linux 64 bit (opensuse 11). It compiled but Enviro crashed whenever an osg-function was called from within VTP. And I solved it now: The reason is (as expected) a discrepency between what OSG does by default and what VTP expects concerning the Matrix: OSG-default is double (which can be changed via the OSG_USE_FLOAT_MATRIX value in CMakeCache.txt - by default it is off) and VTP expects FLOAT (and you should not change it there). So just do the following cmake . -DCMAKE_BUILD_TYPE=Release vi CMakeCache.txt -- change OSG_USE_FLOAT_MATRIX from off to on make make install and then compile the VTP-package. And it works great! Thanks for the splendid software (OSG and VTP)!!! -- Pascal ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Segfault in VTP, but looks like OSG: SOLVED
Hi Pascal, I wonder if you could add something to the VTP build system to detect problems with Matrixd being used for Matrix. The other thing one might be able to do is adapt VTP so that it can handle Matrxf and Matrixd versions of Matrix. With OSG 2.6 onward there now is an include/osg/Config which includes details of the which version of Matrix is used, perhaps this might be of some help. As a general note, I've always been surprised by VTP using float Matrices, as GIS related app I would have expect double Matrices as it solves many of the precision problems associated with real world data. Robert. On Sat, Sep 13, 2008 at 9:40 AM, Pascal Rheinert [EMAIL PROTECTED] wrote: Hi, I got exactly the same error when trying to start VTP (vterrain.org) on my linux 64 bit (opensuse 11). It compiled but Enviro crashed whenever an osg-function was called from within VTP. And I solved it now: The reason is (as expected) a discrepency between what OSG does by default and what VTP expects concerning the Matrix: OSG-default is double (which can be changed via the OSG_USE_FLOAT_MATRIX value in CMakeCache.txt - by default it is off) and VTP expects FLOAT (and you should not change it there). So just do the following cmake . -DCMAKE_BUILD_TYPE=Release vi CMakeCache.txt -- change OSG_USE_FLOAT_MATRIX from off to on make make install and then compile the VTP-package. And it works great! Thanks for the splendid software (OSG and VTP)!!! -- Pascal ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
