Menu
▾
▴
Re: [Visualpython-users] From newbie: major performance woes
|
From: Jiang Q. <jq...@ph...> - 2007-11-08 02:37:56
|
On Wed, Nov 07, 2007 at 10:11:32AM -0500, Bruce Sherwood wrote: > The beta version is called the beta version because there are problems > with it, and it's helpful to have your report of what doesn't work; > there have been almost no reports from Linux users. I normally work on > Windows but have tested the Linux beta version against a variety of > applications. I had concluded that the beta version mostly works okay on > Linux, while there are serious bugs on Windows. You're reporting that it > isn't okay even on Linux, which I was not aware of. In addition to the testing I reported to be not working properly, the stonehenge example crash the beta immediately after launch. And hanoi.py shows a lot of artefacts and becomes slow and pretty much unusable. > > My own testing was done on a more recent version of Ubuntu, though that > seems an unlikely issue, and your hardware is certainly adequate > (presumably your 1.86MHz processor is a typo; it must be 1.86GHz). Yes of course that's a typo. Now one worry about my ancient ubuntu is that my libboost version is still 1.31. Would that be a problem? > > Because I'm trying to get into working on the problems with the beta > version myself, it could be helpful if you would send me an example of a > program where interactivity is compromised, to add to my suite of test > programs. Please find in this URL a program of visualizing 9288 spheres in a 27 unit cell random non-overlapping distributions. Colors marks the grain boundaries. It the stable version, I can rotate the pack at will, zoom in, etc. This is so even when I improve the sphere rendering smoothness by decreasing the two if(size <0.02) if(size<=0.125)... lines to 0.004 and 0.125 respectively. The stock stable 3.2.9 presumably would be even faster. However, in the beta version, I cannot rotate the pack at will, interactivity becomes inadequate. > > There isn't a simple patch to make the production version have > transparency. There was a major change in the fundamental architecture > of Visual to make transparency a possibility. > > As stated in the description of the points object, these are circles or > squares, not spheres, of particular use in plotting graphs. They may or > may not be of any use to you. I would think that they're a lot faster > than spheres. Too bad. > > The late Arthur Siegel did experiment a bit with specifying the level of > detail for the rendering of spheres. There is an undocumented attribute > of the "display" object: > > scene.lod = -3 # range is -6 (low level of detail) to 0 (high) > sphere() > > I don't like having negative values for "level of display" (lod), so > think of this as very experimental, but it will let you try some things. > Alternatively, since you evidently have a lot of experience, you could > make changes of your own to the internal workings of Visual to > experiment with the issues. Thanks much for your report. Unfortunately I really kind of need to actually use visual to help me intuitively grasp some problems in my research, so I can't afford to play it too much. So I decided to forgo transparency for now and revert back to the stable version. Thanks for the help, though. Jiang > > Bruce Sherwood > > Jiang Qian wrote: > >Hi: > > Sorry about the confusing subject line of my last email. I did > >manage to compile the beta, I just have major performance problem. > > I took a look at the code of sphere.cpp, and notice there's > >discussion about levels of details. I wonder, however, are all that > >exposed to the user? Is there anyway I can reduce the level of details > >from my script? Or if that's not possible, is it possible to in fact > >change the code manually and recompile? > > Another issue that may or may not have anything to do with my > >performance woes. There are several demos in the example directory does > >not run properly in my installation. Drape shows only a rod. Double > >pendulum is way slower than in 3.4, Hanoi shows artifacts whenever I > >move a ring. I'm beginning to think there might be something wrong with > >my beta installation. How can I check? > > Thanks in advance for your help. > > Jiang > > > >On Wed, Nov 07, 2007 at 01:35:57AM -0500, Jiang Qian wrote: > >>Hi all: > >> I'm new to this list. I installed the stable visual-3.2.9 and it > >>worked great. But I really would like to to get the transparency to > >>work. So I installed visual 4.beta16. I jumped through all sorts of > >>hoops, and finally got it to work. The transparency works quite well. > >>But I have serious performance issue. > >> I mainly use visual to visualize packed spheres, and small > >>particles inside the spaces between them. In visual 3.2.9, a system with > >>a few hundred spheres packed inside a cube works very smoothly: I can > >>rotate, zoom at will with middle and right button. By combining the two, > >>I can essentially "walk" into the sphere packs.(The spheres rendered are > >>not perfectly smooth, but I manually changed some lines in the source > >>code to make the smoothness ok). It all works fine, the only thing I > >>miss is transparency. > >> However, in visual 4.beta16, using identical code(with *no* > >>transparency yet), I get a spectacularly smooth and well lit spheres, > >>but the tradeoff seem to be heavy performance penalty: I can barely zoom > >>and move, the rotation is slow. Interactive zooming and rotation is the > >>key reason why I use visual: otherwise I can just render things in > >>povray. Needless to say, I would like to find a way to trade away some > >>of the more spectacular visual effects to have better interactive > >>performance. > >> I tried to look through the documentation of beta, but there > >>aren't too much info about performance tuning. The only thing I found is > >>to set shininess to 0, which helps marginally at best. > >> What other tricks can I use to improve interactive performance? > >>For example, can I tune the smoothness of the spheres? Sometimes my cpu > >>usage is not even fully 100%, but dragging, rotating and zooming the > >>image is still very slow. > >> It'll really be a shame if I have to revert to the old version > >>due to performance issue. There must be a way to balance performance > >>and effects. > >> In a related issue. I also need to render particles between the > >>spheres in the pack. In the old version, I basically render them as > >>smaller spheres. But in the new version, there's a mention of point > >>objects, which seems to be just for that purpose. However, there's not > >>too much documentation of the point objects. How do I use them? Do they > >>offer a lighter overhead than the render spheres? Can I give them > >>colors? > >> On the final note, a word about my system: I have a thinkpad > >>T43, with Pentium M Dothan 1.86MHz, 1GB PC2-4200 memory, ATI X300 mobile > >>discrete graphics(M22) with 64MB of VRAM, and 60GB 7200rpm disk. It's > >>not a powerful machine by today's standard, but it runs the old version > >>of visual just fine. Also, runs other 3D accelerated openGL programs > >>like googleearth and stellarium just fine(glxgears score, which I know > >>is not a benchmark, is around 1750 FPS). > >> The software environment is an outdated ubuntu(5.04), with > >>custom compiled 2.6.18 kernel, gcc version 3.3.5, python 2.4, fglrx > >>proprietary driver 8.42.3(latest). Numpy 1.0.3, the latest gtkglxtmm I > >>compiled and installed from source. If you want to know more, let me > >>know what more relevant info I can give. > >> If there's a simple patch for the stable version 3.2.9 that > >>enables transparency, I'll be very happy to use that, too. Because > >>really the stable version satisfies all my need except for > >>transparencies. > >> Thank you all for any hint/help you can give. > >> Jiang > >> > >>------------------------------------------------------------------------- > >>This SF.net email is sponsored by: Splunk Inc. > >>Still grepping through log files to find problems? Stop. > >>Now Search log events and configuration files using AJAX and a browser. > >>Download your FREE copy of Splunk now >> http://get.splunk.com/ > >>_______________________________________________ > >>Visualpython-users mailing list > >>Vis...@li... > >>https://lists.sourceforge.net/lists/listinfo/visualpython-users > > > >------------------------------------------------------------------------- > >This SF.net email is sponsored by: Splunk Inc. > >Still grepping through log files to find problems? Stop. > >Now Search log events and configuration files using AJAX and a browser. > >Download your FREE copy of Splunk now >> http://get.splunk.com/ > >_______________________________________________ > >Visualpython-users mailing list > >Vis...@li... > >https://lists.sourceforge.net/lists/listinfo/visualpython-users |
