I played plenty Quake3 in my time, but I wouldn’t call myself an expert. Still, of the missing textures in your version of the loader, there are only 2-3 that are actually just images… the rest are defined in the .shader files in base Q3 (caulk, hint, noshader, nodraw, missleclip, and lavahell_2000) and in the map we are using’s .shader file (orangesky and pdm_blocks17floor2_trans.) In case you are not familiar with the .shader file, it’s not a shader in the sense of frag and vert shaders, it’s a definition of how to apply, animate and or blend defined texture(s) on a face. Also, there are several things about Q3 textures that have alpha channels (such as making sure they are not back culled and applying a proper alpha blend) that are not handled in the converter.
I created a new class org.xith3d.test.benchmark.Q3FlightTest, which is a little slimmer than BSPLoaderTest (e.g. without the HUD stuff and without input). Please use this version for any benchmark.
Ok, I’ve mostly been waiting as a courtest for you guys to figure out your vertex/index issues on the loader before I put this latest benchmark finally together. Do you have an ETA on that?
Thank you very much for your patience, renanse. I’ve been very busy.
Well I’ve just fixed this issue ;D. At least it was really easy. It was due to the latest changes in the BSPVertex class, which I’ve reverted. They didn’t give any performance or GC boost, so they seemed to be useless in this sense.
Now, feel free to take the most recent version for your webstart tests. And please remember: take the org.xith3d.test.benchmark.Q3FlightBenchmark
Marvin
EDIT: If LWJGL is running faster for you (it doesn’t for me) then please use it for your tests.
You can grab the source I used there, as well as an open office doc of my raw numbers and some charts.
The lwjgl version of Xith wouldn’t run for me due to what appeared to be a opengl-thread issue. (To run in lwjgl mode, I simply changed the JOGL to LWJGL in the test class.) The frame rate of the jogl xith was about 160-165 ish, so about the same, even though the visual look has imroved (proper texture filtering, flame transparent, etc.) You’ll notice the FPS for jME went up a good chunk, that is because we improved our ability to lock down unchanging static sections of the scenegraph. I suggest the next benchmark be something involving animation or interactivity. Maybe capturing and playing back the activities of a hand full of actors?
Anyhow, putting down my benchmark hat for awhile… at least until the next test!
Getting the windows java3d libs involved running an installer… Is there a way to get the linux (or mac) libs for java3d without such a process? I was sleepy enough last night that I didn’t really care to try to find that out.
Well, it should be sufficient to point the the same folder or jar for windows or Linux (never tested it on max, but should be the same). The win libs and linux libs should then be provided at the same place.
Marvin
BTW: Please use the same source, you’re using at the moment.
I’ve never used Java3D. You unfortunately I cannot help you with it. But please start making xith and jme webstartable from linux. I’m very interested in the results on my machine.
Ok, I’ve repacked xith and jme with linux and mac native libs (sorry java3d.) I could only test mac, which works for jME but has the Tuple3f/Vector3f issue I mentioned before upon starting up in Xith. Upgrading my jvm solved that in windows, but there is not an official upgrade for mac above 1.5.0_06. shrug
Anyhow, the point was to get you guys linux native libs, so hopefully that is accomplished now. Same links as before.