TestAutoClodMesh

justin · September 17, 2005, 5:54pm

Hy mojo,

i have made the cvs update and i noticed that TestAutoClodMesh.java gives geometric errors when zooming out. My OS is Linux Kernel v2.6.11 (SuSE v9.3) with an GeForce2 MX card. JDK is v 5.0 Update 4. I don't know if this errors were there before. Sadly i can't test it under Windows.

BTW i see that Update 5 is out.

EDIT:

The same with Update 5.

mojomonk · September 17, 2005, 6:35pm

Just tested it and saw no problems. Also tested TestClodMesh, and TestTerrain (which make use of clod'ing), with no problems.

Can you describe the problem you are seeing?

justin · September 17, 2005, 7:14pm

Yes, the errors start at a number of vertices lower than 540. Some vertices of the mesh go wild. When i go on zooming out, some vertices

justin · September 18, 2005, 9:31am

EDIT:

Ok i have updated to the newest nvidia driver (v1.0 7676) and it didn't help. I have tested TestAutoClodMesh and TestClodMesh in wireframe-mode and the geometric errors start in both with the first reduction (vertices go into the head). The bounding sphere remains constant.

I couldn't notice errors in TestTerrain.

TestDiscreteLOD works fine.

TestAutoClodMesh via webstart from the jme-site runs with no errors.

Would it help when i would make some screenshots?

mojomonk · September 18, 2005, 3:49pm

Would it help when i would make some screenshots?

please do.

renanse · September 18, 2005, 5:35pm

We had an issue like this on your old computer Mojo, iirc. I think it was related to sending more verts than used or something like that. I'll also look into that.

mojomonk · September 18, 2005, 6:49pm

yeah, that's what I was thinking. That's why I wanted the screenshot… so see if it looks like there are lines going off to (0,0,0).

justin · September 18, 2005, 10:32pm

renanse · September 19, 2005, 2:41am

Ok, I've replicated a similar problem. It happens when your card (or driver) does not support OpenGL 1.2 calls. Basically in that case it uses a gl call that doesn't provide a low and high index for what to use from the index buffer (so everything gets used.) I've fixed this by applying a limit to the float buffer in these cases.

My code is borked right now while I refactor bounds, so I'll send the changes to Mojo to implement when he has time. Once he gets them in, please test and let us know how it goes for you. It's possible we may need to do something similar to another part of the code.

mojomonk · September 19, 2005, 4:00am

changes renanse made are in. Test that out, please.

justin · September 19, 2005, 3:44pm

The problem is still there. Nothing has changed. The nvidia driver should support OpenGL up to version 2.0.

I just want to mention that the jme-webstart demo of TestAutoClodMesh runs without errors, but you know that already.

Here are some information i have got from the running nvidia driver. Maybe it helps.

GLX Information

renanse · September 19, 2005, 5:32pm

Yuck, GeForce 2MX… The driver may support 2.0, but I doubt your card supports anything above 1.3 (if that). In any case, I'll look into the other arrays and see what I can find out.

justin · September 19, 2005, 5:54pm

The driver may support 2.0, but I doubt your card supports anything above 1.3 (if that).

Yes, i think it only supports OpenGL 1.2.

renanse · September 19, 2005, 6:08pm

Cool, found another thing we can do that should help. Please give me a day or so though to finish bounds and get things in.

justin · September 19, 2005, 9:46pm

I have time and i am glad to see that you have an idea for a solution

justin · September 20, 2005, 6:14pm

It's working now XD !

Thank you.

renanse · September 20, 2005, 6:48pm

Good to hear.