Pbuffer and GL Error 1282

[michael.bauer]

Hi all,

when I’m trying to render into a (float) pbuffer (tried nvidia 61.11 and 61.06 driver under linux on an FX5950 ultra) I get some problems. I’m able to successfully create a pbuffer and glGetIntegerv says to me that it has
depth bits = 24
red bits = 32
alpha bits = 32
stencil bits = 8
double buffer = 0

Now comes the problem: When I try to render something I get an OpenGL 1282 error after executing the following lines of code

glColor3d(0,1,0);
glBegin(GL_LINES);
glVertex2d(0,0);
glVertex2d(1,1);
glVertex2d(0,1);
glVertex2d(1,0);
glEnd();

Everything works well with the same code if I set up a pbuffer with RGBA 8 bit FBConfig.
(On a Geforce 6800 everything works well with the same code).

Any Ideas what could be wrong?

For glXChooseFBConfig I use the following attribs:

int attrib[] = {
GLX_RENDER_TYPE, GLX_RGBA_BIT,
GLX_DRAWABLE_TYPE, GLX_PBUFFER_BIT,
GLX_DOUBLEBUFFER, 0,
GLX_FLOAT_COMPONENTS_NV, true,
GLX_RED_SIZE, 32,
GLX_GREEN_SIZE, 32,
GLX_BLUE_SIZE, 32,
GLX_ALPHA_SIZE, 32,
GLX_STENCIL_SIZE, 0,
GLX_DEPTH_SIZE, 0,
0 };

[bencbenc]

Yes FX5950 supports floating point pbuffers through the NV_float_buffer extension. 1282 is GL_INVALID_OPERATION and most likely indicates that there isn’t a fragment program bound. The NV_float_buffer extension requires that a fragment program to be bound when rendering to a floating point drawable.

Hope this helps… but not sure why it is simply working on the 6800.

Cheers,
–Ben

[michael.bauer]

Hi Ben,

you are right, I didn’t bind a fragment program. I hoped that the fixed-function pipeline could work with the float buffer. But on FX5950 it didn’t. I didn’t really need it but I cannot see a reason why it should not work.

But on the 6800 it worked without using a fragment program. It seems that nv40 produces floating point values anyway even without binding a fragment program. Am I right? This should be stated (if it is not already) somewhere in the extension or hardware specifications.

Thanks,
Michael

[bencbenc]

NV40 does indeed do 32bit floating point throughout the pipeline. This is burried in the HW specs, but I agree the docs could be updated to make this more clear. Glad things are working for you.

Cheers,
–Ben

[michael.bauer]

Hi Ben,

I’m glad to hear that NV40 does 32bit fp throughout the pipeline. But what about fp blending? I tried to do fp blending with a 32bit fp pbuffer but it was really slooooooow (seems to be software fallback). Is there a FBConfig that allows HW accelerated 32bit fp blending on NV40?

Michael