ATI pbuffer slowdown

OpenGL OpenGL: Advanced Coding
1

Hi all;
Im doing a glow fx using pbuffers, and in NVIDIA hw no problems, but in ATI the fps goes down from 60 to 0.05 …im doing this way:

before loading any texture:

  1. Create a pbuffer_blur 256x256
  2. Create a pbuffer_glow 256x256
  3. Initialize the VBOs and other texures
    -loop-
  4. Render everything to pbuffer_glow, with tex env set to rgb*alpha
  5. Bind pbuffer_glow (as texture_2d) and render it with radial blur to pbuffer_blur
  6. Render the scene
  7. Draw a quad over the whole screen with pbuffer_blur

Im clearing only the depth_buffer_bit of the pbuffer_glow (im rendering a black retangle on the viewport to clear it), and the slowdown is on the step 7…using Visual Studio and debugging, pressing ctrl-f10 (to go to next instruction) last about 3 seconds just for the glEnd() command of the quad rendering…is something wrong with the order oc pbuffer creation, or size of them?

2

hmmm…i find out the problem…im setting the mip level of pbuffers with

glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAX_LOD_SGIS,1);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_LOD_SGIS,1);

,and when i comment it out, the fps return to 60-70…is there any problem using this extension in ATI?

3

My guess is that they have not optimized pbuffers for the scenario of doing mip-map generation. So they are probably just downloading the data from the card into main memory and then re-uploading it to do the generation.

4

Mipmap generation should be fine on ATI hardware. It’s been accelerated for a long time. It seems the problem rather is with clamping the LOD, which I btw wonder what the purpose of that is.

5

yep normal mipmap generation for pbuffers does work fine on my radeon 9700pro… guess really the clamping is the problem

6

just another guess: is the pixelformat for the pbuffers the same as for you mainwindow ?

7

The pixel format is not exactly as the main window one…
int attr_s[] =
{
WGL_SUPPORT_OPENGL_ARB, TRUE, // pbuffer will be used with gl
WGL_DRAW_TO_PBUFFER_ARB, TRUE, // enable render to pbuffer
WGL_BIND_TO_TEXTURE_RGBA_ARB, TRUE, // pbuffer will be used as a texture
WGL_RED_BITS_ARB, 8, // at least 8 bits for RED channel
WGL_GREEN_BITS_ARB, 8, // at least 8 bits for GREEN channel
WGL_BLUE_BITS_ARB, 8, // at least 8 bits for BLUE channel
WGL_ALPHA_BITS_ARB, 8, // at least 8 bits for ALPHA channel
WGL_DOUBLE_BUFFER_ARB, FALSE, // we don’t require double buffering
0 // zero terminates the list
};
i dont requesst for stencil bits and double buffering…
and the texture creation is

int pAttrib[] =
{
WGL_TEXTURE_FORMAT_ARB, // our pbuffer will have texture format RGBA
WGL_TEXTURE_RGBA_ARB,
WGL_TEXTURE_TARGET_ARB, // of texture type GL_TEXTURE_2D
WGL_TEXTURE_2D_ARB,
WGL_MIPMAP_TEXTURE_ARB, TRUE,
0 // zero terminates the list
};

…
crate, share lists…
…
glGenTextures(1, &pbuffer_ID);
glBindTexture(GL_TEXTURE_2D, pbuffer_ID);
glTexParameterf(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
glTexParameterf(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
glTexParameterf(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
glTexParameterf(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR_MIPMAP_LINEAR);
glTexParameteri(GL_TEXTURE_2D, GL_GENERATE_MIPMAP_SGIS, GL_TRUE);

can anybody explain me what is wrong with the clamping?