12
votes

According to ARB_geometry_shader4 it is possible to render a scene onto the 6 faces of a cube map with a geometry shader and the cube map attached to a framebuffer object. I want to create a shadow map using this way. However there seems to be a conflict that I can't resolve:

  1. I can only attach a texture with GL_DEPTH_COMPONENT as internal type to the GL_DEPTH_ATTACHMENT_EXT.
  2. A depth texture can only be 1D or 2D.
  3. If I want to attach a cube map, all other attached textures must be cube maps as well.

So it looks like I can't use any depth testing when I want to render to a cube map. Or what exactly am I missing here?

EDIT: It looks like newer Nvidia drivers (180.48) support depth cube maps.

1
I'm not sure exactly what the rendering to cube map has got to do with the geometry shader, it feels completely independent, nevertheless, I have no answer.falstro
The geometry shader can create 6 output triangles for every input triangle and redirect it to each of the cube map faces (see "Layered rendering"). Without a shader I have to attach each face and render the geometry 6 times.Maurice Gilden
I'm wondering, could you paste the code which creates the frame buffer with the cube texture? I'm interested about this myself.shoosh
Why not just use 6 separate frame buffers for each of the sides of the cube? How are you binding the different sides of the cube map from inside the shader?shoosh

1 Answers

12
votes

Ok, to answer some other questions here:

Of course it is possible to use 6 FBOs, one for each face. Or to use one FBO and attach each face before you draw to it. In both cases the cube map face will be treated like any other 2D texture and you can use it together with normal 2D textures or Renderbuffers. And there's probably not much of a difference in all the possible ways (if the hardware supports them).

However, it's also possible to draw everything in one step and as I was curious as to how this is done I did some research.

To create a FBO with all faces of a cube map attached to a single attachment point I used this code (written in D):

// depth cube map
glGenTextures(1, &tDepthCubeMap);
glBindTexture(GL_TEXTURE_CUBE_MAP, tDepthCubeMap);
glTexParameterf(GL_TEXTURE_CUBE_MAP, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
glTexParameterf(GL_TEXTURE_CUBE_MAP, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
glTexParameterf(GL_TEXTURE_CUBE_MAP, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
glTexParameterf(GL_TEXTURE_CUBE_MAP, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
glTexParameterf(GL_TEXTURE_CUBE_MAP, GL_TEXTURE_WRAP_R, GL_CLAMP_TO_EDGE);
for (uint face = 0; face < 6; face++) {
    glTexImage2D(GL_TEXTURE_CUBE_MAP_POSITIVE_X + face, 0, GL_DEPTH_COMPONENT24,
        width, height, 0, GL_DEPTH_COMPONENT, GL_FLOAT, null);
}

// color cube map
glGenTextures(1, &tColorCubeMap);
glBindTexture(GL_TEXTURE_CUBE_MAP, tColorCubeMap);
glTexParameterf(GL_TEXTURE_CUBE_MAP, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
glTexParameterf(GL_TEXTURE_CUBE_MAP, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
glTexParameterf(GL_TEXTURE_CUBE_MAP, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
glTexParameterf(GL_TEXTURE_CUBE_MAP, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
glTexParameterf(GL_TEXTURE_CUBE_MAP, GL_TEXTURE_WRAP_R, GL_CLAMP_TO_EDGE);
for (uint face = 0; face < 6; face++) {
    glTexImage2D(GL_TEXTURE_CUBE_MAP_POSITIVE_X + face, 0, GL_RGBA,
        width, height, 0, GL_RGBA, GL_FLOAT, null);
}

// framebuffer object
glGenFramebuffersEXT(1, &fbo);
glBindFramebufferEXT(GL_FRAMEBUFFER_EXT, fbo);
glFramebufferTextureARB(GL_FRAMEBUFFER_EXT, GL_DEPTH_ATTACHMENT_EXT, tDepthCubeMap, 0);
glFramebufferTextureARB(GL_FRAMEBUFFER_EXT, GL_COLOR_ATTACHMENT0_EXT, tColorCubeMap, 0);

glDrawBuffer(GL_COLOR_ATTACHMENT0_EXT);

if (!isValidFBO()) {
    glDeleteFramebuffersEXT(1, &fbo);
    fbo = 0;
}
  • If you want to have only a depth map you have to change glDrawBuffer(GL_COLOR_ATTACHMENT0_EXT); to glDrawBuffer(GL_NONE); before validating it (and before drawing to it)
  • MIN and MAG filters must be set to something valid (default would be GL_NEAREST_MIPMAP_LINEAR)
  • width and height of all textures must be the same

To render to the faces of a cube map you need a geometry shader. The following shader misses some rotations but it should be clear what it does. gl_Layer is used to direct the primitive to the correct face (0 = +X, 1 = -X, ...).

#version 120
#extension GL_EXT_geometry_shader4 : enable

void main(void) {
    int i, layer;
    for (layer = 0; layer < 6; layer++) {
        gl_Layer = layer;
        for (i = 0; i < 3; i++) {
            gl_Position = gl_PositionIn[i];
            EmitVertex();
        }
        EndPrimitive();
    }
}