Screen space to world space

[Josh]

This took a very long time to figure out. I finally found a formula in Igor's SSLR shader that helped, then I added the camera position to finish it off. This returns the position in world (global) space of the current pixel being rendered in a fragment shader:

vec4 ScreenPositionToWorldPosition()
{
vec2 icoord = vec2(gl_FragCoord.xy/buffersize);
if (isbackbuffer) icoord.y = 1.0f - icoord.y;
float x = icoord.s * 2.0f - 1.0f;
float y = icoord.t * 2.0f - 1.0f;
float z = gl_FragCoord.z;
vec4 posProj = vec4(x,y,z,1.0);
vec4 posView = inverse(projectioncameramatrix) * posProj;
posView /= posView.w;
posView.xyz+=cameraposition;
return posView;
}

[shadmar]

Yeah we worked hard on that one, but Igor cracked in the end

[Josh]

The division by the w coordinate is beyond my understanding, although I have seen that in a few other algorithms like this. I think that has to do with the projection matrix not using (0,0,0,1) for the right-hand column, which introduces shearing into the equation. I don't know enough about sheared matrices to be able to visualize it like I can an orthogonal matrix.

 

Not just shearing, but the increasing scale of the volume as you move along the z axis...aughhhh!