How does opengl decide which mip level to use?

Question

The question is pretty clear. I ask in terms of using texture coordinates that could come from anywhere (homogeneous, variable, different texture selection).

Say, for example, I sample a texture on a mipmapped (or anisotropically filtered) texture, and I use the square of the variable that was set in the vertex shader. I suppose glsl cannot determine the derivative of an arbitrarily complex function like this, since it knows which mip level to use?

Thanks.

Answer 1

Spatial coherence is commonly used. It calculates the same value on a block of at least 2x2 pixels and calculates the discrete difference between neighbors. This is enough to approximate the derivative.