Also a point light needs the shadow-map to be rendered into a cube-map or at least dualparabolic-map.
Paul needs to implement cube-mapping or there are 2 other texture stages(for dualparabolic-mapping) gone again.
And I think it's nothing you would want to calculate every frame for an average mobile game.
Cube-mapping would be useful for other things like reflections/environment-mapping though.
Small excursion:
For directional shadow-mapping we iterate through the texels and cast a ray towards the sun.
If the sun is blocked by an object, we draw the texel in ambient color and the rest in light, we can check that by comparing the texel depth with the corresponding depth of the shadow-map.
It's the same for point lights, but as it would want to cast a shadow in all directions we need a shadow-map that can represent the depth of all objects viewed from the point light, so we can compare against it.
We can render the cube-map into 6 textures representing each side of the cube, which leaves us with only 2 texture stages for lets say diffuse and normal-mapping...
We could also write the 6 images into a single big texture but it's not possible to do it fast in AGK(I already tried it) and there is a technique in OpenGl ES called cube-mapping(surprise) that does this for us.
