My own approach to this problem would be to consider each tile as a square polygon, then simply project the vertices outward from the light source. Then you take the resulting triangles and rasterize them on to an image scaled such that each pixel represents a tile and use antialiasing (or simply rasterize them on to an image 8 times as large as you need it and scale down, which is essentially the same thing). This gives you roughly the same end result, but would allow you to hardware accelerate almost every single aspect and avoids all the artifacts of shadowmapping. Notably, by restricting each polygon to a square, the projection could be done entirely as a vertex shader, without requiring geometry shaders. This would likely be vastly more efficient due to taking advantage of hardware acceleration.