Static cube mapping is only several vector calculation in GPU at runtime. But dynamic cube mapping is different. The typical way to do this is to combine an updated static cube-mapping for each frame. The required action is to use camera to take 6 pictures as the new cube texture. It is much slower, and it is only a mock of Ray-tracing.
Here is my implementation finished within 4 hours:
