Is it a raymarching loop? How do you cope without the help of `normalize`?

Usual implementation of raymarching uses incremental updates along the ray equal to SDF (SDF is thought as a minimal distance to the object). It requires exact parametrization of the ray according to Euclidean norm. This implementation only requires, that SDF is positive outside the object and negative inside. So the point moves along the ray with the speed proportional to the SDF. It works like a feedback-loop system and does not rely on measuring exact Euclidean distances along the ray. So, this way we can get exact final point position without knowing ray vector norm.