You could try embedding the two codes with an LLM and run any number of similarity measures on the output vectors.