But branch prediction doesn't necessarily need complicated logic. If I remember correctly (it's been 20 years since I read any papers on it), the simple heurist

But branch prediction doesn't necessarily need complicated logic. If I remember correctly (it's been 20 years since I read any papers on it), the simple heuristic "all relative branches backwards are taken, but forward and absolute branches are not" could achieve 70-80% performance of the state-of-the-art implementations back then.