Yeah but you should also then understand what will yield real results, and >>1 instead of /2 has become useless to write manually a long time ago, whereas e.g. continuous memory is more important than ever. You will not optimize a lot by attempting micro techniques from 30 years ago.

Other random example: in some edge cases, integer division replacement by a multiplication can still be relevant today (depends on if its a constant, the compiler, and if nothing optimized also on the exact processor, though, because last models are already ultra-fast with the real integer divide instructions), but I suspect in 15 years (maybe even 10) this will be completely irrelevant, at least for high perf targets.