Binary instrumentation and recompilation of game logic to improve performance.

Fancy title, I know.

I have been interested in recompilation of game logic for specific processors for a while now.

It's probably a no brainer that games are compiled with generic compiler settings for a generic x86 processor, rather than optimizing for a specific microarchitecture(for obvious reasons).
It's also a no brainer that Intel beats AMD in performance when it comes to microarches prior to Zen. This makes a lot of old processors "useless" for gaming as games become more complex and thus require more CPU processing.

But what if you could either reduce or optimize a certain CPU intensive part of the logic so that it executes faster.

I understand that most games today might be more graphically intensive and thus require more GPU horsepower, and thus overclocking the GPU may provide more performance gain, but this won't be the case forever I think.

Battlefield 1 for PC has horrible performance, even on my i5-4670k. It requires so much CPU cycles, that the GPU is actually starved of data. DICE should fix this, but they haven't.