Nice explanation, doing a step-by-step refactoring in C++. So for memory-bound performance optimization, you indeed see that halving memory accesses doubles throughput in practice, more or less.

Curious if this kind of loop fusion would be something you get for free in purely functional language? That is, if you were to write a program (say in Haskell) for min, max plus that same code combined into a single program, would you indeed empirically see double the speed?