> Given that the entire linear chain of comparisons (therefore, every instruction) needs to be evaluated for 2^32-1, it would indeed need to read and execute th

> Given that the entire linear chain of comparisons (therefore, every instruction) needs to be evaluated for 2^32-1, it would indeed need to read and execute the entire 40GB file. It does, but not step by step the way the part I quoted described - the reading from memory can be pipelined with the execution. (And it's an ideal access pattern - reading linearly once - so it wouldn't surprise me if other optimisations kicked in)