Would you need to use a similar mechanism when doing memory-mapped IO in a modern processor? From what I've seen (mostly in the PDP-11, which is admittedly quite old), synchronizing with an IO often involves reading flags from memory, which could be rendered ineffective by reordering.
Would you need to use a similar mechanism when doing memory-mapped IO in a modern processor? From what I've seen (mostly in the PDP-11, which is admittedly quite old), synchronizing with an IO often involves reading flags from memory, which could be rendered ineffective by reordering.