I see that benefit. For code compiled appropriately. The article also thinks this will need to be per-process, so context switching gets even more expensive. M

I see that benefit. For code compiled appropriately. The article also thinks this will need to be per-process, so context switching gets even more expensive. Maybe the mask gets in the way of prefetching the pointer destination, though the machine might speculate that the pointer will be in canonical form. The arithmetic mask itself is very close to free. I guess someone thinks it's worth the silicon and software dev necessary to deploy. If it makes my code faster then I'll probably use it.