I'm guessing the new disambiguated memory stuff will break dependent load ordering. If so, what will be the effect on some of the scalability techniques that use dependent load ordering if they have to start using LFENCE on all dependent loads?
I'm thinking this won't have any effect on dependent loads since in effect the pipeline is going to stall anyway until the load has completed. Is that basically right? I only have a P3 so I can't test it out for now.