By: Dan Downs (retnuh.delete@this.nospam.retnuh.org), January 19, 2011 5:36 pm
Room: Moderated Discussions
>My read of it is that the patent is focused on improving recovery from branch mispredicts
>and maybe holding small loops. The thing I'm curious about is how much you actually
>win by shaving 3-4 cycles from your pipeline on a misprediction caused restart....as
>opposed to something more generalized.
>
Is that cheaper and better power wise than really beefing up the branch predictor? A few more misses but the impact is reduced.
>and maybe holding small loops. The thing I'm curious about is how much you actually
>win by shaving 3-4 cycles from your pipeline on a misprediction caused restart....as
>opposed to something more generalized.
>
Is that cheaper and better power wise than really beefing up the branch predictor? A few more misses but the impact is reduced.
