Do you know of any reliable technique to bust the training out of a branch predictor on current intel and amd CPUs?
I've been working on a Direct3d 11 renderer for some time now for our DAW, and made a tool to profile rendering. Right now it's rudimentary and re-renders the same "heavy" frame over and over.
I'd like to add some kind of "branch predictor busting" code in between the renders, because I'm suspecting I'm could bias the performance measurements by training the branch predictor.
=> More informations about this toot | View the thread | More toots from uucidl@hachyderm.io
text/geminiThis content has been proxied by September (3851b).