A history of branch prediction

The use of previous branch history and branch address as a "context" for prediction reminds me of the very similar technique used for prediction in arithmetic compression as used in e.g. JBIG2, JPEG2000, etc. --- the goal being that, if an event X happens several times in context C, then whenever context C occurs, the probability of X is more likely. Also, since modern CPUs internally have many functional units to which operations can be dispatched, I wonder if, in…

One fun thing to note is that the author doesn’t really talk a lot about anything from this decade, which makes sense considering how much complexity was being added for the tiny improvements towards the end. Modern branch predictors are quite complicated so they can eke out improvements in accuracy by fractions of a percent.

>We also didn’t talk at all about how different workloads affect different branch predictors. Predictor performance varies not only based on table size but also based on which particular program is run. One of my favorite little tradeoffs around this topic is around power usage. This is because, in many cases, adding more hardware (generally) increases total power usage. I.e. adding more execution units makes the processor faster at the expense of increased power usage. Good branch predictors, however, are…