Branch Path Re-Aliasing

Deeper pipelines improve overall performance by allowing more aggressive clock rates, but some performance is lost due to increased branch misprediction penalties. Ironically, with shorter clock periods, the branch predictor has less time to make a prediction and might have to be scaled back to make it faster, which decreases accuracy and reduces the advantage of higher clock rates. We show how the lost accuracy and performance can be reclaimed and even increased, and we show how the predictor can be simplified in the process. We present a new technique, branch path re-aliasing, that moves complexity off of the critical path for making a prediction and into the compiler. The key component of our technique is a hint bit set in each branch instruction that aids in reducing destructive aliasing; unlike other such techniques, the hint bit is not needed until the less critical update stage, so there is no extra delay in accessing the branch predictor. When augmented with our approach, a simple 2048-entry GAg predictor achieves a misprediction rate of 6.5%, 21% lower than the 8.2% misprediction rate of a more complicated gshare predictor of the same size.