SNIP: Scaled Neural Indirect Predictor

This paper proposes an indirect branch predictor based on neural learning. Neural-based conditional branch predictors have been among the most accurate in the literature, so it makes sense to adapt them to the indirect branch prediction problem. However, it is not clear how to use a predictor optimized to produce a true/false output for a problem requiring the prediction of a branch target. My technique, Scaled Neural Indirect Predictor (SNIP) adapts the recently proposed SNAP predictor to predict several bits of the target address, then chooses a known target of the indirect branch with the most matching bits.