Branch transition rate: a new metric for improved branch classification analysis

Abstract

Recent studies have shown significantly improved branch prediction through the use of branch classification. By separating static branches into groups, or classes, with similar dynamic behavior, predictors may be selected that are best suited for each class. Previous methods have classified branches according to taken rate (or bias). We propose a new metric for branch classification: branch transition rate, which is defined as the number of times a branch changes direction between taken and not taken during execution. We show that transition rate is a more appropriate indicator of branch behavior than taken rate for determining predictor performance. When both metrics are combined, an even clearer picture of dynamic branch behavior emerges, in which expected predictor performance for a branch is closely correlated with its combined taken and transition rate class. Using this classification, a small group of branches is identified for which two-level predictors are ineffective.

Modeling program resource demand using inherent program characteristics

“…Note that in order to prevent branch filtering from aggressively impacting the prediction accuracy, we ensure that the total number of filtered dynamic branches is less than 0.1% of the total dynamic branches. Note also that the transition rate buckets used in determining the history length are consistent with those used in branch classification by Haungs et al [10].…”

“…The remaining part is the number of branch misprediction, which is difficult to accurately estimate purely based on program characteristics. However, the branch transition rate proposed by Haungs et al [10] contains a clue as to how many branches are hard to predict, and allows us to roughly estimate the number of mispredicted branches. Branch transition rate measures the frequency at which a branch changes direction between taken and not taken.…”

Workload Modeling for Computer Systems Performance Evaluation

“…For example, a branch that is alternately taken and not taken is completely predictable, whereas one that is randomly taken with the same probability of 50% is not. Therefore the taken rate should be complemented by the transition rate, which quantifies the probability that the branch transitions from being taken to not being taken [321]. Branches with either very low or very high transition rates are highly predictable, regardless of their taken rate.…”

Energy-aware application scheduling on a heterogeneous multi-core system

“…To capture the branch predictability of the program, we use branch transition rate, which is demonstrated to be an appropriate metric for the branch predictability of the program by Huang et.al [14]. Generally speaking, the branch instructions with extremely low and extremely high transition rate are easy to predict since the branch history pattern of these instructions could be captured with short history registers.…”