Data Value Predictor using Stride and Shift

초록

The conventional stride predictor is useful for predicting data values which are varied by a constant stride. However, when the data values of shift, multiplication, and division instructions are predicted, the stride predictor cannot show the feasible performance. Thus, we propose data value predictor using stride and shift to improve predictability. The data value predictor using stride and shift takes advantage of shift values as well as stride values, so that the overall coverage of prediction increases.