Saturating counter design for meta predictor in hybrid branch prediction

  • Authors:
  • Young Jung Ahn;Dae Yon Hwang;Yong Suk Lee;Jin-Young Choi;Gyungho Lee

  • Affiliations:
  • The Dept. of Computer Science & Engineering, Korea University, Seoul, The Republic of Korea;The Dept. of Computer Science & Engineering, Korea University, Seoul, The Republic of Korea;The Dept. of Computer Science & Engineering, Korea University, Seoul, The Republic of Korea;The Dept. of Computer Science & Engineering, Korea University, Seoul, The Republic of Korea;The Dept. of Computer Science & Engineering, Korea University, Seoul, The Republic of Korea

  • Venue:
  • CSECS'09 Proceedings of the 8th WSEAS International Conference on Circuits, systems, electronics, control & signal processing
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

High-performance computer systems have made use of super-pipelining, dynamic scheduling and multiissue superscalar processor technologies. In these systems, branch prediction accuracy has a significant impact on the performance because the penalty for misprediction increases as pipelines deepen and the number of instructions issued per cycle increases. To have a better prediction accuracy, branch predictors utilized in high performance systems are a hybrid type. Hybrid branch prediction employs multiple branch predictors and selects one particular branch predictor per the program context of a given branch instruction instance for prediction. For choosing a particular branch predictor is a job of "meta-predictor". This paper considers type and size of saturating counter design specifically for the meta predictor. State transitions different from a usual saturating counter may choose a particular predictor for prediction better and result a higher prediction accuracy.