Exploration of high-performance branch predictors for softcore implementations
Exploration of high-performance branch predictors for softcore implementations
Branch predictors play a key role in improving the performance of pipelined processors, yet their design choices remain relatively unexplored for FPGA-based softcores. Although modern predictors such as TAGE achieve a low misprediction rate in algorithmic implementations, their complexity introduces significant challenges on FPGAs due to limited resources and routing constraints. This creates a design trade-off between prediction accuracy and hardware efficiency which is not well-examined in the existing literature.
This work presents a comprehensive evaluation of static, dynamic, and correlation-based branch predictors, including TAGE and TAGE with a simplified statistical corrector (TAGE-SC), implemented within a five-stage RISC-V softcore. All predictors are evaluated on an FPGA platform in terms of resource utilization, operating frequency, and misprediction rate. Our TAGE and TAGE-SC designs yield an average misprediction rate improvement of 3.75% and 5.29%, respectively, over pTage, an optimized implementation of TAGE for FPGAs. This improvement comes at the cost of additional hardware. Our results also indicate that, while a 2-bit predictor has a higher misprediction rate, it provides a good balance between accuracy and hardware cost. Our design space also explores finer-grain implementation choices, like the impact of the size and number of TAGE tables. To enable further research, all code is released as open-source.
Israr, Talha
683d082f-9c82-4813-9dff-ec01d8f068fa
Quddus, Ayesha
758fb85b-6e88-4898-a479-a842a00fc9ad
Papaphilippou, Philippos
3da942ad-4093-4b65-a6fc-7e90ca736abe
Israr, Talha
683d082f-9c82-4813-9dff-ec01d8f068fa
Quddus, Ayesha
758fb85b-6e88-4898-a479-a842a00fc9ad
Papaphilippou, Philippos
3da942ad-4093-4b65-a6fc-7e90ca736abe
Israr, Talha, Quddus, Ayesha and Papaphilippou, Philippos
(2026)
Exploration of high-performance branch predictors for softcore implementations.
In 36th International Conference on Field-Programmable Logic and Applications 2026.
IEEE.
9 pp
.
(In Press)
Record type:
Conference or Workshop Item
(Paper)
Abstract
Branch predictors play a key role in improving the performance of pipelined processors, yet their design choices remain relatively unexplored for FPGA-based softcores. Although modern predictors such as TAGE achieve a low misprediction rate in algorithmic implementations, their complexity introduces significant challenges on FPGAs due to limited resources and routing constraints. This creates a design trade-off between prediction accuracy and hardware efficiency which is not well-examined in the existing literature.
This work presents a comprehensive evaluation of static, dynamic, and correlation-based branch predictors, including TAGE and TAGE with a simplified statistical corrector (TAGE-SC), implemented within a five-stage RISC-V softcore. All predictors are evaluated on an FPGA platform in terms of resource utilization, operating frequency, and misprediction rate. Our TAGE and TAGE-SC designs yield an average misprediction rate improvement of 3.75% and 5.29%, respectively, over pTage, an optimized implementation of TAGE for FPGAs. This improvement comes at the cost of additional hardware. Our results also indicate that, while a 2-bit predictor has a higher misprediction rate, it provides a good balance between accuracy and hardware cost. Our design space also explores finer-grain implementation choices, like the impact of the size and number of TAGE tables. To enable further research, all code is released as open-source.
Text
branch_pred_fpl26_accept
- Accepted Manuscript
More information
Accepted/In Press date: 10 June 2026
Identifiers
Local EPrints ID: 513929
URI: https://eprints.soton.ac.uk/id/eprint/513929
PURE UUID: 0bdcb9fc-ce08-4113-a287-af08447f7974
Catalogue record
Date deposited: 02 Sep 2026 16:41
Last modified: 13 Sep 2026 15:51
Export record
Contributors
Author:
Talha Israr
Author:
Ayesha Quddus
Author:
Philippos Papaphilippou
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics