Implementation Of Branch Prediction Mechanism In X-Microprocessor And Research On The Branch Prediction Algorithm Based On Fuzzy Weight

Branch prediction technology is one of the most important field in the research of computer architecture and microprocessor .A high accuracy branch prediction method can largely reduce the stall of microprocessor and raise the efficiency of the microprocessor.Nowadays, the research on branch prediction focuses on dynamic branch prediction.Many new prediction algorithms have been raised based on the merge of different subject.The research on branch prediction includes the use of history information, branch instruction address mapping and state transition. The hardware spending and the complexity of implementation must be considered.This paper presents the research on the branch prediction mechanism of X microprocessor which includes the dynamic prediction , static prediction and the perfect loop exit.The design and implementation of X microprocessor's dynamic branch prediction mechanism has been presented.As X microprocessor is a multi-issue mechanism,a MBPT has been designed and the prediction ,lookup and update can be done.A PHT has been designed to complete branch prediction and the table update together with MBPT.MBPT has two read ports and two write ports .As the limitation of SRAM ,there will be logic and physical conflicts.This paper presents the research on conflict detection.After lots of simulation and synthesis done, a conclusion is drawn that the dynamic branch prediction mechanism achieves the expected target in correctness and timing delay.This paper presents the research and improvement on the common prediction algorithms .The selection method in hybrid branch prediction has been improved through introducing Hamming Distance.A best prediction result can be decided on the minimum value of Hamming Distance.The adaptive dynamic branch prediction mechanism has been improved and fuzzy weight mechanism has been introduced. Every bit of the BHR has a different weight and dynamically changed by a adjust gene. The prediction result is given through fuzzy consequence.The result of SimpleScalar simulation shows that the miss prediction rate of dynamic branch prediction algorithm based on fuzzy weight is 2% lower than the classical gshare prediction mechanism.