System and method for multi-mode branch predictor

Abstract

The invention relates to a system and a method for a multi-mode branch predictor, which are used for predicting the executions of branches capable of executing commands of a computer. In the embodiment, the branch predictor can comprises a program counting register used for storing a program counting value and a branch history register used for storing a branch history value. In addition, the branch predictor can also comprises a prediction Hash table with a plurality of predicted values, and each predicted value uniquely corresponds to a plurality of storage positions. By utilizing the assemblies, the branch predictor can generate a first predicted value corresponding to the program counting value and a second predicted value corresponding to the logical combination of the program counting value and the branch history value. By utilizing the two predicted values obtained from two different prediction modes, the branch predictor can be better suitable for generating an integral predicted value based on the first predicted value and the second predicted value, and the integral predicted value is more exact than a single predicted value based on a single prediction mode.

Description

technical field [0001] The present invention relates to systems and methods for multi-mode branch predictors. Background technique [0002] In computer architecture, superscalar computers (ie, computers with multiple processing components that can execute more than one instruction per clock cycle) can be used for advanced, computationally intensive applications. Typically, superscalar computers employ one or more pipelined processors, which can be used for low-level instruction "crunching" to better handle certain computationally intensive processes. At the level of computer programming, such a process may be implemented as a set of instructions that are part of a subroutine called a branch of instructions or, more simply, a branch. In this way, branches may or may not be executed, depending on the execution of various aspects of the overall program. To help with such computationally intensive branches, the branch's instructions are sometimes executed ahead of time before ...

AI Technical Summary

Problems solved by technology

That is, starting to execute a set of instructions in a branch that was not ultimately taken such that any pipeline processor that started executing a branch that was not ultimately taken may have to flush (i.e., delete) its computations, which is a significant cost to computation time and power. waste of

That is, such early execution of instructions in untaken branches generally does not provide useful results

Images