Dismiss
InnovationQ will be updated on Sunday, Oct. 22, from 10am ET - noon. You may experience brief service interruptions during that time.
Browse Prior Art Database

Method and System For Improved Branch Predictions

IP.com Disclosure Number: IPCOM000214467D
Publication Date: 2012-Jan-30
Document File: 2 page(s) / 44K

Publishing Venue

The IP.com Prior Art Database

Abstract

A method and a system for improved branch prediction is provided wherein an instruction fetching unit utilizes a return address stack. One or more predictions are evaluated for providing a confidence value. The confidence value is used to avoid mis-predictions.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 61% of the total text.

Page 01 of 2

Method and System For Improved Branch Predictions

A method and a system for improved branch prediction is provided. In a microprocessor operation branch prediction may be performed based on a call stack. A call stack may contain data related to return address of plurality of subroutines. A return address call stack may have a stack entry limit associated with number of return addresses stored in the return address stack. The system may determine a confidence level for branch prediction depending on one or more observed conditions related to the return address stack.

One or more observed condition may include one or more of an overflow condition and an underflow condition. The underflow condition may be determined based on one or more events such as, but not limited to, calls, returns, mis-predictions and speculative execution of the microprocessor. The system may assign a low confidence value a branch prediction when the address return stack underflows. A lowest confidence value may be assigned to a subsequent prediction when a mis-prediction is observed. In another embodiment all predictions made with a low confidence value may be considered as mis-predictions. The system may provide a confidence counter to determine the confidence of a branch prediction.

For example, consider a stack diagram as shown in FIG.1. With decreasing number of entries in the stack the confidence counter decreases the associated confidence value. Whenever a misprediction happens the co...