Browse Prior Art Database

Store Re-dispatch Queue

IP.com Disclosure Number: IPCOM000008417D
Publication Date: 2002-Jun-12
Document File: 2 page(s) / 27K

Publishing Venue

The IP.com Prior Art Database

Abstract

Disclosed is a method that improves the store dispatch mechanism by reducing re-dispatch latency.

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 57% of the total text.

Store Re-dispatch Queue

Disclosed is a method that improves the store dispatch mechanism by reducing re-dispatch latency.

Background

The store dispatch mechanism reads information from the store buffer in the MOB, then sends it to the DAC to allocate a request buffer and eventually update the L0 cache. The protocol between the MOB and the DAC dictates that when a store cannot get a request buffer (due to address or structural conflicts) it is cancelled and needs to be re-dispatched. The re-dispatch mechanism involves re-reading the store buffer and shipping to the DAC. This process repeats until the DAC provides a request buffer to the store and writes the L0.

When there is a store cancel, the re-dispatch latency takes two clocks to re-read and dispatch the store buffer. The re-dispatch queue aims to eliminate this latency associated with re-reading the store buffer. This reduces the overall latency of the store in the event the DAC cancels, and also reduces the latency of dependent loads that cannot forward since they are able to get the earlier data form the L0.

General Description

In the disclosed method, stores that are read out on initial dispatch are staged in the re-dispatch queue. The depth of the dispatch queue matches the depth of the pipeline (i.e. from when the store is dispatched to when the cancel information is determined). On a store cancel, the dispatch queue control logic determines whether the store gets re-dispatched from the re-dispatch queue or from the...