Data Store Entry Replacement Using Profile Data

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing data processing systems face challenges in reducing overhead while maintaining prediction accuracy, as large predictor structures are needed to store numerous entries, with many containing non-useful data, leading to inefficiencies and increased circuit area and power consumption.

Innovation Solution

A data processing system with an entry profile store that stores profile data for candidate entries, using performance gain data to preferentially store entries with high associated performance gains, and dynamically updates this data based on cycle counts to optimize storage usage, allowing for efficient use of limited resources.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If large predictor structures are used to store numerous entries, then prediction accuracy is improved, but circuit area and power consumption increase

Engineering Contradiction:

Improveprediction accuracyVSAvoidcircuit area

Core Design Contradiction:

Measurement precisionVSArea of stationary object

Solution Approach 1:

The patent divides the predictor into two separate structures: a small data store for actual prediction entries and a larger entry profile store for candidate entries. This segmentation allows the system to maintain high prediction accuracy through intelligent replacement policies while reducing the critical path length and circuit area by separating prediction functionality from candidate storage.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The entry profile store acts as an intermediary structure that holds candidate entries and provides profile data to guide replacement decisions in the data store. This intermediary mechanism enables the system to achieve performance comparable to full-size data stores while using significantly smaller structures for actual prediction.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Measurement precision

If large predictor structures are used to store numerous entries, then prediction accuracy is improved, but power consumption increases

Engineering Contradiction:

Improveprediction accuracyVSAvoidpower consumption

Core Design Contradiction:

Measurement precisionVSUse of energy by stationary object

Solution Approach 1:

By segmenting the predictor into a small data store and a larger entry profile store, the patent reduces the size of the actively used prediction structure. The data store remains small and fast, while the entry profile store handles candidate management asynchronously, thereby reducing power consumption compared to a single large predictor structure.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The system stores more candidate entries in the entry profile store than can be accommodated in the data store simultaneously. This partial/excessive action allows the system to maintain a small, low-power data store while using the entry profile store to manage additional candidates that are not actively needed, thus reducing overall power consumption.

Inventive Principle:

Principle #16Partial or excessive action

3Area of stationary object

If the number of entries in the data store is reduced to reduce overhead, then circuit area is reduced, but prediction accuracy degrades

Engineering Contradiction:

Improvecircuit areaVSAvoidprediction accuracy

Core Design Contradiction:

Area of stationary objectVSMeasurement precision

Solution Approach 1:

The patent implements dynamic replacement policies that selectively replace entries in the data store based on profile data from the entry profile store. This dynamic approach allows the system to maintain high prediction accuracy with a small data store by intelligently managing which entries are kept and which are replaced, rather than using a static fixed-size structure.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The entry profile store provides feedback in the form of profile data that guides replacement decisions in the data store. This feedback mechanism enables the system to make informed decisions about which entries to retain and which to evict, maintaining prediction accuracy despite the reduced size of the data store.

Inventive Principle:

Principle #23Feedback

4Area of stationary object

If smaller data stores are used to reduce overhead, then circuit area and power are reduced, but useful entries are evicted by non-useful entries

Engineering Contradiction:

Improvecircuit areaVSAvoidprediction effectiveness

Core Design Contradiction:

Area of stationary objectVSReliability

Solution Approach 1:

The entry profile store performs preliminary action by pre-evaluating and storing profile data for candidate entries before they are inserted into the data store. This preliminary sorting and ranking of candidates based on their usefulness allows the replacement mechanism to prioritize retaining useful entries and evicting non-useful ones, preventing the latter from displacing the former in the limited data store capacity.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS8271750B2Entry replacement within a data store using entry profile data and runtime performance gain data

Publication Date: 2012.09.18 ARM LTD

AI summary

A data processing system includes a data store having storage locations storing entries which can be used for a variety of purposes, such as operand value prediction, branch prediction, etc. An entry profile store stores profile data for more candidate entries than there are storage locations within the data store. The profile data is used to determine replacement policy for entries within the data store. The profile data can include hash values used to determine whether predictions associated with candidate entries were correct without having to store the full predictions within the profile data.