Non-Inclusive Cache Tag-Data Decoupling for Snoop Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing multi-level cache systems face challenges in managing data storage efficiently, particularly in non-inclusive caches, which suffer from reduced space usage and high overhead due to duplicate data, while inclusive caches compromise performance with snoop filtering.
Innovation Solution
Implementing a non-inclusive cache with a data pointer system that decouples tags from memory, allowing dynamic reassignment of data associations without copying, and supports snoop filtering through a Non-Inclusive Cache Inclusive Directory (NICID) architecture.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If an inclusive cache architecture is used, then snoop filtering is supported, but space usage increases due to duplicate data storage
Solution Approach 1:
The cache is divided into two separate structures: a tag array that stores address tags and a data array that stores actual data. The tag array is further segmented into valid bits and tag data, allowing independent management of metadata and payload. This segmentation enables the cache to track data presence without duplicating full data copies, reducing space overhead while maintaining snoop filtering capability.
Solution Approach 2:
A directory structure acts as an intermediary between the inclusive cache requirements and the non-inclusive storage implementation. The directory maintains information about data presence and validity, enabling snoop filtering operations without requiring actual data duplication. This intermediary layer decouples the filtering logic from data storage, allowing space-efficient implementation.
2Quantity of substance
If a non-inclusive cache is used, then space usage is reduced, but overhead increases due to data pointer management
Solution Approach 1:
The data pointer management is merged with the existing tag structure. Instead of separate pointer arrays, the valid bits and tag data in the tag array serve dual purposes: traditional cache tagging and data location tracking. This merging eliminates redundant storage structures and reduces overhead while maintaining non-inclusive cache benefits.
Solution Approach 2:
The tag array structure is designed to perform multiple functions simultaneously: address matching, validity tracking, and data location indexing. By making the tag structure universal, the patent eliminates the need for separate data pointer management structures, reducing overall system overhead while enabling non-inclusive cache operation.
3Adaptability or versatility
If data is decoupled from tags using data pointers, then dynamic reassignment is enabled, but access complexity increases
Solution Approach 1:
The cache implementation uses dynamically allocatable arrays for both tag and data storage, allowing the cache to adapt its structure based on workload requirements. The tag array and data array can be independently sized and reconfigured, enabling dynamic reassignment of data associations without fixed structural constraints. This dynamic approach maintains flexibility while keeping access paths straightforward through consistent indexing mechanisms.
Data Source
AI summary
Systems and methods are disclosed for data storage in a non-inclusive cache. For example, an integrated circuit may include a cache that includes a databank with multiple entries configured to store respective cache lines; and an array of cache tags, wherein each cache tag includes a data pointer that points to an entry in the databank. For example, methods may include allocating the entry in the databank to the cache including the array of cache tags from amongst multiple caches in the integrated circuit by writing the data pointer to the cache tag in the array of cache tags.