Graph Neural Network Cache Access via Degree Lookup Table
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing cache access methods for graph neural networks are inefficient, leading to increased training and inference times due to their discrete and random access processes, which hinder the optimization of cache planning in these systems.
Innovation Solution
A cache access method and system that utilize a cache with multiple regions, including a degree lookup table, to optimize access by determining the type of node address and performing lookups based on the node's degree, thereby reducing the average time for graph neural network processor memory access.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If general cache access method is used for graph neural network, then cache structure is simple, but access efficiency is low and training time increases significantly
Solution Approach 1:
The cache is divided into multiple regions including a degree lookup table region, edge information region, and attribute information region. This segmentation allows different types of graph data to be stored in optimized locations, enabling efficient access patterns for GNN operations while maintaining a manageable cache structure.
Solution Approach 2:
The degree lookup table pre-stores the degree information (number of edges) for each node in the graph. By having this information pre-computed and stored in the cache, the system avoids performing degree calculations during training, significantly reducing access time and improving training efficiency.
2Loss of time
If cache lookup is performed for all node addresses, then access completeness is high, but average access time increases
Solution Approach 1:
The system performs cache lookup selectively based on the address type. For degree information, the system checks the degree lookup table first (partial action) and only performs full cache lookup when necessary. This selective approach reduces average access time while maintaining access completeness for all required data.
Solution Approach 2:
The degree lookup table acts as an intermediary structure between the main cache and the GNN processor. It provides a quick reference for node degree information, allowing the system to determine access strategies without querying the entire cache, thus reducing average access time while preserving complete access capability when needed.
Data Source
AI summary
The present application discloses a cache access method and an associated graph neural network system. The graph neural network processor is used for performing computation upon a graph neural network. The graph neural network is stored in the memory in compressed sparse row format. The method includes: receiving an address corresponding to a node of the graph neural network and a type of the address; in response to the type is one of a first type or a second type, performing lookup by comparing the address with a tag field of a degree lookup table to at least obtain a degree of the node; determining whether the degree is greater than a predetermined value to obtain a determination result; and determining whether to perform lookup on a region of the cache corresponding to the type according to the determination result.