CGRA Accelerator Control for Loop-Specific PE Configuration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies face challenges in efficiently managing configuration information for coarse-grained reconfigurable array (CGRA) accelerators, leading to suboptimal performance due to inefficient identification and management of loop-specific configuration data.
Innovation Solution
A method and device that monitor instructions processed by a core processor, identify loops as acceleration targets, and utilize a table to retrieve or generate configuration information for connection status among processing elements (PEs) in the CGRA accelerator, optimizing computation and connectivity based on program counter and hash information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If configuration information is stored and managed efficiently, then processing speed and acceleration performance are improved, but device complexity increases due to the need for tables and management mechanisms
Solution Approach 1:
The system pre-generates and stores configuration information in tables before actual loop execution. When a loop is detected, the accelerator controller queries the pre-prepared configuration data, avoiding time-consuming generation during execution. This preliminary preparation resolves the contradiction by trading off initial setup complexity for significantly improved runtime processing speed.
Solution Approach 2:
The system creates simplified copies of configuration information in lookup tables that can be quickly queried. Instead of managing complex original configuration data structures during execution, the system uses streamlined table representations that enable fast retrieval while maintaining the essential configuration parameters needed for accelerator setup.
2Manufacturing precision
If configuration information is generated and stored for every loop, then acceleration accuracy is improved, but memory usage and loss of time increase
Solution Approach 1:
The system transforms configuration information into hash values as a parameter change. By converting detailed configuration data into compact hash representations for table lookup, the system maintains acceleration accuracy through precise matching while dramatically reducing memory access time and storage requirements.
Solution Approach 2:
The system implements a hybrid approach where configuration information is pre-generated and stored in tables for common cases, but full configuration generation occurs only when table lookups fail or special conditions arise. This partial pre-computation strategy achieves high acceleration accuracy for typical cases while minimizing time loss through selective full processing.
3Productivity
If configuration information is pre-generated and stored in tables, then productivity is improved, but device complexity and memory requirements increase
Solution Approach 1:
The system segments configuration management into distinct components: hash generation modules, table storage structures, lookup mechanisms, and fallback generation paths. This segmentation allows each component to be optimized independently and enables modular complexity management, improving productivity through streamlined operations while containing overall device complexity through structured organization.
Data Source
AI summary
A method and device with accelerator control are provided. The method includes monitoring instructions processed in a core processor, identifying a loop including at least one instruction as an acceleration target while monitoring, comparing at least one piece of information corresponding to the loop that is the acceleration target with information managed in a table, and based on a result of the comparison, identifying configuration information corresponding to the loop, the configuration information controlling connection status of connections among a plurality of process elements (PEs) included in an accelerator.